Amyloid-β₄₂ signals tau hyperphosphorylation and compromises neuronal viability by disrupting alkylacylglycerophosphocholine metabolism

Glycerophospholipids were extracted postmortem from the posterior/entorhinal cortex of AD patients and control subjects (Fig. 1A). PAF isoforms were profiled by high-performance liquid chromatography electrospray ionization mass spectrometry (LC-ESI-MS) (12). Lipids with mass-to-charge ratios (m/z) of 450–600 were analyzed in positive ion mode by MS scan for a protonated molecule at expected m/z followed by precursor ion scan for a diagnostic phosphocholine product ion at m/z 184. Peak intensities were standardized against C13:0 LPC, a synthetic internal standard added at the time of lipid extraction. Nineteen alkylacylglycerophosphocholine species were identified (Figs. S2 and S3). Three of these species were significantly elevated in AD cortex: 1-O-hexadecyl-2-acetyl-sn-glycero-3-phosphocholine (C16:0 PAF), its immediate metabolite/precursor 1-O-hexadecyl-sn-glycero-3-phosphocholine (C16:0 lyso-PAF), and 1-O-oleyl-2-lyso-sn-glycero-3-phosphocholine (C18:1 lyso-PAF) (Fig. 1 B and C and Figs. S2 and S3). As elevations in both C16:0 PAF and C16:0 lyso-PAF suggested a specific disruption in the remodeling of palmitic acid containing-PAFs, we used deuterated standards to quantify tissue concentrations. Clear separation was obtained between C16:0 PAF and isobaric C18:0 LPC with the isoform altered in AD co-eluting with d₄-C16:0 PAF (Fig. S4). Tissue concentrations, expressed as pg/mg in Fig. 1 D and E, represent approximate molar concentrations of 258 ± 30 pM increased to 639 ± 205 pM for C16:0 PAF and 316 ± 25 pM increased to 1035 ± 455 pM for C16:0 lyso-PAF in control and AD cortex, respectively.

To establish the kinetics of palmitic-PAF metabolic disruption, we analyzed the posterior temporal/entorhinal cortex of TgCRND8 transgenic mice (Fig. 1F). TgCRND8 mice express a double mutant form (KM670/671NL+V717F) of the human amyloid precursor protein gene under the control of the prion protein promoter. Increases in cortical Aβ₄₂/Aβ₄₀ ratios are initially observed at 10–12 weeks of age coincident with learning and memory impairment (13). We found that cortical levels of C16:0 PAF and C16:0 lyso-PAF of TgCRND8 mice were comparable to nontransgenic (NonTg) littermates at 8 weeks of age (Fig. 1 G and H), yet were 4-fold and 2-fold higher, respectively, by 14–16 weeks of age (Fig. 1 I and J). Tissue concentrations, expressed as pg/mg in Fig. 1 I and J, represent approximate molar concentrations of 104 ± 25 pM C16:0 PAF in NonTg increased to 407 ± 105 pM in TgCRND8 and 268 ± 135 pM C16:0 lyso-PAF in NonTg increased to 597 ± 320 pM in 14- to 16-week-old TgCRND8 mice.

To determine whether Aβ₄₂ directly disrupts C16:0 PAF metabolism in human neurons, hNT neurons were treated with Aβ₄₂ (25 μM) prepared as soluble oligomers (14). A 24-h exposure signaled neuronal loss (Fig. 2A). Over the course of treatment, neuronal C16:0 lyso-PAF concentrations increased steadily rising to 15-fold that of control cultures (12 ± 0.1 pM to 180 ± 8 pM) (Fig. 2B). Endogenous neuronal C16:0 PAF levels initially dropped from approximately 127 ± 27 pM to 28 ± 4 pM over the first 6 h of treatment and then increased to 5-fold that of control levels within 24 h (619 ± 115 pM) (Fig. 2C). These data are consistent with a specific increase in Land's cycle remodeling of palmitic acid-PAFs in response to oligomeric Aβ₄₂. Also intriguing was the transient decrease in endogenous C16:0 PAF levels suggesting a compensatory increase in intracellular catabolism following acute Aβ₄₂ exposure.

