Glucocerebrosidases catalyze a transgalactosylation reaction that yields a newly-identified brain sterol metabolite, galactosylated cholesterol

β-Glucocerebrosidase (GBA) hydrolyzes glucosylceramide (GlcCer) to generate ceramide. Previously, we demonstrated that lysosomal GBA1 and nonlysosomal GBA2 possess not only GlcCer hydrolase activity, but also transglucosylation activity to transfer the glucose residue from GlcCer to cholesterol to form β-cholesterylglucoside (β-GlcChol) in vitro. β-GlcChol is a member of sterylglycosides present in diverse species. How GBA1 and GBA2 mediate β-GlcChol metabolism in the brain is unknown. Here, we purified and characterized sterylglycosides from rodent and fish brains. Although glucose is thought to be the sole carbohydrate component of sterylglycosides in vertebrates, structural analysis of rat brain sterylglycosides revealed the presence of galactosylated cholesterol (β-GalChol), in addition to β-GlcChol. Analyses of brain tissues from GBA2-deficient mice and GBA1- and/or GBA2-deficient Japanese rice fish (Oryzias latipes) revealed that GBA1 and GBA2 are responsible for β-GlcChol degradation and formation, respectively, and that both GBA1 and GBA2 are responsible for β-GalChol formation. Liquid chromatography–tandem MS revealed that β-GlcChol and β-GalChol are present throughout development from embryo to adult in the mouse brain. We found that β-GalChol expression depends on galactosylceramide (GalCer), and developmental onset of β-GalChol biosynthesis appeared to be during myelination. We also found that β-GlcChol and β-GalChol are secreted from neurons and glial cells in association with exosomes. In vitro enzyme assays confirmed that GBA1 and GBA2 have transgalactosylation activity to transfer the galactose residue from GalCer to cholesterol to form β-GalChol. This is the first report of the existence of β-GalChol in vertebrates and how β-GlcChol and β-GalChol are formed in the brain.

