The phospholipase A₂ pathway controls a synaptic cholesterol ester cycle and synapse damage

The cellular prion protein (PrP^C) is mainly found in specific membrane micro-domains commonly called lipid rafts (Taraboulos et al., 1995). PrP^C acts as a scaffold protein that organises signalling complexes (Linden et al., 2012) and is associated with multiple signalling proteins, including the tyrosine kinase Fyn (Mouillet-Richard et al., 2000) and cytoplasmic phospholipase A₂ (cPLA₂) (Bate et al., 2010), which are linked to synapse degeneration. PrP^C is concentrated at synapses (Herms et al., 1999), and the aggregation of PrP^C (Chiesa et al., 2008) or cross-linkage of PrP^C with monoclonal antibodies (mAbs) (Solforosi et al., 2004) causes synaptic abnormalities. More recently, PrP^C has been identified as a receptor for amyloid-β (Aβ) oligomers (Laurén et al., 2009) which are responsible for the synapse degeneration and cognitive decline in patients with Alzheimer's disease (AD) (Selkoe, 2002; Shankar et al., 2008). The aggregation of PrP^C mediated by Aβ oligomers forms a signalling complex containing activated cPLA₂ and leads to synapse degeneration (Bate and Williams, 2011).

Cholesterol is a key component of lipid rafts (Brown and London, 2000) and, given that the formation and function of lipid rafts depends upon cholesterol concentrations (Rajendran and Simons, 2005), it follows that fluctuations in cholesterol concentrations may alter cell signalling within lipid rafts. We recently reported that soluble Aβ oligomers, highly toxic forms of Aβ (Yang et al., 2017), increased cholesterol concentrations in synaptosomes (West et al., 2017). This observation is consistent with reports showing that Aβ is concentrated in synapses (Lacor et al., 2004; Takahashi et al., 2010) and that the concentrations of cholesterol are increased in Aβ-positive synapses in the cortex of AD patients (Gylys et al., 2007). The increased cholesterol concentrations are dependent upon activated cholesterol ester hydrolases (CEHs) (West et al., 2017), which participate in the cholesterol ester cycle that controls membrane cholesterol concentrations (Chang et al., 2006). Thus, CEH inhibitors block both the Aβ-induced increase in cholesterol concentrations and Aβ-induced synapse damage in cultured neurons. However, the mechanisms leading to the activation of CEHs are poorly understood.

The present study investigated two key ideas. Firstly, we studied the signalling pathways leading to activation of CEHs in synaptosomes. The Aβ-induced activation of CEHs was dependent upon activation of cPLA₂ and the production of platelet-activating factor (PAF). The increased cholesterol concentrations stabilised a signalling platform containing Aβ, PrP^C, cPLA₂, Fyn and cyclooxygenase (COX)-2, and led to the production of prostaglandin E₂ (PGE₂). Concentrations of PGE₂ are raised in AD (Montine et al., 1999) and, in cultured neurons, PGE₂ causes extensive synapse damage (Bate et al., 2010).

In addition, this study examined the dissociation of signalling platforms as a physiological process that limits the intensity of signalling. Acyl-coenzyme A:cholesterol acyltransferase (ACAT)-1 was incorporated into Aβ–PrP^C–cPLA₂–COX-2–Fyn signalling complexes. The esterification of cholesterol by this enzyme reduced cholesterol concentrations, leading to the dissociation of complexes and reduced cell signalling. We hypothesise that the dissociation of signalling complexes is a physiological process that limits the intensity of cell activation. Consequently, conditions that prevent the dissociation of signalling complexes may lead to sustained activation, cell disruption and disease.

