Glial control of sphingolipid levels sculpts diurnal remodeling of circadian circuits

Summary:

Structural plasticity in the brain often necessitates dramatic remodeling of neuronal processes, with attendant reorganization of the cytoskeleton and membranes. While cytoskeletal restructuring has been studied extensively, how lipids might orchestrate structural plasticity remains unclear. We show that specific glial cells in Drosophila produce Glucocerebrosidase (GBA) to locally catabolize sphingolipids. Sphingolipid accumulation drives lysosomal dysfunction, causing gba1b mutants to harbor protein aggregates that cycle across circadian time and are regulated by neural activity, the circadian clock, and sleep. While the vast majority of membrane lipids are stable across the day, a specific subset that is highly enriched in sphingolipids cycles daily in a gba1b-dependent fashion. Remarkably, both sphingolipid biosynthesis and degradation are required for the diurnal remodeling of circadian clock neurites, which grow and shrink across the day. Thus, dynamic sphingolipid regulation by glia enables diurnal circuit remodeling and proper circadian behavior.

eTOC

Vaughen et al. find that brain-enriched lipids called sphingolipids are degraded by glial-derived Glucocerebrosidase (GBA). Glial gba1b knockout in Drosophila causes selective sphingolipid and phospholipid alterations, cyclic protein aggregates, and sleep loss. Specific sphingolipids fluctuate diurnally in the brain, and sphingolipid metabolism instructs structural remodeling of a key circadian circuit.

Graphical Abstract

Introduction

Lifelong brain function requires coordinated biosynthetic and degradative pathways to precisely maintain neural membrane composition, circuit function, and animal behavior. While protein synthesis and degradation have been studied extensively in neurons, our understanding of lipid metabolism in the brain is comparatively limited. Brains are lipid-rich and display regional heterogeneity in lipid species (O’Brien and Sampson, 1965)(Fitzner et al., 2020), which include the CNS-enriched sphingolipids named after the enigmatic sphinx (Merrill, 2011). Moreover, neurons can dynamically reshape their membranes as part of structural and synaptic plasticity. For example, circadian pacemaker neurons in both vertebrates and Drosophila undergo dramatic daily cycles of membrane addition and removal (Becquet et al., 2008)(Fernández et al., 2008) (Herrero et al., 2017) (Krzeptowski et al., 2018)(Petsakou et al., 2015) (Song et al, 2021). In addition, proteins that control membrane lipid composition are associated with Parkinson’s Disease and Alzheimer’s Disease (Yadav and Tiwari, 2014)(Lin et al., 2019) (Futerman and van Meer, 2004), and circadian dysfunction is a common feature of neurodegeneration that can precede disease onset (Leng et al., 2019). However, surprisingly little is known about how specific lipid species are coupled to structural plasticity in neuronal membranes.

A prominent class of neural membrane lipids are the glycosphingolipids, which are built upon a core of Glucosylceramide (GlcCer) from the sphingolipid Ceramide (Cer). GlcCer is degraded by Glucocerebrosidase (GBA) (Figure 1A), and mutations in GBA are linked to sleep disruption and neurodegeneration. Complete loss of GBA activity invariably causes neuropathic Gaucher disease, and GBA carriers are ~3–5 times more likely to develop Parkinson’s disease (Sidransky and Lopez, 2012) characterized by accelerated cognitive decline (Sidransky et al., 2009) (Liu et al., 2016). Mutations in GBA are also associated with isolated REM sleep behavior disorder (Gan-Or et al., 2015)(Krohn et al., 2020). However, the cellular mechanisms underpinning how mutations in GBA cause brain defects are unclear.

Figure 1. Gba1b is required in glia for neuronal lysosome function.

(A) The Drosophila brain consists of neuronal and glial cell bodies arranged in a cortical rind (light green) enclosing dense synaptic neuropils (dark green). Within the Optic Lobe chiasm (OL-iCH), neuronal processes (dark green) are tightly associated with glial cells (yellow). Lysosomes (LysoTracker, red) are localized to cortical cell bodies (see Figure S1), while Ubiquitin (Ubi) aggregates are inside glia in gba1b mutant optic lobes (see Figure S2). Inside lysosomes, Gba1b degrades the lipid GlcCer into Cer (B-E″). LysoTracker labeling in brains from 2 day-old flies. (G-J″) Ubiquitin aggregates in the OL-iCh in 15–20 day-old flies. (B, G) Control. (B′, G′) gba1b^Δ1/+ heterozygotes. (B″, G″) gba1b^Δ transheterozygotes have enlarged lysosomes and accumulate Ubi aggregates in the OL-iCh (arrows). (C-C″, H-H″) RNA-interference, gba1b^RNAi. Control (C, H), neural (C′, H′; nSyb-GAL4), and glial (C″, H″; repo-GAL4). (D-D″, I-I″) Somatic CRISPR, gba1b^cKO. Control (D, I), neural (D′, I′; nSyb-GAL4), and glial (D″, I″; repo-GAL4). Both gba1b^RNAi and gba1b^cKO in glial cells but not neurons cause enlarged Lysosomes and Ubi aggregates. (E-E″, J-J″) Null gba1b^Δ (E, J) can be rescued by nSyb::Gba1b (E′, J′; neurons) or by overexpressing UAS-Gba1b^LAMP in glia (E″, J″; repo-GAL4). (F) Quantification of LysoTracker data. (K) Quantification of Ubi data. *p<0.05, ****p<0.0001, ANOVA, Tukey’s multiple comparisons for normally distributed data, Kruskal-Wallis test for nonparametric data (E-E″, G-J). n>20 optic lobes. Data are represented as mean ± SEM. Scale bar: 25μm. See also Figure S1 and Figure S2, and Table S1 for genotypes.

Drosophila is a powerful model for understanding GBA function: mutations in fly gba1b have evolutionarily conserved phenotypes, including lysosomal impairment, insoluble ubiquitin (Ubi) protein aggregates, and reduced sleep (Davis et al., 2016)(Kinghorn et al., 2016)(Kawasaki et al., 2017). Moreover, core sphingolipid regulatory enzymes (centered around Cer) are conserved in Drosophila, and mutations in these genes can cause neurodegeneration (Acharya and Acharya, 2005) (Acharya et al., 2003). Prior studies identified a role for Gba1b in circulating extracellular vesicles (Jewett et al., 2021) as well as functions for arthropod glycosphingolipids in nervous system development (Haines and Irvine, 2005)(Huang et al., 2018)(Huang et al., 2016)(Soller et al., 2006)(Chen et al., 2007). However, the cellular requirements for the conserved breakdown of GlcCer by GBA in the adult brain remain unknown, and how sphingolipids sculpt neural circuitry and behavior remains mysterious in vivo.

We discover that Drosophila glia produce Gba1b to control lysosome homeostasis and sphingolipid degradation in adult neurons. Removal of gba1b from specific glial types triggered Ubi aggregate formation in both neurons and glia. Glia in both flies and humans have broadly conserved functions in neurotransmitter recycling, metabolism, and membrane phagocytosis (Freeman, 2015)(Yildirim et al., 2019). We identify neuropil ensheathing glia (EG), a glial type that is closely associated with neurites, as well as Perineural glia (PNG) as key mediators of sphingolipid catabolism. Remarkably, Ubi aggregates in young gba1b^Δ mutants shrink during the day and grow larger at night. This cycle of aggregate burden was modulated by the circadian clock and neural activity, as dark-rearing suppressed aggregate formation in the visual system while disrupting the circadian clock blocked aggregate cycling across the brain. Comparative lipidomics studies at dusk and in the night revealed that while the levels of most membrane phospholipids were stable across time, specific sphingolipid species fluctuated diurnally. Moreover, the cycles of neurite growth and retraction in the sLNv circadian pacemaker neurons strongly depended on sphingolipid regulation. Mutations in gba1b, or removal of gba1b from glia, blocked this remodeling and prevented neurite growth. Conversely, inhibiting sphingolipid biosynthesis in sLNv cells abolished neurite retraction. Strikingly, reprogramming the temporal pattern of Gba1b by imposing elevated expression of Gba1b at night in sLNv neurons was sufficient to invert the normal cycle of neurite remodeling, with increased terminal volume at dusk instead of dawn. Thus, glia degradation and neuronal biosynthesis of specific sphingolipids is both necessary and sufficient to sculpt structural plasticity in the brain.

Results

Gba1b is required in glia for neuronal lysosome function

To better characterize brain sphingolipid catabolism and Gba1b function, we generated an early frameshift allele, gba1b^Δ1, which we used in trans to gba1b^ΔTT, a characterized null allele that also deletes an adjacent gene, Qsox4 (Figure S1A, Table S1, (Davis et al., 2016)). We used transheterozygous combinations of these alleles to produce viable adults designated gba1b^Δ (which retain one functional copy of Qsox4 and do not homozygose other chromosomal regions). Consistent with previous work (Kinghorn et al., 2016), we confirmed that gba1b^Δ brains harbor enlarged degradative lysosomes marked by an acidic-compartment label (LysoTracker), active proteases (Cathepsin B), and a lysosomal protein (Lamp) (Figure 1B–B″, Figure S1E–J). Aberrant lysosomes distributed across the cortex of the brain and were enclosed by neural but not glial membranes (Figure S1K–L). We exploited this robust neuronal phenotype to identify where Gba1b is required for lysosome maintenance. Unexpectedly, gba1b knockdown (RNAi) or knockout (somatic CRISPR, gba1b^cKO) in all neurons did not cause any discernible lysosomal phenotypes (Figure 1C′, D′). In contrast, pan-glial gba1b knockdown or knockout triggered lysosomal hypertrophy (Figure 1C″, D″). These data argue that neuronal lysosome function depends on Gba1b expression in glia.

We next attempted to rescue the gba1b LysoTracker phenotype using three Gba1b over-expression constructs (Table S2, Figure S1A): a wild-type Gba1b transgene, a catalytically-inactive transgene (Gba1b^E340K ), and a tethered Gba1b^Lamp transgene designed to restrict Gba1b to the lysosome (Mikulka et al., 2020). Glial overexpression of Gba1b^Lamp or wild-type Gba1b rescued the LysoTracker phenotype (Figure 1E″, Figure S1U). Importantly, expression of Gba1b in glia rescued the ectopic LysoTracker staining in neurons of gba1b^Δ mutants (Figure S1M–P), demonstrating that glial Gba1b is nonautonomously required to maintain neuronal lysosomes. In contrast, overexpressing active Gba1b in neurons caused lysosomal hypertrophy in wild-type animals (Figure S1Q–R). Thus, excess Gba1b activity in neurons is detrimental to lysosome homeostasis. To temper Gba1b induction, we placed Gba1b under the direct control of a neuronal enhancer (nSyb::Gba1b; Figure S1B–D). Interestingly, nSyb::Gba1b fully rescued the lysosomal phenotypes of gba1b^Δ mutants (Figure 1E′). Thus, while Gba1b is selectively produced by glia and sufficient to rescue lysosomes in gba1b^Δ mutants, neurons can also use Gba1b.

