Graphical abstract

   graphic file with name fx1.jpg
   [52]Open in a new tab

Highlights

     * •
       Heparan sulfate (HS) affects autophagy, mitochondria, and lipid
       metabolism
     * •
       AD-related gene transcripts overrepresented in HS-dependent
       transcriptome
     * •
       Reducing HS sulfation rescues neuronal loss in presenilin-deficient
       Drosophila
     * •
       Cell deficits mediated by presenilin knockdown suppressed by
       reduced HS sulfation
     __________________________________________________________________

   Molecular neuroscience; Cell biology

Introduction

   More than 100 years ago, Alois Alzheimer described a dementia condition
   associated with a triad of histopathological abnormalities:
   extracellular deposits of amyloid, intracellular aggregations or
   neurofibrillary tangles, and accumulation of intracellular lipid in
   glia, described as “adipose saccules.”[53]^1 These pathological
   features have guided the strategies to develop a pharmaceutical for
   Alzheimer disease (AD), and amyloid peptide generation and deposition
   (neuritic plaques) has been one of the principal targets of these
   efforts. Recently, two antibody pharmaceuticals that reduce the levels
   of amyloid peptides have been approved by Food and Drug Administration
   (FDA), and one shows demonstrable slowing of cognitive
   loss.[54]^2^,[55]^3 A third antibody pharmaceutical directed against
   Aβ, donanemab, has shown effectiveness in slowing disease progression
   in a recent clinical trial.[56]^4 Despite these successes, there is
   good reason to identify cellular and molecular pathologies that occur
   early in the AD process. Further investment in broad, mechanistic
   approaches to AD development is required to effectively prevent and
   treat this disease at a level that significantly alters the disease
   trajectory.

   Considerable evidence from a variety of experimental approaches has
   shown that there are cellular deficits common to AD, implicated in both
   early, familial AD where causative genetic changes have been
   identified, as well as late-onset AD, where GWAS studies have found
   many risk-associated variants that influence AD
   susceptibility.[57]^5^,[58]^6^,[59]^7 Pathway analyses show enrichment
   for genes involved in cholesterol metabolism, amyloid plaque and
   neurofibrillary tangle formation, membrane trafficking, and the innate
   immune system.[60]^5 Many of the cellular and molecular processes
   embedded within these gene sets are mechanistically connected.[61]^8
   Autophagy, a vital membrane trafficking process compromised in AD, not
   only removes protein aggregates and damaged organelles (such as
   mitochondria via mitophagy) but is required for the catabolism of
   lipids via lipophagy. Mitochondria are also a critical nexus in these
   mechanisms, providing the catabolic machinery to break down fatty acids
   via β-oxidation and providing ATP to support cell energetics in the
   process.

   The cellular and molecular events that accompany AD are represented in
   experimental models of the disease. Many of these are based on human
   genetic data where specific gene deficits produce early onset, severe
   AD and allow detailed examination of the cellular and molecular changes
   that initiate cell demise.[62]^9 PSEN1 mutations are the most common
   cause of familial AD, show dominant inheritance, and produce a
   cognitive decline in the fourth decade with greater than 90%
   penetrance.[63]^10 The vast majority (90%) of clinically identified
   PSEN1 mutations are loss or reduction in function alleles, reducing the
   activity of the γ-secretase, suggesting that the dominant inheritance
   results from reductions in the activity of this critical protein and
   its associated complex.[64]^11 Recently, the intersection of early and
   late-onset AD genetics was further highlighted by the discovery of a
   rare APOE variant, ApoE3 Christchurch (APOE3ch), that was associated
   with the protection of cognitive decline in a 70-year-old with a
   disease-causing PSEN1 mutation.[65]^12 Homozygosity for APOE3ch in this
   individual was also associated with reduced accumulation of Tau,
   whereas amyloid deposition was unabated and significantly higher than
   in younger cohorts with cognitive symptoms but lacking the APOE3ch
   variant. The amino acid change in ApoE3ch provided a clue as to the
   mechanism of this variant providing protection against AD symptoms.
   APOECh bears a change in a lysine residue within the
   heparan-sulfate-binding domain of the protein. Biochemical studies with
   purified APOE proteins demonstrated that APOECh showed dramatically
   reduced affinity for heparin, a highly sulfated form of heparan
   sulfate. Furthermore, APOE4, APOE3, and APOE2 proteins showed heparin
   affinities correlated with their relative risk for developing
   late-onset AD, with higher affinities associated with greater risk.
   Finally, a recent report showed that mice expressing a humanized
   ApoE3ch variant showed altered microglial responses and suppression of
   Aβ-induced Tau seeding and spread.[66]^13 These results suggest that
   heparan-sulfate-modified proteins participate in APOE functions that
   can modulate AD pathogenesis and influence PSEN1-mediated events.

   Heparan-sulfate-modified proteins or proteoglycans (HSPGs) are abundant
   molecules of the cell surface and extracellular matrix. The glycan
   modifications have an important role in their function, serving as
   binding sites for a wide array of proteins, including growth factors
   and growth factor receptors. HSPGs modulate the activity of many
   signaling events, generally serving to increase the activity of
   secreted protein growth factors. HSPGs are an important class of
   coreceptors at the cell surface, promoting the assembly of signaling
   complexes. HSPGs are known to affect signaling via PI3kinase, Akt, and
   mitogen-activated protein kinase (MAPK) systems.[67]^14^,[68]^15 HSPGs
   regulate Wnt, Hh, fibroblast growth factor (FGF), and bone
   morphogenetic protein (BMP) signaling during development and play a
   role in both endocytosis and exocytosis of a variety of
   molecules.[69]^16^,[70]^17^,[71]^18^,[72]^19

   We have previously shown that reducing heparan sulfate chain length or
   sulfation increases autophagy flux in muscle and fat body cells in
   Drosophila.[73]^20 Modest reductions in two critical HS biosynthetic
   enzyme-encoding genes, sulfateless (sfl) and tout velu (ttv), either by
   mutation or RNAi, also rescue cell degeneration in a model of AD
   mediated by expression of a dominant-negative form of presenilin in the
   retina.[74]^21 Cell death and mitochondrial dysmorphology in parkin
   mutants, a model for Parkinson, are also suppressed by reducing HS
   function, and rescue depends on an intact autophagy system.[75]^21
   Autophagy flux and mitochondrial function, both governed by HSPG
   activity, are compromised by PSEN1
   deficits.[76]^22^,[77]^23^,[78]^24^,[79]^25 In this study, we examine
   in detail the effects of HSPGs on autophagy, mitochondrial function,
   and intracellular lipid droplets or liposomes in human cells and mouse
   astrocytes. Transcriptome changes in human cells defective for HS
   synthesis reflect the cellular changes in autophagy, mitochondrial
   biogenesis, and lipid synthesis or transport. AD-associated genes
   identified by large GWAS studies are significantly overrepresented in
   differentially expressed genes in cells lacking HS synthesis. We also
   show that reducing HSPG-mediated signaling affects mitochondrial
   function, autophagy, and lipid metabolism in Drosophila in a manner
   that counters deficits produced by presenilin knockdown and can
   suppress cellular phenotypes and molecular signatures in animals with
   compromised presenilin.

Results

Transcriptomics of human Hep3B cells defective in HS synthesis shows broad
effects on the expression of genes affecting lipid metabolism, autophagy,
mitochondria, and AD-associated genes

   We have previously reported that reducing heparan sulfate synthesis or
   modification alters autophagy flux and both function and morphology of
   mitochondria.[80]^20^,[81]^21^,[82]^26 Reducing HS biosynthesis can
   suppress mitochondrial dysmorphology and muscle cell loss in a fly
   model of Parkinson disease achieved by mutations affecting parkin, the
   fly homolog of human PARK2. To determine if HS synthesis serves a
   conserved role in cellular and molecular events associated with AD
   pathology, we conducted RNA sequencing (RNA-seq) profiling of a human
   cell line (Hep3B) bearing loss-of-function mutations in EXT1,[83]^27 a
   gene required for HS polymerization. Although this hepatocyte-derived
   cell line does not represent a cell type directly involved in AD, lipid
   metabolism and transport are a vital and conserved function implicated
   in AD, and these cells provide the means of exploring regulatory
   networks affected by heparan sulfate biosynthesis. A large number of
   differentially expressed genes was evident, with 1,949 genes showing
   increased expression and 1,294 with decreased expression in EXT1−/−
   cells (p < 0.05, FC ≥ 2.0). We examined genes involved in key cellular
   processes affected in AD, including autophagy, lipid metabolism, and
   mitochondria biogenesis and function.

   Key components of lipid metabolism are represented in the statin drug
   pathway ([84]https://www.wikipathways.org/instance/WP430) and provide a
   broad gene set for assessing differential gene expression affecting
   lipid. Four gene products in this pathway provide critical functions in
   the biosynthesis of cholesterol: (1) HMG-CoA reductase
   (3-hydroxy-3-methyl-glutaryl-coenzyme A reductase (HMGCR), the
   rate-limiting enzyme in the mevalonate pathway; (2)
   farnesyl-diphosphate farnesyltransferase 1 (FDFT1), the first specific
   enzyme in cholesterol biosynthesis, catalyzing the dimerization of two
   molecules of farnesyl diphosphate in a two-step reaction to form
   squalene; (3) squalene monooxygenase (also called squalene epoxidase,
   SQLE), catalyzing the first oxygenation step in sterol biosynthesis and
   considered one of the rate-limiting enzymes in this pathway; and (4)
   sterol O-acyltransferase 1, a resident enzyme of the endoplasmic
   reticulum that catalyzes the formation of fatty acid-cholesterol esters
   (SOAT1). Transcripts encoding all four of these key enzymes are
   dramatically and significantly reduced in EXT1 KO Hep3B cells
   ([85]Table S1), suggesting a significant suppression of cholesterol and
   cholesterol-ester biosynthetic capability. SREBF2, encoding a
   transcription factor that upregulates several genes affecting
   cholesterol biosynthesis, is significantly downregulated in EXT1 KO
   cells. Other genes in this pathway involved in lipid
   transport/lipoprotein assembly are also significantly downregulated in
   EXT1−/− cells including APOE, APOC2, APOC1, and APOB. These proteins
   are involved in the export of lipids from lipid stores into the
   bloodstream and tissues. Another gene affecting lipid
   transport[86]^20^,[87]^21 (MTTP), a protein required for the assembly
   of β-lipoprotein synthesis and lipid export from cells, is also
   markedly downregulated in EXT1−/− cells.

   APOA4, encoding a protein promoting the transport of cholesterol from
   peripheral tissues to HDL, is upregulated in EXT1−/− cells. Other lipid
   transporters with an increased expression upon knockout of EXT1 include
   ABCA7, ABCA10, and ABCA12. ABCA7 is notable for the strong linkage of
   variants in genome-wide association study (GWAS) analysis with AD
   susceptibility,[88]^28^,[89]^29 and loss-of-function alleles are
   associated with an increased risk of AD in both European- and
   African-descent populations.[90]^30^,[91]^31 Another gene of interest
   in lipid biology is PCSK9, which encodes a regulator of low-density
   lipoprotein (LDL) receptor levels on the cell surface by altering
   intracellular trafficking. PCSK9 inhibitors are now FDA-approved drugs
   for lowering LDL levels in plasma. This gene is significantly
   downregulated in EXT1 KO cells. MFSD2 lysolipid transporter A (MFSD2A),
   another gene affecting lipid transport, is dramatically reduced in
   expression levels in EXT1−/− cells. Mice with targeted knockout of
   MFSD2A are leaner and have reduced serum, hepatic, and adipocyte
   triglyceride levels.[92]^32 It is intriguing that in the brains of the
   APP/PS1 mice, expression of the proprotein convertase subtilisin/kexin
   type 9 (PCSK9) and apolipoprotein E (APOE) were all elevated,[93]^33
   and both PCSK9 and APOE are reduced in the Hep3B EXT1 KO cells.

   Recent work has provided additional evidence that lipid accumulation is
   a critical element of AD pathogenesis.[94]^34 Single-nucleus RNA
   sequencing (RNA-seq) analysis of brain cells isolated from AD and
   control samples revealed changes in gene expression that identified a
   class of microglia cells overrepresented in AD brains. These
   disease-associated microglia are notable for their increased expression
   of ACSL1, acyl-CoA synthetase long-chain family member, encoding an
   enzyme essential for lipid droplet biogenesis. Other genes affecting
   metabolic state, NAMPT and DPYD, are also elevated.[95]^34 We find that
   all three of these genes indicative of an altered lipid metabolic state
   in AD-associated microglia are significantly downregulated in EXT1−/−
   Hep3B cells (see [96]Table S1, ACSL1, log2 fold change = −1.2, p adj =
   9.96e^−35). These data emphasize the critical role HS-modified proteins
   have in lipid metabolism, including the formation of intracellular
   lipid stores, lipid droplets. The gene expression changes in cells
   lacking heparan sulfate biosynthesis are all in the direction of
   countering the changes seen in AD.