␤-Glucocerebrosidase (GBA) hydrolyzes glucosylceramide (GlcCer) to generate ceramide. Previously, we demonstrated that lysosomal GBA1 and nonlysosomal GBA2 possess not only GlcCer hydrolase activity, but also transglucosylation activity to transfer the glucose residue from GlcCer to cholesterol to form ␤-cholesterylglucoside (␤-GlcChol) in vitro. ␤-GlcChol is a member of sterylglycosides present in diverse species. How GBA1 and GBA2 mediate ␤-GlcChol metabolism in the brain is unknown. Here, we purified and characterized sterylglycosides from rodent and fish brains. Although glucose is thought to be the sole carbohydrate component of sterylglycosides in vertebrates, structural analysis of rat brain sterylglycosides revealed the presence of galactosylated cholesterol (␤-GalChol), in addition to ␤-GlcChol. Analyses of brain tissues from GBA2deficient mice and GBA1-and/or GBA2-deficient Japanese rice fish (Oryzias latipes) revealed that GBA1 and GBA2 are responsible for ␤-GlcChol degradation and formation, respectively, and that both GBA1 and GBA2 are responsible for ␤-GalChol formation. Liquid chromatography-tandem MS revealed that ␤-GlcChol and ␤-GalChol are present throughout development from embryo to adult in the mouse brain. We found that ␤-GalChol expression depends on galactosylceramide (GalCer), and developmental onset of ␤-GalChol biosynthesis appeared to be during myelination. We also found that ␤-GlcChol and ␤-GalChol are secreted from neurons and glial cells in association with exosomes. In vitro enzyme assays confirmed that GBA1 and GBA2 have transgalactosylation activity to transfer the galactose residue from GalCer to cholesterol to form ␤-GalChol. This is the first report of the existence of ␤-GalChol in vertebrates and how ␤-GlcChol and ␤-GalChol are formed in the brain.
Glycosylation of lipids, including glycerolipids, sterols, and sphingolipids, is a highly-conserved system in living organisms. Sterylglycosides are glycosylated sterols that are found in bacteria, fungi, plants, and animals (1). This evolutionary conservation of sterylglycosides across domains of organisms indicates that they likely play critical roles in fundamental cellular processes. In vertebrates, cholesterol is the major sterol component, and its glucosylation forms ␤-cholesterylglucoside (1-O-cholesteryl-␤-Dglucopyranoside (␤-GlcChol)). 2 Recently, Marques et al. (2) dem- onstrated that ␤-GlcChol is present in human and mouse tissue, including brain, by using liquid chromatography (LC)-electrospray ionization tandem MS (ESI-MS/MS). Because purification of ␤-GlcChol from central nervous system (CNS) tissue has not been reported, and structural analysis based on the ␤-GlcCholcontaining fraction has not been reported, we developed a multistep chromatographic purification protocol capable of isolating ␤-GlcChol from the brain, an organ that contains high levels of glucosylceramide (GlcCer) and galactosylceramide (GalCer). As they exhibit very similar chromatographic behavior to ␤-GlcChol, a method like ours is needed to differentiate it from GlcCer and GalCer. Although ␤-GlcChol is thought to be the sole sterylglycoside in vertebrates, in embryonic chicken brain we found two more sterylglucosides, the plant-type ␤-sitosterylglucoside (1-O-sitosteryl-␤-D-glucopyranoside (␤-GlcSito)) and an as-of-yet structurally-and not fully-identified homologue termed GSX-2 (3). Available structural information on GSX-2 suggests high similarity with plant-type ␤-campesterylglucoside (1-O-campesteryl-␤-D-glucopyranoside) (3). Formation of ␤-GlcChol is mediated by glucocerebrosidase (GCase, GlcCer-degrading ␤-glucosidase) 1 and 2 (GBA1 and GBA2). GBA1 (lysosomal acid GCase, EC 3.2.1.45) is a watersoluble enzyme that can associate with lysosomal membranes. Homozygous mutations in the GBA1 gene cause Gaucher disease (GD), which is the most common lysosomal storage disorder. It is characterized by the accumulation of GlcCer in the lysosomal compartment of macrophages (4,5). Heterozygous mutations in the GBA1 gene are a high-risk factor for Parkinson's disease (PD) and multiple system atrophy (MSA) (6,7). GBA2 (nonlysosomal GCase, EC 3.2.1.45) is a nonintegral membrane-associated protein that localizes at the cytosolic surface of the endoplasmic reticulum (ER) and Golgi apparatus (8 -11). GBA2-deficient mice accumulate GlcCer in the testis, brain, and liver, and they exhibit male infertility (9) and delayed liver regeneration (12). Loss-of-function mutations in the GBA2 gene cause neurological disorders, such as cerebellar ataxia and spastic paraplegia (13)(14)(15). In mammalian cells in vitro, both GBA1 and GBA2 possess not only GlcCer hydrolase activity but also transferase activity, catalyzing a transglucosylation reaction to form ␤-Gl-cChol via transfer of the glucose residue from GlcCer to cholesterol (2,16). GBA1 and GBA2 also possess ␤-GlcChol hydrolase activity, and both enzymes can transfer a glucose residue from ␤-GlcChol to ceramide, indicating that the enzymes catalyze a reversible transglucosylation reaction between GlcCer and ␤-Gl-cChol via a glucose molecule (2,16).
The concentration of ␤-GlcChol is elevated in the liver and the plasma of a GD mouse model. It is also elevated in the plasma of GD patients, the liver of a Niemann-Pick disease type C (NPC) disease mouse model, and the plasma of NPC patients (2). ␤-GlcChol also increases in cells treated with U18666A, an amphiphilic compound that inhibits cholesterol transport and causes accumulation of cholesterol in lysosomes. U18666A-induced ␤-GlcChol elevation is inhibited by conduritol B epoxide (DL-1,2-anhydro-myo-inositol; CBE), a specific inhibitor of GBA1 (2). GBA2-deficient mice have decreased levels of ␤-Gl-cChol in the thymus, the liver, and plasma (2), suggesting that GBA2 is responsible for ␤-GlcChol formation in vivo. Under normal conditions, GBA1 and GBA2 seem to contribute to ␤-GlcChol degradation and formation, respectively. In contrast, when excess cholesterol accumulates in lysosomes, such as that found in NPC disease, GBA1 actively generates ␤-Gl-cChol (2). Although ␤-GlcChol metabolism mediated by GBA1 and GBA2 has been studied in several kinds of cells and tissues and in plasma, the metabolism of sterylglycosides in the brain is not understood.
Although glucose is considered as the sole carbohydrate headgroup in vertebrate sterylglycosides, we previously reported the presence of a sterylglycoside constituent from brain and sciatic nerve tissue exhibiting resistance against digestion by recombinant human GBA1 (rGBA1), Cerezyme (2,3). In contrast, GSX-2 and ␤-GlcSito fractions from embryonic chicken brains (E6 to E18) were digested completely by rGBA1 (3).
In this report, we describe the isolation of sterylglycosides from 12-week-old rat brains, as previous reports suggested a higher content of the rGBA1-resistant sterylglycosides in adult brains compared with early-stage embryonic brains (2,3). Structural analysis of the isolated sterylglycoside fractions revealed the presence of ␤-cholesterylgalactoside (1-O-cholesteryl-␤-D-galactopyranoside, ␤-GalChol), in addition to ␤-GlcChol, GSX-2, and ␤-GlcSito. To understand the biological significance of brain sterylglycosides, we determined their developmental expression and cellular distribution. Using the brains of GBA2-deficient mice and GBA1-and/or GBA2-deficient Oryzias latipes (also known as Japanese rice fish or medaka), we demonstrated that GBA1 is involved in ␤-GlcChol degradation and ␤-GalChol formation and that GBA2 is involved in the formation of ␤-GlcChol and ␤-GalChol and degradation of GSX-2 and ␤-GlcSito in vivo. It has not been reported that GBA1 and GBA2 have GalCer-degrading ␤-gal activity. However, our in vitro enzyme assays revealed that GBA1 and GBA2 degraded GalCer and synthesized ␤-GalChol from GalCer and cholesterol. Our work described here is the first report to demonstrate the existence of ␤-GalChol in vertebrates and to demonstrate the metabolism of sterylglycosides in vertebrate brain.
NMR analysis of GS1 and GS2 revealed a characteristic pattern in the range of 4.5-3.3 ppm ( Fig. 2A), which is typically associated with glucopyranosides and galactopyranosides, respectively. The 1 H chemical shifts of the hexoside portion matches well with those of the references (Table S1). The large splittings (7.7 and 6.9 Hz) of the anomeric protons (Glc H-1 and Gal H-1) at 4.43 and 4.38 ppm in GS1 and GS2, respectively, are archetypical for an axial-axial J coupling consistent with ␤-gly-cosidic linkage. In the low-field region, the aglycon exhibits a deshielded vinyl proton at 5.4 ppm (Chol H-6), which is consistent with the presence of a Chol C5-C6 double bond. The highfield region of the 1 H-13 C HSQC spectrum of GS1 (Fig. 2B) confirms the presence of the expected ␤-cholesteryl aglycon. The carbohydrate region of the homonuclear Hartmann-Hahn (HOHAHA) spectrum of GS2 confirmed the presence of galactopyranoside (Fig. 2C).
To gain further structural insight of the main constituents in GS1 and GS2, each fraction was subjected to reversed-phase LC (RPLC)-ESI-MS/MS analysis. A neutral-loss scan of the carbohydrate moiety following collision-induced dissociation (CID) revealed a single peak for each sample corresponded to a parent ion [M ϩ NH 4 ] ϩ at m/z 566 (Fig. 3, A and B). The peak in GS1 and GS2 exhibited similar elution time compared with authentic ␤-GlcChol and ␤-GalChol standards, respectively. MALDI-TOF/TOF analysis of the lithium adduct of GS1 and GS2 main constituents at [M ϩ Li] ϩ m/z 555.4 (Fig. 3, C and D) followed by high-energy CID (Fig. 3, E and F) revealed a dominant ion at m/z 187 in the product ion spectra, corresponding to [hexose ϩ Li] ϩ . These results revealed the presence of glycosylated sterols in GS1 and GS2 fractions.  To determine the aglycon structure in GS1 and GS2, each fraction was subjected to enzymatic deglycosylation, followed by GC/MS (GC/MS) of the liberated aglycons (Fig. S1). The major sterol liberated from fractions GS1 and GS2 eluted at 20.2 min (Fig. S1, A, peak 1, and D, peak 2). The elution time and fragmentation pattern of the major sterol (Fig. S1, B and E) coincided well with authentic cholesterol. The liberated aglycons from fraction GS2 exhibited an additional peak eluting at 27.2 min (Fig. S1D, peak 3). The absence of the m/z 129 fragment (Fig. S1F), usually present in ⌬ 5 -3-hydroxysteroid TMS derivatives (17), and the late elution time of peak 3 did not allow its identification.
Taken together, structural analysis of the fraction GS1 from adult rat brains confirmed the presence of ␤-GlcChol, and fraction GS2 revealed the presence of ␤-GalChol.