To provide mechanistic insight, we sought to modulate the Aβ₄₂-triggered increase in C16:0 PAF metabolism with minimal perturbation of other lipid networks. Because genetic, molecular, and pharmacological strategies targeting cPLA₂, LPCAT1, or LPCAT2 will also impinge upon the arachidonic acid cascade, phosphatidylcholine and lysophosphatidylglycerol remodeling, as well as structural membrane lipid composition, we used a gain-of-function approach to promote hydrolysis of C16:0 PAF to C16:0 lyso-PAF. PAF-AHs cleave the sn-2 acetyl chain of PAF alkylacylglycerophosphocholines generating alkyl-lyso-glycerophosphocholines (lyso-PAFs) (Fig. S1). There are three PAF-AH isoforms. The 45-kDa secreted PAF-AH isoform is not expressed by hNT neurons (Fig. 2D). Cytosolic PAF-AH II is also is not expressed by hNT neurons (Fig. 2D). Cytosolic PAF-AH Ib is an acetylhydrolase-specific complex composed of two 29-kDa α₁ and/or 30-kDa α₂ catalytic subunits and two 45-kDa regulatory LIS1 subunits (15). We have previously shown that overexpression of the α₂ catalytic subunit enhances PAF hydrolysis in cells that endogenously express all three PAF-AH Ib subunits (16). As observed over the course of CNS development (17), the α₁ catalytic subunit was detected at higher levels before differentiation, while the α₂ catalytic and LIS1 regulatory subunits were maximally expressed following terminal differentiation of hNT neurons (Fig. 2D). Based on this profile, neurons were transfected with EGFP and empty vector, EGFP and PAF-AH II, or EGFP and PAF-AH I α₂. Neurons ectopically expressing PAF-AH II or overexpressing the PAF-AH I α₂ catalytic subunit were protected from Aβ₄₂ (Fig. 2E).

These findings implicated C16:0 PAF and not C16:0 lyso-PAF in Aβ₄₂ toxicity. To test this directly, we used a biochemical loading approach to elevate intraneuronal C16:0 PAF concentrations in the absence of any other toxic stimuli. PAFs can be internalized following binding to PAFR and clatherin-mediated endocytosis, transglutaminase-dependent transmembrane bilayer movement, engagement with low affinity binding sites at the plasma membrane, or direct physicochemical effects at the plasma membrane (16, 18, 19). As hNT neurons do not express PAFR (Fig. S5a), we established the kinetics of PAFR-independent PAF lipid internalization using a Bodipy fluorophore-labeled isoform [1-(O-[11-(4,4-difluoro-5,7-dimethyl-4-bora-3a,4a-diaza-s-indacene-3-proprionyl)amino]undecyl)-2-acetyl-sn-glycero-3-phosphocholine; C11:0 B-PAF]. A progressive increase in cell-associated C11:0 B-PAF was observed over the first 60 min of lipid addition to media (Fig. S6a). Thin layer chromatography (TLC) of lipid loosely associated with the neuronal plasma membrane (acid-labile fractions), intracellular lipid (acid-resistant fractions), and lipid present in extracellular milieu revealed that cell-associated C11:0 B-PAF was rapidly internalized within 5 min (Fig. S6b). An increase in hydrolysis was observed within 15 min with the C11:0 B-lyso-PAF metabolite released into the extracellular milieu (Fig. S6b). This balance between lipid internalization and lipid hydrolysis maintained a constant elevated intracellular concentration of C11:0 B-PAF for at least 60 min (Fig. S6b). Conversely, when hNT neurons were loaded with lyso-PAF, C11:0 B-lyso-PAF was rapidly internalized but converted to C11:0 alkylacylglycerophosphocholine and not C11:0 B-PAF (Figs. S6c and S7). Preferential conversion of C11:0 lyso-PAF to C11:0 B-PAF was only observed when neurons were treated in the presence of Aβ₄₂ (Fig. S7).