Aβ in the soluble fraction of extracts from AD brains (hereafter, soluble AD brain extracts) activates synaptic CEHs, resulting in increased cholesterol concentrations (West et al., 2017). The AD brain extracts used in the assays of the present study contained predominantly Aβ monomers, dimers and trimers, as shown by immunoblotting (Fig. 1A). Initially, we looked at whether AD brain extracts affected the amounts and location of CEHs. CEHs in synaptosomes were found to be predominantly within detergent-resistant membranes (DRMs) (a crude measure of lipid rafts) (Fig. 1B). Synaptosomes incubated with soluble AD brain extract for 1 h did not have different amounts of CEHs (or caveolin, a loading control) compared with that in synaptosomes incubated with control medium (Fig. 1C). Because Aβ activates cPLA₂ (Palavicini et al., 2017; Zhu et al., 2006), we sought to determine whether activated cPLA₂ was involved in the increased synaptic cholesterol concentrations by using two selective cPLA₂ inhibitors [arachidonyl trifluoromethyl ketone (AACOCF₃) and methyl arachidonyl fluorophosphonate (MAFP)]. The activation of cPLA₂ is the first step in the production of bioactive lipids [including prostaglandins (PGs) and PAF] that affect synapse function (Clark et al., 1992; Sang et al., 2005). The addition of soluble AD brain extract to synaptosomes increases the production of PGE₂ in an Aβ-dependent manner (West et al., 2017). Here, we show that incubation with soluble AD brain extract also significantly increased concentrations of PAF in synaptosomes when compared with incubation with control medium (314±56 nM compared with 28±8 nM, n=6, P<0.01, mean±s.d.) or with Aβ-depleted soluble AD brain extract (314±56 nM compared with 32±10 nM, n=6, P<0.01). The Aβ-depleted soluble AD brain extract was prepared by incubation with mAb 4G8 (reactive with 17-24 of Aβ), which reduced the concentrations of Aβ₄₂ (1±0.07 nM Aβ₄₂ compared with 0.03±0.015 nM, n=9, P<0.01) and Aβ₄₀ (4.36±0.22 nM Aβ₄₀ compared with 0.24±0.07 nM, n=9, P<0.01) in AD brain extracts. In contrast, immunodepletion with mAb 3F4 (reactive with human prion proteins) had no significant effect on concentrations of Aβ₄₂ (1±0.07 nM Aβ₄₂ compared with 0.98±0.08 nM, n=9, P=0.38) and Aβ₄₀ (4.36±0.22 nM Aβ₄₀ compared with 4.15±0.28 nM, n=9, P=0.45). Pre-treatment of synaptosomes with 1 µM AACOCF₃ or 1 µM MAFP reduced the Aβ-induced increase in PGE₂ (Fig. 1D) and PAF (Fig. 1E), indicating that these concentrations of drugs inhibited cPLA₂.

Although there were no significant differences in cholesterol concentrations between synaptosomes treated with a vehicle control and those treated with 1 µM AACOCF₃ or with 1 µM MAFP, the Aβ-induced increase in cholesterol concentrations was reduced in synaptosomes pre-treated with 1 µM AACOCF₃ or with 1 µM MAFP (Fig. 1F). Next, the downstream products of cPLA₂ activation were examined in more detail. Pre-treatment of synaptosomes with inhibitors of COXs (5 µM acetyl salicylic acid or 5 µM ibuprofen), the enzymes that metabolise arachidonic acid to PGs, did not affect the Aβ-induced increase in cholesterol concentrations (data not shown). In contrast, pre-treatment of synaptosomes with PAF antagonists [Hexa-PAF (H-PAF), ginkgolide B or CV-6209] reduced the Aβ-induced increase in cholesterol concentrations (Fig. 1G), suggesting that PAF was a key molecule regulating synaptic cholesterol concentrations. Further studies showed that PAF caused a dose-dependent increase in cholesterol concentrations (Fig. 1H). This was accompanied by reductions in cholesterol ester concentrations in PAF-treated synaptosomes, and there was a significant inverse correlation between the concentrations of cholesterol and cholesterol esters (Fig. 1I). Taken together, these results suggest that the Aβ-induced activation of cPLA₂ was a source of PAF that subsequently activated CEHs. This hypothesis was supported by the observation that the PAF-induced increase in cholesterol concentrations was blocked by selective CEH inhibitors [diethylumbelliferyl phosphate (DEUP) and cholesteryl N-(2-dimethylaminoethyl)carbamate; Fig. 1J].

Because PGE₂ causes synapse degeneration in cultured neurons (Bate et al., 2010), the effects of drugs that inhibit COXs, enzymes that convert arachidonic acid to PGs, were studied. There are two major COX isoforms, and in untreated synaptosomes both COX-1 and COX-2 were found to be predominantly within detergent-soluble membranes (DSMs). The addition of soluble AD brain extract, but not Aβ-depleted soluble AD brain extract, caused some COX-2 to migrate into DRMs (Fig. 2B,C). However, the addition had no effect on the amounts of COX-1 or caveolin in DRMs (Fig. 2A,C). There was a significant correlation between the amounts of COX-2 in DRMs and the concentrations of PGE₂ produced by synaptosomes in response to Aβ (Pearson's coefficient=0.85, P<0.01; Fig. 2D). This was consistent with the hypothesis that Aβ activates the COX-2 isoform. Cholesterol is a major factor in the formation of DRMs (Xu et al., 2001), and there was a significant correlation between cholesterol concentrations and the amounts of COX-2 in DRMs in synaptosomes incubated with Aβ (Pearson's coefficient=0.85, P<0.01; Fig. 2E). Pre-treatment of synaptosomes with DFU or DuP-697 (COX-2-selective inhibitors; Copeland et al., 1994; Riendeau et al., 1997), but not valeryl salicylate or SC-560 (COX-1-selective inhibitors; Bhattacharyya et al., 1995; Smith et al., 1998), reduced the Aβ-induced increase in PGE₂ (Fig. 2F). Soluble AD brain extracts cause the Aβ-dependent loss of synaptic proteins, including synapsin-1, vesicle-associated membrane protein-1, synaptophysin and cysteine string protein, from cultured neurons in a tissue culture model of synapse degeneration (West et al., 2017). Here, the addition of soluble AD brain extracts reduced the amounts of synaptophysin in a dose-dependent manner. Pre-treatment of neurons with COX-2-selective inhibitors (DFU or DuP-697), but not COX-1-selective inhibitors (valeryl salicylate or SC-560), protected neurons against the Aβ-induced loss of synaptophysin (Fig. 2G).