Characterization of sterylglycosides in brain
To characterize the expression pattern of brain sterylglycosides, we developed an analytical method using LC-MS. First, ␤-GalChol was chemically synthesized according to the method described under "Experimental procedures." Hexosylcholesterol, including ␤-GlcChol and ␤-GalChol, was monitored using hydrophilic interaction chromatography (HILIC)-ESI-MS/MS. HILIC columns, which are widely used for
In mouse brain lipid samples, HILIC-ESI-MS/MS detected two peaks that exhibited similar retention times as authentic standards of ␤-GlcChol and ␤-GalChol. These peaks were consistent with the peaks for ␤-GlcChol and ␤-GalChol in adult rat brain samples. HILIC-ESI-MS/MS revealed that ␤-GlcChol and ␤-GalChol differ in ionization efficiency (Fig. 4B). This contrasts with the ionization efficiency of GlcCer and GalCer, which was nearly the same (data not shown). To accurately quantify sterylglycosides, we used a deuterium-labeled ␤-Gl-cChol (␤-GlcChol-d 7 ) in further experiments in order to normalize the extraction efficiency of ␤-GlcChol and ␤-GalChol. Because the ionization efficiency of ␤-GlcChol and ␤-GalChol is different, we quantified ␤-GlcChol and ␤-GalChol separately using standard curves of ␤-GlcChol-d 7 and ␤-GalChol, respectively.
To monitor the developmental expression of sterylglycosides in the brain, we used HILIC-ESI-MS/MS to analyze brain lipid extracts obtained from mice of different ages, i.e. embryonic day (E) 12 to postnatal week 10. The lipid extracts were added to internal standards, ␤-GlcChol-d 7 and GlcCer (d18:1-C12:0). Multiple reaction monitoring (MRM) analysis revealed that ␤-GlcChol and ␤-GalChol were present throughout development and varied depending on age (Fig. 5). The amount of ␤-Gl-cChol detected in E12 mouse brain was four times greater than that of ␤-GalChol. In contrast to the modest increase of ␤-Gl-cChol during early development, the amount of ␤-GalChol substantially increased after postnatal day (P) 10. Thus, at postnatal weeks 4 and 10, the concentrations of ␤-GlcChol and ␤-Gal-Chol in mouse brain were comparable.
Next, we investigated the distribution of brain sterylglycosides. We previously reported that in the lipid raft fraction of human fibroblasts, ␤-GlcChol is formed through transglucosylation (20). The lipid composition of lipid rafts is similar to that of exosomes, extracellular vesicles (EVs) ranging in size from 70 to 150 nm in diameter (21)(22)(23)(24). This prompted us to investigate whether sterylglycosides are released from cells via exosomes. EVs containing exosomes were prepared from the culture media of human neuroglioma H4 cells and mouse primary cortical neurons by a sequential centrifugation. The resulting 100,000 ϫ g pellet was separated by continuous sucrose density gradient centrifugation. The exosomal proteins Alix, tumor susceptibility gene 101 protein, and flotillin-1 were detected in sucrose density fractions corresponding to a density range of 1.07-1.18 g/ml (Fig. 6, A and D), similar to previous reports (25,26). We did not detect a marker protein for Golgi and an ER retention sequence in the pellet GM130 and Lys-Asp-Glu-Leu (abbreviated KDEL), respectively.
Analysis of exosomal particle size revealed that the subcellular material in the pellet contained mainly small membrane vesicles of 70 -150 nm in diameter (Fig. 6, B and E). The concentrations of sterylglycosides, ceramide, sphingomyelin (SM), GlcCer, and GalCer were analyzed by LC-ESI-MS/MS (Fig. 6, C and F). Consistent with previous reports (23,24), exosomes derived from H4 cells and mouse primary cortical neurons exhibited higher concentrations of GlcCer and GalCer compared with their cells of origin (Fig. 6, C and F). Although ceramide and SM are reportedly enriched with exosomes (21,23,24,27), we only detected enrichment of ceramide in exosomes derived from H4 cells but not from mouse primary cortical neurons, and SM was not enriched in exosomes (Fig. 6, C and F). Interestingly, ␤-GlcChol, ␤-GalChol, and ␤-GlcSito were detected in exosomes derived from H4 cells and mouse primary cortical neurons. ␤-GlcChol was enriched in exosomes derived from H4 cells, whereas ␤-GlcSito was enriched in exosomes derived from both H4 cells and mouse primary cortical neurons ( Fig. 6, C and F). GSX-2 content was below the detection threshold in all analyzed exosomes and their cells of origin.

Brain sterylglycosides are metabolized by GBA1 and GBA2 in vivo
The metabolism of sterylglycosides has been described in various tissues, but not in brain tissue. Thus, we investigated the involvement of GBA1 and GBA2 in sterylglycoside metabolism. We used the brains of GBA2-deficient (Gba2 Ϫ/Ϫ ) mice and GBA1-and/or GBA2-deficient medakas (O. latipes) for this study.
PCR-based genotyping of mice was used to identify Gba2 ϩ/ϩ and Gba2 Ϫ/Ϫ homozygous offspring (Fig. 7A). The concentrations of sterylglycosides, GlcCer, and GalCer in the brains of WT and Gba2 Ϫ/Ϫ mice were analyzed by HILIC-ESI-MS/MS (Fig. 7B). In the brains of Gba2 Ϫ/Ϫ mice, ␤-GlcChol levels were decreased. This is consistent with the down-regulation of ␤-GlcChol observed in the liver, thymus, and plasma of GBA2deficient mice (2). Unexpectedly, ␤-GalChol levels were also decreased in the brains of Gba2 Ϫ/Ϫ mice. GSX-2 and ␤-GlcSito levels, however, were increased. To date, the levels of GlcCer and GalCer in the brains of Gba2 Ϫ/Ϫ mice have not been reported. Our analysis showed that the level of GlcCer but not GalCer was increased in the brains of Gba2 Ϫ/Ϫ mice. This is consistent with the finding that in Gba2 Ϫ/Ϫ mice GlcCer is up-regulated in the testis, liver, kidney, and small intestine (9,19). GBA1-and/or GBA2-deficient medakas were generated by crossing Gba1 ϩ/Ϫ /Gba2 ϩ/Ϫ heterozygous mutants. 3 The resulting offspring had one of nine possible genotypes:
HILIC-ESI-MS/MS analysis of GalCer and ␤-GalChol showed that GBA1 KD or GBA2 KD reduced ␤-GalChol levels by about 50% (Fig. 8B). GBA1 KD or GBA2 KD did not alter GalCer levels. GALC KD resulted in the up-regulation of ␤-GalChol. We could not detect alteration of GalCer levels in H4 cells after GALC KD. Consistent with the results presented in Fig. 5, CGT KD reduced ␤-GalChol and GalCer levels, suggesting that ␤-GalChol expression depends on GalCer. These results suggest that GBA1, GBA2, and CGT are involved in ␤-GalChol formation and that GALC is involved in ␤-GalChol degradation.

Galactosylated cholesterol in vertebrate brain
formation was not detected using 20 ng of rGBA1 protein, NBD-GalChol was generated when we used more protein.
We repeated the in vitro transgalactosylation activity assay using natural cholesterol as glycosyl acceptor and GalCer (d18: 1-C12:0) as glycosyl donor. Formation of ␤-GalChol was assessed using HILIC-ESI-MS/MS (Fig. 9D). Again, ␤-Gal-Chol was formed after incubating cholesterol for 2 h with rGBA1, homogenates from GBA2-overexpressing HEK293T cells, or rGALC. CBE and NB-DGJ, respectively, inhibited rGBA1-and GBA2-associated ␤-GalChol formation. Unexpectedly, a small amount of ␤-GalChol was formed in the absence of rGALC or without incubation (Fig. 9D, right panel). HILIC-ESI-MS/MS analysis confirmed that there was no ␤-GalChol contamination in any of the rGALC reaction mixtures (data not shown). This result indicates that trace amounts of ␤-GalChol could be formed by nonenzymatic reactions.
To transfer a galactose moiety from GalCer to cholesterol, hydrolysis of GalCer is necessary. Therefore, we analyzed the GalCer hydrolysis activity of rGBA1, of GBA2 in homogenates from GBA2-overexpressing HEK293T cells, and of rGALC using C 6 -NBD-GalCer. HPTLC and fluorescence scanning was used to monitor the formation of the hydrolysis end product, C 6 -NBD-ceramide (Fig. 9E). We also analyzed GlcCer hydrolysis activity using C 6 -NBD-GlcCer. GlcCer hydrolysis occurred in the presence of rGBA1 and the HEK293T cell homogenates. GlcCer hydrolysis activity of rGBA1 and GBA2 (in homogenates) was inhibited by CBE and NB-DGJ, respectively. GalCer hydrolysis occurred in the presence of rGBA1, the HEK293T cell homogenates, and rGALC. The GalCer hydrolysis activity of rGBA1 and GBA2 (in homogenates) was inhibited by CBE and NB-DGJ, respectively. rGBA1 activity toward C 6 -NBD-GalCer hydrolysis was ϳ200 times lower than that toward C 6 -NBD-GlcCer hydrolysis. The GBA2 activity in homogenates toward C 6 -NBD-GalCer hydrolysis was ϳ7 times lower than that toward C 6 -NBD-GlcCer hydrolysis. These results suggest that GBA1, GBA2, and GALC have GalCer hydrolysis activity and transgalactosylation activity. These reactions transfer a galactose moiety from GalCer to cholesterol to form ␤-GalChol in vitro.

