Using neurolipidomics to identify phospholipid mediators of synaptic (dys)function in Alzheimer's Disease

Not all of the mysteries of life lie in our genetic code. Some can be found buried in our membranes. These shells of fat, sculpted in the central nervous system into the cellular (and subcellular) boundaries of neurons and glia, are themselves complex systems of information. The diversity of neural phospholipids, coupled with their chameleon-like capacity to transmute into bioactive molecules, provides a vast repertoire of immediate response second messengers. The effects of compositional changes on synaptic function have only begun to be appreciated. Here, we mined 29 neurolipidomic datasets for changes in neuronal membrane phospholipid metabolism in Alzheimer's Disease (AD). Three overarching metabolic disturbances were detected. We found that an increase in the hydrolysis of platelet activating factor precursors and ethanolamine-containing plasmalogens, coupled with a failure to regenerate relatively rare alkyl-acyl and alkenyl-acyl structural phospholipids, correlated with disease severity. Accumulation of specific bioactive metabolites [i.e., PC(O-16:0/2:0) and PE(P-16:0/0:0)] was associated with aggravating tau pathology, enhancing vesicular release, and signaling neuronal loss. Finally, depletion of PI(16:0/20:4), PI(16:0/22:6), and PI(18:0/22:6) was implicated in accelerating Aβ42 biogenesis. Our analysis further suggested that converging disruptions in platelet activating factor, plasmalogen, phosphoinositol, phosphoethanolamine (PE), and docosahexaenoic acid metabolism may contribute mechanistically to catastrophic vesicular depletion, impaired receptor trafficking, and morphological dendritic deformation. Together, this analysis supports an emerging hypothesis that aberrant phospholipid metabolism may be one of multiple critical determinants required for Alzheimer disease conversion.

Not all of the mysteries of life lie in our genetic code. Some can be found buried in our membranes. These shells of fat, sculpted in the central nervous system into the cellular (and subcellular) boundaries of neurons and glia, are themselves complex systems of information. The diversity of neural phospholipids, coupled with their chameleon-like capacity to transmute into bioactive molecules, provides a vast repertoire of immediate response second messengers. The effects of compositional changes on synaptic function have only begun to be appreciated. Here, we mined 29 neurolipidomic datasets for changes in neuronal membrane phospholipid metabolism in Alzheimer's Disease (AD). Three overarching metabolic disturbances were detected. We found that an increase in the hydrolysis of platelet activating factor precursors and ethanolamine-containing plasmalogens, coupled with a failure to regenerate relatively rare alkyl-acyl and alkenyl-acyl structural phospholipids, correlated with disease severity. Accumulation of specific bioactive metabolites [i.e., PC(O-16:0/2:0) and PE(P-16:0/0:0)] was associated with aggravating tau pathology, enhancing vesicular release, and signaling neuronal loss. Finally, depletion of PI(16:0/20:4), PI(16:0/22:6), and PI(18:0/22:6) was implicated in accelerating Aβ 42 biogenesis. Our analysis further suggested that converging disruptions in platelet activating factor, plasmalogen, phosphoinositol, phosphoethanolamine (PE), and docosahexaenoic acid metabolism may contribute mechanistically to catastrophic vesicular depletion, impaired receptor trafficking, and morphological dendritic deformation. Together, this analysis supports an emerging hypothesis that aberrant phospholipid metabolism may be one of multiple critical determinants required for Alzheimer disease conversion.