The above results suggested that PAF activated CEHs, and that the consequent increased cholesterol concentrations facilitated the translocation of COX-2 to DRMs, activation of COX-2 and the production of PGE₂. In accordance with this theory, pre-treatment of synaptosomes with PAF antagonists (H-PAF, ginkgolide B or CV-6209) significantly reduced the Aβ-induced translocation of COX-2 into DRMs (Fig. 3A) and PGE₂ production (Fig. 3B). Similarly, pre-treatment of synaptosomes with CEH inhibitors [DEUP or cholesteryl N-(2-dimethylaminoethyl)-carbamate] reduced the Aβ-induced translocation of COX-2 to DRMs (Fig. 3C) and PGE₂ production (Fig. 3D). The concentrations of PGE₂ produced by synaptosomes incubated with phospholipase A₂-activating peptide (PLAP, Bachem) were not significantly affected by pre-treatment with PAF antagonists (1 µM H-PAF or ginkgolide B) or CEH inhibitors [20 µM DEUP or 5 µM cholesteryl N-(2-dimethylaminoethyl)carbamate; Fig. 3E], indicating that these drugs did not affect cPLA₂ or COX-2 directly.

We sought to determine whether COX-2 was associated with specific DRMs. PrP^C, a receptor that mediates Aβ toxicity (Laurén et al., 2009), acts as a scaffold protein and organises signalling complexes (Linden et al., 2012). The addition of soluble AD brain extract to synaptosomes caused the formation of Aβ–PrP^C complexes (Fig. 4A), and there was a significant correlation between the amounts of Aβ–PrP^C complexes and the amounts of COX-2 found within DRMs (Fig. 4B), suggesting a possible connection. Pre-treatment of synaptosomes with COX-2 inhibitors (DFU or DuP-697) did not affect the amounts of Aβ–PrP^C complexes formed (Fig. 4C), nor did they significantly alter the Aβ-induced increase in synaptic cholesterol concentrations (Fig. 4D). To isolate specific DRMs, synaptosomes incubated with soluble AD brain extract containing 1 nM Aβ₄₂ for 1 h were homogenised in a non-ionic detergent and incubated with the PrP^C-reactive mAb 4F2 (a gift from Professor J. Grassi, Paris, France). Immunoprecipitates containing PrP^C and associated DRM proteins were collected and analysed. These studies showed that CEHs did not co-precipitate with PrP^C. In contrast, cPLA₂, COX-2 and the tyrosine kinase Fyn all co-precipitated with PrP^C in synaptosomes incubated with soluble AD brain extract, but not in synaptosomes incubated with Aβ-depleted AD brain extract (Fig. 4E). Pre-treatment of synaptosomes with AACOCF₃ or H-PAF prevented the Aβ-induced incorporation of COX-2 and Fyn into PrP^C-containing DRMs (Fig. 4F). Similarly, pre-treatment of synaptosomes with CEH inhibitors reduced the Aβ-induced translocation of COX-2 and Fyn into DRMs (Fig. 4G). It is worth noting that the Aβ-induced translocation of cPLA₂ to DRMs (Bate and Williams, 2011) was not affected by either PAF antagonists or CEH inhibitors, nor did these drugs affect the amounts of caveolin in DRMs (Fig. 4G).

The Aβ-induced changes in synaptic cholesterol concentrations are transient, and concentrations return to basal levels after 4 h, an effect mediated by ACAT-1 (West et al., 2017). The addition of soluble AD brain extract to synaptosomes caused ACAT-1 to translocate from DSMs to DRMs (Fig. 5A). The addition of Aβ-depleted AD brain extract did not significantly alter the amounts of ACAT-1 in DRMs. In addition, ACAT-1 co-precipitated with PrP^C-containing DRMs in synaptosomes incubated with soluble AD brain extract, but not in synaptosomes incubated with control medium or Aβ-depleted brain extract (Fig. 5B). Following the addition of soluble AD brain extract containing 0.6 to 1 nM Aβ₄₂ to synaptosomes, there were significant correlation between cholesterol concentrations and the amounts of ACAT-1 in DRMs (Fig. 5C), and between the amounts of Aβ–PrP^C complexes and ACAT-1 in DRMs (Fig. 5D). Pre-treatment of synaptosomes with a cPLA₂ inhibitor (AACOCF₃), a PAF antagonist (H-PAF) or a CEH inhibitor (DEUP) reduced the Aβ-induced recruitment of ACAT-1 into DRMs (Fig. 5E,F). None of these treatments affected the amounts of caveolin found within DRMs (Fig. 5F).