Sterylglycoside abnormalities in the brains of NPC disease model mice
NPC disease is a neurodegenerative lysosomal storage disease caused by loss-of-function mutations in either the NPC1 or NPC2 gene, genes that encode proteins essential for exporting cholesterol from lysosomes. Impairment of NPC1 or NPC2 leads to the accumulation of cholesterol in lysosomes; this accumulation is the primary defect in NPC disease (29,30). ␤-GlcChol is remarkably elevated in the liver of Npc1 Ϫ/Ϫ mice S.E. Differences were analyzed by unpaired t tests. **, p Ͻ 0.01; ***, p Ͻ 0.001; #, p Ͻ 0.0001. C, effect of Gba1 and/or Gba2 knockout on sterylglycoside formation in medaka brain. Concentrations of sterylglycosides in the 3.5month-old medaka brains were analyzed by HILIC-ESI-MS/MS. These medakas were generated by crossing heterozygous mutants, which had a Gba1 ϩ/Ϫ /Gba2 ϩ/Ϫ genotype. Data (n ϭ 3) are means Ϯ S.E. Differences were assessed by unpaired t tests. *, p Ͻ 0.05; **, p Ͻ 0.01 compared with WT (Gba1 ϩ/ϩ /Gba2 ϩ/ϩ ).

Galactosylated cholesterol in vertebrate brain
(NPC disease model), being 25 times higher compared with WT mice (2). ␤-GlcChol is also elevated in the plasma of patients with NPC disease (2). These findings prompted us to determine the concentration of sterylglycosides in the brains of Npc1 Ϫ/Ϫ mice. The concentration of sterylglycosides, GlcCer, and GalCer was analyzed using HILIC-ESI-MS/MS (Fig. S2). Consistent with the up-regulation of ␤-GlcChol in the liver of Npc1 Ϫ/Ϫ mice (2), the ␤-GlcChol level was elevated in the brains of Npc1 Ϫ/Ϫ mice. Although the ␤-GalChol level was unchanged in the brains of Npc1 Ϫ/Ϫ mice, levels of GSX-2 and ␤-GlcSito increased. As reported previously (31), we also found that GlcCer level was up-regulated and that GalCer level was down-regulated in the brains of Npc1 Ϫ/Ϫ mice.

Galactosylated cholesterol in vertebrate brain
GS2 revealed the presence of ␤-GlcChol and ␤-GalChol, respectively, representing the first report of ␤-GalChol in vertebrates. ␤-GalChol has previously been identified in Borrelia burgdorferi (32,33), a bacterium causing Lyme disease. Lyme disease is the most common tick-borne disease in the Northern hemisphere. Additionally, the presence of GSX-2 and ␤-

Galactosylated cholesterol in vertebrate brain
GlcSito in adult rat brain, albeit at a 40 times lower abundance compared with ␤-GlcChol (data not shown), was confirmed. This low abundance of GSX-2 and ␤-GlcSito precluded further NMR spectroscopic analyses. As shown in Fig. 5, ␤-GalChol expression in the mouse brain relates to the presence of GalCer having long-chain fatty acid (Ͼ18 carbon atoms). GalCer with long-chain fatty acids (Ͼ18 carbon atoms) are enriched in myelin and are important for stabilizing myelin (34). Therefore, it is reasonable to suggest that ␤-GalChol could be a component of myelin. GalCer is quite abundant in the CNS, and its chromatographic behavior is very similar to that of ␤-GlcChol and ␤-GalChol. Because of this overlap with GalCer, it has been extremely difficult to isolate ␤-GlcChol and ␤-GalChol from brains for structural analysis, even when these are present. Our HILIC-ESI-MS/MS analysis revealed aglycon heterogeneity of sterylglucosides in rat, mouse, and medaka brain tissue, as well as in H4 cells, mouse primary cortical neurons, and HeLa cells. However, our HILIC-ESI-MS/MS analysis did not detect aglycon heterogeneity of sterylgalactoside, suggesting that cholesterol is the sole sterol aglycon of sterylgalactoside in vertebrates.
We showed that sterylglycosides are secreted from H4 cells and mouse primary cortical neurons in association with exosomes (Fig. 6). ␤-GlcChol has been demonstrated to be a neurotoxin in vitro (35), and ␤-GlcSito is a neurotoxin in vitro (36,37) and in vivo (37)(38)(39). Thus, the secretion of ␤-GlcChol and/or ␤-GlcSito via exosomes might represent a cellular response to the buildup of toxic lipids in cells, providing a way to remove them from the cells. Exosomes have been hypothesized to participate in transferring pathogenic proteins between cells (40), such as ␤-amyloid peptide and ␣-synuclein, known to be associated with PD and MSA. Because mutations in GBA1 appear to be involved in pathogenesis of PD and MSA (6, 7), it will be interesting to examine what role GBA1-metabolized sterylglycosides play in exosomal distribution of ␣-synuclein.
The levels of sterylglycosides were altered in the brains of GBA2-deficient mice and GBA1-and/or GBA2-deficient medakas. Consistent with a previous report (2), GBA1 seemed to be responsible for the degradation of ␤-GlcChol, and GBA2 for the formation of ␤-GlcChol. The level of ␤-GalChol was reduced in H4 cells in which specific targeting siRNA of GBA1 or GBA2 was introduced. ␤-GalChol was barely detected in the brains of GBA2-deficient mice and GBA2-deficient medakas. This suggests that GBA2 is the responsible enzyme for ␤-Gal-Chol formation in vivo. Although it did not reach a statistically significant decline, ␤-GalChol levels were decreased in the brains of GBA1-deficient medakas, suggesting the involvement of GBA1 in ␤-GalChol formation. GBA2 seemed to be responsible for the degradation of GSX-2 and ␤-GlcSito. However, in the case of GBA2 deficiency, GBA1 seemed capable of degrading GSX-2 and ␤-GlcSito. Because GBA1 and GBA2 are localized to different cellular compartments, further experiments will be needed to determine the subcellular distribution of GSX-2 and ␤-GlcSito in order to understand the GBA1-and GBA2-mediated metabolic pathways.
It remains to be determined whether GSX-2 and ␤-GlcSito are de novo-synthesized in the brain and cultured cells, or whether they are derived from dietary intake or content in the culture media, respectively. The level of ␤-GlcSito did not change in UGCG Ϫ/Ϫ HeLa cells compared with WT HeLa cells. This suggests that GlcCer might not be the glucose donor for ␤-GlcSito formation in vertebrates. In plants, UDP-glucose:sterol glucosyltransferase (UDPG-SGTase, EC 2.4.1.173) is the enzyme responsible for ␤-GlcSito formation (1). Resolution of ␤-GlcSito formation in vertebrates remains stalled, as molecular cloning of the mammalian gene that shares homology with the known UDPG-SGTase gene has been unsuccessful.
Our in vitro enzyme assays revealed that GBA1, GBA2, and GALC exhibited ␤-gal activity and transgalactosylation activity. This reaction transfers a galactose residue from GalCer to cholesterol, leading to the formation of ␤-GalChol. This newlydetected activity against GalCer is significantly lower compared with their known activity against GlcCer. This enzyme's "promiscuity" of GBA1 and GBA2 has not been reported until now. Supporting our findings, it is known that D-galactose-configured deoxynojirimycins inhibit GBA1 and GBA2 (41). Also, in Thermoanaerobacterium xylanolyticum, a thermophilic anaerobic Gram-negative bacterium, not only does GBA2 have ␤-glucosidase activity, but it also has ␤-gal activity toward 4-nitrophenyl ␤-D-galactoside in vitro (42). Further structural analysis will be required to illuminate the differences between GalCer and GlcCer substrate recognition by GBA1 and GBA2.
GalCer is synthesized by CGT at the luminal side of the ER membrane (43) and is degraded by GALC at the luminal side of the lysosome (44). Burger et al. (45) showed transbilayer movement of GalCer in the membrane of cellular organelles, providing a rationale for GBA2-GalCer interaction at the cytosolic side of the ER membrane. In the lysosome, GalCer is located at the inner leaflet of the membrane and thus is readily accessed by GBA1 and GALC for ␤-GalChol synthesis. Although rGALC exhibited transgalactosylation activity to form ␤-GalChol in vitro, the level of ␤-GalChol increased in H4 cells that had specific siRNA targeting of GALC. Additionally, rGALC successfully released the aglycon from the isolated GS2 fraction predominantly containing ␤-GalChol. Together, this suggests that GALC might be responsible for ␤-GalChol degradation in cells. It would be of interest to determine the levels of ␤-GalChol in the Twitcher mouse, a naturally occurring model of Krabbe disease that lacks GALC protein. Although in this study we did not analyze the ability of CGT transgalactosylation in vitro, we found that UDP-Gal was not used as the donor for NBD-GalChol formation. Therefore, CGT might not be the enzyme responsible for ␤-GalChol formation.
Marques et al. (31) showed that GBA2 activity is increased in the brains of Npc1 Ϫ/Ϫ mice. This increased GBA2 activity might also cause up-regulation of ␤-GlcChol in the brains of Npc1 Ϫ/Ϫ mice. In this study, our lipid analysis of GBA2-defi-