NEUROLIPIDOMICS: CATALOGING FUNCTIONAL DIVERSITY IN MEMBRANE BIOLOGY
The field of neurolipidomics seeks to understand how dynamic changes in membrane composition regulate brain cell function. Here, we mined 29 different neurolipidomic datasets generated by 11 independent laboratories for critical changes in neural membrane phospholipid metabolism (Tables S1-S4). We then asked how these changes might mechanistically contribute to synaptic dysfunction in Alzheimer's Disease (AD). Neuronal membranes are enriched in sterols, sphingolipids, glycerolipids, and phospholipids ( Figure 1A). Phospholipids are the most abundant. Their assembly into lipid bilayers, with polar head groups aligning at aqueous interfaces and hydrophobic carbon chains buried within, produces the semi-permeable barriers of cellular (and subcellular) membranes (Figures 1B,C). These bilayers are often conceptualized as undulating fields of identical molecules ( Figure 1C). In fact, they are complex matrices of several hundred molecularly distinct species (Figure 2A). Composition is in constant flux. Carbon chains and defining polar head groups are dynamically exchanged by activated phospholipases and lysophospholipid transferases in response to environmental stimuli (Figures 1D, 2B). Until recently, this diversity has remained largely underappreciated and functional significance unexplored. However, for the first time, significant technological advances in high performance liquid chromatography (LC), electrospray ionization (ESI), and matrix-assisted laser desorption ionization (MALDI) mass spectrometry (MS) are enabling membrane composition to be profiled comprehensively at the molecular level 1 . Coupled with subcellular fractionation and careful consideration of extraction protocols that enrich for different phospholipid families, Phospholipids are the most abundant components of neuronal membranes. (D) Dynamic phospholipid remodeling enables the rapid alterations in membrane form and function required for synaptic transmission while maintaining critical lipid composition essential for synaptic structural integrity. The Land's cycle is depicted. Hydrolysis of the sn-2 arachidonyl chain of a 1-O-ether-linked choline-containing glycerophospholipid by cPLA 2 releases AA (20:4). The residual lyso-PAF backbone can be remodeled by LPCAT1 at the plasma membrane into the powerful PAF family of neuromodulators or reconstructed back into a 1-O-alkyl-linked structural lipid depending upon whether an acetyl group or a long chain fatty acid, primed by the actions of acyl-Co synthetase, is used as a substrate. In turn, the sn-2 acetyl group released by PAF-AH during conversion of PAFs back to lyso-PAFs can be passed to an ester-linked lysophosphatidylcholine, a sphingolipid, or another ether-linked lyso-PAF with either the same or different sn-1 chains. PAF-AH 1b does not have transferase activity; PAFAH II does. This single pathway alone can generate a plethora of different phospholipids through the simple exchange of constituent hydrocarbon chains. Abbreviations: AA, arachidonic acid; cPLA 2 , cytoplasmic PLA 2 ; CoA, coenzyme A; LPCAT1, lysophosphatidylcholine acyltransferase 1; lyso-PAF, lyso-platelet activating factor or 1-O-alkyl lysoglycerophosphocholine; PAF, platelet activating factor or 1-O-alkyl-2-acetyl-glycerophosphocholine; PAF-AH, PAF-acetylhydrolase. species that vary by only one double bond, a single methylene group, or carbon chain linkage can now be quantified directly in synaptic preparations. Further, new high-resolution optical single molecule tracking approaches, notably fluorescence correlation spectroscopy (FCS) coupled with the stimulated emission depletion fluorescence nanoscopy (STED) and fluorescence lifetime imaging microscopy (FLIM), are facilitating, again for the first time, the study of functional interactions specific to lipid microdomains directly in living cells (Eggeling et al., 2009;Kusumi et al., 2010;Sahl et al., 2010;Mueller et al., 2013). Basic unitary conceptions are being challenged. Diacylglycerol (DAG), commonly conceived by cell biologists as a single lipid second messenger, is now recognized to be a family of over 50 structurally distinct isoforms (Callender et al., 2007) each controlling different cellular processes (Deacon et al., 2002).

EXPLORING A "LIPID-CENTRIC" VIEW OF SYNAPTIC FUNCTION AND DYSFUNCTION IN AD CONVERSION FROM A PRE-SYMPTOMATIC TO A SYMPTOMATIC STATE IN AD REQUIRES MULTIPLE METABOLIC DISRUPTIONS
Two central pathologies define AD: (1) intraneuronal accumulation of neurofibrillary tangles composed of hyperphosphorylated tau and (2) aberrant processing of the amyloid precursor protein (APP) to smaller, toxic amyloid β (Aβ) fragments. The most damaging is Aβ 42 . Accumulation is gradual with assembly of soluble Aβ 42 oligomers impairing synaptic function and signaling neuronal loss (Cleary et al., 2005;Palop and Mucke, 2010;Benilova et al., 2012). The "amyloid cascade hypothesis" defines these events as the root cause of AD (Hardy and Selkoe, 2002;Palop and Mucke, 2010;Benilova et al., 2012). Yet new data suggests that these driving Aβ and tau pathologies likely FIGURE 2 | Synaptic phospholipids (and cholesterol). (A) Structural membrane phospholipids are derivatives of sn-glycero-3-phosphoric acid with a 1-O-acyl, a 1-O-alkyl (ether-linked plasmanyl), or a 1-O-alkyl-1 -enyl (vinyl ether-linked plasmenyl) carbon chain, a long-chain fatty acid esterified to the sn-2 position, and a polar headgroup composed of a nitrogenous base, a glycerol, or an inositol unit modifying phosphoric acid at the sn-3 position. The polar head group defines different phospholipid classes with 1-O-acyl linked PS, phosphatidylserine; PA, phosphatidic acid; PI, phosphatidylinositol; PE, phosphatidylethanolamine; and PC, phosphatidylcholine predominant in neuronal plasma membranes. The free diffusion of these species through the lipid bilayer is, in part, influenced by cholesterol, a sterol defined by a four fused-ring core and an alkyl side-chain that preferentially solvates with some but not all sn-1 and sn-2 phospholipid side chains. (B) Phospholipids can be hydrolyzed by different families of PLAs. Cleavage sites are depicted using PC(16:0/16:0) as the target substrate.

www.frontiersin.org
July 2013 | Volume 4 | Article 168 | 3 represent only two of multiple determinants required for AD conversion. This refinement is based on evidence that Aβ 42 accumulation (at AD load levels) can be detected in both cognitively "normal" elderly and humanized mouse models with little learning and memory impairment (Snowdon, 2003;Zahs and Ashe, 2010). These observations have prompted a "re-imagining" of the amyloid hypothesis (Herrup, 2010;Kuller and Lopez, 2011;Nelson et al., 2011). Here, the amyloid deposition cycle, aggravated by chronic neuroinflammation, triggers a critical "change in state" (Herrup, 2010). This "change of state" is envisioned as a convergence of metabolic disruptions resulting in "a new 'normal' biology primed toward neurodegeneration and dementia" (Herrup, 2010).

