Neuronal γ-secretase regulates lipid metabolism, linking cholesterol to synaptic dysfunction in Alzheimer’s disease

SUMMARY

Presenilin mutations that alter γ-secretase activity cause familial Alzheimer’s disease (AD) whereas ApoE4, an apolipoprotein for cholesterol transport, predisposes to sporadic AD. Both sporadic and familial AD feature synaptic dysfunction. Whether γ-secretase is involved in cholesterol metabolism and whether such involvement impacts synaptic function remains unknown. Here we show that, in human neurons, chronic pharmacological or genetic suppression of γ-secretase increases synapse numbers but decreases synaptic transmission by lowering the presynaptic release probability without altering dendritic or axonal arborizations. In search of a mechanism underlying these synaptic impairments, we discovered that chronic γ-secretase suppression robustly decreases cholesterol levels of neurons but not in glia, which in turn stimulates neuron-specific cholesterol synthesis gene expression. Suppression of cholesterol levels by HMG-CoA reductase inhibitors (statins) impaired synaptic function similar to γ-secretase inhibition. Thus, γ-secretase enables synaptic function by maintaining cholesterol levels, while chronic suppression of γ-secretase impairs synapses by lowering cholesterol levels.

eTOC Blurb:

Essayan-Perez and Südhof chronically suppressed γ-secretase in human induced neurons to record their synaptic activity and image their synapses. They show that γ-secretase regulates the probability of neurotransmitter release and synapse number. They identified that the mechanism underlying changes in synaptic transmission occurs via regulation of cholesterol levels by γ-secretase.

INTRODUCTION

γ-Secretase, an intramembrane protease, was named because it cleaves the C-terminal fragments (CTFs) of APP at its intramembranous γ-site after APP had been processed by α- or β-secretases. γ-Secretase cleavage of APP CTFs generates secreted Aβ and p3 peptides as well as the APP intracellular domain (AICD) that may act in transcriptional regulation^1–5. The γ-cleavage of APP CTFs is of critical importance in Alzheimer’s disease (AD) pathogenesis, as Aβ forms soluble oligomers that are neurotoxic and accumulates into characteristic amyloid plaques observed in the brains of AD patients. Amyloid plaques are also observed in some cognitively normal individuals and their pathogenic role is disputed. Moreover, point mutations in the APP and the PSEN1 and PSEN2 genes (encoding the two presenilin isoforms that are the catalytic subunits of γ-secretase) alter Aβ peptide production and cause familial AD, providing further evidence for a close relationship of γ-secretase to AD pathogenesis^6–12.

γ-Secretase is a multiprotein complex comprising presenilins and three additional proteins (nicastrin [NCSTN], PEN2 [PSENEN], and APH1) that are evolutionarily conserved and are universally co-expressed with presenilins in most cells². Besides APP, γ-secretase cleaves multiple additional intramembrane protein substrates, such as Notch, suggesting that γ-secretase has broad additional functions in eukaryotic cells besides cleavage of APP¹³. Indeed, during development and in proliferating cells, the cleavage of Notch by γ-secretase is essential for ligand-induced Notch signaling that controls differentiation of cells, such as the differentiation of progenitor cells into neurons, and that regulates cancer cell division, and that is likely the evolutionarily older function of γ-secretase^14–16. Although a vast number of papers examined the role of γ-secretase in development, AD pathogenesis, and cancer^17–20, and although the atomic structures of γ-secretase and its mutants linked to familial AD have been elucidated²¹, few studies to date have explored the physiological functions of γ-secretase in non-proliferating cells such as neurons, and even fewer studies have probed its role in human neurons.

AD is a complex multifactorial disorder that features not only accumulations of extracellular Aβ, but also formation of intracellular tangles composed of tau and neuroinflammatory responses involving microglial and astrocytic activation. In patients, Aβ accumulation can be observed decades before AD becomes clinically manifest, consistent with a central role for Aβ accumulations in AD pathogenesis^19,22,23. Pathophysiologically, AD leads to synaptic dysfunction and altered neural excitability, followed by neuronal cell death^24–27, but the mechanisms involved are unclear. Recent progress in defining the genetic architecture of sporadic AD that accounts for the majority of AD cases identified variations in a large number of genes predisposing to AD^25,28,29. However, the functions of many AD-linked genes are poorly understood, although their expression patterns suggest that a sizable number of AD-linked genes act in neuroinflammation pathways associated with AD pathogenesis^30–32.

Prominent among the genetic risk factors for AD is ApoE, which encodes an apolipoprotein involved in intercellular cholesterol transport, with the ApoE4 variant being the most impactful genetic risk factor for AD^10,12,33,34. ApoE synthesis is enhanced in astrocytes and microglia during neuroinflammation in AD, and ApoE may act by mediating cholesterol transport or by stimulating signaling pathways via a large number of receptors (>10)^33–36. However, the precise role of ApoE in brain remains to be clarified, as does the possible relation of ApoE-mediated cholesterol transport and signaling to AD pathogenesis and to other AD-associated genes such as PSEN1 and PSEN2. Multiple AD risk genes are involved in cholesterol metabolism in addition to ApoE4^10–12,34. Moreover, lipid accumulations containing cholesterol are common in AD patient brains, as well as abnormal endosomes and lysosomes, suggesting a dysfunction of lipid metabolism in neurons in AD^37,38. How cholesterol and other lipids contribute to the neuronal impairments and neuroinflammatory response in AD, and whether they might be related to γ-secretase, is poorly studied.

In previous studies, we and others described a role for γ-secretase in synaptic transmission, demonstrating that genetic suppression of γ-secretase activity in mice using presenilin or nicastrin mutations causes a profound impairment in neurotransmitter release and in short- and long-term synaptic plasticity^39–44. Extensive evidence suggests that mutations in PSEN1 and PSEN2, the most common causes of familial early-onset AD, represent loss-of-function mutations for γ-secretase^45,46, impairing processes such as APP cleavage⁴⁷ and intracellular calcium homeostasis^48,49. However, at least some of the PSEN1 and PSEN2 mutations are also associated with a gain-of-function production of pathogenic Aβ variants, especially Aβ−42^10,12,34, as are rare mutations in APP that cause familial AD^19,22,23. In a recent study in human neurons, Aβ production was enhanced by the APP Swedish mutation at the β-secretase cleavage site and was associated with increased synapse numbers and synaptic transmission, while APP deletion impaired synapses⁵⁰. Therefore, APP cleavage by γ-secretase is likely involved in AD pathogenesis and in regulating synapses, but how γ-secretase and Aβ are related to the synaptic dysfunction of AD is incompletely understood.

Although a large literature examines the role of γ-secretase, Aβ, and ApoE in mouse models of AD, few mechanistic studies were performed on these molecules in mice and even fewer in human neurons whose dysfunction, after all, drives AD pathogenesis. In mouse embryonic fibroblasts (MEFs), γ-secretase inhibition altered lipid composition by lowering cellular cholesterol ester levels⁵¹ and reduced endocytosis of LDLR, thereby lowering cholesterol uptake⁵². Moreover, in MEFs, Aβ-mediated inhibition of HMG-CoA reductase seems to reduce cholesterol synthesis⁵³. However, no studies were carried out in neurons, in particular human neurons, whose cholesterol metabolism is uniquely different from that of glia or peripheral cells^54,55. Given the central role of γ-secretase in AD, the neuronal dysfunction underlying AD symptoms, and the lack of knowledge of γ-secretase function in human neurons, we investigated the function of γ-secretase in human neurons. Our data uncover a causal relationship between γ-secretase, cholesterol metabolism, and synapses that connects genetic traits in familial and sporadic AD to each other. Specifically, we show that chronic suppression of γ-secretase, either pharmacologically or genetically, causes a loss of synaptic function that is induced by a decrease in cholesterol, revealing a direct causal relationship between γ-secretase, cholesterol, and synapses.

RESULTS

Experimental strategy.

We trans-differentiated human embryonic stem (ES) cells into neurons (‘iN cells’) by forced Ngn2 expression, which produces a comparatively homogeneous population of excitatory forebrain neurons (Figure 1A)^35,56,57. The induced human neurons were co-cultured with mouse glia on day 3 since glial factors are required for maturation and synapse formation of human neurons³⁵. We chronically suppressed γ-secretase activity in the human neurons by two approaches. In our first approach, we incubated the human neurons for four weeks (from day 10 to 37–42) with DAPT or LY411575 (LY), two chemically distinct γ-secretase inhibitors⁵⁸. DAPT (IC₅₀ of 115 nM for total Aβ production) and LY411575 (IC₅₀ of 0.082 nM for total Aβ production) are considered peptidomimetic inhibitors of presenilins^58–61. Aβ levels in the cell media decreased with γ-secretase inhibition, as measured by ELISA (Figure 1B). As expected, APP and neurexin CTFs, conversely, increased manyfold when γ-secretase was chronically inhibited, as monitored by immunoblotting with multiple antibodies (Figure 1C–F; S1F–L). Dilution of cell lysates after chronic γ-secretase inhibition confirmed that the observed large increase in the levels of APP CTFs, although dependent on antibody affinities, was reliable and not distorted by massive non-linear effects. Alternative γ-secretase inhibitor treatments further demonstrated significant increases in APP CTFs (Figure S1F–L).

Figure 1: Chronic γ-secretase inhibition increases synapse numbers in human neurons (iN cells) without altering axonal or dendritic arborization.

(A) Experimental strategy. Human neurons were trans-differentiated from naïve or PSEN1-mutant H1-ES cells by forced expression of Ngn2 (Zhang et al., 2013). Neurons were treated with vehicle or one of two chemically different γ-secretase inhibitors (GSIs) starting at 10 days and analyzed by imaging, electrophysiology, biochemistry, or RNAseq at day 37–42.

(B) Chronic γ-secretase inhibition with DAPT (40 μM) reduces Aβ production in cultured human neurons as monitored by ELISA in the medium at day 8 (2 days prior to treatment), day 14, 18 and 21.