Galactosylated cholesterol in vertebrate brain
cient mouse brain revealed that GBA2 is responsible for degradation of GSX-2 and ␤-GlcSito. Although GBA2 activity is reportedly increased in the brains of Npc1 Ϫ/Ϫ mice (31), we found that GSX-2 and ␤-GlcSito were elevated in the brains of Npc1 Ϫ/Ϫ mice. It remains unclear where GSX-2 and ␤-GlcSito are distributed subcellularly and whether NPC1 exports GSX-2 and ␤-GlcSito from lysosomes. However, one explanation for the up-regulation of GSX-2 and ␤-GlcSito we detected in Npc1 Ϫ/Ϫ mice is that export of GSX-2 and ␤-GlcSito from lysosomes is impaired.
Our results also have other significant implications for the treatment of GD, PD, MSA, cerebellar ataxia, and spastic paraplegia. It will be interesting to determine how these previously unrecognized functions of GBA1 and GBA2, demonstrated in this study, will aid our understanding of the role of GBA1 and GBA2 mutations in GD, PD, MSA, cerebellar ataxia, and spastic paraplegia.

Animals and tissue collection
C57BL/6J mice were purchased from Japan SLC, Inc. (Hamamatsu, Japan). Gba2 Ϫ/Ϫ mice (C57BL/6J-129S6/SvEv mixed background) were generated as described previously (9). Breeding pairs of Gba2 ϩ/Ϫ mice (C57BL/6J-129S6/SvEv mixed background) were a kind gift from Prof. Dr. Dagmar Wachten, University of Bonn (Bonn, Germany). Gba2 ϩ/Ϫ mice were back-crossed to the C57BL/6J background mice for 10 generations. Gba2 Ϫ/Ϫ mice with the C57BL/6J background, along with WT littermates (Gba2 ϩ/ϩ ), were generated by crossing the back-crossed Gba2 ϩ/Ϫ males and females in-house. Offspring mice were genotyped by PCR using genomic DNA as described previously (9). These mice were housed in the RIKEN Center for Brain Science animal housing facilities under a 12-h light/12-h dark on/off cycle and at a constant room temperature of 23 Ϯ 2°C. The mice had free access to water and standard mouse chow (CRF-1, Oriental Yeast Co., Ltd., Tokyo, Japan). Experimental protocols for the animal experiments were approved by RIKEN's Wako Animal Experiments Committee.
For collection and preparation of tissue for biochemical analyses, postnatal mice were deeply anesthetized and perfused with PBS. Whole brains were surgically removed, directly snapfrozen in liquid nitrogen, and stored at Ϫ80°C until lyophilized. After lyophilization, samples were stored at Ϫ80°C until further use. For tissue collection from embryonic mice, we used embryos of pregnant C57BL/6J mice that were purchased from Japan SLC, Inc. Pregnant mice were deeply anesthetized, and the uterus together with the embryos were transferred to icecold PBS. The uterus was cut, and the embryonic sacs were removed to release embryos into the PBS. The released embryos were transferred to ice-cold Hanks' balanced salt solution (Nacalai Tesque, Inc., Kyoto, Japan). Whole brains were surgically removed, directly snap-frozen in liquid nitrogen, and stored at Ϫ80°C until lyophilized. After lyophilization, samples were stored at Ϫ80°C until further use. From P10 to postnatal week 10, male mice were used for tissue collection. From E12 to P2, sex of the mice was not discriminated.
Npc1 Ϫ/Ϫ mice (Npc1 nih , BALB/c background), along with WT littermates (Npc1 ϩ/ϩ ), were generated as described previously (2). These mice (Ϯ3 weeks old) received the rodent AM-II diet (Arie Blok Diervoeders, Woerden, The Netherlands). The mice were housed at the Institute Animal Core Facility in a temperature-and humidity-controlled room with a 12-h light/ dark on/off cycle, and they were given free access to food and water ad libitum. Offspring were genotyped by PCR using genomic DNA, as described previously (48). Experimental protocols for these animal experiments were approved by the Institutional Animal Welfare Committee of the Academic Medical Centre Amsterdam in the Netherlands. For collection and preparation of tissue for biochemical analyses, male mice were first anesthetized with an intramuscular injection of Hypnorm (comprising 0.315 mg/ml phenyl citrate and 10 mg/ml fluani-

Galactosylated cholesterol in vertebrate brain
sone; VetaPharma Ltd., Leeds, UK) and Dormicum (5 mg/ml midazolam; Teva Pharmaceutical Industries Ltd., Parsippany, NJ). The dose was adjusted according to their weight (80 l/10 g bodyweight). These mice were euthanized by cervical dislocation. Whole brains were surgically removed, rinsed with PBS, directly snap-frozen in liquid nitrogen, and stored at Ϫ80°C until lyophilized. After lyophilization, samples were stored at Ϫ80°C until further use.