Time-course studies showed that Aβ–PrP^C complexes (Fig. 6A) and the Aβ-induced targeting of COX-2 (Fig. 6B) and ACAT-1 (Fig. 6C) to DRMs were also transient in synaptosomes. The return of cholesterol concentrations to those of control synaptosomes is blocked by ACAT inhibitors (West et al., 2017). Here, we demonstrate that pre-treatment of synaptosomes with ACAT inhibitors (TMP-153 or YIC-C8-434) significantly increased the longevity of Aβ–PrP^C complexes. Pre-treatment with ACAT inhibitors also increased the amounts of COX-2 and ACAT-1 in DRMs at 2 and 4 h after the addition of Aβ (Fig. 6B,C), indicating that cholesterol esterification mediates the disruption of signalling complexes. This hypothesis was supported by the observation that pre-treatment with ACAT inhibitors also increased the Aβ-induced production of PGE₂ (Fig. 6D) and significantly increases synapse damage in neurons incubated with Aβ (West et al., 2017).

Because PrP^C signalling is dependent upon endocytosis (Caetano et al., 2008), the effects cytochalasin D (100 nM), which inhibits endocytosis into the recycling pathway (Millman et al., 2008), on Aβ–PrP^C complexes was studied. There were significantly higher amounts of Aβ–PrP^C complexes at the surface of synaptosomes (released by the membrane-impermeable enzyme phosphatidylinositol-phospholipase C) pre-treated with cytochalasin D and incubated with soluble AD brain extract containing 1 nM Aβ₄₂ than in synaptosomes pre-treated with a vehicle control (Fig. 7A), showing that cytochalasin D inhibited the endocytosis of Aβ–PrP^C complexes. In addition, pre-treatment with cytochalasin D prevented the dissociation of Aβ–PrP^C complexes (Fig. 7B). In cytochalasin D-treated synaptosomes incubated with soluble AD brain extract, both the cholesterol concentrations (Fig. 7C) and the amount of cPLA₂ within DRMs (Fig. 7D) remained higher than those in synaptosomes pre-treated with a vehicle control. In contrast, pre-treatment of synaptosomes with cytochalasin D reduced the Aβ-induced translocation of COX-2 (Fig. 7E) and ACAT-1 (Fig. 7F) into DRMs, suggesting that these events occur within an intracellular compartment. Co-immunoprecipitation studies with the PrP^C-reactive mAb 4F2 showed that cytochalasin D reduced the Aβ-induced translocation of COX-2 and ACAT-1 to Aβ–PrP^C complexes, but did not affect the translocation of cPLA₂ or Fyn (Fig. 7G). Cytochalasin D also reduced the Aβ-induced PGE₂ production from 265±39 pM PGE₂ to 91±19 pM PGE₂ (n=6, P<0.01). These results suggest that Aβ bound to PrP^C forms a functional signalling complex only in an intracellular compartment. Finally, we report that pre-treatment of cultured neurons with 100 nM cytochalasin D reduced the Aβ-induced loss of synaptophysin (Fig. 7H).

Although CEHs are key enzymes involved in the formation of Aβ-induced signalling complexes (West et al., 2017), little is known about how they are activated. Here, we show that cPLA₂ and PAF are key intermediates in the Aβ-induced activation of CEHs in synaptosomes and that they increase cholesterol concentrations. The increase in cholesterol concentrations facilitates the formation of signalling complexes, leading to synapse degeneration in neurons (Bate et al., 2010).

Activation of cPLA₂ was a key step in the Aβ-induced increase in synaptic cholesterol concentrations. Although activated cPLA₂ can give rise to multiple biologically active lipids, including lyso-phospholipids, our studies were concentrated on the effects of PAF for two reasons. Firstly, because PAF antagonists blocked the Aβ-induced increase in synaptic cholesterol concentrations and, secondly, because PAF also increased synaptic cholesterol concentrations via activation of CEHs. The movement of enzymes to and from membranes containing substrates is a key mechanism of cell regulation. The increase in cholesterol concentrations facilitated the movement of key signalling enzymes, COX-2 and Fyn into DRMs. These results support the hypothesis that Aβ activates the COX-2 isoform. Selective COX-2 inhibitors did not affect the formation of Aβ–PrP^C complexes or the Aβ-induced increase in synaptic cholesterol concentrations. However, COX-2-selective inhibitors reduced the Aβ-induced increase in PGE₂ concentrations and protected neurons against Aβ-induced synapse damage. These results are consistent with reports showing that COX-2 is expressed at the synapse (Kaufmann et al., 1996) and that COX-2 inhibitors reverse Aβ-mediated suppression of long-term potentiation and memory (Kotilinek et al., 2008).

