Blog

By Raja Bhattacharyya, Sophia E. Black, Madhura S. Lotlikar, Rebecca H. Fenn, Mehdi Jorfi, Dora M. Kovacs, Rudolph E. Tanzi

Excerpt from the article published in Cell Rep. May 18, 2021 DOI: https://doi.org/10.1016/j.celrep.2021.109134

Editor’s Highlights

• Axonal generation of Alzheimer’s disease (AD)-associated amyloid-β (Aβ) plays a key role in AD neuropathology
• To transport proteins to the cell surface, axons use endoplasmic reticulum (ER)-rafts or mitochondria-associated ER membranes (MAMs)
• Palmitoylated APP (palAPP) is enriched in mitochondrial- associated ER membranes (MAMs)
• Upregulation of MAM increases cell surface palAPP levels and axonal Ab generation
• Downregulation of MAM attenuates cell surface palAPP levels and axonal Ab generation

Abstract

Axonal generation of Alzheimer’s disease (AD)-associated amyloid-β (Aβ) plays a key role in AD neuropathology, but the cellular mechanisms involved in its release have remained elusive. We previously reported that palmitoylated APP (palAPP) partitions to lipid rafts where it serves as a preferred substrate for β-secretase. Mitochondria-associated endoplasmic reticulum (ER) membranes (MAMs) are cholesterol-rich lipid rafts that are upregulated in AD. Here, we show that downregulating MAM assembly by either RNA silencing or pharmacological modulation of the MAM-resident sigma1 receptor (S1R) leads to attenuated β-secretase cleavage of palAPP. Upregulation of MAMs promotes trafficking of palAPP to the cell surface, β-secretase cleavage, and Aβ generation. We develop a microfluidic device and use it to show that MAM levels alter Aβ generation specifically in neuronal processes and axons, but not in cell bodies. These data suggest therapeutic strategies for reducing axonal release of Aβ and attenuating β-amyloid pathology in AD.

Graphical abstract

Introduction

The subcellular localization of amyloid-b (Ab) generation and downstream effects on neurons are key events in Alzheimer’s disease (AD) neuropathogenesis (Simons 1995; Yamazaki et al., 1995; Koo et al., 1996; Muresan et al., 2009). Synaptic and neuritic Ab can be generated via processing of APP in neuronal processes (Kamal et al., 2001; Cirrito et al., 2008; Tampellini et al., 2009; Das et al., 2013), ~40% of neuronal Ab production takes place in axons (Niederst et al., 2015; Das et al., 2016). However, the molecular mechanisms underlying axonal generation of Ab remain elusive.

To transport proteins to the cell surface, axons use endoplasmic reticulum (ER)-rafts or mitochondria-associated ER membranes (MAMs), found on the ER juxtaposed with mitochondria (Merianda et al., 2009). 5%–20% of the mitochondrial sur- face associates with the ER to form MAMs (Hajno ́ czky et al., 2006; Rowland and Voeltz, 2012). MAMs move bidirectionally as isolated small ER (sER) vesicles in neuronal processes (Pizzo and Pozzan, 2007; Giorgi et al., 2009). Mitochondria are essential for the formation and maintenance of synapses and move along axonal microtubules, neurofilaments, and actin tracks (Kamal et al., 2001; Cirrito et al., 2008; Tampellini et al., 2009; Das et al., 2013).

MAM function has been shown to be significantly increased in fibroblasts from AD patients (Area-Gomez et al., 2012). Smaller MAMs (<10 nm) are found in hippocampal neurons in APP transgenic rats (Martino Adami et al., 2019). APP, BACE1, and g-secretase components are all found in MAMs (Schon and Area-Gomez, 2013; Erpapazoglou et al., 2017), and MAMs bear remarkable similarity to lipid rafts, preferred subcellular microdomains for b-secretase cleavage of APP and Ab generation (Urano et al., 2005; Vetrivel et al., 2005, 2009; Kosicek et al., 2010). Other proteins enriched in MAMs (e.g., inositol-1,4,5-trisphosphate receptor subunit 3 [IP3R3], sigma-1 receptor [S1R], and acyl-coenzyme A:cholesterol acyltransferase [ACAT]) (Schon and Area-Gomez, 2013) have emerged as AD drug targets (Bhattacharyya and Kovacs, 2010).

We recently reported APP is palmitoylated in the ER prior to localizing to lipid rafts, where it serves as a preferential substrate for cleavage by b-secretase (Bhattacharyya et al., 2013). Reduced palmitoylation leads to ER-retention and slower maturation of APP (Bhattacharyya et al., 2013). We previously showed that cortical extracts from mice exhibit age-dependent increases in palmitoylated APP (palAPP). Treatment with palmitoylation inhibitors (e.g., cerulenin [Cer] and 2-bromopalmitate [2-BP]) prevents APP palmitoylation and lowers Ab production. Recently, we reported palAPP forms strong cis-dimers that undergo b-secretase cleavage in detergent resistant membranes (DRMs) or lipid rafts (Bhattacharyya et al., 2016). Loss of the ER/MAM-resident enzyme, acyl-coenzyme A:ACAT (Rusin ̃ ol et al., 1994; Lewin et al., 2002), reduces secreted Ab by up to 92% (Puglielli et al., 2001; Hutter-Paier et al., 2004; Huttunen et al., 2007, 2010; Bhattacharyya and Kovacs, 2010; Bryleva et al., 2010; Murphy et al., 2013; Shibuya et al., 2015). ACAT inhibition decreases levels of lipid raft palAPP and Ab generation by up to 76% (Bhattacharyya et al., 2013). Thus, palmitoylation plays an important role in APP metabolism and Ab generation (Bhattacharyya et al., 2013).