   Earlier work showed that reducing heparan sulfate chain elongation or
   sulfation results in increased autophagy flux to the lysosome in
   Drosophila.[97]^20^,[98]^21 Genes critical for autophagy function are
   significantly elevated in EXT1 KO cells compared to controls, including
   SQSTM1, ULK1, PINK1, ATG9B, SRC, and PRKAG2. GSEA of Hep3B EXT1 KO
   compared to Hep3B+/+ cells also suggests activation of autophagy.
   Reactome pathways, selective autophagy, and aggrephagy are enriched
   (false discovery rate [FDR] = 0.24, and 0.25, respectively) and several
   other related gene sets show similar levels of increased component
   expression (autophagy and lysosome vesicle biogenesis) ([99]Table S3).
   Other pathways with downregulation of gene sets or changes with false
   discovery rate between 0.25 and 0.3 are provided in [100]Tables S4 and
   [101]S5.

   PPARGC1A, a master regulator of mitochondrial biogenesis, and CHD9, a
   coactivator of PPARGC1A, were significantly increased in EXT1−/− cells
   compared to controls. This is consistent with our data showing
   increased MitoTracker signal in NDST1 −/− cells ([102]Figure 1) and
   MitoBlue in EXT1 −/− Hep3B cells ([103]Figure 2). PPARGC1A is required
   for maintaining normal levels of mitochondrial gene expression and
   oxidative metabolism.[104]^35 PPARGC1A also modulates gluconeogenesis
   and the capacity for cells to engage in glucose or fatty acid
   metabolism depending on nutrient availability.[105]^36 CHD9 is
   responsible for remodeling chromatin and is a coactivator of
   PPARGC1A.[106]^37

Figure 1.

   [107]Figure 1
   [108]Open in a new tab

   Flow cytometry analysis of A375 human melanoma cells with NDST1
   knockout

   (A) Flow cytometry analysis of A375 wild-type and NDST1 knockout mutant
   (NDST1^−/−) cell lines after 1-h incubation at 37°C with biotinylated
   guanidinoneomycin (GNeo-biotin) conjugated to streptavidin-Cy5. NDST1
   N-deacetylates and adds N-sulfate to glucosamine residues of heparan
   sulfate, which is required for GNeo binding[109]^46 (t test, n = 4,
   ∗∗∗∗p < 0.0001).

   (B) Flow cytometry analysis of A375 wild-type cells after incubation
   with mTOR inhibitor, rapamycin (10 μM), for 24 h and subsequent 1-h
   incubation with MitoTracker Red. Rapamycin significantly increases the
   MitoTracker Red signal (t test, n = 4, ∗∗∗p < 0.001).

   (C) Flow cytometry analysis of A375 wild-type and NDST1 mutant cells
   after 1-h incubation with MitoTracker Red (t test, n = 4, ∗∗p < 0.01).

   (D) Flow cytometry analysis of A375 wild-type and NDST1 mutant cells
   after a 30-min incubation at 37°C with LipidTox Green (t test, n = 4,
   ∗∗p < 0.01). All error bars represent SEM.

Figure 2.

   [110]Figure 2
   [111]Open in a new tab

   Reduced expression of heparan sulfate biosynthetic enzymes in human
   cell lines significantly increases mitochondria, reduces intracellular
   lipid droplets, and increases oxygen consumption rate measured by
   Seahorse Analyzer

   (A) Representative confocal fields of Hep3B cells stained with MitoBlue
   fluorophore. Knockout of EXT1 significantly increases the number of
   MitoBlue-positive punctae in Hep3B cells (t test, Hep3B EXT1+/+, n = 17
   fields, Hep3B EXT1−/−, 19 fields; ∗∗∗p < 0.001). Error bars show SEM.

   (B) Representative confocal fields of HEK293T cells with stably
   integrated lentiviral vectors expressing scrambled (control), EXT1, or
   NDST1-directed RNAi. Intracellular lipid droplets assessed by staining
   with LipidTox Red. Graph shows average pixel intensity frequency
   distribution for the three HEK293 cell lines shown. Knockdown of EXT1
   and NDST1 significantly decreases LipidTox Red staining intensity in
   HEK293T cells (Kolmogorov-Smirnov test, ∗∗∗p < 0.001). Graph shows
   pixel intensity averages for a sampled area with error bars
   representing the SEM. The experiment was conducted in quadruplicate.

   (C) Representative graph and quantification of A375 cells undergoing a
   Mito Stress Test. Knockout of NDST1 significantly increases the
   baseline and maximal oxygen consumption rate (OCR) of A375 cells (t
   test, n = 30 (10 wells x 3-time points) per genotype, ∗p < 0.05,
   ∗∗p < 0.01, ∗∗∗p < 0.001). The first three time points were taken under
   baseline culture conditions, time points 4–6 after oligomycin addition,
   an ATPase inhibitor, time points 7–9 after FCCP addition, which
   disrupts the proton gradient, and time points 10–12 showing OCR after
   block of electron transport with rotenone and antimycin A addition. The
   experiment was conducted in triplicate. Box and whisker plot shown in
   right panel, outlier points (<Q1 minus 1.5 IQR, >Q3 plus 1.5 IQR)
   excluded.

Effects of EXT1 knockout on Alzheimer disease susceptibility genes

   AD is associated with changes in lipid metabolism and transport, and
   Hep3B cells are derived from hepatocytes, cells critical in these
   processes. We therefore were interested in differential gene expression
   profiles in Hep3B EXT1−/− cells-affected loci identified as AD
   susceptibility genes and their relation to functional gene groups
   implicated in AD pathogenesis. We selected genes identified by GWAS
   analysis classified at Tier 1 or Tier 2, reflecting different levels of
   association across six GWAS studies performed in populations of
   European ancestry and published since 2019[112]^38 ([113]Table S2).
   Overall, 46% of Tier 1 genes showed significant EXT1-dependent changes
   in expression, and Tier 2 representation among the differentially
   expressed genes was 57%. This reflects a statistically significant
   enrichment of AD-susceptibility genes among the differentially
   expressed genes affected by EXT1−/− in Hep3B cells (Chi-square with
   Yates correction, p < 0.0001). g:profiler assessment of the Tier 1
   genes affected in EXT1−/− cells showed significant representation of
   amyloid precursor protein (APP) catabolic process (p-adjusted =
   6.8 × 10^−7). These findings provide evidence for the capacity of
   heparan sulfate function to preferentially affect genes involved in AD
   susceptibility.

   Large and significant downregulation was found for FERMT2 and APOE
   ([114]Table S2). APOE variants represent a substantial genetic risk for
   LOAD[115]^39 and the discovery of the ApoE3ch variant that both rescues
   PSEN1-mediated EOAD and shows greatly reduced heparin binding suggests
   an important relationship between heparan sulfate function and AD
   pathology.[116]^12 FERMT2 encodes a scaffolding protein affecting
   integrin and TGF-β signaling and reducing expression in neural stem
   cells suppresses accumulation of phosphorylated Tau and extracellular
   APP-derived peptides.[117]^40 Two other genes with differential
   expression in EXT1−/− cells include BIN1 and RIN3. BIN1 and RIN3
   interact directly and together affect Rab5-mediated
   endocytosis.[118]^41 GWAS analyses have identified both as AD risk
   loci, and RIN3 levels have been reported to increase in the hippocampus
   and cortex of APP/PS1 mice.[119]^42 Elevated expression of RIN3 in
   basal forebrain cholinergic neurons (BFCNs) from E16 embryos was
   associated with enlargement of early endosomes, suggesting an
   endocytosis abnormality. We have previously reported that compromising
   heparan sulfate biosynthesis increases activity-dependent endocytosis
   at the Drosophila neuromuscular junction,[120]^43 and recent studies of
   tissue culture cells with compromised HS synthesis showed increases in
   endocytosis of GFP-tagged dextran, a measure of clathrin-dependent
   endocytosis (data not shown). These findings indicated that lowering
   heparan sulfate-dependent signaling can increase endocytosis, a process
   that is compromised in AD pathology.

Conservation of heparan sulfate functions in mitochondrial and lipid
metabolism in human cells and mouse astrocytes

   We have examined the role of HSPGs in mitochondrial function and lipid
   metabolism in three different human-transformed cell lines as well as
   mouse primary astrocytes, derived from either wild-type mice or those
   bearing humanized variants of APOE that affect AD risk susceptibility
   in humans. Two independent and previously characterized human cell
   lines were examined: human melanoma A375 cells bearing CRISPR-mediated
   knockout of NDST1[121]^44 and a hepatocellular carcinoma cell line,
   Hep3B, with a directed knockout of EXT1,[122]^27 the cell line used for
   the transcriptomics analysis described earlier. Experiments were also
   carried out with HEK293T cells with lentiviral-mediated knockdown of
   either EXT1 or NDST1. Mouse astrocytes were isolated from wild-type
   animals or those bearing humanized APOE3 or APOE4 variants.[123]^45
   Knockdown of Ndst1 in these mouse astrocytes was achieved with
   transient lentiviral expression in culture.

   Knockout of NDST1 in A375 cells has a dramatic effect on the binding
   and internalization of GNeo, a modified neomycin analog with high
   affinity and specificity for heparan sulfate.[124]^44
   Fluorescence-activated cell sorting (FACS) measures of internalized
   GNeo of A375 NDST1−/− cells compared to wild-type A375 showed an
   approximately 3-fold reduction in internalized GNeo ([125]Figure 1A).
   Autophagy induction has a profound effect on mitochondria levels, and
   the response of A375 cells to rapamycin, a TOR inhibitor, showed a
   significant increase in FACS measures of the mitochondrial marker,
   MitoTracker ([126]Figure 1B). These data demonstrate this cell line
   displays a vigorous autophagy response. The effect of NDST1, and hence
   heparan sulfate structure, on mitochondria was assessed similarly using
   FACS analysis of MitoTracker levels, which increased 1.5-fold,
   comparable to the effects of rapamycin ([127]Figure 1C). We have
   previously shown using transmission electron microscopy that A375
   NDST1−/− cells have altered mitochondrial morphology,[128]^26 and the
   MitoTracker results support the conclusion that compromising the
   structure of heparan sulfate has a profound effect on mitochondria.
   Induction of autophagy is also associated with reductions of
   intracellular neutral lipids, and we employed FACS analysis of LipidTOX
   staining to determine if the state of heparan sulfate affected lipid
   metabolism in A375 cells. LipidTOX level was significantly reduced in
   NDST1−/− cells compared to wild-type controls ([129]Figure 1D).

   The effect of heparan sulfate biosynthesis on mitochondria was also
   assessed in Hep3B cells, comparing wild-type to EXT1 knockout cells. As
   observed for A375 cells with compromised NDST1 function, eliminating
   the activity of EXT1, a chain-elongating heparan sulfate copolymerase,
   significantly increased the mitochondrial signal (MitoBlue,
   [130]Figure 2A). Metric analysis of confocal images showed significant
   increases in mitochondrial number and size in EXT1−/− Hep3B cells
   compared to wild-type controls (data not shown).

   HEK293T cells with integrated lentiviral vectors targeting EXT1 or
   NDST1 were used to evaluate the level of intracellular lipid when
   heparan sulfate elongation or sulfation was reduced ([131]Figure 2B).
   LipidTOX staining was significantly lowered upon knockdown of either of
   these two heparan sulfate biosynthetic enzyme-encoding genes. This is
   notable as a change in the pixel intensity distribution toward a
   greater representation of low-intensity pixels upon EXT1 or NDST1
   knockdown compared to a scramble sequence lentiviral vector
   ([132]Figure 2B).

   The results described earlier document morphological changes for
   markers of mitochondria and neutral lipids stored in intracellular
   lipid droplets upon reduction of key heparan sulfate biosynthetic
   enzyme-encoding genes. To assess the functional consequences of changes
   in heparan sulfate levels or structure on mitochondria, oxygen
   consumption rates were measured with a Seahorse analyzer and a
   mitochondrial stress test. This test provides a measure of both
   baseline and maximal respiratory capacity. Significant increases in
   both baseline and maximal respiration rates were observed for A375
   Ndst1−/− cells compared to wild type A375 ([133]Figure 2C) and Hep3B
   Ext1−/− compared to wild-type Hep3B ([134]Figure S1). These findings
   demonstrate that heparan-sulfate-modified proteins have a functional
   impact on the activity of mitochondria, with loss or partial loss of
   heparan sulfate biosynthesis increasing mitochondrial respiratory
   capacity.

Knockdown of Ndst1 in mouse astrocytes reduces intracellular lipid levels

   To evaluate the capacity of heparan sulfate to modulate attributes in
   cells affected by AD pathology, we examined intracellular lipids in
   wild-type mouse astrocytes as well as those expressing humanized APOE3
   and APOE4 variants. Previous work has shown that astrocytes provide
   important metabolic support for neurons, and APOE variants affect this
   important capacity.[135]^45 Astrocytes expressing the APOE4 variant
   showed elevated accumulation of intracellular lipid compared to
   wild-type or APOE3-expressing cells. Knockdown of Ndst1 (to 60% of
   wild-type levels, measured by qPCR) using a lentiviral vector reduced
   intracellular lipid significantly in wild-type astrocytes, representing
   a greater than 4-fold reduction in lipid droplet volume
   ([136]Figure 3A). This was accompanied by significant reductions in
   Perilipin 2 mRNA, the gene encoding a protein localized to the
   phospholipid shell of lipid droplets, compared to astrocytes
   transfected with a lentivirus bearing a scrambled sequence
   ([137]Figure 3A). Lentivirus-directed knockdown of Ndst1 in astrocytes
   from mice expressing humanized APOE3 and APOE4 variants both display
   comparable and significant reductions in intracellular lipid compared
   to controls expressing a scrambled sequence lentivirus
   ([138]Figure 3B).