(C-F) Chronic γ-secretase inhibition increases the levels of the C-terminal fragments (CTFs) of APP (C & D) and of neurexins (E & F) in human neurons (C, E: representative immunoblot; D, F: summary graphs of CTF levels analyzed at day 37).

(G & H) Chronic γ-secretase inhibition has no effect on axonal or dendritic outgrowth (G, representative images; H, summary graphs of the axon length (top left), dendrite length (top right), number of primary dendrites (bottom left), and dendritic branch points (bottom right)). Neurons treated with vehicle or DAPT starting at day 10 were transfected with tdTomato at day 19 and analyzed on DIV21 to identify early developmental changes in dendritic and axonal architecture.

(I & J) Chronic γ-secretase inhibition robustly enhances synapses numbers (I, representative images of human neurons treated with DAPT or vehicle from day 10 to 37 and stained for presynaptic synapsin, postsynaptic PSD-95, and microtubule-associated protein MAP2 [right, zoomed-in dendrite segments]; J, summary graphs of the density and size of synaptic puncta adjacent to MAP2-positive dendrites stained for synapsin (left) or PSD-95 (middle), or double-labeled for synapsin and PSD-95 (right)). For replication experiments, see Figure S2.

(K & L) Chronic γ-secretase inhibition elevates PSD-95 protein levels (K, representative immunoblots; L, summary graphs of PSD-95 levels as determined in 4 independently replicated experiments). Immunoblots were quantified with fluorescently labeled secondary antibodies; signals were normalized for Tuj1 (Rep1 to 3) or β-actin (Rep 4) as loading controls. For replication experiments, see Figure S2.

Data in (B), (D), (F), (H), (J), (L) are means +/− SEM (numbers in bars show cells/independent biological replicates (H, J), or biological replicates (B, D, F, L). Statistical analyses were performed with two-tailed unpaired t-tests, with * = p<0.05; ** = p<0.01, *** = p<0.001. For additional data, see Figures S1 and S2.

In our second approach, we genetically deleted presenilin-1 (PSEN1) in ES cells, and analyzed neurons derived from multiple PSEN1 knockout ES cell clones (Figure S1A–E). Our PSEN1 gene editing strategy targeted exon 4 with 3 single-guide RNAs. We isolated five cell clones homozygous for +1, −7 or −95 base pair indels causing frameshift mutations (Figure S1B, C). Neurons derived from PSEN1-mutant ES cells exhibited a significant increase in APP CTF levels as monitored by immunoblotting (Figure S1D). Presenilin-1 is subject to auto-cleavage producing a presenilin-1 CTFs that was not detectable in the neurons derived from the PSEN1 KO ES cells, indicating successful disruption of presenilin-1 function in γ-secretase by the genetic deletions (Figure S1E).

Thus, we established independent pharmacological and genetic approaches that chronically suppress γ-secretase function in human neurons. Note that with both approaches, γ-secretase activity is not completely abolished since pharmacological inhibition is intrinsically incomplete and the genetic PSEN1 deletion does not impair expression of the second presenilin gene, PSEN2. In the pharmacological approach, however, γ-secretase activity is suppressed both in neurons and in co-cultured glia, whereas in the genetic approach only neuronal but not glial γ-secretase activity is decreased.

γ-Secretase suppression does not alter neurite extension but increases synapse numbers.

γ-Secretase has multiple substrates with diverse functions¹³, suggesting that by impairing cleavage of these substrates, chronic γ-secretase inhibition could hinder neuronal development even when neurons are directly generated from ES cells. To address this possibility, we analyzed human neurons morphologically after chronic γ-secretase inhibition with DAPT and quantified their axons and dendrites. We observed no significant changes in axonal or dendritic arborization and only a slight decrease in the cell soma size (Figures 1G, H; S2A–D). Given the essential role of γ-secretase in Notch signaling^14,15, it may seem surprising that chronic γ-secretase inhibition does not impair neuronal development, but we bypassed normal early neuronal with the Ngn2-induced trans-differentiation protocol. Also, our pharmacological inhibition begins 10 days after neuronal trans-differentiation was initiated, which may further explain a lack of a Notch phenotype.

We next asked whether chronic γ-secretase inhibition alters synapse formation since AD is thought to be a synaptic disorder^24,26,27. We stained neurons for synapsin and PSD-95 as pre- and postsynaptic markers and quantified the density of synaptic puncta that are positive for either one or both of these markers (Figure 1I; S2F). Unexpectedly, chronic γ-secretase inhibition robustly increased, instead of decreasing, the synapse density (Figure 1J, S2G). Synaptic puncta that are stained by both presynaptic synapsin and postsynaptic PSD-95 exhibited a ~40% increase in density, suggesting that chronic γ-secretase inhibition enhances synapse formation (Figure 1J). Consistent with this conclusion, we detected a 40–60% increase in PSD-95 protein levels and a significant elevation in the levels of an additional subset of synaptic proteins (Figure 1K, L; S2G, H).

γ-Secretase is essential for normal neurotransmitter release.

To determine whether the increase in synapse numbers upon chronic γ-secretase inhibition is associated with increased synaptic transmission, we analyzed spontaneous mEPSCs in human neurons chronically treated with DAPT. In contrast to the increased synapse numbers, we detected no differences in the mEPSC frequency or amplitude (Figure 2A–C). We next monitored evoked EPSCs in chronically DAPT-treated or PSEN1 knockout neurons as a direct measure of synaptic function. Both chronic DAPT treatment and the PSEN1 deletion suppressed synaptic transmission as monitored via the evoked EPSC amplitude and charge transfer, with ~50% and ~30% decrease, respectively (Figure 2D–G). Thus, although chronic γ-secretase inhibition causes an increase in synapse number, it does not change spontaneous mEPSC frequencies and decreases evoked synaptic transmission. There were no changes in the kinetics of mEPSCs or evoked EPSCs (Figure S4E, F, H). The absence of a change in mEPSC frequency in these experiments could be explained as the sum of the increased mEPSC frequency expected from the enhanced synapse numbers, and the decreased mEPSC frequency expected from the weakened synapse function.

Figure 2: γ-Secretase activity is required for normal neurotransmitter release.

(A–C) Chronic suppression of γ-secretase activity produces no major impairments in mEPSCs (A, representative mESPC traces recorded on day 37–42 from human neurons treated chronically with vehicle (0.5%) or DAPT (40 μM); B & C, cumulative probability plots of the interevent intervals (B) and amplitudes (C) of mESPCs [insets: mean mEPSC frequency (B) and amplitude (C)]).

(D & E) Chronic suppression of γ-secretase activity impairs evoked synaptic transmission (D, representative traces; E, summary graphs of the EPSC amplitude (left) and charge transfer (right)).

(F & G) Same as D and E but measured in PSEN1 KO vs. control neurons.

(H & I) Chronic suppression of γ-secretase activity lowers paired-pulse depression, suggesting a decline in release probability (H, representative traces of EPSCs evoked by two closely spaced stimuli [stimulus intervals (20–1000 ms) are indicated above traces; for full traces, see Figure S3]; I, summary plot of the paired-pulse ratios as a function of the interstimulus interval).

(J) Confirmation that chronic suppression of γ-secretase activity lowers synaptic strength as monitored via the amplitude of first evoked EPSCs during paired-pulse stimulation (summary graph of EPSC amplitude).

(K) Chronic suppression of γ-secretase activity increases the coefficient of variation of evoked EPSC amplitudes, consistent with a reduced release probability.

(L–O) PSEN1 mutant vs. control neurons show increased coefficients of variation of evoked EPSC amplitudes, demonstrating that genetic suppression of γ-secretase activity also dramatically decreases the release probability of human neurons. Full traces are shown in Figure S3.

Numerical data are means ± SEM (numbers of cells/experiments are listed in bars). Statistical analyses were performed by two-tailed unpaired t-test (DAPT data) or two-way ANOVA with Bonferroni’s multiple comparison (PSEN1 KO data), with * = p<0.05; ** = p<0.01, *** = p<0.001. For intrinsic properties and controls, see Figure S3 and S4.

A potential cause of the decrease in evoked EPSCs could be a change in neuronal excitability and action potential generation, although this possibility would not explain the unchanged mEPSC frequency. To test the possibility that chronic γ-secretase inhibition alters the electrical properties of neurons, we conducted current clamp recordings. We observed no significant effects of chronic γ-secretase inhibition on action potential firing frequency, excitability, rheobase, peak amplitude, or kinetics (Figure S3A, B). Moreover, we measured the intrinsic electrical properties of the neurons and detected a modest decrease in capacitance and an increase in input resistance, consistent with the decreased soma size observed morphologically (Figure S3C).

An alternative explanation for the decrease in synaptic function upon chronic γ-secretase inhibition is a decline in the neurotransmitter release probability. To test possibility, we monitored EPSCs induced by two closely spaced stimuli (Figure S4B, G). Under our recording conditions in 3 mM extracellular Ca²⁺, human neurons exhibit a high release probability that results in paired-pulse depression (Figure 2H, L). Both chronic pharmacological and genetic γ-secretase inhibition reversed paired-pulse depression, consistent with a decrease in release probability (Figure 2I, M; S4B). As with the single stimuli, chronic γ-secretase inhibition lowered the amplitude of the first of the paired EPSCs by approximately 30% (Figure 2J, N) but significantly increased the coefficient of variation by 50%, which again argues for a decrease in release probability (Figure 2K, O). Membrane properties were also not grossly changed under these conditions, although a modest decrease in capacitance was again detected (Figure S4C–D). Thus, as assessed via two parameters, chronic γ-secretase inhibition impairs synaptic transmission by lowering the neurotransmitter release probability. Given that the mEPSC frequency and amplitude were unchanged upon chronic DAPT treatment whereas the evoked EPSC amplitude was decreased, these data indicate that chronic γ-secretase inhibition generates more synapses whose synaptic strength is decreased because of an impairment in release probability.