Here, we investigated the role of MAMs in APP processing and trafficking and axonal generation of Ab. We employed differentiated neurons from human neuronal progenitor cells expressing APP with familial AD (FAD) mutations (FAD hNPCs) in a three-dimensional (3D) neural cell culture model of AD (Choi et al., 2014). We also employed a biochemical assay to separate palAPP from non-palAPP and total APP (APPtot) and developed a microfluidic system to study MAMs in axonal versus somal microenvironments in neurons. First, we show that association of palAPP with MAMs regulates the processing and cell surface localization of palAPP and Ab generation. Second, we show that modulating MAM levels directly influences b-secretase cleavage of palAPP and Ab generation. Third, we show that MAM-based generation of Ab occurs specifically in neuronal processes and axons. These findings carry major implications for developing AD therapies based on targeting MAM-associated palAPP in axons to alleviate Ab pathology.

Results

palAPP is predominantly localized to MAMs in human neuronal cells and in mouse brains

To quantify the distribution of palAPP in lipid rafts versus MAMs in AD neuronal cells, we compared palAPP levels in flotillin-positive raft fractions (lipid rafts) versus inositol-1,4,5-trisphosphate receptor subunit 3 (IP3R3)-positive raft fractions (MAMs) in ReN-VM neural progenitor stem cells (hNPCs) constitutively overexpressing human APP containing K670N/M671L (Swedish) and V717I (London) FAD mutations (APP_Swed/Lon) (Choi et al., 2014). FAD-APP (APP_Swe/lon) expressing hNPCs (FAD hNPCs) were generated by transfecting hNPCs with IRES-mediated polycistronic lentiviral vectors encoding human APP_Swe/lon with GFP as a reporter. Fluorescence-activated cell sorting (FACS) was employed to enrich the population. FAD hNPCs differentiated in a 3D matrix recapitulate β-amyloid and Tau pathology (D’Avanzo et al., 2015; Kim et al., 2015; Kwak et al., 2020). We metabolically labeled FAD hNPCs with chemically reactive alkylene-palmitic acid (Alkyl-C16) for 18 h, as described previously (Bhattacharyya et al., 2013). Labeled cells were subjected to sucrose gradient centrifugation (Figure 1A), and fractions were probed with the raft marker, flotillin, and MAM-marker, IP3R3. Most flotillin distributed to fraction 3, whereas most IP3R3 partitioned to fraction 4 with some overlap, indicating separation of post-ER rafts and MAMs. Although we detected equal amounts of APP_tot in raft fractions positive for flotillin (PM-rafts) or IP3R3 (MAMs), most APP_tot partitioned to the non-raft fractions (fractions 9 and 10) (Figure 1A), as expected. ABE palmitoylation assay revealed that ~10% of APP_tot is palmitoylated (palAPP) in FAD hNPCs (Figure S1), consistent with our previous findings (Bhattacharyya et al., 2013, Bhattacharyya et al., 2016). To detect palAPP distribution in FAD hNPCs, we subjected each fraction to click-iT addition of TAMRA (tetramethylrhodamine) label, as previously described (Bhattacharyya et al., 2013) (Figure 1A, click-iT palmitoylation). In flotillin-positive lipid rafts, palAPP constituted only 20.7% ± 3.1% of APP_tot, whereas in IP3R3-positive MAMs, palAPP was enriched to 68.12% ± 8.8% of APP_tot (Figure 1B). Meanwhile, <2% of APP_tot was palmitoylated in non-rafts fractions 9 and 10 (Figure 1B).

To confirm the localization of palAPP in MAMs, we extracted crude ER-mitochondria (ER-mito) from FAD hNPCs and used sucrose density gradient centrifugation to isolate MAMs (IP3R3 and ACAT-rich fractions) and non-MAMs (fractions 8–10; Figure 1C). APP_tot distributed into both MAM and non-MAM fractions (Figure 1C). To assess palAPP distribution, the pooled MAM and non-MAM fractions were subjected to the recently developed Badrilla palmitoylation assay (Figure 1D). This assay captures palmitoylated proteins by exchanging palmitic acid with a thiol-bound resin following thioester cleavage and isolates non-palmitoylated proteins in the unbound fractions (Figure S2). We observed strong staining for palAPP in MAM fractions, with little or no evidence of non-palmitoylated APP (non-palAPP) (Figure 1D). Conversely, non-palAPP was mainly distributed in the non-MAM fractions (Figure 1D). Quantitative analysis revealed ~9.8-fold more palAPP was associated with MAMs as compared to non-palAPP. In contrast, ~10.5-fold more non-palAPP distributed in non-MAMs compared to palAPP (Figure 1E). Thus, MAMs almost exclusively contain palAPP.
We next asked whether palAPP localizes to MAMs in mouse brain. Mouse brain homogenates were subjected to Percol-gradient centrifugation as previously described (Wieckowski et al., 2009). Total membrane (TM), crude mixture of ER-mitochondria (ER/mito), ER, mitochondria (mito), and MAMs were isolated (Figure 1F). Equal amounts of proteins were subjected to the acyl-biotin exchange (ABE) palmitoylation assay and probed with an anti-APP antibody (C66) to detect palAPP in the indicated membranes (Figure 2, ABE, IB:C66). In addition to detecting APP, BACE1, IP3R3, and ACAT1 in MAMs, as previously described (Area-Gomez et al., 2009, Area-Gomez et al., 2012), we detected palAPP in MAMs (Figure 2). Thus, palAPP mainly resides in MAMs.