In synaptosomes incubated with Aβ, COX-2 co-precipitated with PrP^C, cPLA₂ and Fyn, suggesting that it formed part of a signalling complex. The Aβ-induced translocation of COX-2 and Fyn to Aβ–PrP^C complexes and the production of PGE₂ were dependent upon the PAF-induced increase in cholesterol concentrations. PAF antagonists and CEH inhibitors did not affect all DRM proteins; for example, they did not affect the amounts of caveolin in DRMs, nor did they affect the Aβ-induced translocation of cPLA₂ to DRMs. Furthermore, neither PAF antagonists nor CEH inhibitors affected the production of PGE₂ by synaptosomes incubated with PLAP, indicating that they do not affect cPLA₂ or COX-2 directly. These observations are consistent with the hypothesis that PAF antagonists and CEH inhibitors modify the membrane environment generated by Aβ that leads to PGE₂ production.

Aβ is produced as part of normal neuronal metabolism (Haass et al., 1992), and plays a role in regular synapse function and memory formation (Abramov et al., 2009; Garcia-Osta and Alberini, 2009; Puzzo et al., 2011). The observations that PAF and PGE₂ affect synapse function and memory formation at physiological concentrations (Chen and Bazan, 2005; Chen et al., 2001; Clark et al., 1992; Kato et al., 1994; Koch et al., 2010) are consistent with the theory that Aβ-induced activation of cPLA₂ and the production of PAF and PGE₂ are part of normal synapse function. The observations that concentrations of PAF (Ryan et al., 2009) and PGE₂ (Montine et al., 1999) are elevated in the brains of AD patients indicate that this pathway is activated in the brains of AD patients. Reports showing that PAF and PGE₂ cause synapse damage in cultured neurons (Bate et al., 2010) suggest that aberrant activation of this pathway causes synapse degeneration. Consequently, it is crucial to understand how this pathway is controlled.

The Aβ-induced translocation of ACAT-1 into DRMs, and more specifically Aβ–PrP^C complexes, was dependent upon PAF and CEH. Aβ–PrP^C complexes in synaptosomes were transient, and there was a close temporal association between cholesterol concentrations and Aβ–PrP^C complexes (West et al., 2017). Notably, pre-treatment of neurons with ACAT inhibitors significantly enhanced Aβ-induced synapse damage (West et al., 2017). Here, we show that pharmacological inhibition of ACAT reduced the dissociation of Aβ–PrP^C complexes, increasing the time COX-2 and ACAT-1 were found in DRMs and significantly increased the Aβ-induced production of PGE₂. ACAT inhibitors have been proposed as AD treatments as they reduce the production of Aβ, and ablation of ACAT-1 reduces pathology in transgenic AD mice (Bryleva et al., 2010; Puglielli et al., 2001). However, in these studies, ACAT was inhibited, or ablated, before pathology developed. Consequently, ACAT inhibitors might be able to prevent the development of AD pathology but may be contraindicated in the latter stages of AD, where concentrations of Aβ are already raised.

Although inhibition of endocytosis by means of cytochalasin D significantly reduced the Aβ-induced production of PGE₂, it had diverse effects upon the Aβ-induced signalling complex. Cytochalasin D disrupted the relationships between Aβ–PrP^C complexes, cholesterol concentrations and enzymes involved in PGE₂ production. Firstly, cytochalasin D maintained the Aβ-induced increase in cholesterol concentrations, stabilised Aβ–PrP^C complexes and increased the time cPLA₂ spent in DRMs. Although these changes might be expected to increase PGE₂ production, cytochalasin D also prevented the Aβ-induced translocation of COX-2 to DRMs/Aβ–PrP^C complexes. In addition, cytochalasin D reduced the Aβ-induced translocation of ACAT-1 to DRMs and Aβ–PrP^C complexes. The lack of ACAT-1 in DRMs explains the maintenance of high cholesterol concentrations and, consequently, the longevity of complexes in synaptosomes.

Collectively, these results suggest that Aβ causes a tri-phasic response. Firstly, aggregation of cell surface PrP^C by Aβ oligomers led to the translocation of cPLA₂ into DRMs, the activation of cPLA₂ and the release of PAF (Fig. 8A). PAF activated CEHs, resulting in the increased cholesterol concentrations that stabilised Aβ–PrP^C–cPLA₂ complexes and recruited Fyn (Fig. 8B). Secondly, endocytosis of Aβ–PrP^C–cPLA₂–Fyn complexes allowed the recruitment of COX-2 into the complex, leading to PGE₂ production (Fig. 8C). The third phase occurred when ACAT-1 was recruited into Aβ–PrP^C–cPLA₂–COX-2–Fyn complexes (Fig. 8D). The subsequent esterification of cholesterol reduced cholesterol concentrations, and led to the dispersal of complexes and the loss of cPLA₂, COX-2 and ACAT-1 from DRMs. These observations indicate that cholesterol esterification led to the dissociation of signalling complexes and reduced cell signalling.