We next tested if Gba1b was required in glia for other gba1b^Δ phenotypes. A hallmark of gba1b^Δ brains is progressive accumulation of polyubiquitinated proteins (Kinghorn et al., 2016)(Davis et al., 2016). We found that older gba1b^Δ mutants accumulated Ubiquitin (Ubi) aggregates in both the inner optic lobe chiasm (OL-iCh) and in the cortex of the Mushroom Body calyx (MB-ca) (Figure 1G, Figure S2A–J), as well as the blood-brain barrier (data not shown). Ubi extensively colocalized with the Ubi-lysosomal adaptor protein p62, a common component of protein aggregates in many neurodegeneration models (Bartlett et al., 2011) (Figure S2E–F). By expressing p62-GFP within neurons or glia in gba1b^Δ brains, we determined that MB-ca puncta were neural, whereas larger Ubi aggregates were glial, including OL-iCh aggregates (Figure S2L, S2M). Similar to the LysoTracker phenotype, glial knockdown or knockout of gba1b induced Ubi aggregates, while neuronal perturbations did not (Figure 1H 1I, Figure S2J). Conversely, glial expression of Gba1b^Lamp in gba1b^Δ mutants fully rescued both Ubi aggregate populations (Figure 1J and Figure S1V, S1W). nSyb::Gba1b also rescued neuronal MB-ca aggregates (Figure S1W) and surprisingly also rescued glial OL-iCh aggregates (Figure 1J′) but not blood-brain barrier aggregates (data not shown). Taken together, Gba1b expression is required in glia to prevent lysosome dysfunction in neurons and subsequent Ubi aggregate appearance in both neurons and glia.

Specific glial subtypes are necessary and sufficient for Gba1b function in the brain

To identify which cells express Gba1b, we queried single-cell RNA-sequencing datasets and found that Gba1b transcripts were produced by multiple glial clusters during pupal development (Kurmangaliyev et al., 2020) (Figure S3A–A′). An adult dataset also revealed low levels of Gba1b expression in glia (Davie et al., 2018). To identify which glia cells express Gba1b, we characterized Gba1b^GAL4, a gene trap in the Gba1b locus (Lee et al., 2018). Gba1b^GAL4 drove expression of a reporter in glia but never in neurons (Figure S3B–C). Combining Gba1b^GAL4 with a panel of LexA-based markers of individual glial types (Pfeiffer et al., 2008)(Kremer et al., 2017) revealed expression of Gba1b in multiple glial lineages, but not astrocyte-like glia (Figure S3D–K). We next removed Gba1b from individual glial subtypes using both RNAi and somatic CRISPR but found no LysoTracker phenotypes (Figure 2A, Figure S3L–M). We reasoned that Gba1b may be redundantly required and screened combinations of glial cell gba1b knockouts. We found that removing Gba1b from both Ensheathing (EG) and perineural glia (PNG) caused lysosomal enlargement (Figure 2A″) and Ubi aggregate accumulation (Figure 2B–C). Conversely, expression of wild-type Gba1b, but not catalytically inactive Gba1b^E340K, in EG was sufficient to rescue the lysosomal and Ubi aggregate phenotypes in gba1b^Δ mutants (Figure 2D–F). However, expression of wild-type Gba1b in PNG failed to rescue these phenotypes (Figure 2D‴–F‴), and expression of Gba1b in astrocytes (which extensively tile the synaptic neuropil) failed to rescue LysoTracker (Figure 2G′). Importantly, EG, PNG, and astrocyte drivers expressed high levels of Gba1b protein in these rescue experiments (Figure S3N–P). Consistent with a critical role for Gba1b in EG, labeling EG membranes in gba1b^Δ mutants revealed that certain Ubi aggregates colocalized with EG membranes (Figure 2H–I). Thus, Gba1b is required in EG and PNG, and expression in EG is sufficient to rescue neuropil and cortex phenotypes.

Figure 2. Specific glial subtypes are necessary and sufficient for Gba1b function.

(A-C″) gba1b somatic CRISPR (gba1b^cKO) in ensheathing glia (EG), perineural glia (PNG), or both EG and PNG. (D-F‴) Rescue of gba1b^Δ by overexpressing Gba1b in EG and PNG. (A-A″, D-D‴) LysoTracker labeling in 2 day-old brains. (B-F‴) Ubi labeling in 20–25 day-old brains. (A, B, C) gba1b^cKO in EG. (A′, B′, C′) gba1b^cKO in PNG. (A″, B″, C″) gba1b^cKO in both EG and PNG. (D′, E′, F′) gba1b^Δ rescue by wild-type UAS-Gba1b expression in EG, but not by enzyme-dead UAS-Gba1b^E340K (D″, E″, F″). (D‴, E‴, F‴) PNG expression of wild-type UAS-Gba1b failed to rescue gba1b^Δ. (G) Quantification of LysoTracker, including the control ISOD1/gba1b^cKO (green), as well as Ubi. (G′) Quantification of rescue data (including data from EG>Gba1b^LAMP (green) and Astrocytes overexpressing Gba1b (yellow) for LysoTracker). (H-H″‴) EG membranes (green) and subsets of Ubi (magenta) in gba1b^Δ colocalize in the MB-ca (arrowheads). (I-I″) EG membranes (green) enclose Ubi (magenta) in gba1b^Δ OL-iCh. (I‴I″‴) Subsets of OL-iCh Ubi structures accumulate bright EG membrane in gba1b^Δ (arrowheads). *p<0.05, ****p<0.0001, ANOVA, Tukey’s multiple comparisons for normally distributed data, Kruskal-Wallis test for nonparametric data (B-C″, E-F‴). n>6 brains. Scale bar: 25μm (A-F); 10μm (H-H″‴, I‴-I″‴), 15μm (I-I″). Data are represented as mean ± SEM. See also Figure S3, and Table S1 for genotypes.

Ubiquitin aggregates cyclically grow and shrink at younger ages

We next examined the onset of Ubi aggregate deposition in young gba1b^Δ flies and found an unanticipated relationship between aggregate burden and circadian time. Specifically, Ubi aggregates in the OL-iCh and the MB-ca were low or absent during the day but accumulated at night in 7 day-old gba1b^Δ mutants (Figure 3A–F). We more closely examined the relationship between aggregates, age, and circadian zeitgeber time (ZT). For both gba1b^Δ and pan-glial gba1b^cKO animals, the nadir of aggregate burden occurred at ZT6 (afternoon), and aggregates grew steadily through ZT12 (evening) and ZT18 (night) in the OL-iCh (Figure 3G–L). Similarly, Ubi aggregates in the MB-ca were smaller during the day and larger at night in young flies (Figure S4A–D). However, in older flies (>15 days, the age used for prior experiments), the diurnal difference in aggregate size was less pronounced (Figure 3H, K and Figure S4C), arguing that diurnal aggregate clearance diminishes with age. In contrast to Ubi aggregates, lysosome markers were not as dramatically modulated by the circadian clock in gba1b^Δ null animals (Figure S4E–G). Subtle changes in control lysosome morphology were detected across time in controls, consistent with circadian autophagosome production (Bedont et al., 2021)(Ryzhikov et al., 2019) (Ulgherait et al., 2021) (Figure S4H–J). Thus, Ubi aggregates undergo significant variations in size in gba1b^Δ mutants across circadian time.

Figure 3. Ubiquitin aggregates cyclically grow and shrink at younger ages.

Ubiquitin labeling (Ubi, white) in controls (A, A′, D, D′) and gba1b^Δ (B, B′, E, E′) in the OL-iCh and Mushroom Body calyx (MB-ca) at zeitgeber time ZT2 (day) and ZT14 (night). (C) Quantification of A-B′. (F) Quantification of D-E′. (G-H) Ol-iCh Ubi in gba1b^Δ and (I-J) glial gba1b^cKO examined at ZT0, ZT6, and ZT12 at younger (G, J) and older (H, K) ages. (I) Quantification of Ubi in gba1b^Δ across days and ZT. (L) Quantification of Ubi in glial gba1b^cKO across days and ZT. Younger gba1b^Δ and glial gba1b^cKO brains accumulate Ubi at night and have a midday nadir (ZT6), whereas older gba1b flies plateau aggregates at an elevated level. *p<0.05, ****p<0.0001, Kruskal-Wallis test for nonparametric data, multiple comparisons. n>15 brains. Scale bar: 25μm. Data are represented as mean ± SEM. See also Figure S4, and Table S1 for genotypes.

Ubiquitin aggregate size is controlled by neural activity and the circadian clock

We hypothesized that the aggregate cycle apparent in gba1b^Δ flies could be directly controlled by the circadian clock, light-evoked changes in neural activity, or both. We tested these possibilities by depriving flies of light and by mutating period (per), a core component of the circadian clock (Figure 4A). Dark-rearing gba1b^Δ flies (DD) dramatically suppressed the number of Ubi aggregates in the OL-iCh, even in 20 day-old flies (Figure 4B, D). In contrast, Ubi aggregates in the MB-ca persisted in DD conditions (Figure 4C, E). This suggests that Ubi aggregates in the OL-iCh are sensitive to light and subsequent neural activity. In contrast, Ubi aggregates in the MB-ca are not responsive to light deprivation, perhaps because light does not strongly modulate neural activity in this region.

Figure 4. Ubiquitin aggregate size is controlled by neural activity and the circadian clock.