Is the decrease in synaptic release probability due to an acute or chronic effect of γ-secretase inhibition? To address this question, we measured the effects of shorter periods of γ-secretase inhibition on evoked EPSCs. We performed a 1-day γ-secretase inhibition in mature neurons on day 37–42 but observed no changes in synaptic transmission (Figure 3A; S5A). However, a 2-day γ-secretase inhibition in mature neurons produced a significant, albeit modest (~25%) decrease in EPSC amplitude (Figure 3B, S5B). The lack of an effect with only 1-day γ-secretase inhibition was not due to a sensitive period in development since γ-secretase inhibition for 1 day at day 21 also did not impair synaptic transmission (Figure 3C; S5C). Conversely, inhibition of γ-secretase from day 10 to 21 for 11 days did impair synaptic transmission, although the effect was smaller than that observed in mature neurons (Figure S5D, E). These results indicate that chronic, but not acute, γ-secretase inhibition impairs synaptic transmission, suggesting an indirect role for γ-secretase in neurotransmitter release.

Figure 3: Acute γ-secretase inhibition does not impair synaptic transmission.

(A) Acute suppression of γ-secretase activity by one-day DAPT treatment at Day 37–42 has no detectable effect on evoked EPSCs (left, representative traces; right, summary graphs of evoked EPSC amplitudes and charge transfer).

(B) Semi-acute suppression of γ-secretase activity by two-day DAPT treatment at Day 37–42 significantly decreases synaptic strength as monitored via evoked EPSCs (left, representative traces; right, summary graphs of evoked EPSC amplitudes and charge transfer).

(C) Acute suppression of γ-secretase activity by a one-day DAPT treatment at Day 20, analyzed on Day 21, has no significant effect on evoked EPSCs (left, representative traces; right, summary graphs of evoked EPSC amplitudes and charge transfer).

Data are means +/− SEM; numbers of cells/replicates are indicated in the bars. Statistical analyses were performed using two-tailed unpaired t-test, with * = p<0.05; ** = p<0.01, *** = p<0.001. For kinetics and intrinsic properties, see Figure S5.

Chronic γ-secretase suppression selectively stimulates transcription of cholesterol synthesis genes.

To explore the mechanisms by which chronic γ-secretase inhibition acts on synapses, we performed RNAseq experiments on human neuron/mouse glia co-cultures after chronic inhibition of γ-secretase with either DAPT or LY411575 and analyzed the neuronal and glial transcriptomes separately based on their species of origin^62,63. Six independent biological replicates were performed to ensure reproducibility. Strikingly, all top upregulated genes induced by chronic γ-secretase inhibition in human neurons, but none in mouse glia, are involved in cholesterol synthesis or transport (Figure 4A, C, E). Expression of nearly every major gene in cholesterol metabolism was stimulated by γ-secretase inhibition in the human neurons (Figure S6A; Luo et al., 2020). In contrast, in mouse glia, no cholesterol metabolism-related gene was upregulated by chronic γ-secretase inhibition (Figure 4B, D, F). Different from upregulated genes, the downregulated genes in human neurons were functionally heterogeneous (Figure 4A, C, G). Here, the largest subgroup in the top 20 downregulated genes included immediate early genes, which are constitutively expressed in cultured human neurons given their spontaneous network activity⁶⁴. The lack of cholesterol metabolism gene changes in mouse glia suggests a cell type-specific transcriptional role of γ-secretase.

Figure 4: Chronic suppression of γ-secretase activity upregulates cholesterol synthesis gene expression in human neurons but not in mouse glia.

(A & B) Volcano plots of differentially expressed genes (DEGs) from co-cultured human neurons (A) and mouse glia (B) that were chronically treated with the γ-secretase inhibitor DAPT (40 μM) or vehicle (DMSO only) from day 10–40, and analyzed by bulk RNAseq. Data resulted from pairwise comparisons of each treatment group versus vehicle, normalized to raw counts, with a standard False Discovery Rate of 0.1; data underwent the Benjamini-Hochberg correction to obtain adjusted p-values (yellow, cholesterol synthesis genes; pink, cholesterol transport genes; blue, activity-dependent immediate early genes; light green, growth factors; dark green, protease inhibitors).

(C & D) Same as A and B, except that the cells were treated with the γ-secretase inhibitor LY411,575 (2.5 μM).

(E & F) Bar graphs for the top 20 upregulated genes identified by RNAseq analyses of DAPT-and LY411575-treated human neurons (E) and mouse glia (F) that exhibited adjusted p-values of <0.0001. The cutoff for log2(fold change) values (y-axis) was >0.3, corresponding to a fold change greater than 20% relative to controls.

(G & H) Same as E & F, but for the top 20 downregulated genes, using adjusted p-values of <0.002 (G) or <0.0001 (H).

(I) qRT-PCR measurement of 4 upregulated cholesterol synthesis pathway genes in three lines of PSEN1 KO human neurons, DAPT- and LY411,575-treated human neurons.

(J) qRT-PCR measurement of 4 upregulated cholesterol synthesis pathway genes in three lines of PSEN1 KO human neurons, DAPT- and LY411,575-treated human neurons.

(K) qRT-PCR measurement of 6 downregulated genes in three lines of PSEN1 KO human neurons in DAPT- and LY411,575-treated human neurons.

Data in (E-J) are means +/− SEM: (A-H) have 6 biological replicates; (I and J) have 3 biological replicates. Statistical analyses for were performed using the Wald test (A-H) and two-tailed unpaired t-test (I-J) and plotted as fold change relative to controls (*p<0.05; **p<0.01, ***p<0.001). For additional validation data, see Figure S6.

In mouse glia (whose γ-secretase activity is also pharmacologically inhibited in these experiments) both up- and downregulated genes were also functionally heterogeneous, except for a preponderance of secreted protease inhibitors among the glial downregulated genes (Figure 4B, D, F, H). In addition, chronic γ-secretase inhibition lowered expression in glia of Hes5, a Notch-target transcription factor⁶⁵, and upregulated expression of colony stimulating factor 1 receptor (Csf1r), indicating possible regulation of microglia⁶⁶. Prior studies suggested that γ-secretase dampens inflammation and Toll-like receptor 4-mediated immune response via the processing of LRP1⁶⁷, but this pathway was not a major component of the genes we identified. Despite the induction of microglial marker Csf1r, chronic γ-secretase inhibition did not appear to induce expression of stress or inflammatory responses in neurons or glia, suggesting that the inhibition did not cause an injury response.

Since the induction of cholesterol metabolism genes by chronic γ-secretase inhibition in human neurons was absent from mouse glia, it does not represent a general cellular pathway controlled by γ-secretase but is specific to neurons, which may explain why it has not been observed previously. To confirm the gene expression changes found by RNAseq and to ensure that these changes are not only caused by pharmacological inhibition of γ-secretase, we performed quantitative RT-PCR analyses of selected genes in human neurons with a chronic pharmacological or a genetic γ-secretase impairment (i.e., PSEN1 deletion). These analyses validated the RNAseq results both for the upregulation of cholesterol metabolism genes and the downregulation of immediate early genes (Figures 4I, J; S6B, C). Cholesterol gene mRNA levels showed an approximately two-fold increase while immediate-early gene mRNA levels exhibited an approximately two-fold decrease with pharmacological inhibition (Figures 4I, J; S6). PSEN1 KO neurons exhibited more modest but still significant changes, possibly because the continued expression of PSEN2 may produce a lesser suppression of γ-secretase activity (Figures 4I, J). Moreover, measurements of the mRNA levels at different times of pharmacological γ-secretase inhibition revealed that γ-secretase inhibition of 48 hours or more was required to induce changes in cholesterol synthesis gene expression (Figure S6D), consistent with the time course of synaptic impairments (Figure 3). Therefore, two transcriptomic approaches in multiple replicates confirmed that nearly all key genes in the cholesterol synthesis pathway were specifically upregulated in neurons upon chronic suppression of γ-secretase activity (Figure S6A).

γ-Secretase inhibition decreases cellular cholesterol levels.

We next asked whether the increased expression of cholesterol synthesis genes upon chronic γ-secretase inhibition might be due to a loss of cellular cholesterol, which would induce cholesterol synthesis gene expression, or conversely cause an increase in cellular cholesterol. To address this question, we used GFP-expressing human neurons that were chronically treated with γ-secretase inhibitors (DAPT or LY411575) or that contained the PSEN1 deletion. We stained the neurons with filipin, a fluorescent dye that inserts into the plasma membrane in a cholesterol-dependent manner^68,69, and with antibodies to L1CAM, a surface marker of human neurons⁷⁰. The co-localization of the filipin fluorescence, GFP, and L1CAM signals then allowed us to specifically quantify the filipin signal from neurons (Figure 5). Both chronic pharmacological γ-secretase inhibition and the decrease in γ-secretase activity induced by the PSEN1 mutation robustly decreased (~25%) the plasma membrane filipin signal in neurons (Figure 5). Given the tight regulation of plasma membrane cholesterol levels in cells, this is a large effect⁷¹.

Figure 5: Chronic suppression of γ-secretase activity reduces plasma membrane cholesterol in human neurons.

(A) Representative images illustrating reduced Filipin staining (blue) in human neurons upon chronic suppression of γ-secretase activity with DAPT or LY411,575. Neurons are labeled by GFP expression (green), and axons are labeled by L1CAM staining (red). Scale bars (white line) represent 10 μm.

(B) Chronic suppression of γ-secretase activity with DAPT or LY411,575 causes a robust decrease in the plasma membrane cholesterol content of human neurons as monitored by Filipin signal in GFP-positive human neurons.

(C & D) Same as A & B, but for PSEN1 KO neurons obtained from three independent KO lines.

Data in (B) and (D) are means ± SEM (numbers of cells/biological replicates) are shown in bars. Statistical analyses were conducted using two-tailed unpaired t-test, with * = p<0.05; ** = p<0.01, *** = p<0.001.