In summary, the key finding of this study is that the Aβ-induced activation of CEH and increase in cholesterol concentrations in synaptosomes was dependent upon activated cPLA₂ and PAF. The increased cholesterol concentrations were associated with the migration of COX-2 and Fyn into lipid rafts and the production of PGE₂, a bioactive lipid that causes synapse degeneration at high concentrations. Crucially, this signalling pathway also facilitated the migration of ACAT-1 into the Aβ–PrP^C–cPLA₂–COX-2–Fyn complex. The subsequent cholesterol esterification reduced free cholesterol concentrations, resulting in the dissociation of this complex and the cessation of cell signalling.

Primary cortical neurons were prepared from the brains of mouse embryos (day 15.5) after mechanical dissociation. A total of 5×10⁵ cells/well were dispensed in 48-well plates in Ham’s F12 containing 5% fetal calf serum for 2 h. Cultures were shaken (600 rpm for 5 min), and non-adherent cells removed by two washes in phosphate-buffered saline (PBS). Neurons were grown in neurobasal medium containing B27 components and nerve growth factor (5 ng/ml) (Sigma-Aldrich) for 10 days. Immunostaining showed that, after 10 days culture, less than 5% of the viable cells stained for glial fibrillary acidic protein (GFAP) or F4/80 (astrocytes or microglial cells). Neurons were pre-treated with test compounds or vehicle controls for 1 h before the addition of test samples, including Aβ preparations or PLAP (Bachem), for 24 h. Stock solutions of these drugs were prepared in ethanol or dimethyl sulphoxide; vehicle controls consisted of equivalent dilutions of ethanol or dimethyl sulphoxide in neurobasal medium and B27 components. The culture medium was used as control medium (neurobasal medium containing B27 components). All experiments were performed in accordance with European regulations (European community Council Directive, 1986, 56/609/EEC) and were approved by the Royal Veterinary College ethical committee.

Synaptosomes were prepared on a discontinuous Percoll gradient, based on methods previously described (Thais et al., 2006). Briefly, 10⁶ neurons were homogenised at 4°C in 1 ml of SED solution (0.32 M sucrose, 50 mM Tris-HCl pH 7.4, 1 mM EDTA and 1 mM dithiothreitol) and centrifuged (1000 g for 5 min). The supernatant was transferred to a four-step gradient of 3%, 10%, 15% and 23% Percoll in SED solution and centrifuged at 16,000 g for 30 min at 4°C. The synaptosomes were collected from the interface of the 15% and 23% Percoll steps and washed in PBS at 4°C. Freshly prepared synaptosomes were pre-treated with test compounds for 30 min and incubated with peptides for 1 h or the time stated. At the end of the experiment, synaptosomes were centrifuged at 16,000 g for 5 min at 4°C, and either total membrane extracts or DRMs were isolated as below.

To access the amounts of Aβ–PrP^C complexes at the surface of synaptosomes, synaptosomes were incubated with 0.5 ml phosphatidylinositol-specific phospholipase C (from Bacillus cereus, Sigma-Aldrich) for 10 min. This enzyme is cell impermeable and, consequently, only released PrP^C expressed on the surface of synaptosomes. Treated synaptosomes were centrifuged (16,000 g for 5 min), and the supernatant was collected and tested for Aβ–PrP^C complexes (see below).

Treated cells or synaptosomes were washed twice in PBS and homogenised in an extraction buffer containing 10 mM Tris-HCl, 100 mM NaCl, 10 mM EDTA, 0.5% Nonidet P-40, 0.5% sodium deoxycholate and 0.2% SDS at 10⁶ cells/ml. Mixed protease inhibitors [4-(2-aminoethyl)benzenesulfonyl fluoride hydrochloride, aprotinin, leupeptin, bestatin, pepstatin A and E-46; Sigma-Aldrich] were added, and large insoluble fragments were removed by centrifugation (1000 g for 5 min).

DRMs were isolated by using their insolubility in non-ionic detergents, as described (London and Brown, 2000). Briefly, synaptosomes were homogenised in an ice-cold buffer containing 1% Triton X-100, 10 mM Tris-HCl, pH 7.2, 150 mM NaCl, 10 mM EDTA and mixed protease inhibitors (as above), and large fragments were removed by centrifugation (1000 g for 5 min at 4°C). The supernatant was incubated on ice (4°C) for 1 h and centrifuged (16,000 g for 30 min at 4°C). The supernatant (DSM) was collected. The insoluble pellet was homogenised in an extraction buffer containing 10 mM Tris-HCl, pH 7.4, 150 mM NaCl, 10 mM EDTA, 0.5% Nonidet P-40, 0.5% sodium deoxycholate, 0.2% SDS and mixed protease inhibitors, and centrifuged (10 min at 16,000 g). The soluble material was reserved as the DRM fraction.

Synaptosomes were homogenised in an ice-cold buffer containing 1% Triton X-100 (as above), and large fragments were removed by centrifugation (1000 g for 5 min at 4°C). The supernatant was incubated with 1 µg/ml of the PrP^C-reactive mAb 4F2 or an isotype control for 1 h at 4°C, followed by the addition of 10 µl of protein G-coated beads (Sigma-Aldrich) for a further 10 min. The mixture was centrifuged at 1000 g for 5 min to collect beads and the attached PrP^C-containing DRMs. Beads were washed twice in ice-cold 1% Triton X-100 before use.