Generation of Gba1 ؉/؊ /Gba2 ؉/؊ medakas and preparation of medaka brains
Medakas (O. latipes) are an ideal vertebrate animal model for our purposes, because deletion of GBA1 in fish does not cause perinatal death (49 -51). Previously, Uemura et al. (49) generated Gba1 nonsense mutant (Gba1 Ϫ/Ϫ ) medaka. In contrast to the perinatal death in humans and mice lacking GBA1 activity, Gba1 Ϫ/Ϫ medakas survived for months, enabling biochemical and behavioral analyses of adult stage.
The ethics statement and maintenance of medakas were described previously (49). Medaka experiments were approved by the Animal Experiments Committee of Kyoto University and were conducted in accordance with Japanese national guidelines. Medakas were maintained in an aquaculture system with recirculating water at 27°C in a 14-h light/10-h dark cycle.
The medaka GBA2 gene consists of 18 exons that encode 858 amino acids. The crRNAs were generated using the CRISPR design tool (RRID:SCR_018159), and the following crRNA was used: 5Ј-GGAGGGCAAAGCACTGTCGGGGG-3Ј. The crRNAs and tracrRNA were constructed by Fasmac Co. (Kanagawa, Japan). The Cas9 RNA was synthesized from a pCS2ϩhSpCas9 vector (catalogue no. 51815, Addgene, Watertown, MA) using a mMessage mMachine SP6 kit (Thermo Fisher Scientific). The RNA mixture was injected into singlecell-stage embryos. The injected founders (F 0 ) were raised to sexual maturity and back-crossed with WT medakas for generating F 1 s. The GBA2 gene of the F 1 s was sequenced, and the novel GBA2-deficient (Gba2 ϩ/Ϫ ) medakas with 21 bases deleted and two bases inserted into exon 5 were obtained. 3 These deletions and insertions resulted in a frameshift mutation, leading to a deficiency in GBA2 protein expression and enzymatic activity of GBA2 in the brain. Off-target candidates were searched using the Medaka pattern match tool (RRID: SCR_018157). No alterations were found in three off-target candidates located on exons.
Gba2 ϩ/Ϫ medakas were back-crossed with WT medakas at least five times, and then crossed with GBA1-deficient medakas to create Gba1 ϩ/Ϫ /Gba2 ϩ/Ϫ medakas. Medaka brains collected by surgery were directly snap-frozen in liquid nitrogen and then stored at Ϫ80°C until use.

Cell cultures
Human neuroglioma H4 cells (HTB-148 TM ) were purchased from the ATCC (Manassas, VA). They were cultured in Dulbecco's modified Eagle's medium (DMEM, FUJIFILM Wako) supplemented with 10% (v/v) fetal bovine serum (FBS). Cultures were maintained in a 5% CO 2 atmosphere at 37°C. HeLa-mCAT#8 cells, a parent cell line, and HeLa-mCAT#8 cell mutants deficient in UGCG (28) were a generous gift from Dr. Toshiyuki Yamaji of the National Institute of Infectious Diseases (Tokyo, Japan). UGCG is a gene that encodes UDPglucose:ceramide glucosyltransferase (GlcCer synthase) (53). These cells were also cultured in DMEM supplemented with 10% FBS and were maintained in a 5% CO 2 atmosphere at 37°C. Stable human GBA2-expressing human embryonic kidney (HEK) 293T cells were generated as described previously (54). HEK293T cells were cultured in DMEM (Sigma-Aldrich) supplemented with 10% (v/v) FBS, 200 g/ml penicillin/streptomycin, and 1% (v/v) GlutaMAX and were maintained in a 7% CO 2 atmosphere at 37°C. For enzyme assays, cell pellets of the stable human GBA2-expressing HEK293T cells were obtained using standard methods. The cell pellets were suspended in McIlvaine buffer (0.1 M citric acid and 0.2 M Na 2 HPO 4 (pH 5.8)) containing 0.1% (w/v) bovine serum albumin (BSA) and were then sonicated.
For primary neuron cultures, we used embryos of pregnant Slc:ICR mice that were purchased from Japan SLC, Inc. These cultures were prepared from the cerebral cortices of Slc:ICR mouse brains on embryonic day 14, according to the methods of Yuyama et al. (26), with some modifications. Briefly, neurons for culturing were obtained from the isolated cerebral cortices using a neuron dissociation solution (FUJIFILM Wako). The dissociated cells were plated onto poly-L-ornithine-coated dishes at a density of 3.16 ϫ 10 5 cells/cm 2 and cultured in Neurobasal TM medium (Thermo Fisher Scientific) supplemented with GlutaMAX TM supplement (Thermo Fisher Scientific), 1 mM sodium pyruvate, 100 units/ml penicillin, 100 g/ml streptomycin, and B-27 TM supplement (Thermo Fisher Scientific). One day after plating, cytosine ␤-D-arabinofuranoside (Sigma-Aldrich) was added to the culture at a final concentration of 10 M to inhibit the proliferation of dividing non-neuronal cells. Three and 6 days after plating, half of the culture supernatant from each dish was collected and replaced with fresh medium. Nine days after plating, all the culture supernatants were collected. The collected culture supernatants were combined and then used for exosome isolation.

Enrichment of sterylglycosides
Frozen brains of 12-week-old male Wistar rats were purchased from Japan Lamb Ltd. (Hiroshima, Japan). Total lipids were extracted from lyophilized brains (100 brains, 45 g dry weight) using 400 ml of chloroform and methanol (C/M at 2:1, v/v) twice and 200 ml of C/M (1:1, v/v). After evaporation, the combined extracts were hydrolyzed for 3 h at room temperature in C/M (2:1, v/v, 240 ml) containing 0.1 M KOH. The reaction mixture was subjected to Folch's partition (55), and the lower phase was evaporated to dryness. The resulting lipid film was resuspended in chloroform (100 ml) and applied to a col-

Isolation of sterylglycosides
The enriched sterylglycoside fraction was further purified by normal-phase HPLC. The lipid film was resuspended in a small volume of mobile phase A (C/M at 98:2, v/v), applied to a PEGASIL Silica SP100 -3 column (4.6 mm inner diameter ϫ 250 mm, particle size, 3 m; Senshu Scientific Co., Ltd., Tokyo, Japan), and eluted with the following gradients of mobile phase B (C/M/W at 80:20:0.2, v/v/v): 10 min, 0%; 60 min, 0 -100% linear gradient; and 10 min, 100% (washing step). The flow rate was kept constant at 0.5 ml/min, and the column was maintained at room temperature. The eluent was collected once per min. The presence of sterylglycosides was evaluated using HPTLC, and positive fractions were pooled and dried.