To independently validate the finding that chronic γ-secretase inhibition lowers cellular cholesterol levels in neurons, we isolated GFP-positive neurons and GFP-negative glia cells at 40 days post induction by fluorescence-activated cell sorting (FACS) (Figure 6A, S7). Cells were isolated after 30 days of pharmacological γ-secretase inhibition with DAPT or LY411575. We then measured the cellular content of unesterified and esterified cholesterol in both neurons and glia (Figure 6A). Again, we observed an approximately 20% decrease in total cholesterol in neurons, whereas cholesterol in glia was not significantly changed (Figures 6B, C). Given the lengthy time of the cell separation procedure, the consistency of the results between filipin staining and chemical cholesterol measurements is reassuring. Viewed together, these data indicate that chronic γ-secretase inhibition lowers neuronal but not glial cholesterol levels, which induces transcription of genes involved in cholesterol synthesis and transport in neurons in response to the decreased cellular content of cholesterol.

Figure 6: Chronic suppression in γ-secretase activity decreases neuronal but not glial levels of unesterified and esterified cholesterol.

(A) Experimental approach: GFP-positive human neurons were separated from GFP-negative mouse glia using FACS and their free and esterified cholesterol contents were measured enzymatically.

(B & C) Summary graphs showing that free and esterified cholesterol levels are diminished in human neurons (B) but unchanged in mouse glia (C). Data in are means +/− SEM (numbers in bars indicate biological replicates). Statistical analyses were conducted using one-sample t-tests with theoretical mean of 1.0, with * = p<0.05; ** = p<0.01, *** = p<0.001. For representative FACS plots, see Figure S7.

Direct suppression of cholesterol synthesis impairs neurotransmitter release.

Does γ-secretase regulate the presynaptic release probability directly, or does it act indirectly by controlling the neuronal cholesterol content? To address this question, we suppressed cholesterol synthesis in neurons using chronic treatment with two unrelated HMG-CoA reductase inhibitors (rosuvastatin and simvastatin; referred to as statins)^72,73, either alone or in combination with DAPT as a γ-secretase inhibitor (Figure S8A). As expected, both statins induced transcription of cholesterol synthesis genes by four-fold since they lower cellular cholesterol levels (Figure 7A). The two statins also decreased immediate-early gene expression in a similar manner to DAPT and LY411575 (Figure 7B). For both transcriptional phenotypes, the effects of chronic cholesterol synthesis inhibition by statins and chronic γ-secretase inhibition by DAPT were not additive, suggesting that the changes operate via the same pathway.

Figure 7: HMG-CoA reductase inhibitors (statins) upregulate cholesterol synthesis gene and downregulate activity-dependent immediate-early gene expression but do not increase synapse numbers.

(A & B) qRT-PCR measurements reveal that chronic inhibition of HMG-CoA reductase, a rate-limiting enzyme in cholesterol biosynthesis, induces upregulation of the same cholesterol-synthesis genes (A) and downregulation of the same set of genes including immediate-early genes (B) as chronic inhibition of γ-secretase. Neuron / glia co-cultures were treated with vehicle (0.5% DMSO), rosuvastatin (1 μM), or simvastatin (1 μM) without or with DAPT (40 μM) from day 10 to 37.

(C) Representative immunocytochemistry images of human neurons treated with vehicle, DAPT, vehicle + Rosuvastatin, or DAPT + Rosuvastatin from day 10 to 37. Neurons were stained for presynaptic synapsin, postsynaptic PSD-95, and microtubule-associated protein MAP2.

(D) Chronic treatment of human neurons with statins, different from chronic treatment of human neurons with γ-secretase inhibitors that robustly enhance synapse numbers, do not induce an increase in synapse density. Panels show summary graphs of the density (top graphs) and size (bottom graphs) of synaptic puncta labeled for synapsin (left) or PSD-95 (middle), or double-labeled for co-localized synapsin and PSD-95 (right). Only puncta adjacent to MAP2-positive dendrites were quantified.

Data in (A), (B), (D) +/− SEM (numbers in bars show cells/independent biological replicates). Statistical analyses were performed with unpaired t-tests for (A) and (B), and with two-way ANOVA with Bonferroni’s multiple comparison for (D), with * = p<0.05; ** = p<0.01, *** = p<0.001.

Next, we tested whether statins had an effect on synapse numbers by measuring the density of synapsin- and PSD95-positive puncta in cells treated with vehicle, DAPT, vehicle and rosuvastatin, or DAPT and rosuvastatin (Figure 7C). Chronic γ-secretase inhibition increased synapse density consistently by 30 to 50%, replicating the results described in Figure 3, whereas statins did not (Figure 7D). Instead, chronic lowering of cholesterol produced a non-significant trend of lowering synapse numbers without preventing the increase in synapse numbers caused by chronic γ-secretase inhibition (Figure 7D). These results suggest that chronic γ-secretase inhibition changes synapse number via cholesterol-independent mechanisms.

Finally, we measured evoked EPSCs to analyze synapse function. Chronic cholesterol synthesis inhibition by statins suppressed synaptic strength, as monitored via the EPSC amplitude and EPSC charge transfer, by 30% and 50% respectively (Figure 8A–C). No effect by statins on membrane properties or EPSC kinetics were detected (Figure S8B, C). The phenotype induced by chronic cholesterol synthesis inhibition was as strong as that of chronic γ-secretase inhibition. Moreover, the effects of the γ-secretase and cholesterol synthesis inhibitions were also not additive, corroborating the notion that they operate in the same pathway (Figures 8A–C). To further compare the effects of chronic γ-secretase and cholesterol synthesis inhibition on synapses, we measured paired-pulse responses (Figure 8D). Chronic γ-secretase and cholesterol synthesis inhibition induced similar effects on the paired-pulse ratio indicative of a decrease in release probability, with a 50% decrease in the amplitude of the first EPSCs compared to controls (Figure 8E–G). Again, their combined application was not additive, indicating that chronic cholesterol synthesis inhibition decreased the release probability (Figures 8E, F). Finally, chronic γ-secretase and cholesterol synthesis inhibition similarly enhanced the coefficient of variation of EPSCs, and here their combined effects were also not additive (Figure 8G). Together, these data support the hypothesis that chronic γ-secretase inhibition suppresses synaptic strength in human neurons by decreased neuronal cholesterol synthesis and reduced neuronal cholesterol content.

Figure 8: HMG-CoA reductase inhibitors (statins) suppress the neurotransmitter release probability in human neurons.

(A) Representative EPSC traces from neurons treated with DAPT and HMG-CoA reductase inhibitors as described for A & B.

(B & C) Chronic inhibition of HMG-CoA reductase causes the same suppression in synaptic strength, measured via the amplitude (B) or charge transfer (C) of evoked EPSCs, as chronic inhibition of γ-secretase. Note that the effects of chronic inhibition of HMG-CoA reductase of γ-secretase are not additive.

(D) Representative traces of EPSCs evoked by two closely spaced stimuli (paired-pulse stimulation), monitored in human neurons as a function of chronic treatments with γ-secretase and/or HMG-CoA reductase inhibitors.

(E) Chronic suppression of the HMG-CoA reductase and/or γ-secretase activity reverses paired-pulse depression, suggesting a decline in release probability. The summary plot depicts the paired-pulse ratios (second/first EPSC amplitude) as a function of the interstimulus interval.

(F) As shown for the isolated EPSCs, chronic suppression of the HMG-CoA reductase and/or γ-secretase activity decreases the amplitude of evoked EPSCs monitored via the first response during paired-pulse stimulation.

(G) Chronic suppression of the HMG-CoA reductase and/or γ-secretase activity also increases the coefficient of variation of evoked EPSC amplitudes, consistent with a reduced release probability.

Data are means +/− SEM (the number of cells/biological replicates are shown in the bars). For (B), (C), (F), (G), statistical analysis was measured using two-tailed unpaired t-test. For (E), statistical analysis was assessed by two-way ANOVA with Bonferroni’s multiple comparison test. Statistical significance: *p<0.05; **p<0.01, ***p<0.001. For additional data, see Figure S8.

DISCUSSION

Here we show that chronic γ-secretase inhibition induced by pharmacological treatments or by deletion of PSEN1 in human neurons causes a decrease in cellular cholesterol levels that results in synaptic dysfunction by suppressing the presynaptic neurotransmitter release probability. The decrease in cholesterol by chronic γ-secretase inhibition is only observed in neurons but not in glia, and strongly induces neuronal cholesterol synthesis gene expression. The fact that cholesterol synthesis genes are the major mRNAs that are upregulated in human neurons by chronic γ-secretase inhibition suggests that cellular cholesterol metabolism is a prominent target of γ-secretase function in neuronal physiology. Our results thus identify a neuron-specific role for γ-secretase in regulating cellular cholesterol levels, thereby defining a functional link between PSEN1, the predominant gene involved in familial AD, and cholesterol metabolism, a major pathway regulated by ApoE that is implicated in sporadic AD. In addition, we show that chronic γ-secretase inhibition causes an increase in synapse numbers that is not induced by the decrease in cholesterol levels with statins. Lowering neuronal cholesterol levels with statins results in the same decrease in neurotransmitter release probability as with chronic γ-secretase inhibition, but does not increase synapse numbers, which is consistent with a multifaceted function of γ-secretase in neurons.

A relation of cholesterol metabolism to the presynaptic release probability was previously described in rodent neurons. Specifically, lowering cholesterol using methyl-beta-cyclodextrin, chronic statin (HMG-CoA reductase inhibitor) treatments, or knockdown of the cytoplasmic cholesterol transport protein ORP2 decreased the release probability of hippocampal neurons^74–78. These findings agree well with our results in human neurons using two chemically distinct statins as cholesterol-synthesis inhibitors, therefore providing a cholesterol-dependent mechanism by which γ-secretase regulates synaptic function. Both chronic γ-secretase and cholesterol synthesis inhibition not only increased cholesterol synthesis gene expression, but also lowered immediate-early gene transcription, consistent with the impairment in synaptic transmission. Somewhat surprisingly, chronic γ-secretase inhibition did not impair development of neuronal axons and dendrites but did produce a small decrease in membrane capacitance and soma size. Since the capacitance decrease was also observed with chronic statin treatments (soma sizes were not measured), it is likely due to a decline in available cholesterol in neurons. Such decline may limit soma expansion during neuronal maturation, suggesting that neurons with a lowered cholesterol content prioritize extension of axonal and dendritic arbors over soma growth.