Silencing of MAM-resident sigma 1 receptor (S1R) reduces MAM levels in FAD NPCs

S1R is a chaperone protein residing in MAMs, where it interacts with proteins such as IP3R3, which anchors the outer mitochondrial membrane protein voltage-dependent anion channel isoform 1 (VDAC1) to the ER-associated molecular chaperone glucose-regulated protein 75 (GRP75) (Hayashi and Su, 2007). Loss or inhibition of MAM-resident S1R disrupts MAM-assembly by destabilizing the MAM-resident IP3R3 (Hayashi and Su, 2007). Genetic ablation of the S1R gene in mouse brain (sigmar1^−/−) results in a robust decrease in IP3R3 levels and significant reduction of ER-mito contact sites (~30% of mitochondria forming MAMs in wild-type neuron versus ~18% in sigmar1^−/− neurons) compared to non-transgenic control mice (Bernard-Marissal et al., 2015).
To study the effect of disrupted MAMs on the processing and β-secretase cleavage of palAPP, we silenced the expression of S1R in FAD hNPCs, using a SMART pool small interfering RNA (siRNA) against sigmar1 (si-S1R) (Amer et al., 2013). We performed the CytoTox-ONE assay to measure lactate dehydrogenase (LDH) levels in the culture media to assess cell viability after introducing si-S1R for 0, 48, and 72 h. We observed ~77% reduction of S1R expression after 48 h, and >90% reduction after 72 h transfection with si-S1R. The LDH-assay revealed that 72 h transfection with si-S1R led to ~53% loss of cell viability, whereas 48 h transfection exhibited little or no reduction in cell viability (Figure S2). Cells transfected with si-S1R for 48 h were fixed and subjected to confocal and EM analyses and immunolabeling with anti-S1R antibody. Confocal microscopy revealed significant reduction of S1R expression in S1R-silenced (si-S1R) cells, compared to control cells (si-non) (Figure 2A). Equal amounts of proteins from si-non and si-S1R cells were probed for IP3R3 and ACAT1. si-S1R cells exhibited dramatic reductions of IP3R3 and ACAT1 levels compared to si-non cells (Figure 2B), indicating that silencing of S1R reduced MAM levels in FAD hNPCs. Meanwhile, S1R-knockdown had little or no effect on VDAC1, a primarily mitochondrial protein, or GRP75, a primarily ER-protein (Figure 2B), suggesting that S1R-silencing specifically affected MAMs, not bulk mitochondria or ER. Quantitation revealed ~85% reduction of S1R expression in si-S1R cells compared to si-non cells, whereas IP3R3 levels decreased by ~67% in the same cells (Figure 2C). Thus, silencing S1R expression effectively downregulates MAMs.
Next, S1R-silenced and control cells were subjected to transmission electron microscopy (TEM) to identify ER-mito contact sites or MAMs. TEM revealed robust reduction of ER-mito contact sites in si-S1R cells compared to si-non cells (Figure 2D). MAMs were quantified directly by counting the number of ER-mito contact sites (5–20 nm size) per mitochondria per field of the electron micrographs (Bernard-Marissal et al., 2015). Quantitation revealed ~67% reduction in ER-mito contact sites or MAMs per mitochondria (MAMs per mito) in si-S1R cells versus si-non cells (Figure 2E). Thus, silencing S1R expression in FAD hNPCs significantly reduces MAM levels.

 S1R-agonist PRE-084 and S1R-antagonist NE-100 regulate MAM levels in FAD hNPCs

S1R function can be regulated using the S1R agonist PRE-084 (PRE) and antagonist NE-100 (NE) (Bernard-Marissal et al., 2015). Total IP3R3 levels were increased in the presence of 5.0 or 10.0 μM PRE and decreased following treatment with 5.0 or 10.0 μM in FAD hNPCs (Figure 2F). 10 μM PRE increased total IP3R3 levels by ~43% (1.43- ± 0.2-fold; p < 0.05, n = 3), whereas treatment with 10 μM NE decreased total-IP3R3 levels by ~20% (0.79- ± 0.08-fold; p < 0.05, n = 3) versus vehicle control (veh) cells (Figure 2G).
To confirm regulation of S1R-activity modulated MAM-associated IP3R3 levels, we isolated MAM, ER, mixed ER/mitochondria, and mitochondria fractions from FAD hNPCs following pretreatment with 10 μM PRE or NE. The fractions were probed with antibodies against IP3R3 and cytochrome C (cyto C) to determine the purity of MAMs and of mitochondria (mito), respectively (Figure 2H). MAM-IP3R3 levels were increased (3.73- ± 0.43-fold) following treatment with PRE and significantly reduced (~55%; 0.45- ± 0.13-fold) following treatment with NE compared to veh cells (Figures 2H and 2I). The significant effects of S1R-activation or -inactivation on MAM-IP3R3 levels as opposed to the more modest effects on total-IP3R3 suggest S1R preferentially modulates MAMs in FAD hNPCs.