(A) Hypothesis that Ubi in young gba1b^Δ flies could be controlled by light and/or the circadian clock. (B-E) Dark-rearing reduced Ubi aggregate formation in Ol-iCh (B) but not MB-ca (C). (B,C) Control. (B′,C′) gba1b^Δ raised under light-dark conditions (LD). (B″, C″) gba1b^Δ raised under dark-dark conditions (DD). (D) Quantification of Ubi in OL-iCh. (E) Quantification of Ubi in MB-ca. Dark-rearing suppressed Ubi aggregate formation in OL-iCh but not MB-ca in gba1b^Δ. (F) Ubi labeling (white) in gba1b^Δ at three different circadian times (CT0, CT6, CT12) in MB-ca after shifting entrained flies to DD conditions. (G) Quantification of MB-ca Ubi burden reveals nadir at CT6. (H-I) Ubi labeling (white) in per⁰¹; gba1b^Δ double mutants (I) at ZT6 and ZT12 compared to gba1b^Δ single mutants. We observed that the midday nadir of Ubi aggregate formation was lost in per⁰¹; gba1b^Δ double mutant animals compared to controls (quantified in J). n>15 brains. *p<0.05, ****p<0.0001, ANOVA, Tukey’s multiple comparisons for normally distributed data, and Kruskal-Wallis test for nonparametric data (B-E). Scale bar: 25μm. Data are represented as mean ± SEM. See also Figure S4, and Table S1 for genotypes.

To test whether aggregate formation is under direct control of the endogenous circadian clock without the confound of light sensitivity, we focused on aggregate formation in the MB-ca. We entrained flies in LD, shifted to constant darkness for one day, and still detected lower aggregate burden at circadian time CT6 (Figure 4F–G). Moreover, introducing the arrhythmic per⁰¹ allele (Konopka and Benzer, 1971) into the gba1b^Δ background abolished cyclic changes in Ubi aggregate burden (Figure 4H–J). While mutating the circadian clock alone did not cause Ubi aggregates to accumulate in the OL-iCh (Figure S4L), we found enhanced Ubi burden in PNG and OL-iCh glia in per⁰¹; gba1b^Δ flies (Figure S4K). Consistent with brain activity modulating aggregate deposition, sleep-depriving gba1b^Δ flies at night plateaued Ubi cycling in both MB-Ca and OL-iCh (Figure S4M–N). Moreover, short-sleeping mutants sleepless and insomniac (Koh et al., 2008)(Stavropoulos and Young, 2011) also progressively accumulated Ubi aggregates in the MB-ca (Figure S4O–P), further suggesting that altered neural activity downstream of sleep disruption influences proteostasis (either directly or indirectly). In sum, gba1b^Δ neuronal aggregate size in the MB-ca is controlled by the circadian clock, and glial aggregate size in the OL-iCh is controlled by both neural activity and the circadian clock.

GBA regulates specific sphingolipids and phospholipids

As Gba1b degrades the lipid GlcCer, we tested if brain lipid levels were altered in gba1b manipulations. We first examined an antibody directed against GlcCer alongside neutral lipids using the lipophilic Nile Red stain. Both GlcCer and Nile Red were increased in gba1b^Δ brains compared to control animals (Figure 5A, B), and GlcCer was increased in pan-glial and EG + PNG gba1b^cKO brains (Fig S5A). Ectopic Nile Red-positive lipids in gba1b^Δ colocalized with LysoTracker (Figure 5C) and also colocalized with GlcCer (Figure 5D). Thus, GlcCer accumulation in gba1b^Δ mutants localizes to engorged, lipid-filled lysosomes.

Figure 5. Gba1b regulates specific sphingolipids and phospholipids.

(A) Staining against GlcCer in OL cortex in control (A) and gba1b^Δ (A′) at ZT12 in 5 day-old brains. (A″) Quantification of GlcCer area. gba1b^Δ mutants have increased GlcCer. (B) Staining against neutral lipids with Nile Red in control (B) and gba1b^Δ (B′) at ZT12 in 5 day-old brains. (B″) Quantification of Nile Red. (C) Co-labeling gba1b^Δ mutants with Nile Red (magenta) and LysoTracker (green) in the OL. LysoTracker particles are Nile Red positive (inset). (D) Fixed brains co-stained for Nile Red (magenta) and GlcCer (green) show GlcCer encircling Nile Red puncta (arrowheads and inset). (E) Lipidomic analysis of fly brains from control and gba1b^Δ animals, dissected in 10 day-old flies at two timepoints, ZT10–12 (dusk, peak of evening activity) and ZT18–20 (midnight, sleeping). Sphingolipids and phospholipids were extracted from two independent experiments and analyzed by liquid chromatography mass spectroscopy (LC-MS). Relative % fraction estimates of analyzed lipids are shown in the bar plot (bottom). (E-E″) Principal Component Analysis (PCA) of sphingolipids z-scored by abundance per sphingolipid species and colored by genotype (green = control, magenta = gba1b^Δ). (E′) GlcCer biplot of PC1 and PC2 separates samples by genotype. (E″) Cer separated gba1b^Δ samples from controls. (E‴) CerPE separated gba1b^Δ samples from controls but more weakly. (F-F″) Total levels of sphingolipid classes (ng/brain) revealed large GlcCer and Cer accumulation (but not CerPE) (yellow bar = ZT10–12, dark bar = ZT16–18). (G-H) Prominent species of GlcCer (G) and Cer (H) accumulated in gba1b mutants. (I) CerPE species modestly decreased in gba1b^Δ mutants. (J-J‴) PCA of phospholipids, z-scored by abundance per species and colored by genotype (green = control, magenta = gba1b^Δ). (J) PI separated samples by genotype, while PE (J′) modestly did so. PC (J″) and PS (J‴) did not separate gba1b^Δ samples from controls. (K) Bidirectional alterations of PI (L) and PE (K) species by gba1b^Δ. (L) *p<0.05, ****p<0.0001, ANOVA, Tukey’s multiple comparisons for normally distributed data. n>10 (A-D) and n=15 brains/point, 8 biological replicates (E-I). Data are represented as mean ± SEM. See also Table S1 for genotypes, and Table S3 for raw lipidomics data.

We next conducted lipidomics of gba1b^Δ and control brains using liquid chromatography/mass spectrometry to detect sphingolipids (GlcCer, Cer and Ceramide-phosphatidylethanolamine (CerPE), the Drosophila sphingomyelin) and phospholipids (Phosphatidylinositol (PI), Phosphatidylethanolamine (PE), Phosphatidylcholine (PC), and Phosphatidylserine (PS)). Given that we detected circadian aggregates and that lipids can vary across circadian time in non-neuronal cells(Katewa et al., 2016)(Gogna et al., 2015) (Rhoades et al., 2018) and head extracts(Schäbler et al., 2020), we dissected brains from two timepoints, ZT10–12 (dusk peak of evening activity) and ZT16–18 (midnight). Sphingolipids and phospholipids are classified by hydrocarbon chain length and the number of carbon-carbon double bonds for each chain. We detected 163 sphingolipid and phospholipid species (Table S3) which we divided into classes (GlcCer, Cer, CerPE, PI, PS, PC, and PE) and then analyzed by Principal Component Analysis (PCA).

PCA revealed large differences in GlcCer and Cer between control and gba1b^Δ mutant animals (73% of variance in GlcCer, 42% of variance in Cer; Figure 5E–E′). CerPE also modestly differentiated samples by genotype (20% of variance, Figure 5E‴). Consistent with the role of GBA in sphingolipid breakdown, total GlcCer and Cer levels were markedly increased in gba1b^Δ mutant brains as well as pan-glial gba1b^cKO brains (Figure 5F–F″, Figure S5B). These changes were largely driven by three GlcCer and Cer species with identical hydrocarbon chains: 14:1/18:0, 14:1/20:0, and 14:1/22:0 (Figure 5G–H), representing ~80% of total GlcCer and ~70% of total Cer in gba1b^Δ mutants. In contrast, CerPE was modestly decreased in a handful of species in gba1b^Δ mutants (Figure 5I).

We next analyzed phospholipids (Figure 5J). Although bulk phospholipid levels were unchanged in gba1b^Δ mutants (Figure S5C), PI and PE separated samples by genotype (Figure 5J–J″). PC and PS did not differentiate samples by genotype in any Principal Components (Figure 5J″–J‴; data not shown). Given that bulk PE and PI levels were unchanged, yet PCA separated PE and PI by genotype, we reasoned that specific species of PE or PI might reciprocally change in gba1b^Δ mutants. Indeed, in gba1b^Δ brains, 18:1/18:1 and 18:1/18:2 PI and PE species were increased, whereas 16:0/18:3 PE and PI were decreased (Figure 5K–L). In total, gba1b^Δ mutants significantly modulated 27 GlcCer, 20 Cer, 11 PI, 6 CerPE, 4 PE, 3 PC, and 3 PS species (Table S4). Notably, the unchanged PS, PC, and PE species account for >65% of total brain lipids (Guan et al., 2013)). Thus, gba1b exerts specific effects on subsets of brain lipids, predominantly GlcCer and Cer species, with weaker effects on CerPE, PI, and PE lipids.

Sphingolipid levels fluctuate diurnally

We next explored whether any lipids changed across time in control animals by PCA (Figure 6A). Strikingly, the first two principal components for GlcCer and CerPE species strongly separated samples taken during at dusk and at midnight, as did Cer, albeit more weakly (Figure 6B). In contrast, time of day accounted for little variance in any of the phospholipids (Figure 6C; data not shown). To directly reveal which lipid species were time-regulated, we constructed mean matrices from dusk and night samples and identified the largest differences between these vectors (FDR<0.05; see STAR methods). This identified ten species with circadian differences in control animals, eight of which were sphingolipids (Figure 6D–O). All ten species displayed reduced diurnal modulation in gba1b^Δ mutants (Figure 6D–O). Crucially, the eight modulated sphingolipids represented 40% of GlcCer, 10% of Cer, and 6.5% of CerPE species (Figure 6E–E″). Intriguingly, two of the GlcCer and Cer species have the same fatty acyl carbon chain compositions (14:1/18:1 and 14:1/20:1), suggesting common functions and/or coupled biosynthesis. Finally, we note that time-modulated species were not those most accumulated in gba1b^Δ mutants. For example, monounsaturated GlcCer 14:1/18:0 strongly accumulated in gba1b^Δ mutants (Figure 5G), whereas time-modulated GlcCer 14:1/18:1 was not elevated in mutants (Figure 6F) and instead failed to fluctuate, as in controls (Figure 6F). Thus, there is substantial diurnal fluctuation in specific sphingolipids that is dependent on Gba1b, but not marked accumulation of these circadian species, suggesting that time-modulated sphingolipids are converted from GlcCer to higher-order glycosphingolipids.

Figure 6: Sphingolipid levels fluctuate diurnally.