Further, our data suggest that γ-secretase controls synapse numbers by a cholesterolin-dependent process since chronic γ-secretase inhibition increased the synapse density, whereas statins did not. In fact, statins produced a trend towards a decrease in synapse number. Moreover, lowering cholesterol with statins also did not prevent the increase in synapse density produced by chronic γ-secretase inhibition. The increase in synapse numbers induced by chronic γ-secretase inhibition is unlikely to represent a compensatory process in response to decreased synaptic transmission, because impairments in neurotransmitter release induced by other mechanisms, including statins, do not increase synapse numbers. Given that γ-secretase has multiple substrates not only in developing but also in mature neurons, the presence of multiple pathways regulated by γ-secretase is plausible. Important regulators of neuronal functions and synaptic organizers, such as neurexins^79,80, neuroligins⁸¹, cadherins⁸², protocadherins⁸³ and Notch pathway ligands⁸⁴ are known to be γ-secretase substrates and may be involved. In this regard, we show that chronic γ-secretase inhibition does not cause a multitude of impairments in developing neurons but appears to have a major positive effect on synapse numbers and a large negative effect on the presynaptic release probability. The restricted range of major γ-secretase functions in committed but developing human neurons emerging from our experiments agrees well with pioneering studies from the Shen lab that also identified a decrease in presynaptic release probability as the major consequence of presenilin and nicastrin deletions in rodent hippocampus^40–42. It is therefore possible that the phenotype of the presenilin deletions in rodent hippocampus was also caused by a lowering of neuronal cholesterol levels.

Since chronic γ-secretase inhibition downregulated cholesterol levels and upregulated cholesterol synthesis and transport gene expression only in neurons but not in glia, the function of γ-secretase in cholesterol metabolism does not represent a general cellular pathway controlled by γ-secretase, but at least in brain appears to be neuron-specific. Chronic γ-secretase inhibition lowered expression in glia of Hes5, a Notch-target transcription factor⁶⁵, and upregulated expression of colony stimulating factor 1 receptor (Csf1r), indicating possible regulation of microglia⁶⁶. However, no upregulation of genes associated with ‘reactive microglia’, such as those observed for the disease-associated microglial gene expression signature^85,86, was detected. Prior studies suggest that γ-secretase dampens inflammation and Toll-like receptor 4-mediated immune response via the processing of LRP1⁶⁷, but this pathway was also not a major component of the genes we identified. Chronic γ-secretase inhibition did not appear to induce expression of stress or inflammatory responses in neurons or glia, such as reactive oxygen species-related genes and cytokine-response genes^87,88. Given the differences in surface proteins of neurons and glia, the repertoire of γ-secretase targets is likely distinct between neurons and various types of glia. It is therefore plausible that γ-secretase activity results in cell type-specific CTFs and ICDs with separate effects on cellular functions, including gene transcription, and that chronic γ-secretase inhibition induces distinct phenotypes in neurons and glia as a result.

The impairment in synaptic function by chronic γ-secretase inhibition that we observed in human neurons agrees well with previous results obtained in mouse models^41,89 and similar impairments induced by mutations of the AD risk gene BIN1⁹⁰, although others have reported a decrease in synapse numbers upon γ-secretase inhibition⁹¹. Early studies reported that the PSEN1 deletion in the hippocampus caused no significant change in paired-pulse facilitation, suggesting no change in release probability⁹², but later studies of the hippocampus with a PSEN1/2 double deletion uncovered a large decrease in release probability⁴¹. The differences in results between these studies and our data, which agree with Zhang et al. [2009], may be due to the extent of the suppression of γ-secretase activity or the time period of suppression. The studies by Zhang et al. [2009] and Yu et al. [2001] on mouse hippocampus analyzed different gene dosages at slightly different time points. However, impaired neurotransmitter release in the mouse hippocampus following suppression of γ-secretase by deletion of nicastrin⁹³ supports the general notion that a decrease in γ-secretase activity causes a decrease in presynaptic release probability.

The finding that statins (HMG-CoA reductase inhibitors), which are taken by millions of people worldwide, cause a large effect on neurotransmitter release in human neurons replicates earlier findings in mouse neurons^74,76. Generally, statins are not known to impair cognition in people, but there appears to be limited consensus on this subject despite the widespread use of statins. A small number of randomized clinical studies with lovastatin and simvastatin, lipophilic statins that can cross the blood-brain barrier, reported cases of short-term cognitive impairment and mild depressive symptoms that correlated with initiation of statins and that subsided once statins were discontinued^94–96. Importantly, a meta-analysis of 25 randomized controlled trials including over 29,000 patients showed no significant adverse cognitive effects due to statin therapy⁹⁷. Our findings raise the question of why the impairment in neurotransmitter release that we observed upon treatment of neurons with these drugs at therapeutic doses does not affect the cognitive performance of human subjects. Likely, for hydrophilic statins, cognitive side effects are not observed in human subjects due to the blood brain barrier that limits the extent to which these drugs can exert their effects on the brain. More generally for all statins, compensatory neural circuit adjustments by other neuronal subtypes may operate in the human brain. Our data also suggest that chronic statin use upregulates cholesterol synthesis genes in neurons, which may offset any changes in the brain following the initiation of these therapies.

The field of AD has long debated the potential use of statins as a treatment strategy, though there is also no consensus to date on this subject⁹⁸. Prior comparative studies across 22,000 medical records revealed a significantly decreased likelihood of an AD diagnosis among individuals taking lovastatin and pravastatin for cardiac reasons, as compared to individuals taking non-statin-based cardiac therapies^99,100. However, a large randomized controlled trial in the general elderly population detected no effect on cognitive decline when statins were used for 42 months¹⁰¹, although this study was not specifically investigating AD. Therefore, the potential of targeting cholesterol for AD therapeutics remains unclear, despite the substantial evidence for cholesterol as a risk factor for AD with ApoE4 as a prominent risk factor. While our study does not focus on AD modeling, our results provide insights into an emerging link between γ-secretase, typically associated with early-onset AD, and cholesterol, relevant for sporadic AD.

Limitations of our study.

Our findings uncover a mechanism whereby γ-secretase regulates synapse function, namely via control of the neuronal cholesterol content, but our data have not identified which γ-secretase substrates among the hundreds described¹³ are responsible for controlling cholesterol metabolism. The same limitation applies to the control of synapse numbers by γ-secretase that is independent of cholesterol.

Moreover, our findings do not provide evidence directly related to AD pathogenesis and we do not know how the control of neuronal cholesterol levels by γ-secretase is related to AD, although the prominent involvement of cholesterol metabolism in sporadic AD suggests a strong connection. Prior studies describe that modifying neuronal cholesterol levels can regulate APP cleavage by γ-secretase, tau phosphorylation, and tau aggregation^73,102–104. These data suggest possible pathways that connect cholesterol metabolism to AD pathogenesis, but other mechanisms are also plausible. Independent of whether the neuron-specific functions of γ-secretase in cholesterol metabolism are relevant to AD pathogenesis, the control of cholesterol by γ-secretase sheds new light on the role of this fascinating enzyme in neurons.

A different limitation of our study is that it does not address a puzzling clinical observation on a rare genetic skin disease, hidradenitis suppurativa. A subset of patients with this disorder harbors loss-of-function mutations in various γ-secretase subunits but these patients do not appear to suffer from an increased risk for AD^105,106. This observation suggests that a decrease in γ-secretase activity does not necessarily lead to AD pathogenesis, although the change in γ-secretase activity caused by the hidradenitis suppurativa mutations may be more modest in neurons than in skin cells and may be qualitatively different. The γ-secretase mutations in hidradenitis suppurativa primarily target NCSTN and PESENEN whereas in familial AD the mutations are observed in PSEN1 or PSEN2, suggesting that various γ-secretase mutations may differentially affect γ-secretase activity in distinct tissues and for diverse substrates.

Furthermore, we validated the pharmacological inhibition of γ-secretase with a genetic PSEN1 deletion, but inhibited cholesterol biosynthesis only pharmacologically and not genetically. Although we used two well-characterized chemically distinct HMG-CoA reductase inhibitors (statins) in our experiments, additional experiments using a CRISPR-mediated deletion of cholesterol synthesis enzymes to confirm the selective action of statins would provide further confidence.

Finally, an additional limitation of our study is that we analyzed a relatively homogeneous population of glutamatergic human neurons produced by trans-differentiation from ES cells with Ngn2 overexpression. It is possible that these neurons are not representative of human glutamatergic neurons in general and may be particularly sensitive to chronic γ-secretase inhibition and statins.

Outlook.

In summary, our data describe a mechanism by which γ-secretase regulates synaptic transmission, namely via its control of neuronal cholesterol levels that in turn determine the presynaptic release probability. We reveal a previously unknown primary function of γ-secretase in maintaining the cholesterol levels of neurons but not glia. Our work demonstrates that decreases in cholesterol levels, produced by suppression of γ-secretase or of cholesterol synthesis, severely impair synaptic transmission. This is a general cell-biological process that may be related to AD pathogenesis, given that cholesterol metabolism is clearly affected by AD-linked genes and by the pathogenesis of AD. The link of γ-secretase to cholesterol metabolism that we identified in the present study connects γ-secretase dysfunction to cholesterol changes and to synaptic dysfunction in AD, opening a new avenue to the identification of AD drug targets.

STAR METHODS

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Dr. Thomas C. Südhof (tcs1@stanford.edu).

Materials Availability

This study did not generate new unique reagents.