 Modulation of MAMs via S1R stabilizes palAPP in FAD-NPC neurons

Palmitoylation localizes and stabilizes transmembrane proteins in lipid-rich membrane compartments (Linder and Deschenes, 2007). We previously reported palmitoylation extends the half-life of APP, from t_1/2 = 2 h for total APP to t_1/2 > 4 h for palAPP, indicating palAPP is twice as stable as total APP (Bhattacharyya et al., 2016). To test whether association with MAMs stabilizes palAPP, we measured the half-life of palAPP in differentiated FAD-NPC neurons following upregulation of MAMs with 10 μM PRE or downregulation with 10 μM NE. MAM-dependent palAPP stability was assessed by cycloheximide (CHX)-chase experiments (Bhattacharyya et al., 2016). To measure palAPP levels, ABE palmitoylation assays were performed after 0, 1, 2, 4, and 6 h of chase with cycloheximide (Figure 3A). ~40% of palAPP levels remained after 4 h chase, confirming our previous studies (Figure 3B) (Bhattacharyya et al., 2016). Treatment with PRE stabilized palAPP: ~90% of palAPP remained after the 4-h chase (Figure 3B). Conversely, treatment with NE dramatically reduced palAPP stability: ~20% of palAPP remained after 4 h of chase (Figure 3B). Meanwhile, neither PRE- nor NE-treatment altered the stability of APP_tot.

Next, we compared the half-life of palAPP versus non-palAPP using the Badrilla palmitoylation assay. We observed a marked decrease in palAPP levels after 4 h of chase following CHX-treatment in the control (veh) experiment (Figure 3C). A similar reduction of palAPP levels was observed only after 6 h of chase in cells pre-treated with PRE (Figure 3C), suggesting stabilization. Conversely, palAPP levels were significantly decreased in NE-treated cells (Figure 3C). 41.1% ± 2.2% of palAPP remained after 6 h of chase in control cells. 68.9% ± 1.7% versus 17.0% ± 1.3% of palAPP remained after 4 h of chase in cells pre-treated with PRE or NE, respectively (Figure 3D), whereas the stability of non-palAPP remained unchanged by either PRE- or NE- treatment after 4 h of chase: ~40% of non-palAPP remained in control, PRE-, or NE-treated cells (Figure 3E). Thus, S1R-activation specifically stabilizes palAPP but not APP_tot or non-palAPP. Conversely, S1R-inactivation specifically destabilizes palAPP.

 MAM assembly regulates levels of cell surface-associated palAPP but not total APP

We next asked whether modulation of MAM assembly affects levels of cell surface-associated palAPP. We used a two-step pull-down assay to simultaneously detect cell surface palmitoylated and non-palmitoylated APP in FAD hNPCs, which were first surface biotinylated with sulfo-NHS-SS-biotin. Surface biotinylated-palmitoylated proteins were isolated from Neutravidin beads using tris(2-carboxyethyl)phosphine (TCEP) prior to carrying out the Badrilla palmitoylation assay (Figure S4). To test the effect of MAM formation on cell surface-associated palAPP, we performed the two-step pull-down assay in untreated FAD hNPCs and those treated with 10 μM PRE or NE (Figure 4A). We detected a robust (~1.5-fold) increase of surface labeled palAPP on PRE-treatment, whereas NE-treatment resulted in ~30% loss of cell surface palAPP (Figure 4B). In contrast, treatment with neither PRE nor NE affected cell surface biotinylation of APP_tot or non-palAPP (Figures 4A and 4B). Cell surface association of the control protein transferrin receptor (Tfr) also remained unaffected following PRE- or NE-treatment (Figures 4A and 4B), consistent with Tfr being nearly absent in MAMs (Sala-Vila et al., 2016). Thus, MAM-assembly specifically promotes cell surface association of palAPP without affecting cell surface association of APP_tot.

Regulation of MAM-assembly in FAD hNPCs via S1R modulates β-secretase cleavage of palAPP

Next, we asked whether silencing the expression of S1R in FAD hNPCs attenuates cleavage of APP by β-secretase. si-S1R decreased ER-mito contact sites by ~70% (Figure 2). The ABE palmitoylation assay revealed that silencing S1R significantly reduced palAPP levels in FAD hNPC cell lysates (Figure 5A, ABE assay, palAPP). In contrast, si-S1R led to no significant differences in levels of palmitoylated flotillin versus control cells (not shown) and only a small reduction in levels of APP_tot(Figure 5A, ABE assay, APP_tot). Quantification of the reduction of palAPP in si-S1R cells in comparison to APP_tot(palAPP/APP_tot) revealed 48.61% ± 14.60% (p < 0.01, n = 3) palAPP/APP_tot in si-S1R cells as compared to si-non (Figure 5B), indicating an ~51% reduction of palAPP levels following S1R-silencing. We detected a smaller (14.72% ± 4.77%; p < 0.03, n = 3) decrease in APP_tot in si-S1R cells versus si-non (data not shown), in agreement with 10%–20% of APP_tot consisting of palAPP.