PATHOLOGICAL MEMBRANE REMODELING MAY REPRESENT A CRITICAL METABOLIC DISRUPTION REQUIRED FOR AD CONVERSION
Our overarching hypothesis is that multiple aberrations in phospholipid metabolism are required for transition from presymptomatic to symptomatic AD. Here, we seek to identify some of these additional critical disruptions by mining existing neurolipidomic datasets. Previous work has correlated the extent of membrane phosphocholine (PC) and phosphoethanolamine (PE) breakdown with severity of dementia and psychosis in AD patients (Klein, 2000;Sweet et al., 2002). Moreover, soluble oligomeric Aβ 42 neurotoxicity is signaled, in part, by enhanced metabolism of ether linked structural phospholipids (Sanchez-Mejia et al., 2008;Ryan et al., 2009). Targeting upstream remodeling or downstream signaling of key metabolites can protect human and murine neurons from Aβ 42 in vitro and improve behavioral indices of learning and memory in vivo (Kriem et al., 2005;Sanchez-Mejia et al., 2008;Ryan et al., 2009). There is also compelling evidence to indicate that the fatty acid substrates available for phospholipid biosynthesis are altered over the course of AD (Table S5). For example, using an unbiased neurolipidomics approach, non-esterfied monounsaturated fatty acid (MUFA) levels have been shown to increase in AD brain whereas polyunsaturated (PUFA) levels generally decrease (Lukiw, 2009;Astarita et al., 2010. These changes are attributed to increased expression of three of the rate-limiting enzymes in MUFA biosynthesis, stearoyl-CoA desaturase-1 (SCD-1) SCD-5a and SCD-5b . Enhanced activity is thought to compensate for decreasing bioavailability of PUFA substrates highlighting the molecular interdependency of lipid metabolic defects that occur over the course of AD . These disruptions may confer AD risk. Recent genome wide association studies (GWAS) have identified variants in sortilinrelated receptor (Rogaeva et al., 2007), clusterin (Harold et al., 2009;Lambert et al., 2009;Jun et al., 2010), bridging integrator 1 (Seshadri et al., 2010), ATP-binding cassette sub-family A member 7 (Hollingworth et al., 2006), and phosphatidylinositol binding clathrin assembly protein (Harold et al., 2009;Jun et al., 2010) genes as AD risk factors. Clearly, these genes do not act through the same biochemical pathways yet each shares in common a regulation of one or more aspects of neural phospholipid (and in some cases cholesterol) metabolism with the canonical ApoE risk allele (Strittmatter et al., 1993).

INTERPRETING SYNAPTIC GLYCEROPHOSPHOLIPIDOME DATASETS
We used a post-hoc neurolipidomics approach to identify potential phospholipid determinants of AD cognitive impairment. We first collated the findings of 29 published neurolipidomic datasets generated using multiple methodologies, technologies and human and murine samples (Tables S1-S4). If not explicitly identified by the studies' authors, we used the online bioinformatics tool VaLid to predict sn-1 and sn-2 carbon chain stereospecificity (Blanchard et al., 2013). Results were combined, heat-mapped, and mined for underlying patterns (Figure 3). This analysis required that we set aside two of fundamental assumptions commonly applied to the interpretation of genomic and proteomic datasets. First, we postulated that critical compositions of synaptic lipids are likely not the ones altered early in disease etiology but rather the species that exhibit a catastrophic depletion in late-stage disease. Unlike mRNA and protein, individual phospholipids do not have their own synaptic halflives and cannot be tracked as single entities from biogenesis to catabolism. Rather, following linkage determination in the endoplasmic reticulum and transport to synaptosomal membranes, their defining sn-1 and sn-2 carbon chains (depending on linkage to the phosphoglyceride backbone) and, to some extent, their polar head groups are passed between species. Each of the free fatty acids released by PLA 1 or PLA 2 hydrolysis ( Figure 2B) has its own synaptic half-life, ranging from 10 days for palmitic acid (16:0) to 60 days for AA (20:4) (Ando et al., 2002). This metabolic consubstantiality, wherein critical chemical compositions are maintained through dynamic remodeling of multiple species, enables the rapid changes in membrane curvature (and fusion) required for neurotransmission while ensuring synaptic structural integrity (Figures 4A-F). We argue that any quantifiable shift in these critical compositions likely indicates that a catastrophic "change in state" has already occurred (i.e., evidence of a membrane biology now "primed toward neurodegeneration and dementia"). Second, we postulated that variations between datasets in the composition of smaller glycerophospholipid intermediates (i.e., the lysophospholipids, free fatty acids, and their downstream metabolites) ( Figure 1C) are not "noise" but likely (1) snapshots of the structural species being actively mobilized to maintain critical structural membrane compositions at time of extraction and (2) identify the neuroactive phospholipid  (Eberlin et al., 2010). The color scheme reflects statistical changes in ApoE −/− , ApoE ε2, ε3, ε4 humanized knock-in (KI), cathepsinD (Ctsd) −/− , APPSw transgenic, PS1 transgenic, PS1/APPSw double transgenic, TgCRND8 transgenic mice, and post-mortem human AD brain tissue relative to cognate controls as published. Where comparisons were not made with disease or experimental conditions, color (green) simply reflects detection in datasets. Where published datasets identified lipids by total number of carbons and degree of unsaturation, assignment of sn-1 and sn-2 chains in this study was performed as indicated in Tables S1-S4 (footnote#1). Tissue abbreviations: Cx, Cortex; Crb, cerebellum; ECx, Entorhinal Cortex; FCx, Frontal Cortex/Prefrontal Cortex; Hip, Hippocampus; PCx, Parietal Cortex; SPM, Synaptosomal membranes; SV, Synaptic Vesicle. All other conditions are as described in Tables S1-S4.