Data and Code Availability

All data and renewable materials reported in this paper will be shared by the lead contact upon request (lead contact, Thomas C. Südhof (tcs1@stanford.edu)). This paper does not report original code. Bulk RNA-seq data have been deposited at GEO and are publicly available as of the date of publication. Accession numbers are listed in the Key Resources table.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Mouse monoclonal anti Syt1	Synaptic Systems	Cat# 105-011C2: RRID:AB_2619760
Chicken polyclonal anti MAP2	Encor	CPCA-MAP2, RRID:AB_2138173
Rabbit polyclonal anti Syn1	Sudhof Lab	Clone YZ6078, RRID: AB_2861225
Mouse monoclonal anti PSD95	Thermo	Cat# MA146: RRID:AB_2092361
Rabbit polyclonal anti APP	Ho et al., 2008	Clone U955
Mouse monoclonal anti APP	Sigma	Cat# MABN380: RRID: AB_2714163
Rabbit polyclonal anti CASK	Neuromab	Cat# O14936: RRID: AB_2877188
Rabbit polyclonal anti Fe65	Cao and Südhof, 2001	Clone 4194
Rabbit polyclonal anti Mint1	Ho et al., 2008	Clone P730
Rabbit polyclonal anti Nlgn1	Jahn Monoclonals	Clone 4C12
Rabbit polyclonal anti pan-Nrxn	Sclip and Südhof, 2020	Clone G394, RRID: AB_2800397
Rabbit polyclonal anti RIM1/2	Acuna et al., 2015	Clone U952
Rabbit polyclonal anti Stx1 a/b	Zhou et al., 2012	Clone 438B
Rabbit polyclonal anti PSD95	Patzke et al., 2019	Clone L667: RRID:AB_2800396
Mouse monoclonal anti L1CAM	Sigma	Cat# L4543: RRID:AB_609903
Rabbit monoclonal anti APP	Abcam	Cat# Ab32136: RRID: AB_2289606
Chemicals, Peptides, and Recombinant Proteins
Accutase	Innovative Cell	Cat# AT104
Gem21 Neuroplex supplement	Gemini Bio	Cat# 400–160
BDNF	PeproTech	Cat# 450–02
Doxycycline	Sigma-Aldrich	Cat# D9891
DMEM media	Thermo Fisher	Cat# 11965092
DMEM/F-12 media	Thermo Fisher	Cat# 11320082
HyClone fetal bovine serum (FBS)	GE Healthcare	Cat# SH30071.03
MEM media	Thermo Fisher	Cat# 51200038
Matrigel	BD Biosciences	Cat# 356230
mTeSR1 media	Stem Cell Technologies	Cat# 85850
N-2 supplement	Thermo Fisher	Cat# 17502048
Neurobasal-A Medium	Thermo Fisher	Cat# 10888022
NT-3	PeproTech	Cat# 450–03
Puromycin	Sigma-Aldrich	Cat# P8833
Thiazovivin	Bio Vision	Cat# 1681
Laminin-1	Invitrogen	Cat# 23017–015
Trypsin / EDTA	Thermo Fisher	Cat# 25300120
Papain, Suspension	Worthington Biochemical	Cat# LS003127
Cytosine arabinofuranoside (AraC)	Sigma-Aldrich	Cat# C6645
Glutamax	Thermo Fisher	Cat# 35050079
TTX	Fisher Scientific	Cat# 50-753-2807
EDTA	Sigma-Aldrich	Cat# E6758-100G
Non-Essential Amino Acids	Thermo Fisher	Cat# 11140076
Human Holotransferrin	Gemini Bio	Cat# 800-131P
Total Human Abeta ELISA	Clonetech	Cat# 27729A
DMSO, Cell Culture Grade	Sigma	Cat# D2650
DAPT	Tocris	Cat# 2634
LY411575	Selleckchem	Cat# S2714
Rosuvastatin	Selleckchem	Cat# S2169
Simvastatin	Selleckchem	Cat# S1792
TRIzol	Invitrogen	Cat# 15596026
RNeasy Plus Mini Kit	Qiagen	Cat# 74104
Filipin	Sigma	Cat# SAE0088
Amplex Red Cholesterol Kit	Invitrogen	Cat# A12216
Experimental Models: Cell Lines
Human H1-ESCs	WiCell Research Institute	Cat# WA01
HEK293T cells	ATCC	Cat# CRL-11268
Experimental Models: Organisms/Strains
CD-1 IGS mouse	Charles River	Cat# Crl:CD1(ICR)
Quantitative qRT-PCR
Human ACAT2	IDT	Hs.PT.56a.26153626
Human ACTB	IDT	Hs.PT.39a.22214847
Human DBH	IDT	Hs.PT.58.39251730
Human DHCR24	IDT	Hs.PT.58.4561516
Human DHCR7	IDT	Hs.PT.58.38763048
Human DUSP4	IDT	Hs.PT.58.1640850
Human DUSP5	IDT	Hs.PT.58.1558918
Human EGR1	IDT	Hs.PT.58.40805543.g
Human FDFT1	IDT	Hs.PT.58.25526506
Human FOS	IDT	Hs.PT.58.15540029
Human HMGCR	IDT	Hs.PT.58.41105492
Human HMGCS1	IDT	Hs.PT.58.4084870
Human IDI1	IDT	Hs.PT.58.19303620
Human JUNB	IDT	Hs.PT.58.27174746.g
Human LDHA	IDT	Hs.PT.58.22929122
Human LDLR	IDT	Hs.PT.58.2004261
Human LSS	IDT	Hs.PT.58.24976485
Human MAP2	IDT	Hs.PT.58.20680759
Human MVD	IDT	Hs.PT.58.39035173
Human NAB2	IDT	Hs.PT.58.24622150
Human NR4A1	IDT	Hs.PT.58.20779713
Human SCD	IDT	Hs.PT.58.45714389
Human SREBF2	IDT	Hs.PT.58.45335433
Human SV2C	IDT	Hs.PT.58.4640250
Human TH	IDT	Hs.PT.58.369742
Human TMEM97	IDT	Hs.PT.58.40917738
Human VEGFA	IDT	Hs.PT.58.1149801
Human VGF	IDT	Hs.PT.58.26072166
Recombinant DNA
FUW-TetO-Ng2-T2A-puromycin	Zhang et al., 2013	52047 Addgene (lentiviral vector)
FUW-rtTA	Zhang et al., 2013	20342 Addgene (lentiviral vector)
FUW-TetO-EGFP	Vierbuchen et al., 2010	30130 Addgene (lentiviral vector)
pREV	Dull et al., 1998	12253 Addgene (lentiviral helper vector)
pRRE	Dull et al., 1998	12251 Addgene (lentiviral helper vector)
pVSVg	Patzke et al., 2019	12259 Addgene (lentiviral helper vector)
Deposited Datasets
Bulk RNA-seq data	GEO	GSE206102
Software and Algorithms
ImageJ	NIH	RRID: SCR_003070
ImageStudio software	Li-Cor	RRID:SCR_013715
pClamp	Molecular Devices	RRID:SCR_011323
Nikon NIS-Elements	Nikon	RRID:SCR_014329
GraphPad Prism	GraphPad Prism	RRID:SCR_002798
RStudio	RStudio Team	RRID:SCR_000432
DESeq2	Bioconductor (Love et al., 2014)	RRID:SCR_015687

EXPERIMENTAL MODEL

ESC Culture.

Male human embryonic stem cells (ESC), line H1, were obtained from WiCell (line WA01). The stem cells were maintained feeder-free in mTeSR1 medium (Stem Cell Technologies), from frozen cell stocks of passage 50. With each passage, cells were detached with Accutase at 37°C, centrifuged and resuspended in mTeSR1 with 2 μM thiazovivin (BioVision), and then replated on Matrigel-coated 6-well plates. The Stem Cell Research Oversight (SCRO) at Stanford University approved the protocols used in this work (SCRO 518).

Mouse Lines.

CD1 wild type female and male pups were used between postnatal days 0 to 2 for primary astrocyte cultures. All experiments were approved by Stanford University’s Administrative Panel on Laboratory Animal Care (APLAC). All mice were housed in the Stanford SIM1 animal facility under supervision of the Stanford animal care unit; all mice were healthy and were not involved in previous experiments.

HEK293T Cell Lines.

HEK293T cells were purchased from ATCC and low-passage cells were expanded and stored. HEK293T cells below passage 20 were cultured in DMEM + 10% FBS and reached confluency every 3 days, at which point they were passaged (1:20 split ratio).

METHOD DETAILS

Gene Targeting for Generating PSEN1 KO ESC Lines.

Three single-guide RNAs (sgRNAs) were designed for PSEN1 (purchased from Synthego). The sgRNA sequences were (5’ to 3’): CACAACGACAGACGGAGCCT, CCAGGGTAACTCCCGGCAGG, CCCTGTGACTCTCTGCATGG. The three sgRNAs were combined with 20 μM recombinant Cas9 (IDT), and Lonza nucleofector solution (P3 4D Nucleofector X kit S, Lonza). The Ribonucleoprotein (RNP) complexes were incubated for 10 minutes at room temperature. H1 ESCs were detached with Accutase, centrifuged, and resuspended in mTeSR1 with thiazovivin; 4.5e5 H1 cells were used for nucleofection with the RNP complexes, per nucleofector cuvette well. Following nucleofection, cells were transferred very gently to 6-well plates with pre-warmed media of mTeSR1 and thiazovivin (2 μM). Once confluent, part of the well was kept for genomic DNA extraction (QuickExtract DNA solution, Lucigen) and Sanger Sequencing to determine the population knockout score, while the remaining cells were plated at 1,000 cells per 6-well plate. When these wells had small and well-defined rounded colonies, the colonies were picked and further expanded prior to freezing. With subsequent genotyping and validation, three PSEN1 KO ESC cell lines (KO A, B, D) with frameshift mutations and 100% knockout scores (ICE Analysis V2, Synthego) were used for experiments.

Glial Cell Culture.

Mouse glia were prepared from cortices of newborn CD1 pups, dissected in HBS on postnatal days 0 to 2 (P0–2). The cortex was digested with 80 μl of papain in 5 ml for 20 minutes at 37°C. Tissues were washed 3 times with DMEM + 10% fetal bovine serum (FBS), then harshly triturated and plated in T75 cell culture flasks with 12 ml of DMEM + 10% FBS media. Once 90% confluent, approximately 7 days after dissection, glial cultures were detached with 0.05% trypsin and replated in 3 new T75 flasks (1:3 split ratio) in DMEM + 10% FBS. The replating step and media prevent mouse neurons from surviving. No antibiotics were used in these cultures.

Plasmids.