We next asked whether decreased palAPP levels following S1R-silencing affects β-secretase cleavage of APP. Conditioned media from si-non and si-S1R cells probed revealed a robust reduction in sAPPβ release relative to total sAPP following si-S1R (Figure 5A, CM, sAPPβ, or sAPP_tot, respectively). The ratio of sAPPβ/sAPP_tot in conditioned media from si-S1R cells was 44.62% ± 16.84% (p < 0.01, n = 3) of that observed in si-non cells (Figure 5B), with an ~58% reduction in sAPPβ release following si-S1R. Thus, ~70% downregulation of MAM-assembly following silencing of S1R leads to significant reduction in β-secretase cleavage of APP, most likely by reducing palAPP levels.

Next, we measured pal-sAPPβ release from FAD hNPCs treated with 10 μM PRE or NE. Figure 4D shows that treatment with PRE increased MAM-IP3R3 levels by ~3.7-fold, whereas NE decreased MAM-IP3R3 levels by ~45%. Next, pal-sAPPβ levels were assessed by the ABE palmitoylation assay as done before (Bhattacharyya et al., 2013). Treatment with PRE led to a significant increase in pal-sAPPβ release compared to veh cells (Figure 7C , ABE, IB:anti-sAPPβ). In contrast, pal-sAPPβ release was dramatically decreased following treatment with NE (Figure 5C, ABE, IB:anti-sAPPβ). Neither PRE nor NE-treatment led to any significant effect on total sAPP (sAPP_tot) or sAPPα release (Figure 5C), nor on APP-C-terminal fragment (CTF) levels (Figure S6). In comparison to sAPPβ_tot-release, pal-sAPPβ (pal-sAPPβ/sAPPβ_tot) was increased by 2.14- ± 0.39-fold (p < 0.001, n = 3) following treatment with PRE (Figure 5D). Conversely, NE-treatment reduced pal-sAPPβ release by ~50% (0.508- ± 0.073-fold as compared to control, p < 0.001, n = 3) (Figure 5D). Thus, modulation of MAM levels via S1R specifically affects β-secretase but not α-secretase cleavage of palAPP and does not affect β- and α-secretase cleavage of APP_tot.

Regulation of S1R-activity specifically modulates MAMs in neuronal processes

To test for S1R-dependent changes in MAM-assembly in neuronal processes versus cell bodies, we performed confocal microscopy to visualize the ER-mito contact sites in cells pre-labeled with cell-permeable GFP-tagged ER (CellLight ER-GFP) and RFP-tagged mitochondrial (CellLight mito-RFP) probes. A MAM-resident Ca²⁺-sensor protein Miro has been shown to be co-transported with mitochondria down axons (Misko et al., 2010). Elevated calcium levels halt axonal transport of Miro, indicating MAM-transport along axons could be regulated by physiological stimuli (Macaskill et al., 2009; Wang and Schwarz, 2009).
We first confirmed that the GFP- and RFP-tagged probes could correctly identify ER-mito contact sites in mouse cortical neurons from C57BL/6 embryonic day 16 (E-16) mice. Confocal microscopy revealed that ER-GFP (green fluorescence) and mito-RFP (red fluorescence) distinctly labeled ER and mitochondria, respectively, in cell bodies as well as in processes (Figure 6A, a–c). The overlapping areas between ER and mitochondria (ER-mito) were identified as MAMs (Figure 6A, d, white arrow). Employing a modified MAM-imaging technique (Filadi et al., 2015) that eliminated all non-overlapping fluorescence (representing ER and mitochondria), we were able to highlight the contact areas (white) (Figure 6A, e–g) and demonstrated MAMs in the processes (Figure 6A, h–j) as well as in cell bodies (Figure 6A, k–m).
ER-mito contact sites >10 nm in both processes (Figure 6A, n–p) and in cell bodies (not shown) were counted as MAMs, using the ImageJ particle measurement tool. Quantitation of MAMs (ER-mito contact sites) revealed that treatment with 10 μM PRE increased MAMs in axons and neuronal processes by ~2.5-fold, without affecting MAMs in the cell bodies or soma of cultured primary neurons (Figure 6B). Conversely, treatment with 10 μM NE decreased ER-mito contact sites by ~50%, only in axons and neuronal processes (Figure 6B). Similar results were obtained when we counted ER-mito contact sites in neuronal processes of FAD hNPCs pre-labeled with the ER and mitochondria probes (Figure 6C). NE-treatment decreased ER-mito contact sites, whereas treatment with PRE increased ER-mito contact sites, primarily in neuronal processes, without altering ER-mito contact sites in cell bodies (Figure 6D). We next used a bidirectional fluorescent compensation (BiFC) assay to assess the effect of S1R-inactivation on axonal MAMs in cells co-expressing split GFP ER and mitochondria probes (ER-GFP(1–10) and mito-GFP11, respectively) (Yang et al., 2018). NE treatment reduced the number of MAMs per axon by ~35% (from 27.87 ± 1.52 MAMs/axon in control to 3.125 ± 1.23 MAMs/axons in NE-treated) (Figure S6). Thus, modulation of S1R activity regulates MAM assembly specifically in axons and neuronal processes, not in the cell body or soma.
To further validate that pharmacological regulation of S1R specifically affects axonal MAMs, we developed a microfluidic device (Figure 6E) that separates bulk neurons from axons following physical axotomy (Figure 6F). Although we accumulated a sufficient number of axons from 10 devices to obtain detectable levels of MAM-resident proteins including IP3R3, ACAT, and VDAC1 (Hedskog et al., 2013; Liu et al., 2019), this number of axons was insufficient for MAM purification. Thus, we assessed MAM levels in axons by measuring total levels of MAM-resident proteins, IP3R3, ACAT, and VDAC1, following axotomy (Figure 6G). We confirmed that the axonal fractions were not contaminated with bulk neurons following axotomy, by probing the fractions with an antibody against the nuclear envelope protein, lamin B1 (Figure 6G, IB:anti-lamin B1). We observed significant increases in all levels of all three axonal MAM-proteins measured following treatment with 10 μM PRE versus veh cells (Figure 6G). Conversely, the levels of all three MAM-proteins were dramatically reduced in axons following treatment with 10 μM NE (Figure 6G). Meanwhile, treatment with either PRE or NE had little or no effect on levels of IP3R3, ACAT, and VDAC1 in bulk neurons (Figure 6G). Quantitative analysis revealed that treatment with PRE increased levels of MAM-resident IP3R3, ACAT, and VDAC1 in the axons by 3.16- ± 0.41-fold, 1.4- ± 0.18-fold, and 1.79- ± 0.11-fold, respectively, as compared to axons from veh cells (Figure 6H). Conversely, axons from NE-treated cells contained 25% (0.25- ± 0.09-fold) IP3R3, 21% (0.21- ± 0.03-fold) ACAT, and only 3% (0.03- ± 0.008-fold) VDAC1 compared to veh cells (Figure 6H).