(A) Principal Component Analysis (PCA) of sphingolipids and phospholipids z-scored by abundance per species in 10 day-old controls from two independent experiments and 8 total biological replicates, colored by time (blue = midnight, ZT16–18; orange = dusk, ZT10–12). GlcCer (B). Cer (B′). CerPE (B″). GlcCer and CerPE strongly split samples by time. (C-C‴) Phospholipids. (C) PI. (C′) PE. (C″) PS. (C‴) PC. Phospholipids did not show strong circadian modulation. (D) Time-modulated species (see STAR methods) plotted for significance in control (y-axis) versus gba1b^Δ samples (x-axis) (sphingolipids = circles, phospholipids = triangles, dusk = orange, midnight = blue). (E) Diurnally modulated species contribution to lipid class totals in controls. (E′-E″) Data from (E) split into ‘dusk-enriched’ and ‘night-enriched’. (F-O) Normalized values of time-modulated species in controls (green) and gba1b^Δ (magenta) for the sample timepoints dusk (yellow bar) and night (black bar); gba1b^Δ blunted diurnal fluctuation. *p<0.05, ****p<0.0001, ANOVA, Tukey’s multiple comparisons for normally distributed data. n=15 brains/point, 8 biological replicates. Data are represented as mean ± SEM. See also Table S1 for genotypes, and Table S3 for raw lipidomics data.

Sphingolipid catabolism is required for adult sleep behavior

As gba1b^Δ mutants have altered lipid profiles with blunted circadian fluctuations in lipid species, we reasoned that gba1b^Δ mutants might display defects in circadian behaviors including activity and sleep. Control flies are crepuscular, with two bouts of activity surrounding dawn and dusk (Figure 7A, (Helfrich-Förster, 2000)). Like controls, gba1b^Δ mutants began sleeping shortly after lights-off, but both sexes displayed diminished sleep characterized by shorter sleep bout duration (Figure 7A, B, Figure S6A, B). Pan-glial gba1b knockouts also displayed sleep loss (Figure 7C, Figure S6C). Moreover, removing gba1b specifically in both EG and PNG also reduced sleep (Figure 7D, Figure S6D). Depleting gba1b only in EG did not cause obvious sleep defects, while depleting gba1b in PNG only reduced daytime sleep (Figure S6I), consistent with barrier glia regulating sleep (Artiushin et al., 2018)(Kozlov et al., 2020). Importantly, expressing Gba1b in both EG and PNG rescued the sleep deficits seen in gba1b^Δ mutants (Figure 7E, Figure S6E), whereas individual PNG or EG drivers had only modest effects (Figure S6J).

Figure 7. Sphingolipid catabolism is required for adult sleep behavior.

Drosophila activity-monitor (DAM) assays for measuring sleep. Adult flies were analyzed from day 5 to day 10 on a 12-hour Light-Dark (LD) cycle. (A-A″) Fraction of flies asleep across ZT for male gba1b^Δ flies (magenta) versus control males (green). (B-B″) Fraction of flies asleep across ZT for female gba1b^Δ flies (magenta) versus control females (green). (C-C″) Fraction of flies asleep across ZT for glial gba1b^cKO males (magenta) versus control males (green). (D-D″) Fraction of flies asleep across ZT for EG&PNG>gba1b^cKO males (magenta) versus control males (green). (E-E″) Fraction of flies asleep across ZT for male gba1b^Δ flies (magenta) versus rescue (EG&PNG>Gba1b) males (blue); control is overlaid (green, from A). (A, B, C, D, E) Fraction sleeping. (A′, B′, C′, D′, E′) Sleep per day. (A″, B″, C″, D″, E″) Total Sleep. Reducing Gba1b activity via either gba1b^Δ, or repo-GAL4>gba1b^cKO, or EG&PNG > gba1b^cKO caused reduced sleep across all days and times. EG&PNG expression of Gba1b was sufficient to partially rescue gba1b^Δ. *p<0.05, ****p<0.0001, ANOVA, Tukey’s multiple comparisons or Welch’s ANOVA depending on equality of data variance. For nonparametric sleep bouts, Mann-Whitney U test was used. n>28 flies. Data are represented as mean ± SEM, as well as the IQR for the Total Sleep boxplots. See also Figure S6, and Table S1 for genotypes.

We next tested for persistence of circadian rhythms when gba1b^Δ mutants were moved to continuous darkness (DD) after entrainment in LD. Although we observed timely evening and morning peaks of activity (Figure S6F–G), arguing against complete abrogation of the circadian clock, gba1b^Δ mutants and glial gba1b removal (repo or EG+PNG) had significantly weaker rhythmicity when free-running in darkness compared to controls (Table S5). Thus, loss of Gba1b caused both sleep and circadian defects, and expression of Gba1b in EG and PNG is necessary for proper sleep and circadian behaviors.

Sphingolipids drive cyclic remodeling of sLNv neurites

Given the importance of Gba1b function for diurnal modulation of lipid composition, we hypothesized that gba1b^Δ mutants might display defects in diurnal remodeling of circadian circuits. sLNv neurons express pigment-dispersing factor (PDF) and undergo dramatic daily remodeling of neurites and terminals (Park et al., 2000) (Fernández et al., 2008) (Herrero et al., 2017) (Herrero et al., 2020). In control animals, sLNv neurites are enlarged at dawn and smaller at dusk(Figure 8A). However, in gba1b^Δ mutants, sLNv neurites were trapped at an intermediate size with reduced sLNv volume and 3-D spread (Figure 8B, Figure S7A–B, Figure 8J). Critically, knocking-out gba1b in EG and PNG also trapped sLNv neurites at a constant reduced size (Figure 8C, Figure S7C). We next examined cell-type specific rescue of sLNv remodeling in gba1b^Δ mutants. nSyb::Gba1b, which expresses Gba1b pan-neuronally, rescued sLNv remodeling in gba1b^Δ mutants (Figure 8D). While expressing Gba1b in EG or PNG alone failed to rescue cyclic remodeling of sLNv terminals in gba1b^Δ mutants (Figure S7D–E), expressing Gba1b in both EG and PNG significantly rescued axonal volume (Figure 8E) but not 3-D spread (Figure 8J–J′). This requirement for Gba1b in both EG and PNG is consistent with the apposition of sLNv neurites to both glial subtypes (data not shown). Thus, gba1b is required nonautonomously in EG and PNG to remodel circadian circuitry, and can act directly in sLNv neurons given rescue by ectopic nSyb::Gba1b.

Figure 8. Sphingolipids drive cyclic remodeling of sLNv neurites.

(A) sLNv neurite remodeling assay following (Fernández et al., 2008) (Petsakou et al., 2015). (B) Confocal projections of sLNv neurites from 5-day old controls labeled by anti-PDF. Neurites are spread at dawn (ZT0) but retracted by dusk (ZT12). (C) gba1b^Δ1 mutants and (D) PNG&EG>gba1b^cKO neurites fail to cycle sLNv neurites compared to controls. (E) nSyb::gba1b fully rescued sLNv remodeling defects of gba1b^Δ. (F) EG&PNG >Gba1b rescued sLNv volume cycling in gba1b^Δ mutants. (G) Simplified schematic of sphingolipid metabolism. (H) Expressing Gba1b in sLNv neurites in gba1b^Δ using Pdf-GAL4 inverted the remodeling cycle, with larger sLNv spread at dusk. (I) Expressing catalytically-dead Gba1b^E340K with Pdf-GAL4 in gba1b^Δ did not invert sLNv cycling. (J) Impairing de novo sphingolipid biosynthesis using Pdf>lace^RNAi drastically increased neurite volume and blocked remodeling. (K) Impairing GlcCer biosynthesis (Pdf>GlcT^RNAi ) modestly impaired cycling in neurite volume and blocked 3-D spread. (L-M) Quantification of sLNv neurites using 3D Spread and axonal volume. Controls from L are duplicated in M; experiments were normalized to highest-volume conditions run in parallel. (N) Model for structural plasticity by glial sphingolipid catbolism. In controls, sphingolipids (GlcCer/Cer) accumulate across the day to promote neurite retraction at dusk. EG/PNG glia express Gba1b (magenta ellipse), and shape the temporal pattern of sphingolipid accumulation. In the absence of Gba1b, sphingolipid levels are high across the day, locking sLNv neurites into a retracted state. Conversely, blocking sphingolipid biosynthesis (Pdf>lace^RNAi) reduces sphingolipid levels, enabling neurites to over-extended. Increasing Gba1b expression at night via Pdf-GAL4 generates increased production of Cer sphingolipids in the morning, inverting the remodeling cycle. *p<0.05, ****p<0.0001, ANOVA, Tukey’s multiple comparisons. n > 25 sLNv neurites (A-I), n>30 flies (L-M), n>10 brains (N-O). Scale bar: 25μm (B-K). Data are represented as mean ± SEM. See also Figure S7, and Table S1 for genotypes.

Given these observations and the circadian flux of sphingolipids in control brains (Figure 6D), we hypothesized that the cycle of growth and retraction of sLNv terminals might reflect cyclic changes in the relative balance of sphingolipid biosynthesis and catabolism. If such cyclic changes play an instructive role in determining when sLNv neurons grow and shrink, altering the timing of Gba1b expression could alter the growth cycle. To do this, we took advantage of the Pdf promoter, which expresses in sLNv and lLNv neurons with higher expression at dawn (Figure S7J; (Park et al., 2000) (Herrero et al., 2020)). Strikingly, upon Gba1b induction in sLNv neurons with Pdf-GAL4 in gba1b^Δ mutants, cyclic remodeling was restored but occurred antiphase to wild-type controls, with larger neurite spread at dusk and reduced spread at dawn (Fig 8F). As expected, lysosomal and aggregate phenotypes were not rescued in the rest of the brain (Figure S7H–I). Thus, Gba1b activity is sufficient to directly control the temporal phase of sLNv membrane remodeling.

To further test the hypothesis that sphingolipid flux is critical for sLNv remodeling, we impaired de novo sphingolipid biosynthesis using RNAi against lace, the Serine palmitoyltransferase subunit (SPTLC2), the rate-limiting enzyme required for sphingosine (and thus Cer, GlcCer, and CerPE(Acharya and Acharya, 2005)). Remarkably, sLNv neurons with reduced Lace activity were locked into a “morning-like” state, with an aberrantly large neurite volume that did not significantly cycle across the day (Figure 8H, Figure 8K). To test whether GlcCer is necessary for sLNv retraction, we knocked-down Glucosylceramide Synthase (GlcT), the enzyme responsible for GlcCer synthesis, using Pdf>GlcT^RNAi. Here, sLNv neurons with reduced GlcT were mildly enlarged and impaired in diurnal cycling (Figure 8 I–K), but less dramatically than the effect of lowering all sphingolipids via lace knockdown. Thus, cell-autonomous changes in sphingolipid biosynthesis and degradation impair structural plasticity in circadian circuits. Moreover, we infer that elevated levels of sphingolipids are associated with neurite shrinkage, while reduced levels of sphingolipids are associated with neurite growth. These data demonstrate that altering either sphingolipid biosynthesis or catabolism in sLNv circuitry and neighboring glia is necessary and sufficient to alter diurnal remodeling of sLNv neurite structure.