Lentiviral packaging plasmids Rev, RRE, and VSV were previously described in Zhang et al. (2013). The Ngn2 overexpression plasmids were FUW-TetO-Ngn2-T2A-puromycin or FUW-TetO-Ngn2-P2A-GFP-T2A-puromycin⁵⁶. The FUGW-GFP plasmid was a gift from David Baltimore (Addgene plasmid #14883). The tdTomato overexpression used the pCAG-tdTomato¹⁰⁷.

Virus Generation.

Transfection for lentivirus production was performed in HEK293T cells at 70% confluence on the day of transfection, using calcium phosphate and HBS (Takara). In total, 12 μg of lentiviral packaging DNA were transfected per T75 flask of HEK293T cells: Rev (4 μg), RRE (8 μg), and VSV (6 μg). 48 hours after transfection, cell media was harvested and centrifuged at 19,000g for 2 hours. Pellets were resuspended overnight at 4°C in 100 μl of DMEM, aliquoted and frozen at −80°C. For each 6-well plate, 3 μl of each lentivirus were used.

Generation of iN Cells and Pharmacological Treatments.

H1 ESCs were directly reprogrammed into iN as previously described⁵⁶. In summary, H1 ESCs were plated in mTeSR1 at a ratio of 1:18 and infected with two lentiviruses, TetO-Puro-Ngn2 and rtTA (prepared as described above) in mTeSR1 media with thiazovivin. For a subset of experiments requiring green fluorescent labeling of neurons, a third lentivirus carrying GFP was added at the time of infection. On day 0, Ngn2 expression was induced with doxycycline in the induction medium containing: DMEM-F12, doxycycline (2 μg/ml, Sigma), Non-Essential Amino Acids, BDNF (10 ng/ml, PeproTech), human NT3 (10 ng/ml, PeproTech), mouse Laminin-1 (0.2 μg/ml, PeproTech). On day 1 and 2, puromycin (1 μg/ml) was added to the induction medium. On day 3, iN were dissociated from the 6-well plates in which they were induced using Accutase, and were replated on Matrigel-coated coverslips in 24-well plates. The media contained Neurobasal with B27 (Gemini21), doxycycline, BDNF, NT3, Laminin-1, and Glutamax. Glia were simultaneously detached (as previously described), resuspended in the same Neurobasal-based media with the neurons, and the human neuron-mouse glia cell mixture was replated in each well with a Matrigel-coated coverslip. Half-media changes were performed with the Neurobasal-based media every 2 days until day 10 in vitro.

On day 10, cells were transitioned to Growth Medium: 900 ml MEM, 100 mg transferrin, 50 ml FBS, 25 ml of 20% D-Glucose, 2.5 ml Glutamax, 2.5 ml of 8% NaHCO3, 20 ml of B27 (Gemini21), doxycycline (2 μg/ml) and 4AraC (2 μM). Between day 10 and the experimental day, half-media changes were performed once a week. If iN were made from PSEN1 knockout ESCs (previously described), no additional components were added to the media starting on day 10. For iN made from wild-type ESCs for experiments with γ-secretase inhibition, the co-cultures were treated with vehicle (Cell Culture-Grade DMSO, Sigma, 0.5%) or with the GSIs DAPT (40 μM, Tocris) or LY411575 (2.5 μM, SelleckChem) starting day 10. The treatments were subsequently provided in the weekly media changes. For co-culture experiments treated with both GSIs and statins, rosuvastatin (1 μM, SelleckChem) or simvastatin (1 μM, SelleckChem) was added starting day 10, and treated weekly until the experimental day. Unless noted otherwise, experiments in this work were performed at days 37 to 42 in vitro.

Immunofluorescence.

Human iN were washed 3 times with PBS and fixed with 4% PFA and 4% sucrose in PBS for 15 minutes at room temperature. Cells were then washed three times with PBS, followed by permeabiliziation with 0.1% Triton-X-100 for 15 minutes at room temperature. Primary antibodies used for immunocytochemistry were: rabbit polyclonal anti-Synapsin (Südhof Lab, YZ6078 RRID: AB_2920537, 1:1000 dilution), mouse monoclonal anti-PSD95 (Thermo, 7E3–1B8, 1:250 dilution), chicken polyclonal anti-MAP2 (Encor, CPCA-MAP2, 1:1000 dilution), mouse monoclonal anti-L1CAM (Sigma, clone UJ 127.11, 1:300 dilution). For synapsin-only staining, blocking was performed in 5% goat serum (Sigma); for synapsin and PSD95 co-labeling, blocking was performed in 1% goat serum with 4% BSA for one hour at room temperature. Incubation with primary antibodies was performed overnight at 4°C in the corresponding blocking buffers. Cells were washed three times with PBS and incubated for one hour at room temperature with Alexa conjugated secondary antibodies: goat anti-chicken Alexa 647, goat anti-rabbit Alexa 488, and goat anti-mouse Alexa 546, all 1:500. Coverslips were washed once with PBS and once with water to remove salts before mounting, then were mounted using Fluoromount (with DAPI) on glass slides. All images were taking with the Nikon A1RSi confocal microscope at 60x for analysis of synaptic puncta using Nikon Analysis Software. For L1CAM staining, live cells were incubated with L1CAM antibody (mouse, Millipore-Sigma L4543, 1:500) at 37°C for 15 minutes, then fixed and incubated with goat anti-mouse Alexa 546 (1:500). Permeabilization (as described above) was only performed on these L1CAM-stained coverslips if co-labeling with another antibody was needed.

Cell Morphology and Tracing.

Human iN co-cultured with mouse glia were sparsely transfected with 0.5 μg of TdTomato per well of a 24-well plate. Transfections were performed using 2M calcium phosphate and HBS on day in vitro 19. Two days later (day 21), cells were fixed using 4% paraformaldehyde and 4% sucrose in PBS, for 15 minutes at room temperature, and mounted on glass slides using Fluoromount with DAPI. Images were taken with the Nikon A1RSi confocal microscope on 20x magnification, with stitching to follow isolated axons and dendrites. Axons and dendrites were traced using Fiji software.

Immunoblotting and Protein Quantification.

Neurons were lysed in ice-cold RIPA buffer (50 mM Tris-HCl pH 8.0, 150 mM NaCl, 0.1% SDS, 0.5% sodium deoxycholate, 1% Triton-X-100, plus protease inhibitors (Roche). Lysates were then centrifuged at 14,000 g for 30 minutes. Protein levels were measured by BCA (Pierce) per the manufacturer’s instructions, or loaded on stain-free 10% gels (Bio-Rad), along with standards of known concentrations. The supernatant was diluted in 4X Laemmli Buffer (Bio-Rad) with beta-mercapto-ethanol. Lysates were analyzed by SDS-PAGE on 4–20% polyacrylamide Criterion gels (Bio-Rad), followed by semi-dry transfer for mixed molecular weights. Blocking was performed using 5% BSA in TBST at room temperature. Primary antibodies used were: rabbit polyclonal anti-APP (Südhof Lab, U955, 1:1000 dilution); mouse monoclonal anti-APP (Millipore, MABN380, 1:1000 dilution); rabbit monoclonal anti-APP (Abcam, Y188, 1:1000 dilution); rabbit polyclonal anti-CASK (Neuromab, clone K56A/50, 1:1000 dilution); rabbit polyclonal anti-Fe65 (Südhof Lab, 1:1000 dilution); rabbit polyclonal anti-Mint1 (Südhof Lab, P730, 1:1000 dilution); rabbit polyclonal anti-Nlgn1 (Südhof Lab, K138, 1:500), rabbit polyclonal anti-pan-Neurexin (Südhof Lab, G394, 1:1000 dilution); rabbit polyclonal anti-PSD-95 (Südhof Lab, L667, 1:1000 dilution); rabbit polyclonal anti-RIM1/2 (Südhof Lab, 1:1000 dilution); rabbit polyclonal anti-synapsin (Südhof Lab, YZ6078, 1:1000 dilution); mouse monoclonal anti-Synaptotagmin-1 (Synaptic Systems, clone 41.1, 1:1000 dilution); rabbit polyclonal anti-Syntaxin-1a/b (Südhof Lab, 438B, 1:1000 dilution). Primary antibodies were diluted in 5% BSA in TBST and incubated overnight at 4°C. Following 3 washes with TBST, membranes were incubated with fluorescently labeled secondary antibodies (800CW and 680LT) for one hour at room temperature in 5% BSA in TBST. Signals were normalized for Tuj1 (mouse, Biolegend, 1:2,500). Blots were imaged using the Odyssey Infrared Imager CLX and analyzed using Image Study 5.2.5. (LI-COR Biosciences).

ELISA.

Conditioned culture media was harvested on days in vitro 8 (baseline, prior to γ-secretase inhibitor treatment), 14, 18, and 21 to measure the secreted total Ab content. Media was stored at −20°C until usage, to allow for all timepoints to be measured on the same ELISA plate. Once thawed, media was loaded in the ELISA plates and the wash steps were performed per the manufacturer’s instructions for total human Aβ (Clonetech, #27729A). Absorbance was measured using the Apollo LB915 (Berthold Technologies), with background subtraction prior to analysis.

Electrophysiology.

Whole-cell patch-clamp electrophysiology was performed on coverslips of human iN grown on mouse glia on days 37 to 42 (unless specified for day 21). For all experiments, coverslips were mounted in a recording chamber with an inverted phase-contrast microscope (Nikon) with DIC. Extracellular bath solution contained: 145 mM NaCl, 5 mM KCl, 2 mM MgCl2, 10 mM HEPES, 3 mM CaCl2, 10 mM glucose monohydrate, pH 7.4 and osmolarity 305 mOsm. Borosilicate glass pipettes were prepared from the Narishige PC-10 puller (Japan), with resistances of 3–3.5 MΩ. For voltage-clamp recordings, the recording pipette was filled with internal solution containing: 135 mM CsCl, 5 mM NaCl, 4 mM ATP-Magnesium, 10 mM HEPES, 0.3 mM GTP-Sodium, 5 mM EGTA, 5 mM QX314-Bromide, pH 7.2 and osmolarity 300 mOsm. For current-clamp recordings, the recording pipette contained internal solution with: 123 mM K-gluconate, 10 mM KCl, 1 mM MgCl₂, 10 mM HEPES, 1 mM EGTA, 1.5 ATP-Magnesium, 0.3 mM GTP-Sodium, 5 mM glucose, pH 7.2 and osmolarity 300 mOsm. Electrical signals were recorded with a two-channel Axoclamp 700B amplifier (Axon Instruments), digitalized with a Digidata 1440 digitizer (Molecular Devices), controlled under Clampex 10.1 software. Whole-cell capacitance, membrane resistance, and series resistance were recorded and compensated more than 80%. All recordings were performed at room temperature.