www.frontiersin.org
July 2013 | Volume 4 | Article 168 | 5 signaling molecules that transiently accumulate in different disease states. We argue that these changes will emerge earlier than catastrophic alterations in overall structural phospholipid composition. Taken together, we maintain that the interpretation of lipidomic datasets depends less upon a determination of absolute phospholipid levels (which will vary considerably from laboratory to laboratory and methodology to methodology) than upon identification of the patterns in lipid composition that are altered over the course of disease progression (Brown and Murphy, 2009). Here, we present three such patterns in membrane metabolism identified through analysis of 29 independent datasets predicted to signal synaptic dysfunction in AD.

PATTERN 1: EVIDENCE FOR SYNAPTIC POISONING: A PRIMARY ROLE FOR ETHER-LINKED PHOSPHOLIPIDS IN MEDIATING AD VESICULAR DEPLETION SYNAPTIC MICRODOMAINS EXHIBIT DIFFERENT PHOSPHOLIPID PROFILES
Enrichment in PC, PE, PI, PS, specifically isoforms with (16:0) and (18:1) carbon chains at both their sn-1 and sn-2 positions, is detected in synaptosomal membranes prepared from healthy mouse, rat, and human brain (Tables S1-S4) 2 . Given the dietary abundance of palmitic acid and oleic acid, this pattern is, of course, expected. Yet there is an intriguing molecular specificity that distinguishes different synaptic microdomains ( Figure 4A). In healthy murine tissue, the relative abundance of PCs and 1-O-alkyl-PC isoforms ( but only a small percentage of the more elastic PS (and PI) (16:0/16:0) species compared to non-raft domains (Takamori et al., 2006;Martin et al., 2010). Finally, fewer acyl-linked PC, PE, and PS species with docosahexaenoic acid (DHA, 22:6) at their sn-2 positions are found in isolated synaptic vesicles relative to other synaptosomal microdomains and brain tissue (Williams et al., 2000;Han et al., 2001;Takamori et al., 2006;Wurtman et al., 2009;Axelsen and Murphy, 2010;Martin et al., 2010).