Figure 3.

   [139]Figure 3
   [140]Open in a new tab

   Knockdown of Ndst1 in primary mouse astrocytes lowers levels of
   intracellular liposomes

   (A) Micrographs showing levels of lipid droplets (LD), a marker for
   lentiviral transfection (green), and nuclear stain (DAPI, blue) in
   control, cells transfected with either a scramble sequence shRNA vector
   or one targeting Ndst1. Ndst1 shRNA produced reduced levels of LD
   number and volume (panels on the right), as well as decreases in Plin2
   expression measured by qPCR.

   (B) The effect of Ndst1 shRNA was evaluated on astrocytes from animals
   expressing humanized variants of APOE3 and APOE4. LD volumes were
   significantly reduced by Ndst1 knockdown compared to scramble controls
   in astrocytes from either of these two genotypes [images show
   individual cells, nuclei (blue), and LD (red), and graphs provide
   quantitation of LD volumes. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001). Error
   bars represent SEM.

Presenilin/Psn deficit in Drosophila neurons produces degeneration suppressed
by reducing heparan sulfate modification

   Presenilin is one component of an enzyme complex, γ-secretase, a
   protease with a number of important substrates, including amyloid
   precursor protein (APP) and Notch.[141]^47^,[142]^48 Mutations
   affecting presenilin-1 (PS1) function are responsible for a familial,
   early onset form of AD. Although PS1 mutations show dominant
   inheritance, these variants are loss or partial loss-of-function
   variants, established by the demonstration that approximately 90% of
   138 mutant proteins derived from human-disease-associated variants and
   reconstituted into an active enzyme complex showed reduced Aβ peptide
   production.[143]^11^,[144]^49 These studies demonstrate that PS1
   partial loss-of-function models are relevant to understanding AD
   mechanisms. In vertebrate systems, compromising presenilin function
   results in mitochondrial dysfunction, accumulation of intracellular
   lipid, and disruption of autophagosome to lysosome trafficking.[145]^50
   With these observations in mind, the cellular and molecular phenotypes
   of partial loss of function changes in Drosophila γ-secretase were
   examined. These experiments were made possible by the development of
   effective RNAi constructs directed against two essential subunits of
   the γ-secretase complex, Presenilin (Psn), the catalytic subunit, and
   Nicastrin (Nct), an evolutionarily conserved component proposed to
   affect substrate selection.[146]^51 All four protein components of
   human γ-secretase are represented in Drosophila, and these two
   homologous enzyme complexes share common substrates. RNAi knockdown of
   either Psn or Nct mRNAs in neurons produces neurodegeneration in the
   adult brain and retina and has effects even when functional changes are
   conditionally restricted to adult neurons.[147]^52

   We began by assessing the phenotypic severity of three RNAi constructs,
   two different Psn and one Nct-directed UAS-RNAi vector expressed in the
   brain using a neuron-specific Gal4 driver, elav-Gal4. Animals were
   evaluated for a startle-induced negative geotaxis assay
   ([148]Figure 4E), a locomotor reactivity response where animals move
   upward after a gentle tapping to the bottom of a plastic tube.[149]^53
   These findings confirmed previously published phenotypes, with shPsn2
   showing more severe behavioral deficits and shPsn3 demonstrating a
   modest but significant deficit in behavior when compared to controls
   ([150]Figure 4E, shPsn3 data shown). Expression of Nct RNAi in neurons
   when animals were reared at 25°C was nearly completely lethal. The
   majority of the analyses described below therefore focused on shPsn3,
   the RNAi construct with an intermediate phenotype.

Figure 4.

   [151]Figure 4
   [152]Open in a new tab

   Reduction of sfl function reduces Psn-knockdown-induced
   neurodegenerative vacuole formation and apoptosis levels

   (A) Representative images of Drosophila brains stained with Alexa Fluor
   594-Phalloidin and imaged by confocal microscopy. Top row panels show
   low magnification images of the entire brain; bottom row: higher
   magnification (see scale bars). White arrows indicate neurodegenerative
   vacuoles. Black arrows indicate flares showing the higher staining
   intensity of phalloidin associated with and near vacuoles. Black
   arrowheads indicate tracheoles, air conducting tubules, discernable in
   the serial confocal sections.

   (B) Bar graphs showing average vacuole area and count per brain in
   female Drosophila. Vacuole area and number of vacuoles per brain are
   significantly increased in shPsn3 animals compared to controls and
   significantly elevated compared to shPsn3;sfl^RNAi animals 7 days
   post-eclosion (ANOVA Tukey test: w^RNAin = 10, shPsn3 n = 24,
   shPsn3;sfl^RNAin = 14, sfl^RNAi = 20; ∗p < 0.05, ∗∗p < 0.01,
   ∗∗∗p < 0.001); error bars represent standard error of the mean (SEM).
   Open symbols denote significance values for 7-day-old animals and
   closed symbols for 30-day-old animals. Number of symbols indicates
   levels of significance between the two groups with those symbols.
   Vacuole area and number of vacuoles per brain are significantly
   increased in shPsn3 animals 30 days post-eclosion compared to controls,
   and double shPsn3;sfl^RNAi animals showed a significant reduction in
   vacuole number compared to shPsn3. Vacuole area was reduced in
   shPsn3;sfl^RNAi animals compared to shPsn3 but did not achieve
   significance (ANOVA Tukey test: w^RNAin = 13, shPsn3 n = 20,
   shPsn3;sfl^RNAin = 8, sfl^RNAi = 20; ∗p < 0.05, ∗∗p < 0.01,
   ∗∗∗p < 0.001).

   (C) Representative images of Drosophila brains stained with the
   apoptosis marker, acridine orange. Micrographs are pixel maximum
   intensity stacks from confocal serial sections.

   (D) Bar graphs quantifying apoptotic staining in optic lobes of adult
   female Drosophila, age matched 1–3 weeks. Apoptosis is significantly
   higher in shPsn3 than controls (w^RNAi) and significantly lower in
   shPsn3;sfl^RNAi compared to shPsn3 animals (ANOVA Tukey test: w^RNAin =
   10, shPsn3 n = 10, shPsn3;sfl^RNAin = 10, sfl^RNAin = 10; ∗p < 0.05,
   ∗∗p < 0.01, ∗∗∗p < 0.001).

   (E) Bar graphs demonstrating average performance on negative geotaxis
   assays in adult female Drosophila, aged 4–7 days. Performance is
   statistically significant between w^RNAi and all other genotypes (ANOVA
   Tukey test: w^RNAin = 25, shPsn3 n = 20, shPsn3;sfl^RNAin = 25,
   sfl^RNAi = 35; ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001). shPsn3;sfl^RNAi
   show significantly improved geotaxis compared to shPsn3 animals
   (p < 0.05).

   Earlier work has shown the capacity of reduced heparan sulfate
   biosynthesis to suppress cell degeneration, restore dysmorphic
   mitochondria, and reduce accumulation of ubiquitin-modified proteins in
   parkin mutants, a Drosophila model of PARK2-mediated Parkinson
   disease.[153]^21 Partial reductions in the function of either of two
   genes involved in heparan sulfate biosynthesis, sulfateless (sfl),
   encoding a homolog of N-deacetylase N-sulfotransferase (NDST1), or tout
   velu (ttv), a glycosyl transferase responsible for heparan sulfate
   chain elongation (homolog of human EXT1), had dramatic effects on
   parkin mutants, restoring muscle cell morphology and flight
   activity.[154]^21 We, therefore, examined the capacity of reduced
   heparan sulfate biosynthesis to affect neurodegeneration in the brain
   of shPsn3 RNAi-expressing animals. Adult flies expressing shPsn3 RNAi
   under the control of elav-Gal4 were aged for 7 or 30 days prior to
   phalloidin and Hoechst staining of adult heads to visualize actin and
   nuclei in whole-brain preparations.[155]^54 Serial confocal sections of
   the entire brain were taken to identify vacuoles, areas of acellularity
   produced by neuronal loss ([156]Figure 4A, white arrows). Consistent
   with earlier published work, expression of shPsn3 in the adult brain
   produced a marked increase in neurodegeneration measured as the area of
   vacuolization or number of vacuoles per brain compared to control
   animals expressing a w^RNAi construct (the white gene governs eye
   pigmentation) ([157]Figures 4A and 4B). Knockdown of sfl transcripts
   with a sfl^RNAi construct showed no difference from control animals
   with respect to levels of degenerative vacuoles. However, RNAi of sfl
   significantly reduced the level of degeneration in the brain mediated
   by shPsn3 knockdown in animals aged 7 or 30 days ([158]Figure 4B,
   compare shPsn3 to shPsn3;sfl^RNAi). These findings demonstrate that Psn
   neurodegenerative phenotypes can be suppressed by modest reductions in
   sfl function and hence HSPG-mediated activities.

   Disturbed presenilin function in Drosophila produces an elevated level
   of apoptosis,[159]^55 reflecting activities observed in human cells
   bearing PSEN1 mutations.[160]^56^,[161]^57 We therefore explored the
   capacity of reduced sfl function to suppress apoptosis mediated by Psn
   knockdown. A fluorescent vital dye, acridine orange (AO), is a widely
   used marker for apoptosis in Drosophila[162]^58 and vertebrate
   cells[163]^59 and provided a clear indication of elevated apoptosis in
   adult brains with Psn knockdown ([164]Figures 4C and 4D). The level of
   AO-positive cells was dramatically reduced in shPsn3 animals with
   coexpression of sfl^RNAi in neurons, reflecting a remarkable capacity
   of reduced HS sulfation to abrogate apoptosis in adult brains
   ([165]Figures 4C and 4D). Behavioral assessment of adult animals also
   showed suppression of presenilin knockdown-mediated defects by reducing
   sfl mRNA levels. shPsn3;sfl^RNAi animals performed significantly better
   in the geotaxis assay than shPsn3 but were not restored to control
   levels ([166]Figure 4E).

Cellular processes regulated by Psn and heparan-sulfate-modified proteins

   The capacity of reduced sfl function to rescue neurodegeneration and
   behavioral deficits mediated by presenilin deficits prompted an
   analysis of cellular phenotypes with a focus on mitochondria,
   intracellular lipid, and autophagosome-related structures, all features
   known to be disrupted by PS1 mutations in humans.[167]^50^,[168]^60
   Given the documented effects of PS1 mutations on many cell types and
   their involvement in conserved cellular functions, we examined the
   cellular phenotypes of presenilin and sfl knockdown in the larval fat
   body. The fat body is an essential metabolic regulatory organ where
   autophagy serves a critical role in balancing anabolic and catabolic
   processes.[169]^61 It therefore serves as an ideal cell type to monitor
   conserved elements of lipid metabolism, autophagy, and mitochondrial
   function affected by presenilin and its modulation by
   heparan-sulfate-modified coreceptors. The fat body has also shown to be
   an integral element of mechanisms affecting aging,[170]^62 a topic
   relevant to the studies here.

   Fat body cells are large (∼60 μm) polyploid cells where intracellular
   organelles can be visualized in detail with specific fluorescent
   markers using serial confocal microscopy. Genes affecting mitochondria
   and neurodegeneration, including parkin and pink1, have been identified
   using RNAi screens and visualization of mitochondria in fat body
   cells,[171]^63 demonstrating the utility of this preparation for
   assessing cellular phenotypes relevant to neural function and disease.
   Selective knockdown of genes in fat body cells is possible using
   UAS-RNAi-producing transcripts expressed under the transcriptional
   control of r4Gal4, a fat-body-specific Gal4 transgene. r4Gal4 produces
   a reduction of Psn and sfl mRNA by 32% and 28% of levels measured in
   control animals (w^RNAi), respectively, determined by RNA-seq (see
   later discussion).

Psn and sfl knockdown in Drosophila fat body have opposing effects on
mitochondrial structure and liposome organization

   In order to evaluate the effect of presenilin function on mitochondria
   and intracellular lipids, Psn-RNAi expression was directed to the fat
   body. Mitochondria and intracellular lipids were visualized using the
   fluorescent probes MitoTracker Red and LipidTOX by serial confocal
   microscopy, respectively. Psn knockdown had profound effects on the
   level and morphology of MitoTracker-labeled organelles in fat body
   cells ([172]Figures 5A and 5B). MitoTracker signal intensity was
   significantly reduced in shPsn3 preparations compared to control
   animals expressing w^RNAi ([173]Figure 5B). The size of
   MitoTracker-labeled organelles was also affected by Psn knockdown
   ([174]Figure 5B), indicating that mitochondria are smaller upon
   reductions of Psn function. The effect of shPsn3 on mitochondrial
   morphology was confirmed by transmission electron microscopy (TEM) and
   serial block face scanning electron microscopy (SBFSEM), revealing
   small, ellipsoid mitochondria in animals with shPsn3, compared to
   larger, branched mitochondria in control preparations (RNAi of w gene)
   or animals with sfl^RNAi ([175]Figure 5E).

Figure 5.

   [176]Figure 5
   [177]Open in a new tab

   psn and sfl function alter mitochondrial number and morphology in fat
   body cells, and sfl knockdown suppresses psn^RNAi-mediated
   abnormalities

   Mitochondria labeled with MitoTracker Red and visualized by confocal
   serial section.