Enzymatic deglycosylation of sterylglucosides
Aglycon release from the GS1 fraction was performed as described previously (3). Aglycon release from the GS2 fraction was performed as described previously (56), with minor modifications as follows. A portion of the GS2 fraction was dissolved in C/M (2:1, v/v) and taken to dryness. This portion was then suspended in 10 l of 0.1% Triton X-100 followed by sonication. Next, 0.4 mg of oleic acid (Sigma-Aldrich) dissolved in C/M (2:1, v/v) was taken to dryness and then dispersed by sonication in 40 l of 70 mg/ml sodium taurocholate containing 0.1% Triton X-100. To release aglycon from the GS2 fraction, we made a reaction mixture in a total volume of 100 l, which contained 10 l of the dispersed oleic acid, 10 l of the dispersed GS2 fraction, and 860 ng of rGALC in 50 l of McIlvaine buffer (pH 4.1). The reaction mixture was incubated at 37°C for 20 h. The reaction was terminated by addition of 2 ml of C/M (2:1, v/v) and 400 l of water (adjusted to a total volume of 500 l, including the reaction mixture) to facilitate lipid extraction by Folch's partition. The organic layer was separated and dried. The sterol fraction was purified by HPTLC on Silica Gel 60 using hexane/diethyl ether/acetic acid (80:20:1, v/v/v) as an eluent. Lipids were stained by primuline reagent and visualized by long-wave UV. The band that co-migrated with standard cholesterol (FUJIFILM Wako) was collected and extracted by Folch's partition. The organic layer containing the released aglycons was separated and dried under a stream of N 2 .

GC/MS analysis
The lipid containing released aglycons was suspended in 25 l of TMS and was incubated at room temperature for 30 min. The resulting trimethysilylated material was subjected to GC/MS analysis on a GC system 7890A (Agilent Technologies Inc., Santa Clara, CA) attached to a mass spectrometer, JMS-T100GCV AccuTOF GCv 4G (JEOL Ltd., Tokyo, Japan). An HP-5ms capillary column (30 m ϫ 0.25 mm, film thickness of 0.25 m; Agilent Technologies Inc.) was interfaced with the GC system. We employed the following temperature gradient: from 130 to 250°C at a heating rate of 20°C/min, and from 250 to 300°C at a heating rate of 5°C/min.

MALDI-TOF/TOF analysis
The GS1 fraction, GS2 fraction, and authentic standards of ␤-GlcChol and ␤-GalChol were each dissolved in C/M (2:1, v/v) at a concentration of ϳ0.5 g/l, then mixed with MALDI matrix A (10 g/l of DHB in C/M, 1:1, v/v) and B (1 g/l LiCl in methanol) at a ratio of 1:1:1 (v/v/v). One l of the resulting mixture was spotted onto an MTP 384 target plate of polished steel BC (Bruker, Billerica, MA) and dried. The samples were analyzed with an autoflex speed TOF/TOF (Bruker) equipped with the TOF/TOF option. A Smartbeam II Nd:YAG laser pulse of 355 nm was operated at 2 kHz. For product-ion mass spectrum acquisition, argon collision gas was introduced. The collision energy of 6 keV was used to induce high-energy CID.

LC-ESI-MS/MS analysis
The LC-ESI-MS/MS analysis was performed on an LC system Nexera X2 (SHIMADZU, Kyoto, Japan) attached to a triple-quadrupole linear ion trap mass spectrometer, QTRAP4500 (SCIEX, Tokyo, Japan). The LC-ESI-MS/MS datasets were analyzed with MultiQuant TM (version 2.1) and Analyst software (SCIEX). Target lipids were monitored in MRM mode using specific precursor-product ion pairs, as detailed in Table S2.

NMR spectroscopy
Highly-purified sterylglycoside fractions and authentic standards were dissolved in CDCl 3 containing tetramethylsilane as an internal chemical shift reference. One-dimensional 1 H NMR and two-dimensional double-quantum filtered correlation spectroscopy, homonuclear Hartmann-Hahn (HOHAHA) spectra, and 1 H-13 C multiplicity-edited HSQC spectra were recorded on an Avance-500 spectrometer (Bruker BioSpin, Yokohama, Japan) equipped with a triple-resonance (TXI) cryogenic probe and an Avance-600 spectrometer with a TXI probe. Probe temperature was set at 25°C. The NMR data were processed with TopSpin software (version 2.1; Bruker BioSpin), and the spectra were displayed using TopSpin Plot Editor (version 2.1; Bruker BioSpin).

Chemical synthesis of 1-O-cholesteryl-␤-D-galactopyranoside
All solvents and reagents were purchased from commercial suppliers as reagent grade and were used as is. Reactions with anhydrous solvents (Kanto Chemical Co., Inc.) were performed under an argon atmosphere. Flash column chromatography was performed with Silica Gel 60 N (40 -100 mesh, Kanto Chemical Co., Inc.) and the indicated solvent systems. Analytical TLC was performed on Silica Gel 60 F256 plates (Merck Millipore). Mass spectra were recorded on an AccuTOF JMS-T700LCK (JEOL) using ESI. NMR spectra were obtained on an ECX-500 spectrometer (JEOL) in the indicated solvent system at 25°C using residual nondeuterated solvent signals as chemical shift references. Cholesteryl 2,3,4,6,-tetra-O-acetyl-␤-D-galactopyranoside was synthesized as follows. 2,3,4,6-Tetra-O-acetyl-␣-D-galactopyranosyl 2,2,2-trichloroacetimidate (2.0 g, 4.06 mmol, 1.2 Eq) and cholesterol (Chol) (1.3 g, 3.37 mmol, 1.0 Eq) were dissolved in anhydrous dichloromethane (20 ml) at Ϫ40°C. After 10 min, the coupling was initiated by addition of TMS (20 l, 24.6 mg, 0.11 mmol) and stirred for 2 h at Ϫ40°C. The reaction was quenched with triethylamine (3 ml); the organic layer was increased with dichloromethane (20 ml); and the resulting mixture was extracted twice against water (20 ml) and once against brine (20 ml). The organic layer was dried over Na 2 SO 4 , concentrated under reduced pressure, and subjected to flash chromatography (hexane/ethyl acetate  ␤-GalChol was synthesized as follows. Cholesteryl 2,3,4,6,tetra-O-acetyl-␤-D-galactopyranoside (0.7 g, 0.98 mmol, 1.0 Eq) was dissolved in anhydrous dichloromethane (10 ml) and treated with freshly-prepared sodium methoxide solution (10 ml of anhydrous methanol with 10 mg of sodium). The suspension was stirred for 2 h at room temperature, quenched with acetic acid (52 l, 50 mg), and evaporated to dryness. The resulting solid was suspended in hot ethanol and filtered, and the filter cake was further rinsed with hot ethanol. The filter cake was dried under high vacuum to give ␤-GalChol (300 mg, 0.547 mmol, 56%) as a white solid.