 Modulation of MAMs in neuronal processes/axons regulates axonal Aβ generation from FAD hNPCs

APP and amyloidogenic processing enzymes, β- and γ-secretases, are localized in axons and dendrites (Kaether et al., 2000), and anterograde axonal transport delivers Aβ and sAPPβ to neurites around amyloid plaques. Axons have been heavily implicated in AD-related Aβ-generation and tau phosphorylation (Kamal et al., 2001; Cirrito et al., 2008; Tampellini et al., 2009; Das et al., 2013). Thus, we next investigated whether MAM-assembly in neuronal processes and axons affects Aβ generation. To measure axonal Aβ release, we employed a microfluidic chamber system (Xona Microfluidics) to perform fluidic separation of cell somas from axons in differentiated FAD hNPCs (Niederst et al., 2015). In this system, capillary channels draw axons from the somal chamber into the axonal chamber. To prevent cells from migrating into the axonal chamber, we plated FAD hNPCs in a 3D matrix on the somal side prior to differentiation. Both somal and axonal chambers were probed either with Hoechst dye to label the nuclei or with antibody against tau and/or neurofilament heavy (NFH) antibodies to label axons (Figure 7A, tau and NFH). We then carried out microfluidic isolation of the axonal microenvironments from bulk neuronal environments by establishing a small (30 μL) volume difference between axonal and somal compartments (Taylor et al., 2005). After 10–15 days in differentiating media, conditioned media from each chamber was subjected to Aβ ELISA to detect Aβ₄₀ levels (Aβ₄₂ levels were below detection limit). Levels of Aβ₄₀ in the somal chamber reached 372.4 ± 50.7 pM per chamber in 24 h. To account for axonal Aβ₄₀ in the somal chambers, we determined axonal numbers by subtracting the number of tau-positive axons in the axonal chambers from the number of cells (~30,000 DAPI-stained nuclei in the somal chambers) in the somal chambers, assuming each cell produced only one axon. Aβ₄₀-release in the axonal chambers was then calculated as the number of Aβ₄₀ molecules per tau-positive axon, using a previously described method of measurement (Niederst et al., 2015). We found differentiated FAD hNPC neurons generated 10.69 × 10⁸ ± 4.2 × 10⁸ Aβ₄₀ per cell bodies and 2.97 × 10⁸ ± 5.3 × 10⁶ Aβ₄₀ per axon in 24 h (Figure 7B).
Next, we pharmacologically modulated MAM assembly and tested for effects on Aβ generation in neuronal processes/axons versus neuronal cell somas. We measured Aβ₄₀ levels in the somal and axonal chambers of 10- to 15-day differentiated FAD hNPCs in microfluidic systems that received either vehicle (veh), 10 μM PRE, or 10 μM NE for 24 h. Neither PRE- nor NE-treatment altered Aβ₄₀ levels in the somal chamber. However, PRE treatment increased the number of Aβ₄₀ molecules released per axon by an additional 11.63 × 10⁸ ± 9.2 × 10⁷. In contrast, NE treatment reduced axonal Aβ₄₀ release to nearly undetectable levels (Figure 7B). Thus, modulation of MAM levels governs Aβ generation, specifically in axons and neuronal processes, but not in the cell soma.