We exploited this loading approach to elevate acutely C16:0 PAF and C16:0 lyso-PAF levels in hNT neurons without directly impacting on other glycerophosphocholine second messengers. Intracellular lipid concentrations were determined by LC-ESI-MS and neuronal viability was assessed by TUNEL after a 24-h exposure. Significant neuronal death was only observed when intracellular C16:0 PAF levels reached molar concentrations comparable to those detected in postmortem AD tissue, 14- to 16-week-old TgCRND8 mice, and Aβ₄₂-treated hNT neurons (i.e., >400 pM/cell) (Fig. 2F). To establish specificity, fold changes in other changes in other endogenous alkylacylglycerophosphocholine concentrations were profiled and quantified by LC-ESI-MS. Nine PAF species were identified in human neurons. No significant change in any other isoform was observed beyond the expected sustained elevation in the target lipid and a progressive increase in C16:0 lyso-PAF likely the result of compensatory hydrolysis (Fig. 2G). Only direct 24-h application of C16:0 PAF (Fig. 2F) not C16:0 lyso-PAF (Fig. 2H) signaled neuronal death. To demonstrate that neurotoxicity was not the result of nonspecific lytic effects of extracellular PAF, plasma membrane integrity was assessed using ethidium bromide homodimer (EthD-1), a membrane impermeant dye. Direct application of C16:0 PAF to hNTs did not increase EthD-1 uptake over a 24-h treatment period (Fig. S8b). Loss in membrane integrity was only observed using methyl-carbamyl PAF (mc-PAF), a PAF analog resistant to hydrolysis, applied to hNT neurons at concentrations that exceeded critical micelle concentration (2.5–5 μM) (Fig. S8c). These data provided converging evidence that intraneuronal accumulation of C16:0 PAF signals a loss of viability in human neurons. To confirm these data, hNT neurons were transfected with EGFP and empty vector or EGFP and PAF-AH I α₂ and treated with 1 μM C16:0 PAF. As observed in Aβ₄₂-treated cultures (Fig. 2E), enhancing cytosolic hydrolysis of C16:0 PAF to C16:0 lyso-PAF prevented neuronal loss (Fig. 2I).

To establish how intraneuronal C16:0 PAF accumulation signals Aβ₄₂ toxicity, we first assessed calcium influx. A 180-s exposure to C16:0 PAF or C16:0 lyso-PAF did not acutely elevate intraneuronal calcium levels in the absence of PAFR as would be expected by a receptor-mediated event (Fig. S5b). We next assessed ER stress responses. In murine neurons, Aβ₄₂ activates both calpains and ER-associated caspase-12 (20). The human homolog of caspase-12 has a frameshift mutation resulting in a premature stop codon and an amino acid substitution within a critical site for caspase enzymatic activity, rendering it catalytically inactive (21). Caspase 2 is a candidate human ER caspase, while caspase 7 translocates to the ER following apoptotic challenge (22). To test whether rising C16:0 PAF levels impair ER function, we assessed changes in BIP (GRP78) expression. BIP increased rapidly in hNTs in response to C16:0 PAF loading (Fig. 3A). ER stress-associated calpain/caspase activation was next assessed by examining spectrin cleavage. Spectrin is a scaffold protein that anchors the plasma membrane to the cytoskeleton. The kinetics of cleavage can be used to track sequential calpain/caspase activation. Proteolytic processing by calpains (m and μ) yields 149- and 136-kDa fragments, whereas caspase 2, 3, and 7 cleavage yields 148- and 137-kDa fragments. Both the calpain/caspase-generated 148/149-kDa fragments can be further processed to 114-kDa and 34-kDa products by caspases 3/7 but not caspase 2 (23). We observed an increase in spectrin cleavage, generating both 148/149-kDa and 136/137-kDa fragments, 15 min after C16:0 PAF application (Fig. 3B). A further increase in the levels of the 136/137-kDa fragment was observed within 60-min coincident with the appearance of the caspase 3/7-specific product (Fig. 3B). To confirm these data, calpain and caspase activation were assessed directly. Activation of m-calpain is associated with cleavage of the 80-kDa calpain pro-protein (24). Using an antibody that detects the inactive pro-protein but not the cleaved product, a marked decrease in pro-m-calpain levels was detected 5–45 min after C16:0 PAF challenge consistent with activation (Fig. 3B). By direct enzyme assay, significant caspase 2 and caspase 3/7 activation was only observed at 60 min, following calpain activation (Fig. 3C). To establish whether this calpain-caspase signaling pathway was sufficient to compromise neuronal viability, hNT neurons were treated with C16:0 PAF in the presence of the caspase 2 inhibitor Z-VDVAD-FMK or the caspase 3/7 inhibitor Z-DEVD-FMK. Inhibiting either caspase 2 or caspase 3/7 protected hNTs from C16:0 PAF neurotoxicity (Fig. 3D).