   (A) Images are projections of a stack of confocal images, showing the
   organization and morphology of mitochondria. The red dashed circle
   indicates the margins of one cell. White arrows mark individual
   mitochondria. Spheroid-shaped areas without MitoTracker Red signal are
   the large liposomes found in fat body cells.

   (B) The measure of mitochondrial size shows Psn knockdown significantly
   reduces the size of MitoTracker-positive structures compared to
   controls (w^RNAi), whereas sfl^RNAi has the opposite effect (increased
   size of MitoTracker-positive structures); error bars represent SEM (t
   test, p values are in relation to w^RNAi except where marked by
   bracket, ∗∗p < 0.01, ∗∗∗p < 0.001). sfl^RNAi rescues the mitochondrial
   phenotypes produced by psn knockdown, reflected both in morphology and
   size of MitoTracker-labeled structures (p values in relation to w^RNAi,
   ∗∗∗p < 0.001). Significant differences were also measured for pixel
   intensity, with shPsn3 showing reduced average pixel intensity compared
   to controls and sfl^RNAishPsn3 animals restoring pixel intensity to
   control levels (w^RNAi).

   (C) Reducing sfl function with a characterized P element mutant,
   sfl^03844 also rescues mitochondrial number and morphology in fat body
   cells with Psn knockdown. Mitochondria labeled with MitoTracker Red and
   shown in a single optical confocal section (individual mitochondria
   marked with arrow).

   (D) Measures of mitochondrial count and size shows Psn knockdown
   significantly reduces the size and number of MitoTracker-positive
   structures compared to controls (w^RNAi), whereas sfl^RNAi
   significantly increases the size and number of MitoTracker-positive
   structures compared to Psn knockdown (t test, n = 10, ∗p < 0.05,
   ∗∗p < 0.01).

   (E) Top row panels show serial block-face EM assembly of mitochondria,
   providing a 3D rendering of mitochondrial morphology. Mitochondria in
   shPsn3 are small spheroidal structures, whereas both control (w^RNAi)
   and sfl^RNAi show more elongated mitochondria. The TEM images in the
   second and third rows of images show mitochondria (highlighted in
   translucent color) in these genotypes. Of note is the very elongated
   mitochondrion found in a sfl^RNAi animal and the large, lobed
   mitochondrial found in sfl^RNAishPsn3 preparations.

   We have previously shown that reductions in heparan sulfate
   biosynthetic enzyme-encoding genes affect mitochondrial morphology in
   muscle, producing larger mitochondria via a process that requires an
   intact autophagy system.[178]^21 We observe a similar phenotype
   with RNAi of sfl in fat body cells, with a more intense MitoTracker
   signal and larger, labeled structures compared to w^RNAi controls
   ([179]Figures 5A and 5B). SBFSEM and TEM confirmed the organization of
   mitochondria detected with MitoTracker in sfl^RNAi animals, with some
   large mitochondria evident in electron microscopic (EM) sections
   ([180]Figure 5E).

   The morphology of liposomes was evaluated by confocal microscopy of
   neutral lipid fluorophore staining (LipidTOX) as well as TEM. Control
   animals (w^RNAi) showed very regular liposomes, largely filling the
   entire cell ([181]Figure 6A). shPsn3 animals had larger, spherical
   liposomes with an increase of irregularly shaped lipid-staining
   structures ([182]Figure 6A, red circled structures) compared to the
   control. TEM confirmed the presence of large and some fused liposomes
   in shPsn3 animals ([183]Figure 6C). Also apparent were spherical
   structures smaller than liposomes, intermixed among them, and evident
   by their absence of staining ([184]Figure 6A). These structures were
   not prevalent in control or sfl^RNAi animals. To determine the nature
   of these intracellular organelles, we examined the distribution of an
   autophagosome marker, GFP-Atg8a ([185]Figure 6B). Staining with both
   LipidTOX and GFP-Atg8a showed these features to be Atg8a-positive and
   hence derived from an autophagosome-related organelle (green
   arrowheads, [186]Figure 6B). The accumulation of these structures in
   shPsn3 fat body cells indicates a change in autophagosome structure or
   traffic in animals with Psn knockdown. TEM sections from shPsn3 animals
   also show an accumulation of double membrane organelles consistent with
   autophagosome morphology as well as multivesicular structures
   ([187]Figure 6B). These are cellular phenotypes characteristic of
   limited autophagosome maturation[188]^64 and further suggest that Psn
   knockdown compromises autophagy flux and results in an accumulation of
   Atg8a-positive structures. sfl^RNAi-expressing animals also showed
   changes in liposome morphology, with a lower density of
   LipidTOX-staining structure compared to controls ([189]Figure 6A) and a
   greater representation of smaller liposomes. These findings were
   confirmed by TEM analysis ([190]Figure 6C).

Figure 6.

   [191]Figure 6
   [192]Open in a new tab

   sfl^RNAi partially reverts psn-knockdown-induced alterations in
   liposome morphology

   (A) Representative images of fat body stained with LipidTox Red. Top
   row shows lower magnification image, whereas lower row shows higher
   magnification, more detailed view. Control fat body cells (w^RNAi) show
   large, regular liposomes (L) filling the cell body. shPsn3 results in
   changes in the spacing of liposomes and the appearance of irregularly
   shaped LipidTox staining structures (red circles and arrowheads).
   sfl^RNAi produces a reduced density of liposomes and also some
   irregularly shaped LipidTox staining structures. Arrows indicate
   vesicular-shaped gaps in the LipidTox Red signal (see details in C).
   The letter “L” labels liposomes.

   (B) Fat body cells stained with LipidTox in animals expressing
   Atg8a-GFP transgene and shPsn3. The vesicular structures devoid of
   LipidTox staining show a GFP signal, indicating an accumulation of
   Atg8a (green arrowheads). The far-right panel shows a TEM micrograph of
   a shPsn3 fat body cell with double membrane structures (white
   arrowheads), consistent with autophagosome morphology (relatively
   electron-lucent compared to autolysosomes [AL] as well as
   multi-vesicular bodies (arrows). These structures are found in fat body
   cells with compromised autophagy flux.[193]^64

   (C) TEM micrographs of fat body cells, showing the organization of
   autolysosomes (ALs) and liposomes (L). ALs appear as electron-dense
   structures, 2–5 μm in control animals (w^RNAi) but smaller and less
   electron-dense in animals with RNAi of Psn. Shown are TEM images from
   two independent Psn RNAi lines (shPsn2, shPsn3). Smaller ALs indicated
   by white arrows. In sfl^RNAi animals, ALs display a different
   appearance, with electron-dense “cables” within a structure that can be
   seen fused with liposomes in some cases. Two micrographs of
   shPsn3,sfl^RNAi preparations are shown (both higher magnification),
   with some restoration of AL size and electron density, as well as the
   appearance of ALs fusing with liposomes (black arrows) as seen in
   sfl^RNAi.

   PSEN1 deficits are known to disrupt autophagosome traffic, and this
   cellular abnormality is common to several neurodegenerative disorders.
   Conversely, reductions in heparan sulfate chain length or sulfation
   increase the trafficking of autophagosomes to the lysosome.[194]^20 It
   was therefore of interest to examine autophagosome-related structures
   in fat body cells with presenilin or sfl knockdown. In TEM micrographs
   of wandering third-instar larvae fat body cells, autolysosomes are
   prominent structures with a characteristic size of 2–5 μm, a single
   encompassing membrane, and containing varied electron-dense
   materials.[195]^64 These are readily seen in TEM sections of control
   cells ([196]Figure 6C). In two different Psn RNAi-producing lines, TEM
   showed smaller, less electron-dense autolysosome structures, indicating
   that knockdown of Psn alters autophagy in these cells. sfl^RNAi had a
   very different effect on autolysosomes, producing a membrane bound
   structure with a network of electron-dense "cables" within. These
   autolysosomes were often seen in association with or having coalesced
   with liposomes ([197]Figure 6C). These structures are consistent with
   the induction of autophagy and the liberation of lipids from liposomes
   in these animals.[198]^65

Reduction of sfl function suppresses cellular abnormalities mediated by Psn
knockdown

   The cellular changes mediated by Psn or sfl knockdown were distinct and
   in an apparently divergent manner for both mitochondrial and liposome
   morphology. Autophagy-related structures were also altered in different
   ways in Psn versus sfl RNAi-bearing animals. We examined these
   subcellular features in cells with reductions in both Psn and sfl and
   found a restoration toward wild-type morphology. Mitochondria were
   significantly larger, and MitoTracker signal levels were higher in the
   shPsn3 sfl^RNAi double RNAi compared to shPsn3 alone ([199]Figures 5A
   and 5B). The capacity of reduced sfl expression to rescue shPsn3
   deficits was confirmed by testing the capacity of a characterized null
   allele of sfl, sfl^03844, to also suppress mitochondrial phenotypes
   mediated by reduced presenilin function ([200]Figure 5C). Likewise, the
   distribution of liposomes was restored in the double knockdown toward
   the wild-type pattern ([201]Figure 6A). TEM supported the same
   conclusion with the appearance of nearly normal size autolysosomes
   along with those resembling those characteristic of sfl knockdown in
   the double shPsn3 sfl^RNAi knockdown. This collection of phenotypes
   demonstrated the block or reversal of cellular deficits mediated by
   loss of Psn function by compromising the activity of
   heparan-sulfate-modified proteins.

Transcriptomics signatures of Psn, sfl, and Psn sfl double knockdown show
that sfl knockdown restores wild-type patterns for key metabolic regulatory
genes

   The capacity of reducing heparan sulfate function to rescue cellular
   abnormalities of presenilin deficits prompted a search for molecular
   signatures that accompany that rescue. This strategy could identify
   genes that are functionally important for the pathology mediated by
   presenilin knockdown. Quadruplicate samples of RNA were isolated from
   the fat body of climbing third-instar larvae of the following
   genotypes: shPsn3, sfl^RNAi, w^RNAi (controls), and shPsn3;sfl^RNAi
   double knockdown. Genes with significantly changed expression levels in
   either shPsn3 or sfl^RNAi compared to w^RNAi controls were identified,
   and those with opposing directions of change were determined
   ([202]Figure 7A). It is notable that the greatest number of genes with
   significant and opposite differential gene expression are downregulated
   in shPsn3 ([203]Figure 7B). For example, in the 200 genes with the
   greatest fold difference, 18 show opposite expression changes for the
   shPsn3 down, sfl^RNAi up gene set and only one gene for shPsn3 up,
   sfl^RNAi down group. This suggests that the capacity of sfl knockdown
   to rescue Psn deficits lies largely with the correction of shPsn3
   downregulated genes.

Figure 7.

   [204]Figure 7
   [205]Open in a new tab

   RNA-seq analysis of fat body transcriptome shows sfl^RNAi-mediated
   reversal of a set of differentially expressed genes from Psn knockdown
   animals

   (A) Experimental scheme for identifying differentially expressed genes
   with opposing changes in shPsn3 versus sfl^RNAi and their expression in
   double shPsn3;sfl^RNAi.

   (B) Volcano plots of genes significantly downregulated in shPsn3 and
   upregulated in sfl^RNAi (y axis: probability of expression changes; x
   axis: fold change (FC) as log[2]. The far-right panel shows the volcano
   plot for those genes in the shPsn3;sfl^RNAi animals, demonstrating
   restoration of all but one gene to levels of control animals or higher.

   (C) A breakdown of what genes show rescue of gene expression levels by
   functional class and listing of the affected transcripts, with their
   human homologs.

   Twenty-six genes show this pattern of differential gene expression,
   significantly downregulated in shPsn3 animals and upregulated in
   sfl^RNAi. The expression of this set of genes was normalized in the
   shPsn3;sfl double knockdown ([206]Figure 7B). Gene set analysis of
   these genes using g:profiler or PANGEA revealed significant enrichment
   after correction for multiple testing in GO biological processes
   transmembrane transport and Reactome solute carrier (SLC)-mediated
   transporter (R-DME-425407, corrected p value = 9.079e-4). A sizable
   proportion (37%) of these genes whose expression is decreased with Psn
   knockdown and displayed corrected expression mediated by sfl knockdown
   are members of the solute carrier gene family ([207]Figure 7C). CG42235
   (kumpel) encodes a homolog of the SLC5A family of transporters,
   Na+-dependent monocarboxylate cotransporters. This family of SLCs in
   Drosophila includes kumpel, rumpel, and bumpel, and the latter has been
   directly demonstrated to transport both pyruvate and
   lactate.[208]^66^,[209]^67 Overexpression of bumpel, a gene expressed
   in largely the same glial cell set as kumpel, rescues autophagy defects
   and neuronal death in a Drosophila model of C9orf72
   neurodegeneration.[210]^66 Two members of the SLC2A family of hexose
   transporters are also represented in the gene set rescued by sfl,
   CG33281, and CG7882. Three members of the SLC22A family were also
   downregulated in Psn knockdown animals and restored to normal levels of
   expression with sfl^RNAi. This group of carriers transports organic
   ions, including carnitine, a molecule critical for fatty acid transport
   and oxidation in mitochondria.[211]^68^,[212]^69 Another group of
   transcripts suppressed upon Psn knockdown and restored to wild-type
   levels by simultaneous reduction of sfl expression affect metabolic
   regulation broadly and includes neuropeptide-like precursor 2 (Nplp2),
   insulin-like peptide 5 (Ilp5), and urate oxidase (Uro). There are
   genetic interactions between insulin-like pathway signaling and uric
   acid accumulation in Uro mutants, suggesting the gene expression
   changes in Ilp5 and Uro may be functionally connected.[213]^70 Nplp2
   has recently been shown to affect lipid transport and homeostasis,
   serving as an exchangeable lipoprotein in the hemolymph.[214]^71 Ilp5
   encodes a member of the insulin-peptide family and affects a number of
   metabolic processes.[215]^72 These results collectively show that
   modest reductions in Psn function have dramatic effects on genes
   affecting metabolic processes that can be countered by modulation of
   HSPG function.