Discussion

We have previously shown that palmitoylation of APP targets palAPP to lipid rafts where it becomes a preferred substrate for b-secretase. Here, we show that palAPP is primarily localized to specialized lipid rafts known as MAMs, in both FAD hNPCs and in mouse brain. We also show that genetically silencing or pharmacologically inactivating the MAM-resident receptor, S1R (with the S1R-antagonist NE100) to downregulate MAM assembly (by destabilizing MAM-resident IP3R3), leads to greatly decreased levels of palAPP, reduced stability and cell surface trafficking of palAPP, decreased b-secretase cleavage of pal- APP, and reduced generation of sAPPb and Ab. In contrast, the S1R-agonist PRE-084 upregulated MAM levels and increased palAPP stability and cell surface trafficking, b-secre- tase cleavage of palAPP, and Ab generation. Furthermore, we show that modulation of S1R regulates MAM assembly, stability, and cell surface trafficking of palAPP, b-secretase cleavage ofpalAPP, and Ab generation specifically in axons. Thus, we demonstrate a clear connection between MAM-assembly and Ab generation from palAPP, specifically in axons and neuronal processes.

Previous studies have shown that SigR1 knockout (Sigmar1
/
) decreases MAM-assemblies by nearly 50% (Bernard-Marissal et al., 2015). We demonstrated that silencing of S1R expression not only reduces MAMs (Figure 2) but also re- duces the levels of palAPP with minimal to no effects on APPtot levels (Figure 5A). We have previously shown that palAPP and/ or palAPP-dimers are better substrates for b-secretase cleavage in lipid-rich microdomains as compared to APPtot (Bhattachar- yya et al., 2013, 2016). Here, we show that levels of sAPPb were reduced with silencing of MAM-resident S1R expression, and regulation of S1R activity by an agonist (PRE-084) or an antagonist (NE-100) not only modulates MAMs (Figure 2), but also specifically promotes or reduces b-secretase cleavage of palAPP, respectively (Figures 5C and 5D). Thus, MAMs and pal- mitoylation of APP directly modulate b-secretase cleavage of palAPP and subsequent Ab generation.

The observation that S1R activation and inactivation oppositely modulate the stability of palAPP with no effects on APPtot (Figure 3) is consistent with previous studies reporting evidence for palmitoylation-dependent stability of transmembrane proteins in lipid-rich membranes (Linder and Deschenes, 2007). Here, we show that modulation of S1R-activity affects cell-surface association of palAPP without affecting that of APPtot (Figure 4). Although it remains unclear how association with MAMs promotes the cell surface association of palAPP, it is worth noting that ~11% of MAM-proteins are found in the plasma membrane (Poston et al., 2013). Additionally, MAM-like ER-rafts fusing with plasma membrane, called PAMs, have been detected in several tissues involved in regulating Ca2+ homeostasis, lipid trafficking, and mitochondrial structures (Szymanski et al., 2017).

We also showed S1R activity mainly modulates MAM levels in axons and neuronal processes as opposed to neuronal cell bodies and soma (Figure 6). Thus, S1R-activity specifically regulates the assembly of a novel subset of MAMs enriched in axons and neuronal processes, while the assembly of MAMs in cell bodies is independent of S1R-activity. A previous study reported that S1R knockout only reduced MAM-assemblies in neurons by >50%, suggesting that more than half of MAM-as- semblies were independent of S1R activity (Bernard-Marissal et al., 2015). Moreover, MAMs also differ in size and motility (Volpe et al., 1991; Bannai et al., 2004). We discovered that modulation of MAMs in neuronal processes via alteration of S1R-activity specifically affects b-secretase cleavage of palAPP without affecting cleavage of total APPtot (Figures 5 and S5). Small molecule agonists and antagonists of S1R are currently being considered as therapeutics for numerous neurodegenerative diseases (Ryskamp et al., 2019).

Using a system incorporating FAD gene-expressing hNPCs grown in a 3D matrix inside microfluidic chambers, we were able to reveal that S1R regulates MAM-dependent Ab generation exclusively in axons and neuronal processes, but not in soma or from bulk neurons (Figure 7). Axons and synapses are critical sites for APP-processing and Ab production. (Lazarov et al., 2002; Sheng et al., 2002; Brendza et al., 2003; Spires et al., 2005). Previous studies have shown that intra-axonal Ab is generated ahead of extracellular Ab especially when axons are under stress (Suo et al., 2004). Moreover, amyloid deposition has been reported to decrease when b-secretase processing of APP and Ab generation is shifted away from axonal to somal compartments (Lee et al., 2005). A recent study reported that ER-mitochondria tethering is increased in axons following injury (Lee et al., 2019). Here, we show that axons specifically use MAMs, containing ~70% of preferred b-secretase substrate, pal- APP, to generate Ab. Other studies have reported that Ab and/or the b-cleaved C-terminal fragment of APP (C99 or APP-bCTF) modulate MAM function in AD (Schreiner et al., 2015; Pera et al., 2017). Knocking down the expression of MAM-resident mitofusin 2 (Mfn 2) resulted in increased ER-mito contact and decreased levels of Ab (Leal et al., 2016). Our data reveal an additional converse mechanism whereby modulation of MAMs regulates Ab generation in axons and neuronal processes. Collectively, these findings suggest a possible feed-forward mechanism for MAM-dependent AD pathogenesis.