Calpains can activate “tau kinases,” including glycogen synthase kinase 3β (GSK3β) and cyclin-dependent kinase 5 (Cdk5) (25, 26). Alkylacylglycerophosphocholine second messengers have already been shown to stimulate GSK3β activity (27). Their impact on Cdk5 or tau has not been examined. Here, we assessed indicators of both tau kinase activation and inhibition in hNTs loaded with C16:0 PAF. A transient increase in phosphorylation of the GSK3β regulator AKT at Thr-308, commonly associated with the inhibition of GSK3β activity, was observed after 5 min (Fig. 4A). These changes were followed by an activity-associated decrease in GSK3β phosphorylation at Ser-9 after 45 min (Fig. 4A). These data indicate that GSK3β is not activated within the first 45 min of intracellular C16:0 PAF accumulation. More rapid increases in the levels of Cdk5 and its activators p35 and p39 were observed within 15 min of C16:0 PAF application (Fig. 4A). Despite apparent m-calpain activation (Fig. 3B), cleavage of p35 and p39 to the more potent p25 and p29 Cdk5 activators was not detected (Fig. 4A). This p25/p29-independent induction of Cdk5 activity, as a result of acute C16:0 PAF accumulation, was validated by assessing transfer of radiolabeled P³² to histone-1 following immunoprecipitation of Cdk5 (Fig. 4B).

To evaluate AD-associated tau phosphorylation, we examined two of the main sites aberrantly phosphorylated in AD brain (Ser-202 and Thr-205) (28). AT8-reactive sites were phosphorylated within 15 min of C16:0 PAF exposure (Fig. 4C). To establish whether phosphorylation was mediated by Cdk5, cultures were treated with a general Cdk5/p35 inhibitor roscovitine. Roscovitine prevented the C16:0 PAF-dependent phosphorylation of tau on AD-specific epitopes (Fig. 4D) and protected hNT neurons from C16:0 PAF toxicity (Fig. 4E).

Taken together, these data provide strong evidence that a pathogenic accumulation of C16:0 PAF in response to oligomeric Aβ₄₂ participates in signaling the hyperphosphorylation of tau and, ultimately, causes caspase-dependent neuronal death. However, pharmacological strategies targeting C16:0 PAF must also consider the amnesiatic impact of PAFR inhibition on learning and memory (29, 30). PAFR is expressed by subpopulations of neurons, notably in the hippocampus (30). Thus, strategies to inhibit C16:0 PAF signaling must be able to pharmacologically distinguish between PAFR-dependent (neuroprotective/functional) and PAFR-independent (neurotoxic) PAF pathways. Orsellinic acid is a fungal-derived benzoic acid that inhibits C16:0 PAF-induced ER caspase activation without affecting PAFR-neuroprotective signal transduction (31). As with gain-of-function approaches to promote C16:0 PAF hydrolysis and restore physiological intracellular concentrations (Fig. 2 E and I), inhibition of intracellular PAF signal transduction with orsellinic acid protected hNT neurons from Aβ₄₂ toxicity (Fig. 4F). As a control, we treated cells with the PAFR antagonist (and structural PAF analog) CV3988. CV3988 did not impact on C16:0-mediated tau phosphorylation in PAFR-negative human neurons or Aβ₄₂ toxicity (Fig. S5d).

Flash-frozen parahippocampal gyri without fixation from deceased AD patients (74 ± 8 years) and controls (78 ± 16 years) who suffered sudden death due to non-neurological complications were obtained from the Douglas Hospital Research Centre Brain Bank (Montreal, Canada). Postmortem delay was between 5 and 25 h. TgCRND8 mice (13) and NonTg littermate controls were killed at 8 or 14–16 weeks of age. All animal manipulations were performed according to the strict ethical guidelines for experimentation established by the Canadian Council for Animal Care.

Glycerophospholipids were extracted according to a modified Bligh and Dyer procedure (16). LC-ESI-MS and TLC analyses were performed as described in SI Materials and Methods.

NT2/D1 cells were terminally differentiated to hNT neurons as described in ref. 36. RT-PCR conditions and primer sequences were as described in ref. 16. All treatment details are detailed in SI Materials and Methods. Transfections, quantification, calcium imaging methodologies, activity assays, and antibodies used for Western analyses are described in SI Materials and Methods.