Discussion

Our data demonstrate that glia produce Gba1b to non-autonomously control brain sphingolipids, protein degradation, and neurite remodeling in a circadian circuit (Figure 8N). We identified two specific glial subtypes, EG and PNG, as critical sources of Gba1b required for lysosomal function, proteostasis, circadian behaviors, and neurite remodeling. While previous genetic studies in vertebrate models did not determine whether GBA was required in glia or neurons (Enquist et al., 2007)(Keatinge et al., 2015)(Uemura et al., 2015), lysosomal GBA expression in mice and humans is ~5-fold higher in glia and microglia compared to neurons (Zhang et al., 2016)(Zhang et al., 2014), and gba mice harbor aggregates in astrocytes (Osellame et al., 2013). Given these expression patterns and the striking similarities in brain phenotypes seen across GBA mutants in flies, fish, and mammals, there appears to be an evolutionarily ancient role for glia in regulating sphingolipid metabolism in the brain.

Glia sculpt neurites to temporally control circuit remodeling

Previous work identified the cytoskeletal effector Rho1 and the transcription factor Mef2 as important for sLNv neurite remodeling (Petsakou et al., 2015)(Sivachenko et al., 2013). Our work demonstrates that these changes must be coordinated with membrane remodeling, particularly sphingolipid degradation and biosynthesis, both of which are required for membrane retraction and growth (Figure 8N). In control animals, sLNv neurites are enlarged at dawn, retract across the day to a stunted dusk state, and then re-extend in a daily cycle. Remarkably, this cycle coincides with a diurnal sphingolipid cycle in which specific GlcCer and Cer species are elevated during the retraction phase (ZT10–12) but reduced during the growth phase (ZT16–18). In gba1b^Δ mutants, sphingolipids are elevated and fail to cycle, locking sLNv neurites into a stunted state. Conversely, in animals in which sphingolipid biosynthesis is reduced (lace^RNAi), the cycle of sLNv growth and retraction is also blocked but sLNv neurites remain in an extended state across the day. Notably, lace is bound by the transcription factor Clock (Abruzzi et al., 2011) and undergoes circadian changes in expression, increasing before dusk and peaking at midnight in sLNv cells (Abruzzi et al., 2017)(Figure S7K). Finally, changing the timing of Gba1b expression in sLNv neurites preserved the diurnal cycle of neurite growth and retraction, but inverted its phase, such that sLNv neurites in these animals were reduced at dawn but enlarged at dusk. Thus, carefully timed cycles of sphingolipid degradation and biosynthesis pattern neurite remodeling.

We identified specific sphingolipids as elevated during neurite retraction at dusk, including Cer and GlcCer 14:1/18:1 and 14:1/20:1, suggesting that these specific species might play an important regulatory role. Although sphingolipids are substantially less abundant than phospholipids (with Cer and GlcCer species representing < 0.5% of neural membranes), their unique effects on membrane biophysics makes them well-poised to exert strong effects on membrane remodeling and structural plasticity (Castro et al., 2014) (Carvalho et al., 2010)(Holthuis et al., 2001); alternatively, secondary messenger roles or protein modifications via lipidation could regulate circuit structure and function (Merril, 2011) (Goyal et al., 2019). Given that GlcT depletion (which blocks GlcCer but not Cer formation) did not increase sLNv volume compared to lace manipulations (which blocks GlcCer, Cer, and CerPE), we favor a model whereby Cer species drive membrane retraction. This hypothesis would be consistent with recent work revealing that diurnal changes in Cer species trigger retraction of microglia processes (Liu et al., 2021) and that synaptic boutons overgrow upon global sphingolipid depletion in lace mutant neuromuscular junctions (West et al., 2018).

Sphingolipids regulate the cytoskeleton in many contexts. Cer and GlcCer specifically tune neurite growth cone branching in rat primary hippocampus cultures, with Cer driving growth followed by GlcCer refining branch-point number (Schwarz and Futerman, 1997)(Schwarz et al., 1995). Here, the relative balance of GlcCer synthesis and degradation was critical, as reducing GlcCer abolished branching whereas GlcCer accumulation (by GBA inhibition) promoted branching. Separately, blocking sphingolipid synthesis in fibroblasts acutely reduced lamellipodia formation (Meivar-Levy et al., 1997), acid ceramidase knockdown shrunk neurites via RhoA upregulation in neuroblastoma lines, and loss of a dihydroceramide desaturase activated Rac1 (Tzou et al., 2021). Similarly, GBA2 (non-lysosomal) knockout mice harbor cytoskeletal defects, shorter neurites, and dysregulated Rho GTPase localization (Raju et al., 2015), (Woeste et al., 2019). These and our data point to sphingolipids as critical lipid regulators of neural morphology during development and adult structural plasticity, and suggest that sphingolipids could function coordinately with or even upstream to cytoskeletal remodeling in sLNv circuitry.

While sLNv neurites are a dramatic example of membrane remodeling, many neurons grow and shrink across circadian cycles and varied environmental conditions (Heisenberg et al., 1995)(Barth et al., 1997)(Pyza and Meinertzhagen, 1999). Membrane turnover is likely important during structural plasticity at synapses, and indeed gba1b flies have memory defects (Davis et al., 2016). As EG and PNG are broadly distributed throughout the brain (Kremer et al., 2017), glia-mediated sphingolipid degradation may be central to membrane remodeling in many neural circuits. Daily neurite remodeling is also a feature of the mouse suprachiasmatic nucleus (Becquet et al., 2008). Interestingly, EG engulf neuronal membranes to clear damage in a sleep-dependent fashion (Doherty et al., 2009)(Musashe et al., 2016)(Stanhope et al., 2020), a cellular response potentially co-opted from a role in the daily remodeling of neurite membranes. Moreover, microglia, which express GBA in mice and humans, locally prune neurites during development and injury (Schafer et al., 2012)(Cangalaya et al., 2020), regulate sleep (Liu et al., 2021), and are enriched for sphingolipid catabolizing genes (Fitzner et al., 2020). GBA and sphingolipids may therefore control many forms of structural plasticity across species.

Protein aggregation is dynamically controlled by circadian state and neural activity

Our characterization of gba1b mutant animals revealed a surprising circadian cycle of protein aggregate accumulation and removal. We observed both activity-dependent and direct circadian control of aggregate burden. Protein aggregates have been described in wide-ranging disease models from flies to mice (Suresh et al., 2018) and are a prominent feature of neurodegeneration in humans (Ross and Poirier, 2004), including cells derived from GBA mutant mice and patients (Mazzulli et al., 2011)(Awad et al., 2015)(Schöndorf et al., 2014) (Osellame et al., 2013). Given our observations, it may prove critical to characterize aggregate accumulation (and lipid abundance) with respect to circadian time and neural activity. Indeed, circadian phagocytosis of amyloid-beta, a component of Alzheimer’s aggregates, was recently observed in cultured macrophages to be driven by circadian biosynthesis of heparan sulfate proteoglycans (Clark et al., 2022).

Lipids as linchpins of disease: a central role for glia, circadian cycles, and sleep

Our data provide a direct mechanistic connection between the enzymatic activity of Gba1b in glia to circadian and sleep behavior. Notably, gba1b mutant animals display deficits in activity and sleep prior to overt accumulation of aberrant protein aggregates. Similarly, many Parkinson’s patients have sleep defects prior to development of clinical characteristics typically associated with neuropathological aggregate accumulation (Gan-Or et al., 2015)(Krohn et al., 2020). Many genes that impinge on lysosomal function and sphingolipid degradation are linked to Parkinson’s disease (Pan et al., 2008) (Lin et al., 2018), and glia are key regulators of Parkinson’s and other neurodegenerative diseases (Zuchero and Barres, 2015)(Liddelow et al., 2017). Moreover, precise control of lipid species is central to many neurodegenerative models and is often modulated by neural activity (Guttenplan et al., 2021)(Liu et al., 2015)(Tsai et al., 2019)(Jung et al., 2017) (Dasgupta et al., 2009) (Acharya et al., 2003) (Sellin et al., 2017) (Valadas et al., 2018) (Yin et al., 2018). Recently, long-chain saturated lipids were found to mediate neurotoxicity by reactive astrocytes (Guttenplan et al., 2021), and we also observed increased nighttime long-chain saturated Cer/GlcCer species in gba1b mutants, which coincided with circadian aggregate burden (Table S4). Thus, mutations in lipid-regulating genes could impair glial remodeling of sleep circuits. Sleep has been proposed to drive clearance of aggregates from the brain (Xie et al., 2013)(Zhang et al., 2018), and glia regulate sleep and circadian rhythms in flies and mice (Ng and Jackson, 2015) (Brancaccio et al., 2019)(Herrero et al., 2017). Moreover, cell-type specific functions for sphingolipids occur in glia (Ghosh et al., 2013)(Kunduri et al., 2018)(Dahlgaard et al., 2012), and regional lipid heterogeneity pervades the human brain (O’Brien and Sampson, 1965). Unraveling the complicated mechanisms of cell-type specific lipid synthesis and degradation may provide crucial insights into the connections between sleep, circadian rhythms, neurodegenerative diseases, and neuronal membrane dynamics.

STAR Methods Text:

LEAD CONTACT

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Tom Clandinin (trc@stanford.edu).

MATERIALS AVAILABILITY

Flies and plasmids are available from the lead contact, and flies will be deposited at Bloomington Drosophila Stock Center (BDSC).