The subcellular location in which APP primarily undergoes b-secretase cleavage in neurons has remained a topic of intensive investigation. We previously reported that APP undergoes palmitoylation targeting it to lipid rafts where it serves as a preferential substrate for b-secretase. In addition to lipid rafts in plasma membranes, intracellular MAMs have been increasingly investigated in neurodegenerative diseases such as AD and amyotrophic lateral sclerosis. We previously reported that reduced palmitoylation leads to ER-retention of APP. Here, we expand on these findings by showing that most palAPP is partitioned to MAMs. We and others have previously shown that the loss or inhibition of the MAM-resident enzyme acyl-co-enzyme A:ACAT reduces cell surface localization of APP and Ab generation (Huttunen et al., 2009; Bhattacharyya and Kovacs, 2010; Murphy et al., 2013). More recently, we showed that inhibition of ACAT reduces palAPP in lipid-rafts by ~76%, in vitro (Bhattacharyya et al., 2013). Palmitoylation inhibitors 2-bromopalmitate and cerulenin also reduced palAPP level and Ab generation in vitro (Bhattacharyya et al., 2013, 2016). Confocal imaging revealed that cerulenin (Cer) treatment reduced APP distribution in axonal MAMs (IP3R3-labeled), but not in somal MAMs in differentiated PC12 and NPCs (Figure S7). Based on these data, we propose that palAPP partitions into axonal MAMs, whereas total or de-palmitoylated APP resides in somal MAMs. Future studies will be needed to explore whether palmitoylation inhibition as well as ACAT inhibition ameliorates b-amyloid pathology by reducing palAPP levels specifically in axonal MAMs.

In this study, we used FAD hNPCs expressing the FAD mutant of APP (APPSwe/Lon) to examine the role of MAM-association of palAPP in axonal Ab generation. Lipid raft fractionation, cell sur- face biotinylation, and ABE palmitoylation assays (Figures 1A, 4A, and S1) have confirmed that the overall cellular localization and palmitoylation of total APP in FAD hNPCs remain similar to those observed in vitro and in vivo (Bhattacharyya et al., 2013, 2016). These observations suggested that the Swedish and the London mutations in APP do not affect APP’s overall cellular localization or palmitoylation. Several reports have shown that Swedish and/or London mutations in APP in non-neuronal (MDCK and COS) and neuronal (SN56) cells altered the endocytosis of APP without altering its cell surface association. The APPswe mutant has been shown to undergo b-secretase cleavage in secretory vesicles en route to the cell surface, whereas APPwt undergoes b-secretase cleavage in post-Golgi compartments after endocytosis (Haass et al., 1995). The Swed- ish and London mutations in APP slowed the rapid transport from cell surface to lysosomes required for APP-degradation (Lorenzen et al., 2010). In a recent study using the iPSC neurons, Kwart et al. (2019) have shown that CRISPR/Cas9-edited cells expressing APPSwe have promoted enlargement of Rab5-positive endosomes, which is an early hallmark of AD (for review, see Peric and Annaert, 2015). Therefore, it would be interesting in future studies to investigate whether the Swedish/London mutations in APP (APPSwe/Lon) affect endocytosis of palAPP to lysosomes or alter endosome size in AD.

The degree of dementia in AD is primarily correlated with loss of synapses. Synaptic dysfunction, preceded by reduced syn- aptic transmission and loss of dendritic spines, is largely driven by neurotoxic Ab42-oligomers (Cleary et al., 2005; Haass and Selkoe, 2007). Physiological levels of Ab42-oligomers have also been shown to suppress long-term potentiation (LTP) in hippocampal slices (Mango et al., 2019). In contrast, Ab in the picomolar range has also been shown to be required for LTP in- duction (Koppensteiner et al., 2016). Thus, a key question regarding AD pathology is how Ab is generated in axons and neuronal processes. Previous studies have shown that APP can be transported anterogradely in axons (Koo et al., 1990; Si- sodia et al., 1993; Buxbaum et al., 1998), and Ab can be made in axonal terminals (Cirrito et al., 2005). To date, the cellular and molecular mechanisms by which this critical pool of axonal Ab is generated have remained unclear. Our microfluidic chamber system showed that Ab may be released from axons even in the absence of synapses. This observation is in line with several reports showing focally increased toxic Ab species in axonal swellings of dystrophic neurons in AD (Stokin et al., 2005; Chevalier-Larsen and Holzbaur, 2006; Kanaan et al., 2013). We reason that MAM-upregulation in axons promotes focally increased secretion of Ab species from regions along the axons. Axonal swelling along axons has been reported at the early stage of AD pathology leading to axonal dysfunction and Ab de- posits surrounding dystrophic neurons in later stage of the disease (Teipel et al., 2007; Cross et al., 2008; Serrano-Pozo et al., 2011). The mechanisms in the early phase leading to axonal damage in AD pathology remain unclear. MAM dysregulation is an early event in AD pathogenesis (for review, see Yu et al., 2021). Axonal BACE1 levels were increased in axonal swellings in an AD mouse model by genetic deletion of an adaptor protein GGA3, which regulates lysosomal degradation of BACE1, leading to increased production of Ab (Lomoio et al., 2020). Thus, Ab released from axons (even in the absence of synapses) is likely to impact AD pathogenesis. In summary, we have demonstrated that axonal Ab generation is specifically modulated by MAMs via stabilization and cell surface trafficking of palAPP, followed by b-secretase cleavage. These data strongly suggest that modulation of MAM-associated palAPP, specifically in axons (e.g., by regulation of S1R), may be considered a therapeutic strategy for ameliorating Ab-induced neurodegeneration in AD.