Quantification of lipids in animal tissue
Lyophilized tissue and frozen tissue were homogenized, and total lipids were extracted with C/M (2:1, v/v, 5 ml) mixture added to 5 pmol/mg lyophilized tissue or to 1 pmol/mg frozen tissue of GlcCer (d18:1-C12:0), and ␤-GlcChol-d 7 (3) served as internal standard. Extracts were dried under a flow of N 2 and hydrolyzed for 2-3 h at room temperature in C/M (2:1, v/v, 2-3 ml) containing 0.1 M KOH. After neutralization with 7.5 l of glacial acetic acid, the reaction mixture was subjected to Folch's partition, and the lower phase was dried under a flow of N 2 . The resulting lipid film was suspended in C/M (2:1, v/v) at a concentration of ϳ40 g of lyophilized tissue/l or ϳ100 g of frozen tissue/l, and diluted 10-fold with mobile phase A for HILIC-ESI-MS/MS analysis. Aliquots (10 l) were subjected to HILIC-ESI-MS/MS analysis. The resulting peak areas were integrated and quantified relative to the associated internal standard.

Quantification of lipids in cultured cells and exosomes
Frozen cells and frozen exosomes were resuspended in 250 l of water and sonicated. Protein concentration was determined by a Pierce TM BCA protein assay kit (Thermo Fisher Scientific). Water was added to the homogenate up to 400 l, and then 1.5 ml of C/M (1:2, v/v), ϳ10 pmol of GlcCer (d18:1-C12:0), ␤-GlcChol-d 7 , ceramide (d18:0-C12:0), and SM (d18:1-C12:0), each as internal standards, were added. The sample was sonicated and centrifuged at 3,000 rpm for 10 min at room temperature, and the supernatant was collected. A volume (1.9 ml) of C/M/W (1:2:0.8, v/v/v) was added to the remaining pellet, and then the sample was sonicated and centrifuged as described above. The supernatant was collected and combined with the first supernatant.
Next, lipids were extracted according to the method of Bligh and Dyer (57) by adding 1 ml of chloroform and water each; the lower phase was taken to dryness under a flow of N 2 . For glycolipid analysis, a portion of the sample was hydrolyzed with KOH as described above. The resulting lipid film was suspended in C/M (2:1, v/v) at a concentration of ϳ150 g of protein/l and diluted 10-fold with mobile phase B for RPLC-ESI-MS/MS analysis or mobile phase A for HILIC-ESI-MS/MS analysis. Aliquots (10 l) were subjected to RPLC-ESI-MS/MS or HILIC-ESI-MS/MS analysis, respectively. The resulting peak areas were integrated and quantified relative to the associated internal standard. For comparing lipid concentrations of exosomes and their cells of origin, the lipid concentrations were normalized with lipid phosphate, which was determined using a total phosphorous assay (58).

Exosome isolation
Exosomes were prepared from culture supernatants of H4 cells and mouse primary cortical neurons as described previously (26). For H4 cells, culture medium was replaced with serum-free medium 1 day before exosome isolation; culture supernatants were collected 24 h later. For the primary cortical neurons, culture supernatants were collected according to the method described under "Cell cultures." The collected culture supernatants were sequentially centrifuged at 3,000 ϫ g for 10 min at 4°C, 4,000 ϫ g for 10 min at 4°C, and 10,000 ϫ g for 30 min at 4°C to remove cells, dead cells, and debris. To obtain exosomes as pellets, the material was centrifuged again at 100,000 ϫ g for 1 h at 4°C. The resulting pellets were then subjected to lipid extraction, particle size analysis, and sucrose gradient centrifugation.
For sucrose gradient centrifugation, each exosome pellet was suspended with 1 ml of 2.3 M sucrose in 20 mM HEPES and transferred into centrifugal tubes (Ultra-Clear TM , 14 ϫ 89 mm; Beckman Coulter K.K., Tokyo, Japan). The sucrose gradient (11 ml of 0.25-2.3 M sucrose in 20 mM HEPES) was subsequently layered onto 1 ml of resuspended exosomes and then centri-fuged in an SW41TI swinging rotor (Beckman Coulter K.K.) at 100,000 ϫ g for 18 h at 4°C. After centrifugation, fractions of 1 ml (typically 11-12 fractions in total) were collected from the top layer of the gradient, diluted with ϳ2 ml of 20 mM HEPES, and pelleted by centrifugation for 1 h at 100,000 ϫ g. The resulting pellets were resuspended in PBS and subjected to Western blotting.

Analysis of exosomal particle size
A qNano System (Izon Science, Ltd., Christchurch, New Zealand) was used to analyze the particle size of H4Ϫ and mouse primary cortical neuron-derived exosomes resuspended in 50 mM HEPES, 150 mM KCl (pH 8.0). We followed the manufacturer's instructions.
The assay of transglycosylase activity was also performed with natural cholesterol as the acceptor. The reaction mixture in a total volume of 40 l contained 40 M cholesterol, 80 or 200 M GalCer (d18:1-C12:0) dissolved in the appropriate buffer as described above, 2% ethanol, and a desired amount of enzyme. The volume of McIlvaine buffer in the reaction mixture was 2ϫ greater than described above. After incubation at 37°C for 2 h, the reaction was terminated by adding chloroform/methanol (2:1, v/v) containing 800 fmol of ␤-GlcChol-d 7 as an internal standard to normalize extraction efficiency. Lipid extraction was performed by Folch's partition, and the organic phase was pooled and evaporated. The extracted lipids were separated by HPTLC with chloroform/methanol (85:15, v/v). Lipids were stained with primuline reagent and visualized using long-wave UV. The bands co-migrating with standard ␤-GlcChol and ␤-GalChol were collected, combined, and extracted using Folch's partition. The organic layer containing sterylglycosides was collected and dried under a stream of N 2 . The extracted lipids were subjected to HILIC-ESI-MS/MS, and formation of ␤-GalChol was determined.

In vitro assay of GlcCer or GalCer hydrolase activity
The GlcCer and GalCer hydrolase assay used in this study was carried out according to the method we described previously (16), with some modifications. The reaction mixture in a total volume of 20 l contained 100 pmol of C 6 -NBD-GlcCer or C 6 -NBD-GalCer dissolved in appropriate buffer and a desired amount of enzyme. The assay for rGBA1 was performed with 10 l of McIlvaine buffer (pH 5.3), Triton X-100 (final 0.25%, v/v), and sodium taurocholate (final 0.6%, w/v). The assay for the homogenate of human GBA2-expressing HEK293T cells was performed with 16.6 l of McIlvaine buffer (pH 5.8) containing 0.1% (w/v) BSA and ethanol (final 1%). The assay for rGALC was performed with 10 l of McIlvaine buffer (pH 4.1), Triton X-100 (final 0.01%, v/v), oleic acid (final 2 g/l), and sodium taurocholate (final 7 g/l). After incubation at 37°C for 30 min, the reaction was terminated by adding chloroform/methanol (2:1, v/v); the lipids were extracted, and C 6 -NBD-ceramide formation was analyzed as reported before (16).

Statistical analysis
Statistical analyses and univariate descriptive statistics were performed using GraphPad Prism, version 5.02 for Windows (GraphPad Software, San Diego, CA). Group means Ϯ S.E. was calculated. To assess group differences, at least three independent experiments were assessed using unpaired t tests.