DATA AND CODE AVAILABILITY

Sphingolipid lipidomics are deposited as Supplemental Table 3. Python code used for sleep analysis is available at https://github.com/ClandininLab/sleep_analysis and deposited at Zenodo (https://doi.org/10.5281/zenodo.6816531). All FIJI and R scripts are available upon request. All data and information required to analyze the data in this paper are available from the lead contact upon request.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Drosophila Maintenance

Flies were maintained on standard molasses food. Freshly prepared food was used for all behavior, lipidomics, and circadian IHC/live imaging experiments. Stocks were maintained in light/dark (LD) incubators for 12:12 light:dark cycles at 24–25°C, for all experiments except those using gba1b^RNAi, which were performed at 29°C. For dark-dark rescue experiments (DD), crosses were set in DD incubators wrapped in foil, and experimental F1s were only temporarily exposed to light every 3 days to change food. Females were used for all experiments, except as noted; however, no obvious sex differences were observed (data not shown). For circadian experiments with age-matched genotypes, F1 experimental genotypes were collected en masse and then split into independent vials housed longitudinally in the same incubator, on the same shelf (to ensure similar lighting). Individual vials were removed immediately prior to dissection.

Drosophila Genetics

See Table S1 for a complete list of genotypes and sample sizes. Unless otherwise stated, the control genotype was an isogenized lab-stock (ISOD1, 75) placed in trans to an independent control, CantonS (CS) (“+”, ISOD1/CS). The gba1b^Δ1 frameshift allele (Fig. S1A) was induced by crossing nos-Cas9 (germline Cas9) to a ubiquitously expressed gba1b guide (BDSC#77100) targeting an early exon to generate males carrying nos-Cas9/gba1b^sgRNA on the 2nd chromosome. These males carrying potential gba1b^Δ? alleles were crossed to an isogenic +;+;pr/TM6b virgin stock, and 10 males (+;+;gba1b^Δ??/Tm6b) were outcrossed to a double-balancer stock. F3 homozygotes were screened for LysoTracker phenotypes; 9/10 lines had strong gba1b loss-of-function LysoTracker enlargement and subsequent sequencing revealed deletions at the predicted guide locus of 1,3,6, and 10nt. The 10nt deletion, gba1b^Δ1, was used in trans to gba1b^ΔTT (Davis et al., 2016) (“gba1b^Δ″, gba1b^Δ). In certain experiments, gba1b^Δ1 was used in trans to an alternative gba1b allele, gba1b^Δ5, which deletes most exons (but not the start codon). gba1b^Δ5 was isogenized by backcrossing 5 generations into the ISOD1 background and recombined with GAL4 lines on chromosome III. repo-GAL4 was also isogenized by backcrossing 5 generations into ISOD1 for repo>gba1b^cKO sleep experiments. For all rescue assays with gba1b alleles, the presence of two gba1b alleles was confirmed by taking F1 siblings that failed to carry a component of the rescuing constructs (GAL4/UAS or otherwise) and checking for brain LysoTracker enlargement. Rescuing transgenes were screened for GAL4-independent rescue by checking UAS-Gba1b; gba1b^Δ controls for gba1b loss-of-function phenotypes using LysoTracker and anti-Ubi staining.

METHOD DETAIL

Molecular Genetics

All UAS-Gba1b overexpression constructs were synthesized and cloned (Genscript) and injected into attP landing sites (BestGene; Table S2). attp2 was used for all 3^rd chromosome inserts, while attP16 and attP30 were used for 2^nd chromosome, selected based on inducibilty and lower background expression than attP40 (Markstein et al., 2008). The pJRF7–20XUAS-IVS-mCD8::GFP overexpression backbone harboring attB (Addgene # (26220, Pfeiffer et al., 2010)) was digested with XbaI and XhoI to remove intervening mCD8-GFP, which was replaced by a gba1b minigene (exons1–9 including signal sequence, omitting introns, stop codon, and UTRs); a Kozak sequence CAAA proceeds the ATG start codon before the gba1b minigene, and all constructs terminate with 3XSTOP and a SV40 terminator. In UAS-Gba1b^mCherry constructs, a NotI site and 10X-Proline linker (GCGGCCGCGAGGTGGCGGAGGTGGCGGAGGTGGCGGAGG) were inserted prior to the mCherry sequence harboring an 11 N-terminal nucleotide deletion to improve mCherry stability and prevent cleavage from Gba1b (Huang et al, 2014). For UAS-Gba1b^V5 constructs, a NotI site and 6XGly linker were inserted (GCGGCCGCGCAGCAGCAG) before a V5 epitope tag. A cloning error caused a frameshift immediately after the NotI linker and rendered the V5 epitope unusable. For the UAS-Gba1b^E340K enzyme-dead variant, the human-equivalent of E340K, E404K in Drosophila (AACACGAAGTCCTGC→ AACACGGAGTCCTGC), was induced by site-directed mutagenesis and validated by sequencing (Sequetech; see Table S2 for sequencing primers). For UAS-Gba1b^LAMP, the LAMP1 transmembrane sequence from Drosophila (GAAGCTCATGTAACCGCGGGAGGTGGAACGACGCCTAGCACACAGATATGATATTA ATACCACCAAAATTAATGCAGCTAAAGCAATTCCAACGGCAATGGGAACCAC) was appended to the C-terminal of a UAS-Gba1b-V5 backbone, following a NotI-(GS)₅-(V5)₃ linker and epitope tag. For nSyb::Gba1b, the 824bp promoter used in the pan-neural nSyb enhancer GMR57C10 was synthesized and cloned into UAS-Gba1b^E340K by digesting with HindIII and BGIII, which deletes the 20XUAS upstream (5XGAL4-DBD)₄ sequences as well as the HS promoter but leaves the mhc IVS intact. This places the Gba1b minigene ~100bp downstream of a strong pan-neural nSyb promoter/enhancer construct. Expression of constructs were validated by IHC against Gba1b, V5, or mCherry, as appropriate.

Polyclonal rabbit Gba1b antibody generation

The entire Gba1b protein minus the signal peptide (AA #22–566) was cloned into a GST-vector containing plasmid, induced and purified from inclusion bodies by Abclonal. Three New Zealand rabbits were selected based on low-background expression of pre-immunization bleeds and then immunized with Gba1b with 4 immunizations spanning 40 days. Sera was collected at 80 days and checked by Elisa for antigen reactivity. We screened test-bleeds against Gba1b-overexpressing flies and identified two bleeds which gave strong reactivity against Gba1b-specific overexpressing cells via IHC. One serum was selected for purification by antigen affinity chromatography, which labeled the Gba1b antigen at 89KD by Western Blot. While anti-Gba1b detects overexpressed Gba1b well, endogenous Gba1b expression is extremely low (Davie et al., 2018) and did not emerge above background in gba1b^Δ1/TT or gba1b^ΔTT/ΔTT mutant brains; it is also possible that anti-Gba1b cross-reacts with Gba1a, which is expressed at low levels in Kenyon Cells (Davie et al., 2018).

LysoTracker Brain Explant Imaging

Flies in batches of 15 were anesthetized on ice and transferred to an immobilizing plastic collar. Brains were dissected under cold 1X dissection saline (103mM NaCl, 3mM KCL, 5mM TES, 1mM NaH₂PO₄, 4mM MgCl) and placed in Terasaki plates containing 1X dissection saline before individual transfer to 12μL of freshly diluted LysoTracker solution (1:500 dilution from stock LysoTracker Red DND-99, 2μM, Thermofisher) for 2 minutes before immediate transfer to 100μL of saline on a microscope slide. Brains were pressed to the bottom of the saline bubble and oriented with dorsal side up (apposed to the coverslip). Brains were immediately imaged on a Leica SP8 confocal using a 40X Lens (N.A. 1.30) at 3X digital zoom. Z-stacks of 5 slices through the cortical region were acquired from optic lobes. Batches of 5 brains (interleaving control and experimental brains) were transferred to individual wells of Saline, LysoTracker, and individual slides in parallel. For co-staining LysoTracker with MagicRed, Nile Red, or in Lamp^3XmCherry (Hegedűs et al., 2016) backgrounds, LysoTracker Green DND-26 was used (also at 2μM, Thermofisher). Here, brains were incubated 2 minutes in LysoTracker green and then 2 minutes in either MagicRed (ab270772 Abcam, reconstituted in 200uL of DMSO) or Nile Red (N1142 Thermo Fisher Scientific, 1mg/mL acetone stock used 1:100 (10μg/mL final) in dissection saline) before mounting as above and immediately imaging. For imaging myrTdT and GFP flies, we used Lysotracker DeepRed, L12492 (Thermofisher) at 1:200 in dissection saline with native fluorescence for TdTomato and GFP. We were unable to reliably fix LysoTracker signals from gba1b brains.

IHC Dissections and Staining

As with LysoTracker, flies were anesthetized on ice and immobilized in dissection collars. The proboscis was removed under cold dissection saline and then freshly diluted 4% paraformaldehyde (PFA) was added to the dissection collar (32% EM grade PFA (EMS)). Brains were fixed for 25 minutes; after 5 minutes of immersion in 4% PFA, the remaining head cuticle and surrounding fat was gently removed. Post-fixation, brains were washed three times in 1X PBS before completing the dissection in collars and removing brains into Terasaki wells with 0.5%PBSTx (TritonX-100 0.5% in 1X PBS; for GlcCer and Nile Red staining, 0.05% PBSTx was used to better preserve lipids). Brains were permeabilized for 30 mins then transferred to blocking solution (10%NGS in 0.5%PBSTx) for 40 minutes before adding primary antibodies in 0.5%PBSTx+10%NGS (1:10 CSP, 1:200 anti-FK2 polyubiquitin, 1:500 anti-p62 (Abcam ab178440), 1:20 nc82, 1:50 anti-PDF, 1:2000 anti-Gba1b, 1:10,000 anti-GFP (Abcam ab13970), 1:500 anti-GlcCer (RAS_0010 anti-rabbit Glycobiotech) for 24–48 hours at 4°C while rocking gently. Brains were then washed three times in 0.5%PBSTx before transfer to appropriate secondaries (1:500, Thermo Fisher Scientific) and rocked at 4°C overnight. Brains were washed three times in 0.5%PBSTx before placing in 70% glycerol for clearing, then mounted and imaged in Vectashield. Images were collected using a Leica SP8 confocal microscope equipped with a 40X lens (N.A. 1.30) at 3X digital zoom. Z-stacks of 15 slices through the OL-iCh and MB-ca a were acquired to quantify Ubi aggregate formation. For circadian experiments, IHC experiments were independently repeated three times.

The following antibodies were obtained from the Developmental Studies Hybridoma Bank, created by the NICHD of the NIH and maintained at The University of Iowa, Department of Biology, Iowa City, IA 52242: CSP (6D6) developed by S. Benzer; brp (nc82) developed by E. Bruchner; and PDF (C7) developed by J. Blau.