Samples were mixed with Laemmli buffer containing β-mercaptoethanol and heated to 95°C for 5 min; proteins were separated by electrophoresis on 15% polyacrylamide gels. Proteins were transferred onto a Hybond-P polyvinylidene membrane by semi-dry blotting and membranes blocked using 10% milk powder. Antibodies used are listed in Table S1. These were visualised using a combination of biotinylated anti-mouse/goat/rat/rabbit IgG (Sigma-Aldrich), extravidin–peroxidase and enhanced chemiluminescence.

Soluble extracts from brain tissue (temporal lobes supplied by Asterand; informed consent was given to Asterand, and samples were collected according to the Declaration of Helsinki, 2000) of three patients with a clinical and pathologically-confirmed diagnosis of Alzheimer's disease were prepared, using methods as previously described (Shankar et al., 2008). Briefly, brain tissue was cut into approximately 100 mg pieces and added to 2 ml tubes containing lysing matrix D beads (Q-Bio). Tris-buffered saline (20 mM Tris-HCl, pH 7.4, 150 mM NaCl,) was added, so that there was the equivalent of 100 mg brain tissue/ml. Tubes were shaken for 10 min (Disruptor Genie, Scientific Industries) and centrifuged at 16,000 g for 10 min to remove cell debris. Soluble material was passed through a 50 kDa filter (Sartorius) to remove any protease activity, and desalted (3 kDa filter; Sartorius) to remove drugs and small molecules. The retained material contained peptides from 3-50 kDa and were stored at −80°C. For biological experiments, the soluble AD brain extracts were diluted in neurobasal medium containing B27 components. For immunoblot analysis, extracts were mixed with an equal volume of 0.5% NP-40, 5 mM CHAPS, 50 mM Tris-HCl, pH 7.4 and separated by polyacrylamide gel electrophoresis. Proteins were transferred onto a polyvinylidene membrane by semi-dry blotting and blocked using 10% milk powder. Aβ was detected by incubation with mAb 6E10 (reactive with an epitope 1-16 of Aβ; Covance, SIG 39340), biotinylated anti-mouse IgG, extravidin–peroxidase and enhanced chemiluminescence.

Soluble AD brain extracts were incubated with 1 μg/ml mAb 4G8 (reactive with epitope 17-24 of Aβ; Covance, SIG-39220) or 1 μg/ml mAb LN27 [reactive with epitope 45 to 53 of the amyloid precursor protein; Covance, SIG-39188 (mock-depletion)] and incubated at 4°C on rollers for 1 h. Protein G microbeads (Sigma-Aldrich) were added (10 µl/ml) for 30 min and protein G bound-antibody complexes removed by centrifugation at 16,000 g for 5 min. The treated AD brain extracts were filtered (0.2 µM; Sartorius) before use.

The amount of synaptophysin in neuronal extracts was measured by enzyme-linked immunosorbent assay (ELISA), as previously described (Bate et al., 2010). Maxisorb immunoplates (Nunc) were coated with 1 µg/ml of a mouse anti-synaptophysin mAb (MAB368; Millipore) as a capture antibody. Samples were applied (1 h at room temperature), and bound synaptophysin was detected using 1 µg/ml rabbit polyclonal anti-synaptophysin (Abcam, ab53166), followed by an anti-rabbit IgG conjugated to alkaline phosphatase (Sigma-Aldrich, A3687) and 1 mg/ml 4-nitrophenol phosphate solution (Sigma-Aldrich). Absorbance was measured on a microplate reader at 405 nm. Samples were expressed as ‘units synaptophysin’, where 100 units was defined as the amount of synaptophysin in 10⁶ untreated neurons.

The amounts of cPLA₂ in extracts was measured by ELISA, as previously described (West et al., 2017). Maxisorb immunoplates were coated with 0.5 µg/ml of mouse mAb anti-cPLA₂ (clone CH-7; Upstate, 05-568) and blocked with 5% milk powder in PBS+0.1% Tween-20 (PBST). Samples were added for 1 h and the amount of bound cPLA₂ was detected using a goat polyclonal anti-cPLA₂ (Santa-Cruz Biotechnology, sc-4049), followed by biotinylated anti-goat IgG, extravidin–alkaline phosphatase and 1 mg/ml 4-nitrophenol phosphate solution. Absorbance was measured at 405 nm and the amount of cPLA₂ protein expressed in units; 100 units was defined as the amount of cPLA₂ in control preparations.

The concentrations of PGE₂ and PAF in synaptosomes were determined using competitive enzyme immunoassay kits (R&D Systems, Abingdon, UK), according to the manufacturer's instructions.