MEMBRANE REMODELING REQUIRED FOR NEUROTRANSMITTER RELEASE IS REGULATED, IN PART, BY THE SUBSTRATE SPECIFICITIES OF SYNAPTIC PHOSPHOLIPASES
Vesicle diameter, measured by cryo-electron microscopy, ranges between 30 and 60 nm, with an average diameter of 42 nm (Takamori et al., 2006;Castorph et al., 2010). A single phospholipid occupies a space of ∼65 Å 2 (0.65 nm 2 ) (Takamori et al., 2006;Castorph et al., 2010). With phospholipid composition calculated as 50-75% of the total lipid content, each 42 nm diameter vesicle is estimated to contain around 7000 individual species (McMahon and Gallop, 2005;Takamori et al., 2006). Structure, in part, dictates topography. The extreme curvature of vesicular membranes places more phospholipids in the outer and fewer in the inner leaflet to balance the spatially larger exterior positive curvature with the spatially smaller interior negative curvature (McMahon and Gallop, 2005;Takamori et al., 2006) ( Figure 4F). Phospholipid identity within these leaflets must also reconcile the need to stabilize vesicular structure with the requirement to rapidly reverse membrane curvature and fuse with the plasma membrane on demand (Figure 4F). In a proteincentric view, these dynamics are conceptually better suited to pliant fusogenic phospholipids whose physicochemical natures are "released" once the structural constraints enforced by integral membrane proteins are relaxed (McMahon and Gallop, 2005). Such a model makes the predominance of saturated diacyl PC(16:0/16:0) lipids in isolated synaptic vesicles somewhat surprising (Takamori et al., 2006). Reconstitution studies indicate that the cylindrical geometries of dipalmitoyl-PCs are more suited to the generation of planar membranes or, in presence of water, to the formation of vesicles with a longer-lived stable structure under negative tension (McMahon and Gallop, 2005;Shinoda et al., 2010). The enrichment of PC(16:0/16:0) would therefore be expected to render synaptic vesicles more fusion-resistant yet these neurolipidomic profiles are supported by small angle neutron scattering analysis detecting cylindrically shaped lipids within mini-microdomains of ∼15 nm in length along the perimeter of reconstituted vesicles (Vogtt et al., 2010). Form reveals function when placed in context with the dynamics of synaptic phospholipase substrate specificities ( Figure 4A). Secretary phospholipase A 2 isoforms (sPLA 2 , Group IIA and possibly Group V) and PlsEtn-selective phospholipase A 2 (PlsEtn-PLA 2 ) are preferentially activated at sites of vesicular fusion (Kolko et al., 2003;Wei et al., 2003;Takamori et al., 2006). Group IIA sPLA 2 , released at the synapse, displays a low affinity for ester-linked acyl-PCs and a high affinity for ether-linked 1-O-alkyl-PCs (Pruzanski et al., 1998;Boyanovsky and Webb, 2009). Intracellular calcium-independent PlsEtn-PLA 2 preferentially hydrolyzes DHA (22:6) from PlsEtns species (Ramadan et al., 2010). Hydrolysis of the outer leaflets of vesicular membranes (by intracellular PlsEtn-PLA 2 ) and of pre-synaptic densities (by sPLA 2, ), releasing AA (20:4) from PC(O-16:0/20:4) and DHA (22:6) from PE(P-16:0/22:6), would indeed generate highly fusogenic wedge-shaped lyso-PAFs and lyso-plasmalogens without impacting on neighboring diacyl-PCs. The geometries of lyso-PAF PC(O-16:0/0:0), lyso-PlsEtn PE(P-16:0/0:0), and lyso-PlsEtn PE(P-18:0/0:0) are predicted to favor the reversal from negative to positive curvature required for the fusion pore formation (Piomelli et al., 2007;Shin et al., 2010) ( Figure 4F). These lyso-lipids have also been shown to regulate soluble N-ethylmaleimide-sensitive factor attachment receptor (SNARE) protein complexes embedded both in vesicular (v-SNARE) and pre-synaptic target (t-SNARE) membranes. SNARE complexes direct vesicular fusion with activities modulated by their phospholipid binding partners. Association with acyl-linked lysophospholipids promotes the formation of larger of t-/v-SNARE protein complexes slowing down vesicle fusion while association with lyso-PlsEtns enhances the formation of smaller complexes speeding up fusion (Glaser and Gross, 1994). Moreover, the free DHA (22:6) and AA (20:4), released from the sn-2 position of ether-linked structural membrane lipids by the actions of sPLA 2 and PlsEtn-PLA 2 , can themselves signal enhanced basal release of noradrenaline, again apparently through converging feedback mechanisms that activate SNAREs (Darios et al., 2010;Geraldine et al., 2010).

ENHANCED HYDROLYSIS OF VESICULAR PAF PRECURSORS AND MEMBRANE PLSETNS IS MECHANISTICALLY CONSISTENT WITH THE HYPOTHESIS THAT INCREASED VESICULAR RELEASE PRECIPITATES VESICULAR DEPLETION OVER THE COURSE OF AD
MS profiling of post-mortem AD brain as well as tissue and synaptosomes isolated from transgenic models of AD and AD genetic risk factors provide strong evidence that this pattern is disrupted over the course of AD (Figures 3A-D). Specifically, a depletion in DHA (22:6), an increase in free AA (20:4), and an accumulation of ether-linked fusogenic lyso-phospholipids defined by (16:0) carbons at their sn-1 position are consistently detected across datasets (Sanchez-Mejia et al., 2008;Ryan et al., 2009;Astarita et al., 2010. This particular pattern is remarkably similar to the acute changes in free fatty acids and phospholipid profiles observed following toxin-induced synaptic depletion in snake venom poisoning. Certain snake venoms act to stimulate vesicular fusion, partially by enhancing phospholipase-dependent structural membrane lipid hydrolysis, while simultaneously inhibiting the remodeling of the ether and vinyl-ether linked lyso-phospholipid metabolites required for the subsequent reconstitution of vesicular membranes. The net effect is vesicular depletion precipitating acute synaptic failure (Valentin and Lambeau, 2000). This phenomenon can also be artificially induced simply by the addition of the lyso-phospholipids at concentrations that inhibit their own remodeling (Rigoni et al., 2005;Rossetto et al., 2006). The phospholipid profiles detected in AD datasets mirror this pattern of synaptic poisoning (Figures 3A-D) albeit over a much longer time course. Changes are consistent with the hypothesis that enhanced vesicular release in patients suffering from mild cognitive impairment (MCI), precedes (and likely precipitates) the vesicular depletion seen in moderate to late stage AD (DeKosky et al., 2002;Truchot et al., 2007). Certainly, acute exposure of primary neurons to Aβ oligomers enhances vesicular fusion and increases the rate of neurotransmitter release while chronic exposure depletes neurons of synaptic vesicles and impairs neurotransmission (Dante et al., 2008;Nimmrich and Ebert, 2009;Parodi et al., 2010).