LysoTracker/Ubiquitin particle analysis

Confocal LIF files were imported into FIJI and analyzed with a custom macro. Maximum intensity projections (MIPS) of slices (equal number between control and experimental) were made and thresholded based on pixel intensity (preserving values 2 standard deviations greater than the mean). Particles were analyzed: for LysoTracker and Ubi puncta, we imposed a circularity requirement of 0.4–1 and size requirement of > 20pixels. For chiasm ubiquitin, we imposed a circularity requirement of 0.1–0.7 and size >100pixels. This FIJI macro worked well for puncta particles but occasionally missed chiasm ubiquitin or erroneously called peripheral cells as chiasm ubiquitin; thus, all analyzed MIPS and FIJI-identified particles were validated by eye post-FIJI pipeline. Particle metrics were exported to excel and graphed and analyzed by ANOVA corrected via Tukey’s test for multiple comparisons in Prism GraphPad or Kruskal-Wallis test for comparisons of nonparametric data (generally Ol-iCh chiasm Ubi).

Sleep/Activity (DAM) assay and analysis

We used Drosophila Activity Monitors (DAMs, Trikinetics). 1–2 day-old flies (typically males but for select experiments virgin females) were loaded into tubes, and sleep data were recorded until day 10. We used data from day 5 to day 10 after developmental sleep completed. Cessation of activity for 5 mins was used as a proxy for sleep; flies that were inactive for 12h prior to endpoint of the analyzed period were inferred to be dead and omitted from analysis. Data were analyzed in ClockLab as well as with a custom Python library that is available at https://github.com/ClandininLab/sleep_analysis. We confirmed that our python script produced comparable results to both ClockLab and ShinyR Dam (https://karolcichewicz.shinyapps.io/shinyr-dam/ ) (Table S5). Statistical analysis was done by ANOVA between control and mutant cumulative sleep or activity binned on daytime vs nighttime; for data that was not equally variate, Welch’s ANOVA was used. For sleep bout length and number, the Mann-Whitney U test was used.

For testing endogenous circadian rhythms in control, gba1b^Δ, and glia>gba1b^cKO, 5 day old flies were shifted into DD (following LD entrainment) and analyzed for days 6–11 after 1 day of adaptation using Clocklab’s FFT and Chi-squared analyses. For Chi-squared, p<0.01, bin =6, and 0–28 time-windows were set. For %Rhythmic, we report data using both FFT amplitude <0.001 as well as Chi-squared power <0.01 (Table S5).

For sleep deprivation, a VWR 2500X shaker delivered a randomized 2s pulse every 60s from ZT12 to ZT0.

sLNv assay and analysis

We entrained flies in LD (12h-light, 12h-dark), dissected them at dawn (ZT0) and dusk (ZT12), and stained brains for anti-PDF (DSHB C7 1:50) and the neuropil counterstain brp (DSHB nc82 1:20) using secondary antibodies Alexa Fluor 633 (A21126) and Alexa Fluor 488 (A21141) (following protocol in “IHC Dissections and Staining”). Both anti-PDF and CD8-GFP genetic labeling of sLNv cells faithfully report sLNv morphology and circadian remodeling (Fernández et al., 2008) (Petsakou et al., 2015). We acquired 1024×1024 resolution confocal stacks of ventral sLNv terminals using a Leica SP8 confocal microscope with a 40X immersion lens (N.A. 1.30) at 2X digital zoom, acquiring image planes 1μm apart. Confocal images were exported to FIJI. Analysis of 3D spread was conducted blind to genotype or ZT using Matlab code kindly provided by Justin Blau (Petsakou et al., 2015). We excluded brains where PdfTri neurons were not fully pruned following metamorphosis, as these PDF+ cells overlap with sLNv terminals (Helfrich-Förster, 1997). Experiments were normalized to highest-volume control condition (typically ZT0 controls) run in parallel with the genetic manipulation, as batch ICH effects changed baseline control volume parameters. All ZT0/ZT12 experiments were dissected on the same day.

Lipidomics and Lipid Analyses

Fly cohorts of control (CS/ISOD1) and gba1b^Δ1/TT mutants were housed in antiphase incubators and dissected at ZT16–18 (midnight, asleep) or ZT10–12 (dusk, evening peak of activity) at day 3 and day 10. For figures were ZT is not indicated, genotypes were pooled across the two dissection ZTs. Crosses were maintained in 12:12 light/dark (LD) incubators that were at 25°C and 50–70% humidity. Light intensity was estimated to be ~400 lumens. For dissections during ‘night’, flies were removed from incubators in aluminum-foil enclosed boxes and not exposed to light until dissection.

15 brains per condition (genotype/timepoint) were dissected in quadruplicate (4 separate tubes). Dissections were in 1X dissection saline (see “IHC dissections” above). Flies were anesthetized on ice; all non-brain tissue (fat/hemocytes, larger trachea) was removed, as well as retina (entirely) and most lamina. Single dissected brains were immediately transferred to Eppendorf tubes containing 20μl saline on ice; after 15 brains were added (10–15mins), 180μl of methanol was added (90% methanol v/v) and brains were snap-frozen on dry ice and stored at −80°C until further analysis. The main lipidomics experiment conducted across day3/day10 was duplicated two months later with independent genetic crosses.

From each group of 15 brains, 10 brains were analyzed for sphingolipids and phospholipids. Brain total lipids containing internal standards (ISTD) were extracted as described by (Tsai et al., 2019). ISTD used included GlucosylßCeramide d18:1/8:0 (Avanti #860540), CerPE d18:1/24:0 (Avanti #860067) and Equisplash™ LIPIDOMIX® Quantitative Mass Spec Internal Standard (Avanti #330731). Lipid extracts were separate by normal phase chromatography (Agilent Zorbax RX-Sil column 3.0 × 100 mm, 1.8 μm particle size) using an Agilent 1260 Infinity LC system (Santa Clara, CA). The total LC run time was 38 min (with 5 min post-run) at a flow rate of 0.3 μL/min. Mobile phase A was composed of isopropanol/hexane/water (58:40:2,v/v) with 5mM ammonium acetate and 0.1% acetic acid. Mobile phase B consisted of isopropanol/hexane/water (50:40:10,v/v) with 5mM ammonium acetate and 0.1% acetic acid. Gradient elution consisted of increase from 34 to 36 min until 100% of mobile phase B, and after 36 min the mobile phase decreased to 0% for 2 more min. All MS measurements were obtained using an Agilent 6430 Triple-Quad LC/MS system (Santa Clara, CA) operating in positive ion mode. Gas-phase ions of various lipid species were obtained using electrospray ionization (ESI). The MS was operated in multiple reaction monitoring (MRM). The collision energies vary from 15 to 45 eV. Data acquisition and analysis was carried out using the Agilent Mass Hunter software package.

For lipid species, total ng/lipid species were calculated per brain per tube as well as relative % of species class. We prioritized ng/brain counts for sphingolipid subspecies but only obtained relative % counts for phospholipid subspecies. We report raw values in Table S3 but for certain figures report values normalized to the highest ng/brain value per experiment run for that sphingolipid class. For initial bulk and subspecies analyses, we normalized values to highest ng/brain value for individual lipid species across the experimental run; total lipids detected (ng/brain) was unchanged between mutants and controls in both experiments.

For principal component analysis (PCA), lipid species were split into groups based on class (Cer, GlcCer, CerPE, PS, PI, PC, PE), filtered for being common to experimental runs and in the top 99% cumulative fraction, and combined after z-scoring within experiment for individual species (e.g. z-scoring GlcCer 14:1/18:0 across all tubes from experiment #1, then combining with z-scored values for GlcCer 14:1/18:0 from experiment #2). Z-scored matrices were used for PCA analysis in R. To ascertain effects of gba1b on lipid species, biplots for Principal Components #1–10 from day 10 control and mutant brains (all ZTs) were plotted and colored by genotype. To test for time-effects, control samples from day 10 were analyzed by PCA, and Principal Components #1–10 biplots were colored by time. To directly identify time-modulated species, we constructed mean matrices for ‘dusk’ and ‘midnight’ z-scores from controls or mutants, and subtracted these matrices to find species that changed most across time. We then calculated t-tests for these species and corrected for multiple comparisons by the Benjamini-Hochberg FDR method. Species with FDR<0.05 were plotted in ggplot2; we confirmed that each identified species showed circadian fluctuations in both raw and normalized data in both experiments.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analysis was done in Graphpad Prism 9, Python 3.71, or R Studio (Build 443). Relevant tests are listed in figure legends. For determining differences in Ubi, LysoTracker, and normalized lipid species, ANOVA by Tukey’s multiple comparisons for normally distributed data was used (Graphpad), and Kruskal-Wallis test for nonparametric data was used (Graphpad). For quantifying sleep differences, ANOVA by Tukey’s multiple comparisons (equal data variance) or Welch’s ANOVA (unequal variance) was used in Python. For quantifying nonparametric sleep bouts, the Mann-Whitney U test was used. For identifying time-modulated or sleep-modulated lipids, t-tests corrected for multiple comparisons by the Benjamini-Hochberg FDR method was used in R (species with FDR<0.05 were considered statistically significant).