Transcriptome patterns affected by heparan-sulfate-modifying enzymes and
comparisons between human and Drosophila cells

   We have used a diverse set of systems and cells to understand how
   heparan-sulfate-modified proteins, via the fine structure of their
   glycosaminoglycan chains, affect cell processes known to participate in
   neurodegenerative pathophysiology. We have taken advantage of the broad
   conservation of these processes both across species and cell types. The
   transcriptome analysis conducted for a human hepatocyte-derived tumor
   cell line, Hep3B, bearing knockout of EXT1, a gene required for HS
   chain polymerization, provided a broad molecular confirmation of the
   changes observed with cellular markers of autophagy, mitochondria, and
   lipid metabolism. Knockdown of sfl in fat body cells in Drosophila
   provide a parallel system to evaluate the degree of conservation of
   these pathways and molecular signatures. Despite the differences
   between the cell types, as well as the genes targeted, sfl/NDST in
   Drosophila and EXT1 in the Hep3B cells, there are shared cellular
   phenotypes as described earlier. The assessment of RNA-seq data from
   fat body cells with RNAi knockdown of sfl compared to control animals
   also provides some interesting and shared features relevant to
   vertebrate systems. Gene set analysis using the top 500 differentially
   expressed genes from sfl^RNAi compared to controls (w^RNAi) revealed
   one functional human disease set that achieved significance (p value
   0.04), neurodegenerative disease (Human Disease sets from FlyBase
   2017). Pathway analysis among KEGG 2019 groups showed a number of
   metabolic pathways affected by knockdown of sfl, including glycerolipid
   metabolism (Adj. p value 0.007), metabolism of xenobiotics by
   cytochrome P450 (Adj. p value 2.5e-7), folate biosynthesis (Adj. p
   value 0.001), and galactose metabolism (Adj. p value 0.02).

   A few individual genes with significant changes in expression are
   notable. Spargel, the Drosophila homolog of PPARGC1A, the master
   transcription regulator of mitochondrial biogenesis, is elevated in
   sfl^RNAi cells, providing a molecular signature for the effect of sfl
   on mitochondrial numbers. Genes affecting autophagy and membrane
   trafficking are also among those with significant changes, including
   Cp1, a homolog of mammalian CSTL1, that is significantly upregulated.
   Cp1 in Drosophila is an important component promoting autolysosome
   clearance.[216]^73 Genes affecting lipid accumulation are also altered
   in sfl^RNAi animals, the Dpp/TGB-β pathway in particular. In Drosophila
   this conserved signaling system suppresses lipid accumulation, and its
   downregulation is required for catabolic release of lipids.[217]^74 The
   Dpp-activated transcription factor Mad is downregulated in sfl^RNAi fat
   body cells, as is a marker of Dpp-activity, kek1, indicators that Dpp
   signaling is reduced, a known heparan-sulfate-regulated
   pathway.[218]^75^,[219]^76 These gene expression changes are consistent
   with the morphological changes in liposomes observed in sfl^RNAi
   animals. Overall, the differential transcriptome signatures documented
   here indicate that heparan sulfate modification affects autophagy,
   lipid metabolism, and mitochondrial function across species and cell
   types.

Discussion

   A number of findings over the years have implicated
   heparan-sulfate-modified proteins in AD pathogenesis.[220]^16^,[221]^77
   Recently, a dramatic example was reported for a member of a large
   kindred with a dominant disease-causing mutation in PSEN1.[222]^12 This
   individual was cognitively intact well beyond the typical early onset
   of symptoms and was remarkable for being homozygous for a rare variant
   of APOE3, APOE3 Christchurch. This form of APOE3 carries a mutation in
   the heparan-sulfate-binding domain, a change that essentially abrogated
   binding to heparin, a highly sulfated form of heparan sulfate. It was
   also reported in this study that heparin-binding strength for other
   disease-associated APOE variants was directly proportional to disease
   susceptibility (APOE4>APOE3>APOE2>>ApoE3ch). Recent work with mice
   bearing a humanized APOE3ch allele has provided evidence that the
   findings from that single individual are broadly applicable and point
   to a critical role for heparan sulfate function in PSEN1-mediated
   pathogenesis. The APOE3ch protein expressed in mice showed profound
   effects on microglial reactivity and Aβ-induced Tau seeding and
   spread.[223]^13 Thus, the involvement of heparan-sulfate-modified
   proteins in PSEN1 and APOE function as well as AD pathogenesis seems
   evident. In this context, it is important to bear in mind the diverse
   and important functions served by heparan-sulfate-modified proteins.
   HSPGs serve to assemble protein complexes to modulate a number of
   cellular processes, including endo- and exocytosis as well as signaling
   at the cell surface, providing coreceptor function for a myriad of
   protein growth factors. How then do HSPGs participate and influence AD
   pathogenesis?

   The studies reported here begin with exploring the relationship between
   heparan-sulfate-modified protein function and cellular and molecular
   signatures of AD pathogenesis in human cell lines and mouse astrocytes.
   We find that compromising heparan sulfate synthesis or modification has
   profound effects on autophagy, mitochondrial function, and lipid
   metabolism in these vertebrate systems. The changes in these cellular
   functions were reflected in heparan-sulfate-dependent changes in the
   transcriptome, including genes affecting cholesterol synthesis, lipid
   transport, autophagy, and mitochondrial biogenesis. There was also a
   remarkable enrichment of AD-associated genes identified by multiple
   GWAS studies in differentially expressed genes affected by loss of
   EXT1, a gene encoding a heparan sulfate polymerase. The cellular and
   molecular mechanism for HSPG modulation of AD-related changes was
   examined using a Drosophila model of presenilin-mediated pathology. The
   results from this analysis were clear: reducing heparan sulfate
   function increased autophagy, mitochondrial function, and biogenesis
   and lowered intracellular lipid droplets, as observed in human cells
   and mouse astrocytes. Furthermore, reducing heparan sulfate
   modification by partial loss of function of sfl, an NDST1 homolog in
   Drosophila, rescued cellular and molecular abnormalities in
   Psn-defective animals, as well as the age-dependent neurodegeneration
   produced with partial reduction of Psn function.

   The extensive genetic analysis of AD using GWAS, as well as whole-exome
   and -genome sequencing technologies, has made remarkable progress in
   identifying genes that confer susceptibility or causation of this
   disease. Collectively, these findings have identified shared genes and
   pathways affecting both familial, early onset as well as late-onset
   AD.[224]^5^,[225]^6 The current challenge is to identify the cellular
   and molecular events that are not only associated with disease but also
   serve a functionally critical role in pathogenesis. Model organisms
   provide the potential for addressing this challenge by virtue of the
   genetic tools that can identify genes that modify or suppress disease
   phenotypes. The conservation of the genes and their functions shown to
   be critical in AD in the fruit fly Drosophila provides an opportunity
   to identify functionally critical events in the molecular and cellular
   pathogenesis. This model system also allows for the study of
   interactions suggested by human genetic findings that have important
   implications for identifying potential targets for pharmaceutical
   intervention.

Transcriptome patterns and cellular events affected by heparan sulfate
biosynthesis

   Evaluating heparan-sulfate-dependent transcription patterns using Hep3B
   cells bearing an EXT1 knockout was informative with regard to the
   pathways and cellular processes governed by HSPG-mediated signaling.
   First, it provided molecular evidence for the profound effects of HSPG
   function on cholesterol biosynthesis and lipid transport. These
   findings link the effects of reducing HS synthesis on intracellular
   lipids stored in liposomes that we have documented for a variety of
   cell types, including mouse astrocytes. We also observed effects on
   transcripts affecting mitochondrial biogenesis and autophagy, all
   processes involved in neurodegenerative disease. Finally, there was an
   enrichment of genes with differential expression affected by EXT1 among
   AD-associated genes identified by extensive GWAS study. HSPG-controlled
   transcripts overlap to a remarkable degree with this set of genes
   implicated in the development and susceptibility to late-onset AD.

   Functional analysis of several human cell lines and mouse astrocytes
   demonstrated that HSPGs serve critical functions in autophagy,
   mitochondrial physiology, and lipid metabolism. Modest changes in HS
   structure can produce marked changes in mitochondria, intracellular
   lipid, and autophagy flux. The changes affected by reducing heparan
   sulfate function counter the pathological shifts seen in AD and other
   neurodegenerative diseases, namely increases in autophagy, increased
   number and function of mitochondria, and lowered levels of
   intracellular lipid droplets.

   We have used Drosophila to determine the functional interaction between
   presenilin and HSPGs as one model for evaluating the impact of HSPGs on
   neurodegeneration. In particular, the cellular and molecular phenotypes
   of reducing the function of the Drosophila presenilin gene, Psn, and a
   gene critical for heparan sulfate biosynthesis, sfl, were examined.
   Heparan-sulfate-modified proteins are known to regulate processes
   disrupted by presenilin deficits, autophagy in particular. In both
   Drosophila[226]^20^,[227]^21 and mice,[228]^78^,[229]^79 compromising
   the function of either heparan sulfate biosynthetic enzyme-encoding
   genes or individual heparan-sulfate-modified core proteins produces an
   increase in autophagy flux. For example, reductions in Perlecan (Hspg2)
   in mice has profound effects on autophagy,[230]^78 as well as
   mitochondria and lipid metabolism,[231]^79 events linked to changes in
   autophagy. In human hepatocellular carcinoma, elevated Glypican-3
   expression is associated with the suppression of autophagy, and
   reducing GPC3 elevates autophagy levels and suppresses cell
   growth.[232]^80^,[233]^81^,[234]^82 In contrast to the effects of
   lowering heparan-sulfate-modified protein function, mutations in PSEN1
   compromise autophagosome to lysosome
   flux,[235]^83^,[236]^84^,[237]^85^,[238]^86 and activating autophagy
   can restore autophagy levels and rescue mitochondrial
   abnormalities.[239]^87 These findings point to the regulation of
   autophagy as a key element of AD pathogenesis and
   heparan-sulfate-modified proteins as an important determinant of
   autophagy flux.

Psn and sfl knockdown have opposing effects on key cellular processes, and
modest reductions of sfl suppress Psn-mediated cell deficits and
neurodegeneration

   Previous work established that reductions in presenilin/Psn or
   Nicastrin, another component of the γ-secretase complex conserved in
   Drosophila, in adult neurons produce age-dependent
   degeneration.[240]^52 We used this system to not only explore the
   interaction of Psn with heparan sulfate biosynthesis in
   neurodegeneration but also evaluate the cellular and molecular
   signatures in fat body cells, the principal metabolic storage and
   transport organ in Drosophila. Psn and sfl knockdown showed divergent
   effects on mitochondrial number and morphology, as well as
   intracellular lipid and autolysosome structure seen with EM and
   confocal visualization of fluorescent-tagged markers. Consistent with
   these different effects on cellular features, sfl knockdown was able to
   suppress both neurodegeneration and the spectrum of cellular phenotypes
   produced by partial loss-of-function Psn knockdown. It is notable that
   modest changes in the function of either of these genes produced
   readily discernable phenotypes. RNA-seq analysis of fat body cells with
   knockdown achieved with RNAi vectors showed 30%–40% reductions in mRNA
   and clear effects on cell phenotypes and cell loss in the nervous
   system.

   Knockdown of Psn in fat body cells produced differential gene
   expression changes, as did sfl^RNAi. Comparing the differential
   expression patterns showed a subset of genes with opposite directions
   of change. The vast majority of these genes were downregulated in Psn
   knockdown animals and rescued to normal levels with simultaneous
   reduction of sfl mRNA levels. The identity of these genes, reflecting a
   molecular rescue signature, showed a large representation of solute
   carrier transporters, molecules inherently affecting metabolism and
   energy balance. One SLC family member was of particular interest,
   CG42235 (kumpel), encoding a homolog of the SLC5A family of
   transporters, Na^+-dependent monocarboxylate cotransporters. Kumpel and
   another member of this family, bumpel, have been shown to transport
   lactate and pyruvate[241]^67 and when overexpressed, can rescue neuron
   loss mediated by expression of C9orf72, a human disease gene affecting
   frontotemporal dementia and amyotrophic lateral
   sclerosis.[242]^66^,[243]^88 Monocarboxylate and lipid transporters
   have also been shown to be vital in neuron-glia interaction in
   responding to reactive oxygen species (ROS) damage affected by
   mitochondrial dysfunction in neurons.[244]^89 One gene affected by Psn
   knockdown and rescued by compromising HS sulfation is Nplp2, a lipid
   transport protein that responds to thermal stress, promoting
   survival.[245]^71 Nplp2 has functions that parallel APOE in vertebrate
   systems, namely serving as an exchangeable lipoprotein that can be
   soluble in hemolymph or blood and also associate with VLDLs.[246]^71
   Our transcriptomic analysis of Psn rescue by sfl has identified classes
   and gene family members shown to be critical in two other models of
   neurodegeneration in Drosophila. These findings emphasize the
   participation of HSPGs in cellular signaling systems affecting neuronal
   survival and point to processes critical in neurodegeneration, namely
   monocarboxylate and lipid transport.