REDUCTIONS IN PE(P-16:0/22:6) AND PE(P-16:0/20:4) CORRELATE WITH SEVERITY OF AD COGNITIVE IMPAIRMENT
Not all profiles, however, detect sparing of PE(P-16:0/22:6) and PE(P-16:0/20:4) in AD ( Figure 3A, Table S1). In studies where these species are reduced, loss correlates with clinical dementia scores but, interestingly, not always with post-mortem AD pathology (Han et al., 2001). Patients with less Aβ and tau pathology but a greater reduction in PlsEtns are more impaired than those with exacerbated primary pathology but little change in PlsEtns profile (Han et al., 2001). Similarly, enhanced turnover of structural PlsEtns and PCs, manifested by an accumulation of smaller lyso-plasmalogens and cholinecontaining metabolites detected by magnetic resonance spectroscopy, correlates with more severe cognitive impairment, deterioration, and psychosis (Sweet et al., 2002). Together, these data point to a primary impact of PlsEtn deficiency on synaptic function and highlight PE(P-16:0/22:6) and PE(P-16:0/20:4) and to a lesser extent lyso-PAF PC(O-16:0/0:0) as critical phospholipids species regulating vesicular fusion at the membrane level. Metabolism of structural membrane PIs and 1-O-alkyl PAF precursors generates powerful lipid second messengers. Differential distribution of phosphorylated PI metabolites is required for synergistically assembling and remodeling the SNARE complexes providing a second means by which synaptic phospholipids may directly regulate neurotransmission. PI(4,5)P 2 and PI(3,4,5)P 3 are enriched in plasma membrane with 16:0/18:1, 18:0/20:4, and 18:0/22:6 isoforms predominating (Igbavboa et al., 2002). PI(4)P is found more frequently in Golgi membranes while PI(3)P and PI(3,5)P 2 are enriched in endosomal membranes (Lasiecka et al., 2009). In yeast, PI(3)P both enhances the capacity of membrane-bound SNARES to drive fusion in the absence of SNARE chaperones as well as synergistically activates SNARE chaperones to recruit Vam7p into fusion-competent complexes. PI(3)P has also been shown to be required for the subsequent SNARE complex disassembly once fusion is complete and cargo released (Mima and Wickner, 2009a,b). In neurons, this process further depends upon interaction with PI(4,5)P 2 found in highest concentrations at pre-synaptic densities. Decreasing PI(4,5)P 2 abolishes the ability of the v-SNARE-regulator, synaptotagmin-1, to interact with both the pre-synaptic membrane and t-SNARE precomplexes thereby preventing calcium-dependent release of neurotransmitter from the pre-synaptic active zone (Lee et al., 2010).

PROGRESSIVE DEPLETION OF PI(4,5)P 2 TRANSIENTLY INCREASES VESICULAR FUSION AND CHRONICALLY IMPAIRS NEUROTRANSMITTER RELEASE
Metabolic interdependency is, however, underlined by the fact that the familial AD mutations in presenilin 1 and 2, in turn, impair PI(4,5)P 2 metabolism effectively reducing levels in vesicular and pre-synaptic membranes (Landman et al., 2006). Moreover, in experimental models of AD, PI(4,5)P 2 is downregulated in response to synaptotoxic oligomeric Aβ 42 (but not monomeric or fibrillar configurations). This decrease is mediated, in part, by increases at the pre-synaptic active zone in the Ca 2+ dependent activation of PLC hydrolyzing PI(4,5)P 2 to DAG and IP 3 as well as increased activity of synaptojanin 1, the primary PI(4,5)P 2 phosphatase present at neuronal synapses (Berman et al., 2008). In axons, the hydrolysis of PI(4,5)P 2 by synaptojanin 1, drives the inner leaflet to adopt a negative curvature that embraces trafficking vesicles during fusion (Cremona et al., 1999). Such activity provides another lipidcentric mechanism by which oligomeric Aβ 42 acutely increases vesicular fusion then chronically impairs neurotransmitter release (Dante et al., 2008;Nimmrich and Ebert, 2009;Parodi et al., 2010).