Neurons were maintained at −70 mV holding potentials for whole-cell voltage-clamp recordings. To record miniature excitatory postsynaptic currents (mEPSCs), TTX (0.5 μM) was applied to the bath solution. To record evoked excitatory postsynaptic currents (eEPSC) or paired-pulse recordings, extracellular stimulation was achieved with a bipolar electrode (WPI) placed 200 μm away from patched cells. Square pulses (100 μs) of voltage were applied via a stimulus isolation unit (Model 2100 Isolated Pulse Stimulator, A-M Systems). Single evoked currents were recorded at −70 mV holding potentials in voltage clamp, and with 10 sweeps per recording. Coefficients of variation were calculated from the standard deviation of the amplitudes, divided by the mean EPSC amplitudes of each cell. For paired-pulse recordings, two identical stimuli were applied in succession, with interstimulus intervals (ISIs) of 20 ms, 50 ms, 100 ms, 200 ms, 500 ms, and 1000 ms between stimuli. The first response amplitude and coefficients of variation were measured for recordings with ISIs of 20 ms. At each ISI, 3 sweeps were recorded. The ratio between peak amplitude of the second and the first responses were computed to obtain the Paired Pulse Ratio (PPR). To trigger action potentials, cells were held in current clamp near −70 mV resting potential. The recording Current steps were injected through the recording pipette over 1 second in 10 pA increments from −20 pA to 240 pA.

RNA Sequencing.

Co-cultures of human neurons and mouse glia (day 40) in 6-well plates were lysed using TRIzol (Invitrogen). Samples were homogenized in TRIzol, then mixed with chloroform to achieve phase-separation. The upper layer containing RNA was loaded onto columns of the RNeasy Plus Mini Kit (Qiagen) and purified per manufacturer’s instructions. RNA was eluted in 30 ul of RNAse-free water (Invitrogen) and the concentration was measured with the NanoDrop 1000 Spectrophotometer (ThermoFisher). For RNAseq experiments, RNA concentration and integrity was verified independently at the Stanford PAN Facility; all RNA integrity numbers were greater than 9. RNA was stored at −80°C and only thawed once for either RNAseq or cDNA synthesis.

Bulk RNAseq was performed for a total of 6 biological replicates. Illumina TruSeq Stranded mRNA was used for library preparation, and sequencing was performed with the Illumina NovaSeq with 100 bp paired-end reads. Sequencing depth was 40 million total reads (20 million paired). Raw data was processed using Kallisto for pseudo-alignment of the reads to a chimeric mouse and human transcriptome, as described in Bray et al. (2016) and used in Marro et al. (2019). The R package DESeq2 was used to analyze differential gene expression¹⁰⁸. The DESeq data set included the matrix of raw, un-normalized counts. The DESeq2 model internally corrected for library size. The mean expected counts of each gene within each group were adjusted by normalization factors. Batch effects were controlled for in the DESeq2 data set by using “batch” and “condition” variables. The “results” function filtered the mean normalized counts, followed by the Log2FoldChange (LFC) shrinkage function to ensure that the largest fold changes were not due to genes with unreasonably low counts. P-values were also corrected for low counts using the Benjamini-Hochberg correction. The False Discovery Rate (FDR) threshold for adjusted p-values was set to alpha < 0.1, and the adjusted p-value threshold used to determine significance was 0.05. The log2(Fold Change) threshold was greater than 0.25 for upregulated DEGs and was less than −0.25 for downregulated DEGs; this corresponded to an approximately 20% increase or decrease in gene expression, respectively. The top 20 genes across both DAPT and LY411575 groups, relative to vehicle, were determined first by significance of the adjusted p-value, followed by the log2(Fold Change) cutoff. Volcano plots were generated using the VolcaNoseR package and the “EnhancedVolcano” package in R Studio.

Quantitative RT-PCR.

Total RNA was extracted as described above, at day in vitro 40 following acute or chronic GSI treatment. qRT-PCR measurements were performed on cDNA synthesized from RNA (RNA to cDNA kit, ThermoFisher). The IDT 5X Master Mix was used along with PrimeTime qPCR probe assays from IDT, which consist of two primers and one FAM-labeled ZEN/IBFQ-quenched 5’ nuclease probe. The PrimeTime qPCR primer-probes were human-specific probes for: ACAT2 (Hs.PT.56a.26153626); ACTB (Hs.PT.39a.22214847); DBH (Hs.PT.58.39251730); DHCR24 (Hs.PT.58.4561516); DHCR7 (Hs.PT.58.38763048); DUSP4 (Hs.PT.58.1640850); DUSP5 (Hs.PT.58.1558918); EGR1 (Hs.PT.58.40805543.g); FDFT1 (Hs.PT.58.25526506); FOS (Hs.PT.58.15540029); HMGCR (Hs.PT.58.41105492); HMGCS1 (Hs.PT.58.4084870); IDI1 (Hs.PT.58.19303620); JUNB (Hs.PT.58.27174746.g); LDHA (Hs.PT.58.22929122); LDLR (Hs.PT.58.2004261); LSS (Hs.PT.58.24976485); MAP2 (Hs.PT.58.20680759); MVD (Hs.PT.58.39035173); NAB2 (Hs.PT.58.24622150); NR4A1 (Hs.PT.58.20779713); SCD (Hs.PT.58.45714389); SREBF2 (Hs.PT.58.45335433); SV2C (Hs.PT.58.4640250); TH (Hs.PT.58.369742); TMEM97 (Hs.PT.58.40917738); VEGFA (Hs.PT.58.1149801); VGF (Hs.PT.58.26072166). qRT-PCR was performed with the Applied Biosystems 7900HT machine. Both ACTB and MAP2 were used as endogenous controls (reference genes) in every qPCR experiment. Data were normalized to the mean of ACTB and MAP2 using the 2^−ΔΔCt method. Ct values greater than 35 were omitted.

Filipin Staining.

Coverslips with GFP-positive human neurons grown on GFP-negative mouse glia were incubated with L1CAM primary antibody for 15 minutes at 37°C, as described above, to live-stain the axons of human neurons. Filipin stock solution (Sigma, 1 mg/ml) was diluted to a final concentration of 0.05 mg/ml in PBS with 5% goat serum. Coverslips were then rinsed three times with PBS at room temperature, and then were fixed with 4% paraformaldehyde and 4% sucrose in PBS. Coverslips were not permeabilized to avoid loss of free cholesterol with detergent use. Coverslips were incubated with glycine (1.5 mg/ml) in PBS for 10 minutes at room temperature to quench paraformaldehyde. Diluted filipin was added to coverslips for 2 hours at room temperature, in the dark. Cells were rinsed once with PBS and once briefly with water to remove salts before mounting with Fluoromount (no DAPI). Images were taken with the Nikon A1RSi confocal microscope on 60x magnification, with NyQuist zoom. Nikon Analysis software was used to measure Filipin fluorescence intensity within the GFP-positive (neuronal) area.

FACS.

Co-cultures of GFP-positive human iN and GFP-negative mouse glia were dissociated with Accutase for 20 minutes at 37°C on day in vitro 40, in order to ensure single-cell suspensions. The Accutase was quenched with Neurobasal cell media. The dissociated cells were passed through 100 μm filters, and were then sorted using a BD Bioscience FACS Aria II SORP flow cytometer. GFP-negative human iN were used as a negative control in order to set the gating scheme for isolation of GFP-negative and GFP-positive cell populations. Gating was conducted using BD Biosciences FACS Diva Software and was further analyzed using FlowJo flow cytometry analysis software. Cells were sorted into individual tubes containing Neurobasal media with supplements (as described above). Sorted cells were then centrifuged at 1.4 rpm for 5 minutes to obtain cell pellets of human iN or of mouse glia, which were then resuspended in 1X reaction buffer of the Amplex Red Cholesterol Kit (Invitrogen, A12216).

Amplex Red Cholesterol Quantifications.

Sorted human neurons and sorted mouse glia from co-cultures at day 40 were lysed for cholesterol analysis by sonication in 1X reaction buffer of the Amplex Red Cholesterol Assay Kit (Invitrogen, A12216). The homogenized mixture underwent snap-freezing and was subjected to enzymatic reactions per the Amplex kit protocol, as previously described (2). Cholesterol esterase was omitted for the free cholesterol measurements with the same kit. Measurements were normalized using cholesterol standards of known concentrations, as well as Resorufin standards, supplied in the Amplex kit. Fluorescence was measured using the Tecan Infinite M1000 plate reader at the Stanford High-Throughput Bioscience Center.

QUANTIFICATION AND STATISTICAL ANALYSIS

Results are reported as means with SEM, with the number of cells and number of biological replicates reported on each graph or in the figure legend. Statistical analyses were performed with GraphPad Prism 9. Statistical tests were performed using unpaired two-tailed Student’s t-tests for two groups, one-way ANOVA with Dunnett’s multiple comparison tests for multiple groups, or two-way ANOVA with Bonferroni’s multiple comparison tests for multiple groups with two variables (* p < 0.05, ** p < 0.01, *** p < 0.001). RNA-seq statistical analysis were performed using DESeq2¹⁰⁸, using the Wald test and are described in the “RNA Sequencing” section of the Method Details.

Highlights:

• Pharmacologic and genetic suppression of γ-secretase decrease synaptic transmission

• Chronic γ-secretase inhibition decreases cholesterol levels in neurons but not glia

• Regulation of synaptic release probability by γ-secretase is cholesterol-dependent

• γ-Secretase suppression increases synapse numbers independently of cholesterol