HSPGs and AD pathogenesis

   The experiments we have presented here assessing the role of HS
   structure on presenilin-mediated phenotypes in Drosophila have relied
   on reducing sfl mRNA. sfl is the sole Drosophila homolog of vertebrate
   NDST genes. sfl/NDST genes encode enzymes that remove an acetyl group
   and replace it with a sulfate on one of the two sugars that comprise
   the HS polymer, N-acetyl glucosamine. This sulfation step is critical
   for later sulfotransferase modifications and therefore serves as a
   means of broadly altering the sulfation pattern without altering chain
   length. The structure of heparan sulfate and its biosynthetic machinery
   are highly conserved across species, and structural analysis of
   Drosophila heparan sulfate has demonstrated its conserved sulfation
   patterns compared to vertebrate HS.[247]^90^,[248]^91 Indeed, a mouse
   Ndst1 completely rescues sfl mutant phenotypes when expressed correctly
   from the endogenous sfl regulatory elements,[249]^92 whereas mouse
   Ndst2 has limited rescue activity. The 2-O, 6-O and 3-O
   sulfotransferases are conserved in Drosophila but the role of these
   modifications in autophagy, mitochondrial function, and lipid
   metabolism has not been investigated. HSPGs have been shown to affect
   the transcellular propagation of misfolded protein aggregates, and HS
   chain length and sulfation patterns do affect Tau, α-synuclein, and Aβ
   differentially although Tau appears to have the most stringent
   structural requirements.[250]^93^,[251]^94 Recently, it has been found
   that increases in 3-O-sulfated HS are found in AD brains.[252]^95 These
   results support the identification of HS3ST1 as an AD-associated risk
   gene and its role in Tau internalization.[253]^96^,[254]^97 It is worth
   noting that NDST activity is required for 3-O sulfation to take
   place.[255]^98

   It has been known for some time that HSPGs are components of
   AD-pathological structures, namely neuritic plaques containing
   APP-derived peptides. Compromising HS biosynthesis reduces the levels
   of these pathological structures in mouse models of AD as does ectopic
   expression of HS degradative enzymes.[256]^99^,[257]^100 It is also
   known that HSPGs affect the internalization and propagation of Tau
   aggregates, providing structures that seed intracellular
   fibrils.[258]^93 These studies all suggest mechanisms where HSPGs
   contribute to AD pathogenesis and therefore point to HS synthesis as a
   potential therapeutic target. The studies here provide another
   dimension; HSPGs affect signaling events that influence cellular
   processes known to be early and important for the development of
   neurodegenerative disease. Our studies show that HSPGs regulate
   autophagy, mitochondrial structure and function, and lipid metabolism,
   all important processes in neurodegeneration. We have provided direct
   evidence for HSPGs affecting neurodegeneration and cellular deficits
   mediated by reductions in presenilin function. HSPGs are cell surface
   signaling proteins, and their effect on presenilin-mediated cell
   defects indicates that changes in signaling are an important
   contributor to AD pathogenesis. It is intriguing that GSEA pathways
   with suppressed signaling in EXT1−/− cells are linked to aberrant
   signaling ([259]Table S4: Signaling by PDGFR in Disease and Signaling
   by BRAF and RAF1 Fusions). Perhaps reducing heparan sulfate
   biosynthesis suppresses multiple signaling processes that promote AD
   pathogenesis.

Considerations of model systems to understand AD pathology

   We have used a number of model systems to explore the function of
   heparan-sulfate-modified proteins in AD-related pathogenesis. These
   studies were prompted in some measure by human genetic findings on gene
   variants affecting PSEN1-mutation-mediated pathology, namely APOE3
   Christchurch, which is remarkable for its lack of heparan sulfate
   binding and capacity to suppress cognitive decline. Nonetheless, it is
   important when using model systems to consider their relevance to the
   human disease. This is particularly true when the models are
   evolutionarily distant from humans, in organisms that do not normally
   exhibit the disease. These considerations apply to the work in fruit
   fly Drosophila, as well as cell types not directly involved in AD
   pathology. Drosophila do not show amyloid deposits or Tau accumulation
   unless directly engineered to do so. Yet, presenilin and all the other
   components of the γ-secretase complex are conserved between vertebrates
   and Drosophila, and deficits in presenilin produce similar cellular
   phenotypes, including mitochondrial deficits and increased apoptosis.
   Both of these cellular phenotypes were rescued by reductions of sfl
   function in Drosophila, assessed both in the brain and in a critical
   lipid regulatory organ, the fat body.

   What can we learn about AD pathogenesis from cellular and molecular
   studies of non-CNS cell types, such as hepatocytes? For the work
   described here, the answer lies in the conserved mechanisms affected by
   both presenilins and heparan-sulfate-modified proteins. These molecules
   affect many cell types outside the nervous system but perform conserved
   regulatory functions. Cells that have a major role in lipid metabolism
   can be particularly instructive. This useful relationship has been
   noted by a number of studies, many focusing on the function of APOE, an
   important determinant of AD
   susceptibility.[260]^101^,[261]^102^,[262]^103 There is also evidence
   that expression of APP restricted to the liver disrupts lipid
   metabolism and can produce some pathological features of AD, including
   increased neutral lipid accumulation and neurodegeneration.[263]^104

   There is growing evidence that disturbances in lipid metabolism and
   APOE, a protein affecting lipid transport, are important elements of AD
   pathogenesis.[264]^45 Recent work profiling transcripts using
   single-nucleus RNA sequencing from brain tissue of AD patients and
   controls bearing different APOE alleles is telling. AD-associated
   transcriptome signatures in microglia, were characterized by elevated
   levels of ACSL1 expression, a gene affecting lipid droplet formation.
   These disease-associated microglia, called LDAM
   (lipid-droplet-associated microglia), were highest in AD patients with
   two APOE4 alleles. In addition, the levels of neutral, intracellular
   lipid were proportional to cognitive decline and anatomical extent of
   disease. The central role that lipid metabolism and APOE play in
   pathogenic processes argues that examining gene functions in other cell
   types, such as hepatocytes, that are also critical for the regulation
   and transport of lipid and express APOE can reveal important
   mechanistic relationships.

Limitations of the study

   Some of the analysis of heparan sulfate influences on autophagy,
   mitochondria, and lipid was conducted in non-CNS human tumor cells or
   glycogen and lipid-storing cells in Drosophila. The relevance of these
   findings to AD pathogenesis in vertebrates is inferred based on
   conservation of these mechanisms across organisms and cell types.

STAR★Methods

Key resources table

   REAGENT or RESOURCE SOURCE IDENTIFIER
   Bacterial and virus strains
     __________________________________________________________________

   Lentivirus Negative Scramble shRNA
   Target Sequence: GCACTACCAGAGCTAACTCAGATAGTACT Origene Cat#TR30021
   Lentivirus NDST1 KD
   Target Sequence: TTCCACAACCAATCTGTGCTGGCAGAGCA Origene Cat#TL509817A
     __________________________________________________________________

   Chemicals, peptides, and recombinant proteins
     __________________________________________________________________

   Alexa Fluor™ 594 Phalloidin Thermofisher Cat#A12381
   MitoTracker™ Red CMXRos Thermofisher Cat#M7512
   HCS LipidTOX™ Red Neutral Lipid Stain Thermofisher Cat#[265]H34476
   HCS LipidTOX™ Green Neutral Lipid Stain Thermofisher Cat#[266]H34475
   Biotracker MitoBlue Live Cell Dye Sigma-Aldrich Cat#SCT086
   Prolong™ Gold Antifade Mountant with DAPI Thermofisher Cat#[267]P36941
   Vectashield Antifade Mounting Medium Fisher Scientific Cat#NC9532821
   GNeo-biotin Synthesized as described[268]^44 N/A
   Hoescht 33342 Solution Thermofisher Cat#62249
   Rapamycin ApexBio Cat#A8167
   Acridine Orange Thermofisher Cat#A1301
     __________________________________________________________________

   Critical commercial assays
     __________________________________________________________________

   Macherey-Nagel NucleoSpin® RNA Kit VWR Cat#MANA740955.50
   Adult Brain Dissociation Kit Miltenyi Biotech Cat#130-107-677
     __________________________________________________________________

   Deposited data
     __________________________________________________________________

   FASTQ of Drosophila melanogaster RNAseq Data NCBI SRA Database
   Accession# PRJNA1106548
   FASTQ of Hep3B cells RNAseq Data NCBI SRA Database Accession#
   PRJNA1107266
     __________________________________________________________________

   Experimental models: Cell lines
     __________________________________________________________________

   Human: A375 Cells: Wildtype Gift from J. Esko, UCSD N/A
   Human: A375 Cells: NDST1 KO Gift from J. Esko, UCSD N/A
   Human: HEK293T Cells: Scramble This paper: transfected by Anoop
   Narayanan N/A
   Human: HEK293T Cells: NDST1 KD This paper: transfected by Anoop
   Narayanan N/A
   Human: HEK293T Cells: EXT1 KD This paper: transfected by Anoop
   Narayanan N/A
   Human: Hep3B Cells: Wildtype Gift from J. Esko, UCSD[269]^27 N/A
   Human: Hep3B Cells: EXT1 KO Gift from J. Esko, UCSD[270]^27 N/A
     __________________________________________________________________

   Experimental models: Organisms/strains
     __________________________________________________________________

   D. melanogaster: UAS-shPsn2 Gift from J. Kang, Harvard
   University[271]^52 N/A
   D. melanogaster: UAS-shPsn3 Gift from J. Kang, Harvard University N/A
   D. melanogaster: UAS-Nct^RNAi Gift from J. Kang, Harvard University N/A
   D. melanogaster: c∗155 elav-gal4; UAS-DcrII Bloomington Drosophila
   Stock Center (NIH P40OD018537) RRID: BDSC 25750
   D. melanogaster: UAS-GFP-Atg8a Gift from T. Neufeld, University of
   Minnesota, Minneapolis, MN[272]^105 N/A
   D. melanogaster: UAS-w^RNAi Vienna Drosophila Resource Center (VDRC,
   [273]www.vdrc.at). RRID: BDSC 30033
   D. melanogaster: R4-Gal4 Bloomington Drosophila Stock Center (NIH
   P40OD018537) RRID: BDSC 33832
   D. melanogaster: sfl^03844 Bloomington Drosophila Stock Center (NIH
   P40OD018537) RRID: BDSC 5575
   D. melanogaster: UAS-sfl^RNAi Bloomington Drosophila Stock Center (NIH
   P40OD018537) RRID: BDSC 34601
   D. melanogaster: UAS-GFP-Atg8a; R4-Gal4 Made in-lab[274]^20 N/A
   D. melanogaster: shPsn3; sfl^03844 This manuscript N/A
   D. melanogaster: shPsn3; UAS-sfl^RNAi This manuscript N/A
   Mouse: humanized ApoE3 astrocytes Obtained as described.[275]^45 N/A
   Mouse: humanized ApoE4 astrocytes Obtained as described.[276]^45 N/A
     __________________________________________________________________

   Recombinant DNA
     __________________________________________________________________

   Scramble shRNA
   Hairpin Sequence:
   CCTAAGGTTAAGTCGCCCTCGCTC
   GAGCGAGGGCGACTTAACCTTAGG Sarbassov et al.[277]^106 RRID: Addgene 1864
   NDST1 shRNA
   Target Sequence:
   TCTCGGCCTACTACCTATATG Sigma-Aldrich Cat#TRCN0000232002
   EXT1 shRNA
   Target Sequence:
   CCCAACTTTGATGTTTCTATT Sigma-Aldrich Cat#TRCN0000039995
     __________________________________________________________________

   Software and algorithms
     __________________________________________________________________

   ImageJ Fiji Described in Schindelin et al.[278]^107
   [279]https://imagej.net/software/fiji/downloads
   MiniTab MiniTab N/A
   Psomagen mRNA Sequencing Psomagen N/A
   BasePair RNA Seq Analysis BasePair N/A
   GraphPad Prism v10.0
   FlowJo Analytical Software v10.8.0 FlowJo N/A
   Avizo Thermofisher N/A
   Imaris 9.3.0 Oxford Instruments N/A
   [280]Open in a new tab

Resource availability

Lead contact

   Further information and requests for resources should be directed to
   and will be fulfilled by the Lead Contact, Scott Selleck
   (sbs24@psu.edu).

Materials availability

   Fly and cell lines used in this study are available upon request.

Data and code availability

     * •
       Data: Raw RNAseq data (FASTQ) of Drosophila melanogaster fat body
       have been deposited in NCBI Sequence Read Archive (SRA):
       [281]https://www.ncbi.nlm.nih.gov/sra/PRJNA1106548, with data
       identification number [282]PRJNA1106548. Raw RNAseq data (FASTQ) of
       Hep3B cells RNAseq have been deposited in NCBI Sequence Read
       Archive (SRA): [283]https://www.ncbi.nlm.nih.gov/sra/PRJNA1107266,
       with data identification number PRJNA1107266. Differential Gene
       Expression Data from this study are provided in [284]Tables S7 and
       [285]S8.
     * •
       Code: This paper does not report any original code.
     * •
       Additional Information: Any additional information required to
       reanalyze the data reported in this paper is available from the
       [286]lead contact upon request.