PATTERN 3: THE FRANKENSTEINIAN SYNAPSE: DENDRITIC DEFORMATION AND DHA LIPID RAFT MOVEMENT IS REGULATED BY DYNAMIC CHANGES IN PERI-AND POST-SYNAPTIC PHOSPHOLIPID COMPOSITIONS
Finally, neurolipidomic profiling combined with high resolution imaging approaches are also revealing a reorganization of postsynaptic microdomains in AD mechanistically associated with the dendritic spine deformation and dysfunction (Tackenberg et al., 2009). Like pre-synaptic and vesicular membranes, dendritic peri-and post-synaptic densities are enriched in PE and PlsEtn isoforms (Figures 4B-F) but, as discussed above, with a functional topography (Han et al., 2001;Igbavboa et al., 2002;Takamori et al., 2006;Ryan et al., 2009;Axelsen and Murphy, 2010;Brand et al., 2010;Eberlin et al., 2010;Lohmann et al., 2010;Sharman et al., 2010) (Figure 3A, Table S1). Application of STED-FCS and single molecule optical tracking approaches to the study of lipid dynamics in living cells further reveals a remarkable territoriality to the free diffusion of structural lipids through cell membranes suggesting that lipid composition may regulate the direction of lipid raft movement between peri-and post-synaptic densities. Phospholipids appear to be limited in their free diffusion within membranes to subdomains of ∼20 nm in diameter (Eggeling et al., 2009;Sahl et al., 2010). The speed at which membrane lipids patrol these territories is cholesterol-assisted and backbone-specific. Sphingolipids, for example, freely diffuse traversing a 3 nm membrane radius of their territory within 3 ms in the absence of cholesterol. When cholesterol is present, they become trapped for up to 17 ms in these same regions. PE(16:0/16:0) pays little heed to cholesterol, diffusing at rates of less than 4 ms regardless of the presence or absence of cholesterol (Sahl et al., 2010). Thus, the enrichment of sphingolipids within cholesterol-rich lipid rafts is likely, in part, ensured by their lethargy in the presence of cholesterol while the presence of PEs (and PlsEtns) and alkylacylPCs at the borders of lipid rafts may reflect a functional interaction with cholesterol (Tables S1, S2). Here, isoform specificity comes into play. Converging artificial membrane reconstitution studies using 2 H NMR, nuclear Overhauser enhancement spectroscopy in 1 H magic angle spinning NMR, X-ray diffraction, and solid-state 2 H NMR strongly suggest that cholesterol exhibits an aversion for DHA and the unsaturated 1-O-alkyl chains of PAF precursor lipids. Rather, cholesterol favors solvation with saturated (16:0) or (18:0) chains (Brzustowicz et al., 1999(Brzustowicz et al., , 2002aShaikh et al., 2002;Pitman et al., 2004;Kusumi et al., 2010).  (Tables S1-S3). One could also envisage that the radial rotation of PE(16:0/22:6), for example, pirouetting to solvate its 16:0 sn-1 hydrocarbon with cholesterol could also create eddies and currents in dendritic microdomains promoting directional raft movement through the bilayer schematically presented at a single time point in Figure 5A.

COMPOSITIONAL SPECIFICITY IN DHA DEPLETION MORPHOLOGICALLY DEFORMS DENDRITIC SPINES
Disruptions in these patterns are associated with both impairment of lateral trafficking of receptors along dendritic membranes and morphological deformation of peri-and post-synaptic densities in AD and animal models of AD ( Figure 5B). It is well-established that DHA (22:6) concentrations decrease in AD brain, liver, and AD risk models (Farooqui et al., 2007;Lukiw and Bazan, 2008;Pomponi and Bria, 2008;Pauwels et al.,  2009; Astarita et al., 2010;Seshadri et al., 2010 (Igbavboa et al., 2002) ( Figures 3A-C, Tables S1-S3). Further, in some MCI patients, most AD patients, and APP/PS-1 double transgenic mice, a reduction in flippase activity, responsible for translocating PS isoforms within synaptic bilayers is reduced resulting in the accumulation of PS at the outer leaflet of synaptosomal membranes (Bader Lange et al., 2008, 2010. Because phospholipid length is directly proportional to the number of carbon atoms yet inversely proportional to the number of double bonds in each carbon chain (Shaikh et al., 2009), the combinations of the regional enrichment of PS(22:6/22:6), PE(22:6/22:6), and PC(22:6/22:6), the depletion of other DHA-containing isoforms throughout postsynaptic membranes, and the reduction of PS(18:0:/22:6) at the inner leaflet would most likely thin regions of peri-and postsynaptic densities while making integral lipid rafts more bulky and less mobile ( Figure 5B). These changes are predicted to impair synaptic transmission. This neurolipidomic interpretation inferred from phospholipid composition is consistent with previous hypotheses based on observations of physicochemical lipid properties (Piomelli et al., 2007;Pomponi and Bria, 2008). Certainly transgenic mice ectopically expressing human tau exhibit thinner apical and basal dendritic spines associated with a reduction in synaptic strength (Dickstein et al., 2010). This pattern is also consistent with the morphological deformation, dystrophy, and ultimate loss of dendritic spines detected by Golgi impregnation and electron microscopy of post-mortem AD cortex (Baloyannis et al., 2007).