Maxisorb immunoplates were coated with 1 µg/ml mouse mAb to amino acids 575 to 602 of COX-1 (H-1, sc-166572, Santa Cruz Biotechnology) overnight and blocked with 5% milk powder. Samples were added for 1 h, and the amount of bound COX-1 was detected with 0.5 μg/ml of a rabbit polyclonal IgG to amino acids 63 to 124 of COX-1 (H-62, sc-7950, Santa Cruz Biotechnology), followed by a biotinylated anti-rabbit IgG (1:2000), extravidin–alkaline phosphatase and 1 mg/ml 4-nitrophenol phosphate solution. Absorbance was measured at 405 nm, and results were expressed as the percentage of COX-1, where 100% was defined as the amount of COX-1 in control synaptosomes.

Maxisorb immunoplates were coated with 1 µg/ml mouse mAb (5E10/D10; Enzo Life Sciences) overnight and blocked with 5% milk powder. Samples were added for 1 h, and the amount of bound COX-2 was detected with 0.5 μg/ml of rabbit polyclonal IgG to epitope 50 to 111 (H-62; Santa Cruz, sc-7951), followed by a biotinylated rabbit anti-goat IgG (Sigma, 1:2000), extravidin–alkaline phosphatase and 1 mg/ml 4-nitrophenol phosphate solution. Absorbance was measured at 405 nm, and results were expressed as the percentage of COX-2, where 100% was defined as the amount of COX-2 in control synaptosomes.

Maxisorb immunoplates were coated with 1 µg/ml rabbit monoclonal anti-ACAT-1 (EPR10359; ab168342, Abcam, 1:1000) overnight and blocked with 5% milk powder in PBS+2% Tween-20. Samples were added for 1 h, and the amount of bound ACAT-1 was detected with 0.5 μg/ml of a goat polyclonal IgG (G-15; sc-161307, Santa Cruz Biotechnology), followed by a biotinylated rabbit anti-goat IgG (1:2000), extravidin–alkaline phosphatase and 1 mg/ml 4-nitrophenol phosphate solution. Absorbance was measured at 405 nm, and results were expressed as the percentage of ACAT-1, where 100% was defined as the amount of ACAT-1 in control synaptosomes.

The amounts of Aβ–PrP^C complexes in synaptosomes was measured by ELISA, as previously described (West et al., 2017). Maxisorb immunoplates were coated with mAb 4F2 reactive with PrP^C. Plates were blocked with 5% milk powder, and samples were added for 1 h. Aβ bound to PrP^C was detected by using biotinylated mAb 6E10 (Covance). Bound Aβ was detected using extravidin–alkaline phosphatase and 1 mg/ml 4-nitrophenol phosphate solution, and absorbance was measured at 405 nm. Samples were expressed as a percentage of maximum optical density in control synaptosomes.

To detach Aβ₄₂ from membrane components that blocked specific epitopes, samples (300 µl) were mixed with 700 µl of propan-2-ol and sonicated. Proteins were precipitated by addition of 250 µl 100% w/v trichloroacetic acid, incubation on ice for 30 min and centrifugation (16,000 g for 10 min at 4°C). The pellet was washed twice with ice-cold acetone, dried, suspended in a buffer containing 150 mM NaCl, 10 mM Tris-HCl, pH 7.4, 10 mM EDTA and 0.2% SDS, and sonicated.

Nunc Maxisorb immunoplates were coated with mAb 4G8 in carbonate buffer overnight. Plates were blocked with 5% milk powder in PBST, and samples were applied. The detection antibody was an Aβ₄₂-selective rabbit mAb BA3-9 (Covance, SIG-39168), followed by biotinylated anti-rabbit IgG and extravidin–alkaline phosphatase. Total Aβ was visualised by addition of 1 mg/ml p-nitrophenol phosphate solution, and optical density at 405 nm was measured in a spectrophotometer.

The concentrations of cholesterol in samples were measured using the Amplex Red cholesterol assay kit (Life Technologies, Paisley, UK) (Robinet et al., 2010), according to the manufacturer's instructions. Briefly, cholesterol was oxidised by cholesterol oxidase to yield hydrogen peroxide and ketones. The hydrogen peroxide reacted with 10-acetyl-3,7-dihydroxyphenoxazine (Amplex Red reagent) to produce highly fluorescent resorufin, which was measured by excitation at 550 nm and emission detection at 590 nm. By performing the assay in the presence or absence of cholesterol esterase, we were also able to determine the amounts of esterified cholesterol within samples. Each experiment contained cholesterol standards and solvent-only controls. Cholesterol concentrations were calculated by reference to the cholesterol standards.

Differences between treatment groups were assessed using paired Student's t-tests, and one-way and two-way ANOVA with Bonferroni's post hoc tests (IBM SPSS statistics 20). Error values are ±s.d. Correlations were determined using bivariate analysis, and significance was set at 0.01%.

We would like to thank Professor Alun Williams (Cambridge University, UK) for supplying human brain material.