Experimental model and study participant details

Mouse

   All animal experiments were approved by the University of Arizona
   Institutional Animal Care and Use Committee (IACUC). Humanized ApoE3
   (029018) and ApoE4 (027894) knockin mice (homozygous) and the wild-type
   C57BL/6J mice were obtained from the Jackson Laboratory and bred at the
   University of Arizona animal facility. All mice were housed in a
   temperature (23C) and humidity-controlled room with a 12-h light and
   12-h dark cycle with ad libitum access to water and standard laboratory
   global 2019 diet (23% calories from protein, 22% calories from fat, 55%
   calories from carbohydrate, 3.3 kcal/g; Envigo) and cared in facilities
   operated by University Animal Care. 3-month male mice were used for
   primary astrocyte isolation and in vivo/ex vivo assays. Timed pregnant
   mice of designated genotype were prepared for the isolation of E17
   embryonic hippocampal neurons.

Drosophila melanogaster

   Fly strains were raised on standard cornmeal/sucrose/agar media at 25°C
   under a 12 h day/12 h night cycle. w^1118 served as the wild-type
   stock. [UAS-Atg8a RNAi (43097)] and [UAS-w RNAi (30033)] RNAi strains
   were obtained from the Vienna Drosophila Resource Center (VDRC).
   [UAS-sfl RNAi (34601)], [sfl^03844 (5575)], and [c∗155 elav-gal4;
   UAS-dcrII (25750)] were obtained from the Bloomington Drosophila Stock
   Center (BDSC). [UAS-shPsn2] and [UAS shPsn3] flies were a generous gift
   from J. Shen and Norbert Perrimon, Harvard Medical School, Boston, MA.
   [shPsn3; UAS-sfl RNAi] and [shPsn3RNAi; sfl^03844] stocks were
   generated in lab. Where sex, developmental stage, and age is an
   influence it is noted in the results.

Human cell lines

   A375, Hep3B, and HEK293T cell lines were cultured in DMEM supplemented
   with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin at
   37°C, 5% CO[2] (Gibco). All human cell lines were tested for mycoplasma
   at initiation of all experiments. A375 cells were first isolated from a
   54-year old female. Hep3B cells were first isolated from an 8-year old
   black male. HEK293T cells were first isolated from a female fetus.

Method details

Drosophila brain dissection and assessment of neurodegeneration

   Following eclosion, flies were sorted by sex and aged for 7 or 30 days.
   Whole flies were fixed in 4% PFA containing 0.5% Triton X-100 (0.5%
   PBS-T) for 3h with nutation at 20°C. The samples were washed 4 times
   with 0.5% PBS-T for 15 min with nutation at 20°C. Brains were dissected
   from whole flies in 0.008% PBS-T and placed in staining solution (DAPI
   at 1:1000 and Phalloidin at 1:100 in 0.5% PBS-T) for 16-24h with
   nutation at 4°C. Following staining, brains were washed 4 times with
   0.5% PBS-T for 15 min with nutation at 20°C and once with 1xPBS for
   30 min with nutation at 20°C to remove any residual detergent or stain.
   A SecureSeal imaging spacer was applied to each microscope slide, and
   brains were mounted anterior side up with SlowFade Gold Antifade
   Mountant (without DAPI). A more detailed version of this procedure can
   be found in.[287]^54 Full brain samples were imaged non-sequentially on
   an Olympus FV3000 confocal microscope at a step size of 2 μm. Images
   were analyzed in Fiji ImageJ using the trace and measure tools. The
   Look Up table was set to Fire, and the maximum pixel intensity was
   adjusted to provide optimal visualization of vacuoles. Once a vacuole
   could be distinguished from a tracheole, it was traced in the slice in
   which the diameter was the largest. The total number of vacuoles per
   brain and the total area of vacuoles per brain were summed,
   respectively. ANOVA statistical analysis (one-way, Tukey comparison)
   was performed in Minitab to confirm significance.

Acridine orange brain staining of Drosophila adult brain[288]^108

   Following eclosion, flies were sorted by sex and aged 1 to 3 weeks.
   Female flies were knocked out with CO[2] and placed in 70% ethanol for
   1 min before being transferred to a dissection tray containing Ringer’s
   solution. After dissection, the brains were stained with 30 μL of
   5 μg/mL Acridine Orange for 5 min on a glass microscope slide. The
   staining solution was removed, the brains were washed 1 time with 50 μL
   of Ringer’s solution, and 50 μL of fresh Ringer’s solution was added as
   a mounting medium. Full brain samples were imaged on an Olympus FV3000
   confocal microscope at a step size of 2 μm. Images were analyzed in
   Fiji ImageJ as z stack maximum intensity projections. The Freehand
   Select tool was used to trace the outline of the most anatomically
   complete lobe and the area of this selection was measured. The image
   threshold was adjusted to remove background signal and the area outside
   the outlined tracing was deleted. The resulting image was converted
   into a binary pixel scheme, and the number of white pixels was
   measured. This value was normalized by the area of the tracing, and an
   average number of pixels per area per genotype was calculated.

Negative geotaxis assay

   The behavioral abilities of Drosophila were measured via a
   startle-induced negative geotaxis protocol.[289]^53 Upon eclosion,
   flies were separated by sex and aged for 3–6 days. For each geotaxis
   assay, 10 flies were placed into the apparatus and a startle response
   was induced by firmly tapping the apparatus on a mat until all flies
   had fallen to the bottom. The number of flies above the 8 cm line after
   10 s was recorded and the flies were given a 60-s rest period before
   the next trial was performed. An average of 5 repetitions per vial of
   flies was taken as a single data point.

MitoTracker red staining of fat body

   The fat body of third instar Drosophila melanogaster larvae were
   dissected in PBS and transferred to a 100 nM solution of MitoTracker
   Red CMXROS. They were incubated at room temperature for 30 min out of
   light. The fat bodies were washed in PBS and then fixed in a 4% v/v PFA
   solution where they were incubated at room temperature for 30 min out
   of light. After fixation, the fat body was washed in PBS and
   transferred to the glass imaging slide. ProLong Gold Antifade Mountant
   with DNA Stain DAPI was applied to the fat bodies as a mounting medium
   with counterstain. The fat bodies were imaged with an Olympus FV3000
   confocal microscope.

   Images of fat body stained with MitoTracker Red CMXRos were processed
   in Fiji ImageJ on a maximum intensity projection. An auto-threshold was
   applied to the images through Max Entropy. A watershed was applied and
   segmented mitochondria. Particles were selected for those between 1 and
   10 μm in area and between 0.25 and 1.00 in circularity.

   The size of mitochondria was analyzed using the built-in particle
   analysis tool in Fiji ImageJ on a maximum intensity projection. A
   200×200 pixel box was used as a sample area, with the upper left pixel
   being the locator pixel that corresponded to the coordinates of the
   box. A random number generator was used to select two numbers between 0
   and 824 to generate a random (x,y) coordinate to place the box. If the
   box was not entirely over a region where signal was present, the
   coordinates were reselected. Each sample was analyzed using this method
   three times, ensuring that the analyzed areas did not overlap.

LipidTox red staining of fat body

   The fat body of third instar Drosophila melanogaster larvae were
   dissected in PBS and then fixed in a 4% v/v PFA solution where they
   were incubated at room temperature for 30 min out of light. After
   fixation, the fat bodies were washed in PBS. Fat bodies were then
   transferred to a LipidTox Red solution diluted 1:999 in buffer and
   incubated for 30 min at room temperature out of light. After staining,
   the fat bodies were transferred to a glass imaging slide. ProLong Gold
   Antifade Mountant with DNA Stain DAPI was applied to the fat bodies as
   a mounting medium with counterstain. The fat body tissue was imaged
   with an Olympus FV3000 confocal microscope.

Transmission electron microscopy

   Fat body tissues were dissected out of third-instar larvae and stored
   overnight in a refrigerator in Modified Karnovsky’s fixative. Samples
   were washed three times in 0.1M cacodylate buffer and subsequently
   fixed for 1 h in 1–2% Osmium tetroxide in 0.1M cacodylate buffer.
   Samples were washed two times with 0.1M cacodylate buffer and one time
   with milliQ water before undergoing an en bloc stain for 1 h in
   filtered 2% aqueous Uranyl acetate. Samples were dehydrated in
   progressively increasing ethanol concentrations (50%, 70%, 85%, 90%,
   100%, 3xEM grade ethanol) followed by three acetone dehydrations.
   Samples were infiltrated with Spurr’s resin and imaged.

   Samples were examined using FEI Tecnai G2 Spirit BioTwin and FEI Talos
   200C microscopes.

RNA isolation

   RNA was isolated according to the Macherey-Nagel NucleoSpin RNA Kit
   protocol for cultured cells and tissue. Briefly, Hep3B or Drosophila
   melanogaster fatbody cells were lysed in 350 μL Buffer RA1 and 3.5 μL
   β-mercaptoethanol and then filtered through a NucleoSpin Filter column
   through centrifugation for 1 min at 11,000 × g. The lysate was then
   mixed with 350 μL 70% ethanol and passed through a NucleoSpin RNA
   Column at 11,000 × g for 30 s in order to bind the RNA to the column.
   The binding step was repeated three additional times for lysate from
   fatbody. 350 μL of Membrane Desalting Buffer was added to the
   NucleoSpin RNA Column and centrifuged for 1 min at 11,000 × g. 95 μL of
   a DNase reaction mixture was incubated on the column for 15 min at room
   temperature to digest DNA. After digestion, 200 μL of Buffer RAW2 was
   applied to the column and centrifuged for 30 s at 11,000 × g to
   inactivate the rDNase. 600 μL of Buffer RA3 was then added to the
   column and centrifuged for 30 s at 11,000 × g. 250 μL of Buffer RA3 was
   applied to the column and centrifuged for 2 min at 11,000 × g to dry
   the membrane. 60 μL of RNase-free H[2]O was applied to the column and
   centrifuged at 11,000 × g for 1 min in order to elute the RNA. Hep3B
   cells were grown to 70% confluency in a six-well plate at 37°C and 5%
   CO[2] in DMEM with 10% FBS and 1% Penicillin/Streptomycin.

RNA sequencing

   RNA sequencing was performed by Psomagen using the Illumina Truseq
   stranded mRNA library, a 40 bp sequencing coverage, 40 million total
   reads, and a read length of 150 bp PE on a NovaSeq6000 S4. All samples
   passed Psomagen Quality Control checks for both RNA and cDNA. RNAseq
   quality was validated by checking for a Q30 greater than or equal to
   85% through NovaSeq 6000 and NovaSeq X Plus analysis, a Q30 greater
   than or equal to 80% through HiseqX analysis, and a Q30 greater than or
   equal to 70% for 2 × 300 bp through MiSeq analysis. Low-quality bases
   with a Q less than 20 were trimmed. Subsequently, reads with more than
   10% low-quality bases were trimmed. Adapter sequences were then trimmed
   from remaining reads. Ribosomal RNAs and RNAs mapped to incorrect
   genomes were filtered out of the dataset. Alignment coverage, depth,
   and pair-end mapping information were generated based on the percentage
   of mapped reads, percentage of reads mapped onto specific genomic
   regions defined by GTF or GFF, 3′ and 5′ bias patterns, the properly
   paired read percentage, and the strandness of the alignment.

   Transcriptomics analysis was performed in BasePair, in which one
   Principal Component covered 90.53% of variants in HEP3B data and in the
   mid-40 percentile for different genotypes in Drosophila data with
   quadruplicate samples.

RNA sequencing data analysis

   RNA sequencing data was analyzed, and Differential Expression data was
   generated by Basepair. Data was aligned to the appropriate genome using
   STAR and gene expression levels were quantified using featureCounts.
   Differential Expression data was generated using DESeq2 and pathway
   enrichment analysis was performed using GSEA.

MitoBlue staining of human cultured cells

   Tissue culture cells were plated in DMEM with 10% FBS and 1%
   Penicillin/Streptomycin on Ibidi 12-well chamber slides at 30,000
   cells/well the day before they were to be stained. The cells were kept
   overnight at 37^oC and 5% CO[2]. On the day of the experiment, the
   complete DMEM was removed and a 5 μM MitoBlue solution in DMEM
   containing no FBS or PenStrep was applied to the cells. The cells were
   incubated for 30 min at 37°C out of light. The staining solution was
   aspirated, and the cells were washed with PBS. 4% v/v PFA was applied
   to the cells, which were then incubated for 30 min at room temperature
   out of light. The PFA and rubber chambers were then removed and
   Vectashield was applied as a mounting medium. The cells were imaged
   with an Olympus FV1000 microscope. Images of Hep3B cells stained with
   MitoBlue were processed in Fiji ImageJ on a maximum intensity
   projection. An auto-threshold was applied to the images through Max
   Entropy. A watershed was applied and segmented mitochondria. Particles
   were selected for those between 1 and 10 μm in area and between 0.25
   and 1.00 in circularity. All images included in [290]Figure 2 were
   brightened 20% from originals for best reproduction.