CONVERGING METABOLIC DISRUPTIONS IN PI, PE, PLSETNS METABOLISM AND DHA BIOAVAILABILITY IS PREDICTED TO IMPAIR RECEPTOR TRAFFICKING BETWEEN PERI-AND POST-SYNAPTIC DENSITIES
DHA (22:6) depletion may also directly impact upon receptor density and chemical reception at post-synaptic densities. Aβ oligomers contribute to the impairment of synaptic transmission, in part, by disrupting the endocytosis and trafficking of AMPA and NMDA receptors at dendritic spines (Durakoglugil et al., 2009;Rui et al., 2010). In vitro, receptors are both rapidly removed from post-synaptic densities and fail to re-insert during synaptic potentiation following exposure of hippocampal neurons to oligomeric Aβ (Rui et al., 2010). Although underlying mechanisms have only begun to be elucidated, the reduction in PI metabolites and the depletion of DHA from PE and PlsEtns isoforms detected across multiple neurolipidomic datasets are implicated (Figures 3A,D, Tables S1, S4). In dendrites, PI(3,4,5)P 3 , generated, in large part, through the actions of class I phosphatidylinositol-3-kinases (PI3Ks), is required to maintain AMPA receptors at post-synaptic densities (Arendt et al., 2010). Inhibition of PI(3,4,5)P 3 synthesis enhances AMPA receptor mobility such that they drift from post-to peri-synaptic sites (Arendt et al., 2010). This biology is likely reinforced by the depletion of PI precursors, notably PI(16:0/20:4), PI(18:0/20:4), PI(16:0/22:6), and PI(18:0/22:6) observed in ApoE-deficient and AD transgenic mice (Igbavboa et al., 2002;Chan et al., 2012). Within this Frankensteinian synapse-a term we have coined to describe the result of this catastrophic transformation brought about by aberrant changes in lipid composition-rafts would be no longer "corralled" within post-synaptic densities but rather would be free to diffuse to and from peri-synaptic regions ( Figure 5B). In support of this hypothesis, post-synaptic density protein 95 (PSD-95), a specialized scaffolding protein that complexes receptors with cytoskeletal elements at the synapse, is reduced at post-synaptic densities when cellular membranes are experimentally depleted of DHA yet enriched following DHA supplementation (Wurtman et al., 2009;Langelier et al., 2010). Further, recent evidence that ongoing systemic DHA depletion (Astarita et al., 2010) is compensated for by increased MUFA biosynthesis  suggest that these critical metabolic disruptions are likely mutually negatively reinforcing thereby playing a determinative role in precipitating the "critical change of state" required for AD conversion.

CONCLUSIONS: WHAT DO CHANGES IN PHOSPHOLIPID COMPOSITION TELL US ABOUT AD SYNAPTIC DYSFUNCTION?
Advances in genomics have identified genetic determinants of neurodegenerative disease. Direct biochemical investigations have elucidated multiple signaling pathways altered by these genetic determinants leading to cognitive deterioration (Kim and Tsai, 2009;Nimmrich and Ebert, 2009). The combination of genomics with proteomics is being used to map the temporal changes in gene and protein expression that occur during transition from pre-symptomatic to symptomatic disease states. We argue that the next major advance in rational therapeutic design will come from tying the dynamics of the susceptible cellular metabolome into these genomic and proteomic maps of disease. The emerging field of neurolipidomics is identifying patterns of membrane disruption predicted to confer AD risk. In this analysis, three overarching metabolic disturbances were detected by post-hoc analysis of 29 independent datasets. These impairments include an increase in the hydrolysis of PAF precursors and membrane PlsEtns coupled with a failure to regenerate these relatively rare alkyl-acyl and alkenyl-acyl structural phospholipid compositions. Initially, pathological disruptions appear to affect specific phospholipids defined by carbon chain length, linkages, and degree of unsaturation. and morphological dendritic deformation. It will be essential to test whether intervention into one or more of these metabolic pathways can delay conversion from pre-symptomatic to symptomatic AD in the face of ongoing Aβ 42 and tau pathology. A better understanding of how cellular bioactive lipids alter susceptibility to driving AD pathologies represents a new, potentially transformative, means of identifying and targeting metabolic determinants of Aβ susceptibility and, in the long-term, an exciting means of potentially promoting AD resistance. We would also like to thank and acknowledge the outstanding neurolipidomic researchers whose work has not been included in this article due to space constraints. The authors wish to dedicate this article in heartfelt memorial to neuro-architect Dr. Marco Frascari (1945, our mentor and friend.