Supplementary Material

2

Table S1: detailed genotypes and n, related to Figures 1–7

3

Table S2, sequences for primers and UAS-Gba1b plasmids, related to Figure 1

4

Table S3, lipidomics data (sphingolipids/phospholipids, ng/brain), related to Figures 5–6

5

Table S4, differentially expressed lipids based on time or genotype, related to Figure 6

6

Table S5, sleep and circadian analyses, related to Figure 7

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Gba1b (anti-rabbit polyclonal)	This paper, abclonal	Anti-Gba1b
Csp (anti-mouse IgG2b)	Developmental Studies Hybridoma Bank	Cat#DCSP-2 (6D6) RRID: AB_528183
Ubi (anti-mouse IgG1)	Enzo Life Sciences	Cat# BML-PW8810, RRID:AB_10541840
p62 (anti-rabbit polyclonal)	Abcam	Cat# ab178440
GFP (anti-rabbit polyclonal)	Abcam	Cat# ab13970
GlcCer (anti-rabbit polyclonal)	Glycobiotech Gmbh	Cat# RAS_0010
PDF (anti-mouse IgG2b)	Developmental Studies Hybridoma Bank	Cat# PDF C7, RRID: AB_760350)
Bruchpilot (anti-mouse IgG1)	Developmental Studies Hybridoma Bank	Cat# nc82; RRID: AB_2314866
Goat anti-Chicken IgY (H+L) Secondary Antibody, Alexa Fluor 488	moFisher Scientific	Cat# A-11039, RRID:AB_2534096)
Goat anti-Rabbit IgG (H+L) Cross-Adsorbed Secondary Antibody, Cyanine3	ThermoFisher Scientific	Cat# A10520, RRID:AB_2534029
Goat Anti-Mouse IgG2b Antibody, Alexa Fluor 488 Conjugated	ThermoFisher Scientific	Cat# A-21141, RRID:AB_141626)
Goat anti-Mouse IgG1 Cross-Adsorbed Secondary Antibody, Alexa Fluor 647	ThermoFisher Scientific	Cat# A-21240, RRID:AB_2535809)
Goat anti-Mouse IgG2b Secondary Antibody, Alexa Fluor 633	ThermoFisher Scientific	Cat# A-21146, RRID:AB_2535782)
Critical commercial assays
LysoTracker Red DND-99	ThermoFisher Scientific	Cat# L7528
LysoTracker Deep Red	ThermoFisher Scientific	Cat#  L12492
LysoTracker™ Green DND-26	ThermoFisher Scientific	Cat# L7526
MagicRed Cathepsin Assay	Abcam	Cat# ab270772
Nile Red	ThermoFisher Scientific	Cat# N1142
Experimental models: Organisms/strains
D. melanogaster: +(ISOD1)	Clandinin lab	N/A
D. melanogaster: +(CantonS)	Clandinin lab	N/A
D. melanogaster: P{y[+t7.7] v[+t1.8]=TRiP.HMS01893}attP40	Bloomington Drosophila Stock Center	BDSC#38977, FBst0038977
D. melanogaster: gba1b[ΔTT]	Davis et al., 2016	N/A
D. melanogaster: gba1b[Δ1]	this paper	N/A
D. melanogaster: gba1b[Δ5]	Perlara (Joshua Mast)	N/A
D. melanogaster: P{y[+t7.7] v[+t1.8]=TKO.GS00895}attP40	Bloomington Drosophila Stock Center	BDSC#77100, FBst0077100
D. melanogaster: P{y[+t7.7] w[+mC]=UAS-Cas9.P2}attP2/TM6B, Tb[1]	Bloomington Drosophila Stock Center	BDSC#58986, FBst0058986
D. melanogaster: nSyb::Gba1b {attP16}	this paper	N/A
D. melanogaster: UAS-gba1b-LAMP-V5 {attP16}	this paper	N/A
D. melanogaster: UAS-gba1b-V5 {attP16}	this paper	N/A
D. melanogaster: UAS-gba1b-V5 {attP2}	this paper	N/A
D. melanogaster: UAS-gba1b[E340K]-V5 {attP2}	this paper	N/A
D. melanogaster: UAS-gba1b[E340K]-V5 {attP16}	this paper	N/A
D. melanogaster: UAS-gba1b-mCherry-V5 {attP16}	this paper	N/A
D. melanogaster: UAS-gba1b-mCherry-V5 {attP2}	this paper	N/A
D. melanogaster: TI{GFP[3xP3.cLa]=CRIMIC.TG4.0}gba1b1b[CR00541-TG4.0]/TM3, Sb[1] Ser[1]	Bloomington Drosophila Stock Center	BDSC#78943, FBst0078943
D. melanogaster: P{y[+t7.7] w[+mC]=8XLexAop2-FLPL}attP40	Bloomington Drosophila Stock Center	BDSC#55820, FBst0055820
D. melanogaster: P{13XlexAop-myr::tdTomato}attP2	Gerald Rubin	FBtp0093486
D. melanogaster: w[*]; P{w[+mC]=UAS-Stinger}2, PBac{y[+mDint2] w[+mC]=13XLexAop2-IVS-tdTomato.nls}VK00022	Bloomington Drosophila Stock Center	BDSC#66680, FBst0066680
D. melanogaster: w[*]; P{w[+mC]=UAS-mCherry.NLS}2; MKRS/TM6B, Tb[1]	Bloomington Drosophila Stock Center	BDSC#38425, FBst0038425
D. melanogaster: P{10XUAS-IVS-mCD8::GFP}attP2	Bloomington Drosophila Stock Center	BDSC#32185, FBst0032185
D. melanogaster: P{y[+t7.7] w[+mC]=GMR57C10-GAL4}attP2	Bloomington Drosophila Stock Center	BDSC#39171, FBst0039171
D. melanogaster: P{y[+t7.7] w[+mC]=GMR57C10-lexA}attP40/CyO	Bloomington Drosophila Stock Center	BDSC#52817, FBst0052817
D. melanogaster: w[1118]; P{y[+t7.7] w[+mC]=GMR77A03-GAL4}attP2	Bloomington Drosophila Stock Center	BDSC#39944, FBst0039944
D. melanogaster: w[1118]; P{y[+t7.7] w[+mC]=GMR77A03-lexA}attP40	Bloomington Drosophila Stock Center	BDSC#54108, FBst0054108
D. melanogaster: w[1118]; P{y[+t7.7] w[+mC]=GMR85G01-lexA}attP40	Bloomington Drosophila Stock Center	BDSC#54285, FBst0054285
D. melanogaster: w[1118]; P{y[+t7.7] w[+mC]=GMR85G01-GAL4}attP2	Bloomington Drosophila Stock Center	BDSC#40436, FBst0040436
D. melanogaster: w[1118]; P{y[+t7.7] w[+mC]=GMR54C07-GAL4}attP2	Bloomington Drosophila Stock Center	BDSC#50472, FBst0050472
D. melanogaster: w[1118]; P{y[+t7.7] w[+mC]=GMR54C07-lexA}attP40	Bloomington Drosophila Stock Center	BDSC#61562, FBst0061562
D. melanogaster: w[1118]; P{y[+t7.7] w[+mC]=GMR53H12-GAL4}attP2	Bloomington Drosophila Stock Center	BDSC#50456, FBst0050456
D. melanogaster: w[1118]; P{y[+t7.7] w[+mC]=GMR53H12-lexA}attP40	Bloomington Drosophila Stock Center	BDSC#53573, FBst0053573
D. melanogaster: w[1118]; P{y[+t7.7] w[+mC]=GMR56F03-GAL4}attP2	Bloomington Drosophila Stock Center	BDSC#39157, FBst0039157
D. melanogaster: w[1118]; P{y[+t7.7] w[+mC]=GMR56F03-lexA}attP40	Bloomington Drosophila Stock Center	BDSC#53574, FBst0053574
D. melanogaster: w[*]; P{y[+t7.7] w[+mC]=GMR56F03-GAL4}attP24/CyO	Bloomington Drosophila Stock Center	BDSC#77469, FBst0077469
D. melanogaster: w[*]; P{w[+mC]=nrv2-GAL4.S}3; P{nrv2-GAL4.S}8	Bloomington Drosophila Stock Center	BDSC#77469, FBst0077469
D. melanogaster: w[1118]; P{y[+t7.7] w[+mC]=GMR86E01-GAL4}attP2	Bloomington Drosophila Stock Center	BDSC#45914, FBst0045914
D. melanogaster: w[1118]; P{y[+t7.7] w[+mC]=GMR86E01-lexA}attP40/CyO	Bloomington Drosophila Stock Center	BDSC#54287, FBst0054287
D. melanogaster: w[*]; P{w[+m*]=alrm-GAL4.D}3/CyO; Dr[1]/TM3, Sb[1]	Bloomington Drosophila Stock Center	BDSC#67031, FBst0067031
D. melanogaster: w[1118]; P{y[+t7.7] w[+mC]=GMR83E12-GAL4}attP2	Bloomington Drosophila Stock Center	BDSC#40363, FBst0040363
D. melanogaster: w[1118]; P{y[+t7.7] w[+mC]=GMR83E12-lexA}attP40	Bloomington Drosophila Stock Center	BDSC#54288, FBst0054288
D. melanogaster: w[*]; P{w[+m*]=moody-GAL4.SPG}2	Bloomington Drosophila Stock Center	BDSC#90883, FBst0090883
D. melanogaster: P{w[+m*]=GAL4}repo/TM3, Sb[1]	Bloomington Drosophila Stock Center	BDSC#7415, FBst007415
D. melanogaster: per 01	Bloomington Drosophila Stock Center	BDSC#80928, FBst0080928
D. melanogaster: +; Pdf-GAL4;+	Dragana Rogulja	FBtp0011844
D. melanogaster: LAMP 3xmCherry	Gábor Juhász	FBal0325101
D. melanogaster: inc 1	Nicholas Stavropoulos	FBal0266013
D. melanogaster: inc 2	Nicholas Stavropoulos	FBal0162225
D. melanogaster: qvrEY04063 (sleepless)	Bloomington Drosophila Stock Center	FBst0016588
D. melanogaster: P{GawB}9–1379-137	Amita Sehgal	FBal0344513
D. melanogaster: P{GMR83E12-GAL4}	Bloomington Drosophila Stock Center	FBtp0063377
D. melanogaster: y1 v1; P{TRiP.HMC03219}attP40 [lace-RNAi]	Bloomington Drosophila Stock Center	FBti0157638
D. melanogaster : P{KK107304}VIE-260B [GlcT-RNAi]	Vienna Drosophila Resource Center	FBst0479877
Oligonucleotides
Primers for Gba1b locus sequencing, see Table S2	This paper
Recombinant DNA
Sequences of Gba1b plasmids, see Table S2	This paper
Software and algorithms
ImageJ (FIJI	ImageJ	RRID: SCR_003070
GraphPad Prism 9	GraphPad	RRID: SCR_002798
MATLAB_R2021a	MATLAB	RRID: SCR_001622
Python 3.7.1	Python	RRID: SCR_008394
R Studio 2022.02.0 Build 443	R	RRID: SCR_000432
Clocklab Analysis 2 (Version 6.1.02)	Actimetrics	RRID:SCR_014309
Python Sleep Analysis	Python	https://doi.org/10.5281/zenodo.6816531

Highlights.

• Glia produce Glucocerebrosidase for sphingolipid degradation in Drosophila

• Glial gba1b knockout causes diurnal protein aggregate formation and sleep loss

• Diurnal fluctuations in lipid subsets, especially sphingolipids, depend on Gba1b

• sLNv circuit remodeling is spatiotemporally controlled by sphingolipid metabolism