LipidTOX red staining of human cultured cells

   Tissue culture cells were plated in DMEM with 10% FBS and 1%
   Penicillin/Streptomycin on Ibidi 12-well chamber slides at 30,000
   cells/well the day before they were to be stained. The cells were kept
   overnight at 37°C and 5% CO[2]. On the day of the experiment, the DMEM
   was removed and 4% v/v PFA solution, a fixative, was applied to the
   cells for 30 min at room temperature out of light. The PFA was
   aspirated, and the cells were washed with PBS. LipidTOX Red diluted
   1:1000 in buffer was applied to the cells and they were incubated for
   30 min at room temperature out of light. The LipidTOX Red solution and
   rubber chambers were then removed and Vectashield was applied as a
   mounting medium. The cells were imaged with an Olympus FV1000
   microscope. Maximum projection confocal images of HEK293T cells were
   analyzed for differences in LipidTOX Red signal intensity. A 145×145
   pixel box was used as a sample area, approximately ten cells were
   included in each pixel box. A histogram was collected of each sample
   area, with 256 bins collected in each histogram. The scale of pixel
   intensities was set from 0 to 4096. Each sample was analyzed using this
   method three times, ensuring that the analyzed areas did not overlap.
   All LipidTOX images included in [291]Figure 2 were brightened equally
   for better reproduction. The graph in [292]Figure 2 represents average
   pixel intensities for sampled regions for one experiment with error
   bars representing SEM. The experiment was conducted in quadruplicate.

Preparation of fluorescently tagged guanidinylated neomycin-C5 conjugate and
uptake experiments

   Biotinylated guanidinylated neomycin (GNeo-biotin) was synthesized, as
   previously described,[293]^44 and stored in water at −20°C. After
   thawing at room temperature, GNeo-biotin was diluted into PBS to
   2.5 μM. To this solution, streptavidin-Cy5 (ST-Cy5, Thermo Fisher) was
   added to achieve a final molar ratio of 1:5 of fluorophore to biotin.
   To ensure completion of the biotin-streptavidin reaction, the solution
   was gently mixed and allowed to incubate at room temperature, shielded
   from light, for 20 min. Following this incubation, the GNeo-ST-Cy5
   conjugate was diluted to the desired concentration in DMEM media. A375
   wildtype or NDST1−/− knockout cells were plated at 200,000 cells/well
   in 6-well tissue culture plates and let adhere overnight. The next day,
   the cells were washed with PBS and incubated with fluorescent-tagged
   GNeo-biotin (10 nM) for 1 h at 37°C under an atmosphere of 5% CO2.
   Control cells were treated with ST-Cy5 only. After incubation, cells
   were washed twice with PBS, released with 0.05% trypsin/EDTA (Corning),
   and analyzed by flow cytometry.

MitoTracker red and LipidTOX green uptake studies and flow cytometry

   A375 wildtype or NDST1−/− knockout cells were plated at 200,000
   cells/well in 6-well tissue culture plates and let adhere overnight.
   For drug treatments, DMSO or rapamycin (10 μM) was added the next day,
   then cells were incubated for 24 h at 37°C under an atmosphere of 5%
   CO2. The next day, the cells were washed with PBS and incubated with
   MitoTracker Red (1:1000, Thermo) or HCS LipidTOX Green (1:500, Thermo)
   diluted in DMEM media for 1 h at 37°C under an atmosphere of 5% CO2.
   After incubation, cells were washed twice with PBS, released with 0.05%
   trypsin/EDTA (Corning), and analyzed by flow cytometry. Flow cytometry
   experiments were performed using a CytoFLEX S (Beckman Coulter) flow
   cytometer (≥10,000 events/sample), and raw data were analyzed using
   FlowJo Analytical Software v10.8 (Becton Dickinson). Cells were gated
   according to forward and side scattering. The extent of uptake was
   quantified using the geometric mean of the fluorescence intensity
   versus control samples. These values were plotted and further analyzed
   using GraphPad Prism v10.0.

Seahorse measurement of oxygen consumption rate

   The cell’s rate of oxidative metabolism of glucose was evaluated with
   XF Cell Mito-stress Assay in a Seahorse XFe24 analyzer which allows
   real-time metabolic analysis in live cells. Three cell lines, A375,
   Hep3B, and HEK293T, were plated on Agilent Seahorse XF24 Cell Culture
   Microplates (30,000 cells/per well) and were placed in a CO2 incubator
   overnight. The sensor cartridge used from an Agilent Seahorse XFe24
   Extracellular Flux Assay kit was hydrated with Seahorse XF Calibrant
   (1000 μL/well) overnight in a non-CO2 incubator. Cells were
   subsequently washed with XF assay media [0.5 mM Glutamine (200 mM
   solution), 0.5 mM Glucose (1.0 M solution), 0.5 mM Pyruvate (100mM
   solution), 48.5 mM Seahorse XF DMEM assay medium, pH 7.4]. After
   washing, cells were incubated in respective media (37°C, 1 h) in a
   non-CO2 incubator, allowing for temperature/pH equilibration before
   each measurement in the metabolic analyzer. Each plate contained four
   wells not seeded with cells to serve as blank controls for
   temperature-sensitive fluctuations. In the case of HEK293T, there were
   six control wells, to divide specified wells evenly between three
   genotypes. The lyophilized drugs from the Seahorse XF Mito Stress kit
   were diluted in Seahorse XF assay media and loaded into the
   corresponding injection ports in the sensor cartridge. Following
   measurements of resting respiration, cells were subsequently treated
   with Oligomycin (1.5 μM) for the assessment of non-phosphorylating
   oxygen consumption rate (OCR), with the mitochondrial uncoupler FCCP
   (1.0 μM) for the evaluation of maximal OCR, and with a combination of
   Rotenone and Antimycin A (both at 0.5 μM) for estimating
   extra-mitochondrial OCR. In all cases, three measurements were
   recorded, each over a 5–8 min interval including mixing and incubating.
   At the end of the Seahorse run, to allow comparison among different
   experiments, data were normalized to the total cell amount per well. In
   all cases, the metabolic parameters of the assay–basal and maximal
   respiration, proton leak, and ATP production through oxidative
   phosphorylation– were calculated with Agilent/Seahorse XF Report
   Generator software and expressed as OCR in pmol/min. The results are
   illustrated as a representative from at least three independent
   experiments performed in triplicate or means ± standard error from at
   least three independent experiments performed in triplicate.

Isolation and culture of primary astrocytes

   Primary astrocytes were isolated as described as follows: Fresh mouse
   brain tissue without cerebellum or brain stem tissue was enzymatically
   digested then mechanically dissociated with the gentleMACS Dissociater
   with Heaters. Debris Removal Solution and Red Blood Cell Removal
   Solution were then used to remove myelin, cell debris, and
   erythrocytes. Astrocytes were isolated using Anti-ACSA-2 MicroBeads.
   Microglia were removed by shaking in an incubator overnight and
   astrocytes were verified as negative for microglia through Iba1
   staining. Isolated astrocytes were cultivated in DMEM/F12 with 10% FBS
   penicillin/streptomycin for two weeks before use.[294]^45

Astrocyte lipid droplets staining and quantification

   Lipid droplets in astrocytes were stained and quantified as follows:
   HCS LipidTOX Red Neutral Lipid Stain was used to stain astrocytes
   following the manufacturer’s instruction. Isolated astrocytes were
   cultivated in DMEM/F12 with 10% FBS penicillin/streptomycin for two
   weeks before use. Medium was removed and cells were washed with PBS 3
   times. Cells were then fixed in 4% Paraformaldehyde for 30 min and
   stained with 1X LipidTOX for 30 min at room temperature. A Zeiss LSM
   880 with Airyscan was used to collect Z section images and generate
   three-dimensional images. The surface module of Imaris 9.3.0 was used
   to quantify the number and volume of lipid droplets from at least 3
   animals or batches of cell cultures; n = 3–8 coverslips/condition; 8 ±
   2 cells/coverslip. Each data point represents the averaged value of all
   cells on a coverslip.[295]^45

Lenti-Ndst1shRNA transfection

   Astrocytes isolated from E18 pregnant C57BL/6J, APOE3 and APOE4
   knock-in mice were seeded on Poly-D-Lysine coated 12-well or 6-well
   plate with coverslips at a density of 3 × 10^5 or 5 × 10^3 well with
   complete medium (DMEM/F12 supplemented with 10% Fetal Bovine Serum and
   1% Penicillin & Streptomycin), respectively. The lentivirus stocks
   (Lenti-Ndst1shRNA and scramble) were thawed on ice before transfection.
   The appropriate volume of virus was diluted into the media (DMEM/F12
   supplemented with 10% FBS only) in order to achieve the desired MOI
   (Multiplicity of infection, 5–10 cells) of virus according to the stock
   concentration. After 48h incubation, the medium was aspirated, and the
   cells were washed once with PBS and incubated with complete medium for
   another 3 days. Astrocytes from 12-well plates were collected for
   RT-PCR and 6-well plates with coverslip were used for immunostaining.
   Results were obtained from 4 to 6 independent experiments performed in
   duplicate wells. Transduction efficiency was assessed by analyzing the
   number of cells that showed positive for GFP fluorescence after
   fixation and by staining all cell nuclei with DAPI.

RNA extraction and RT-qPCR from Drosophila and human cultured cells

   Total RNA was isolated from cultured cells described above using TRIzol
   and PureLink RNA Mini Kit (Cat# 12183018A, Thermo Fisher). Extracted
   RNAs were reverse transcribed to cDNA using the SuperScript IV VILO
   Master (Cat# 11756050, Thermo Fisher). Relative mRNA levels of target
   genes (Ndst1, Plin2, Gfap, Fasn, Acc, Pparg, Cpt1α, Pgc1α) were
   amplified by SYBR Green real-time PCR assays in QuantStudio 6 System.
   The sequence for the Ndst1f primer is CCACAACTATCACAAAGGCATCG. The
   sequence for the Ndst1R primer is GAAAGGTGTACTTTAGGGCCAC. The sequence
   for the Plin2F primer is GACCTTGTGTCCTCCGCTTAT. The sequence for the
   Plin2R primer is CAACCGCAATTTGTGGCTC. The sequence for the HprtF primer
   is AGTGTTGGATACAGGCCAGAC. The sequence for the HprtR primer is
   CGTGATTCAAATCCCTGAAGT. The sequence for the GapdhF primer is
   TGACCTCAACTACATGGTCTACA. The sequence for the GapdhR primer is
   CTTCCCATTCTCGGCCTTG. The relative gene expression level was calculated
   by the comparative Ct (ΔΔCt) method when compared to the internal
   control genes (Hprt, Gapdh, β-actin, RPL13) for individual samples.
   Results were obtained from 4 independent experiments performed in
   duplicate wells.

LipidTOX staining

   Lipid droplet staining was performed according to the product manual
   (HCS LipidTOX Neutral Lipid Stains, Cat# [296]H34476 HCS LipidTOX Red
   neutral lipid stain). Briefly, the incubation medium was aspirated, and
   enough formaldehyde fixative solution was added to a 6-well plate to
   cover cells for 30 min at room temperature. Then the fixed cells were
   rinsed with PBS 3 times and incubated with LipidTOX neutral lipid stain
   (1:200 dilution) at room temperature for 30 min. The cells were covered
   with mounting medium containing DAPI (Cat#H-2000, Vector Laboratories,
   Inc). Images were then captured with Zeiss Laser-Scanning Microscopy
   (LSM) 880 with Airyscan detector. z stack images were captured to
   generate three-dimensional volumetric data. The number and volume of
   intracellular lipid droplets were quantified using the surface module
   of Imaris 9.8.0.

Quantification and statistical analysis

Avizo image processing

   Three-dimensional reconstructions of mitochondria were generated using
   Avizo software. Outlines of mitochondria were traced through all slices
   where the structure was visible and interpolated between slices using
   the built-in Avizo function set to the appropriate slice depth of
   100 μm. Images were taken of each mitochondrial reconstruction at an
   approximately equal zoom for analysis. Quantitative data was obtained
   using the built-in measurement function of Avizo and analyzed.

Drosophila gene set analysis

   FlyEnrichr was used to assess gene set representation for the top 500
   significant differentially expressed genes in sfl^RNAi of fat body
   cells[297]^109^,[298]^110 compared to controls (w^RNAi).

   Flow Cytometry statistics were analyzed using GraphPad Prism v10.0
   ([299]https://www.graphpad.com/) and other statistics were analyzed
   using Minitab ([300]https://www.minitab.com/en-us/). Details of
   statistical tests, sample sizes, n, what n represents, and levels of
   significance can be found in figures and their corresponding figure
   legends.

Details of statistical tests

   ANOVA Tukey Tests and Kolmogorov-Smirnov Tests were used in MiniTab.
   Two-tailed heteroscedastic T-Tests were performed in Excel. ANOVA Tukey
   Tests and T-tests were used to analyze collections of data points from
   multiple experiments. Kolmogorov-Smirnov Tests were used to analyze
   averaged distributions of pixel intensity from an experiment, with data
   presented as a representative graph. Data are expressed as mean ± SEM.
   p values of ∗<0.05, ∗∗<0.01, and ∗∗∗<0.001 were considered as
   significant.

Acknowledgments