Cellular cholesterol homeostasis and Alzheimer’s disease

Alzheimer’s disease (AD) is the most common form of dementia in older adults. Currently, there is no cure for AD. The hallmark of AD is the accumulation of extracellular amyloid plaques composed of amyloid-β (Aβ) peptides (especially Aβ1-42) and neurofibrillary tangles, composed of hyperphosphorylated tau and accompanied by chronic neuroinflammation. Aβ peptides are derived from the amyloid precursor protein (APP). The oligomeric form of Aβ peptides is probably the most neurotoxic species; its accumulation eventually forms the insoluble and aggregated amyloid plaques. ApoE is the major apolipoprotein of the lipoprotein(s) present in the CNS. ApoE has three alleles, of which the Apoe4 allele constitutes the major risk factor for late-onset AD. Here we describe the complex relationship between ApoE4, oligomeric Aβ peptides, and cholesterol homeostasis. The review consists of four parts: 1) key elements involved in cellular cholesterol metabolism and regulation; 2) key elements involved in intracellular cholesterol trafficking; 3) links between ApoE4, Aβ peptides, and disturbance of cholesterol homeostasis in the CNS; 4) potential lipid-based therapeutic targets to treat AD. At the end, we recommend several research topics that we believe would help in better understanding the connection between cholesterol and AD for further investigations.

. The 24S-hydroxycholesterol (24S-OH) is the most abundant oxysterol in the brain, with the enzyme responsible for its biosynthesis, cholesterol 24S-hydroxylase (CYP46A1), playing a key role in the degradation and excretion of cholesterol from the brain (13). Due to the tight junction that exists between the endothelial cells of the blood-brain barrier (BBB) [see (14) for a review], the efflux of macromolecules between the brain interior and the blood vessels within the brain is severely limited; a much higher concentration of 24S-OH is found in the brain than in the blood, while the opposite is true for 27-hydroxycholesterol. However, under conditions when the BBB becomes leaky, more 24S-OH may enter the blood more readily, while 27-hydroxycholesterol may move into the brain interior more readily, as suggested by Björkhem and colleagues (15,16). Under various oxidative conditions, cholesterol can be auto-oxidized and converted to 7-ketocholesterol. In steroidogenic tissues, such as the adrenal glands, testes, ovaries (17), and the hippocampi of the brain (18), cholesterol is first converted to pregnenolone, which can then serve as the precursor for the synthesis of many steroid hormones. In the animal models of various neurodegenerative diseases, neurosteroids (such as those produced in the brain hippocampus) are shown to exhibit neuroprotective activities (19). In the liver, cholesterol serves as the precursor for bile acid biosynthesis. Bile acids are the major catabolites of cholesterol, and play important roles in lipid digestion and absorption. The chemical structures of cholesterol, various oxysterols, pregnenolone, and cholesteryl ester (CE) (shown here as cholesterol oleate as an example) are shown in Fig. 1.

Macromolecules involved in cellular uptake, storage, and removal of cholesterol
In addition to synthesizing cholesterol de novo, essentially all mammalian cells can acquire cholesterol from exogenous sources. Cells in systemic tissues acquire their cholesterol mostly from LDL, the major cholesterol-carrier in the blood, via its internalization through the LDLR (20). Cells in the CNS, on the other hand, acquire their cholesterol mainly from ApoE lipidated with phospholipids and cholesterol (ApoE/PL/C), the major lipoprotein in the CNS (21), through its internalization via multiple ApoE receptors (also see the section, Lipoproteins in the blood versus lipoproteins in the CNS). To prevent the excessive accumulation of free (i.e., unesterified) cholesterol in cells, the excess cholesterol either gets esterified to form CEs for storage, or gets removed from the cell altogether. Cholesterol esterification is catalyzed by the enzymes, ACATs, also called sterol O-acyltransferases (ACAT SOAT s); these enzymes use various long-chain fatty acyl-CoAs, and various sterols with 3--OH, including cholesterol and various oxysterols (22,23) as their substrates. There are two ACAT SOAT genes (24,25). The ACAT1 SOAT1 gene is ubiquitously expressed in essentially all cells, including cells in systemic tissues and in the CNS; while the ACAT2 SOAT2 gene is expressed in intestinal enterocytes and in hepatocytes; low levels of ACAT2 SOAT2 are also detectable in various peripheral tissues examined. Both ACAT1 SOAT1 and ACAT2 SOAT2 are integral membrane proteins located at the ER region, and both are allosterically activated by their own substrate (cholesterol or oxysterols) (26,27). Neither Acat1 Soat1 nor Acat2 Soat2 is controlled by the transcription factor, SREBP2. Regarding the cellular sterol removal process, it involves the release of sterols and certain phospholipids at the cell surface. This process depends on the ABCA1, which is a multi-span membrane protein mainly enriched at the plasma membrane (PM) [reviewed in (28)]. The sterol efflux process also requires the presence of certain helical apolipoproteins located at the cell exterior, such as ApoA-I, the major apolipoprotein in HDL in the blood, or ApoE, the major apolipoprotein in the CNS (also see the section, Lipoproteins in the blood versus lipoproteins in the CNS). ABCA1 and a related protein, ABCG1, which is a separate member of the ABC transporter family (29,30), work in concert to prevent the buildup of excess cholesterol (31)(32)(33). ABCA1 exhibits a preference to use newly biosynthesized sterols, especially lanosterol, as its sterol substrates (34). The expression of the ABCA1 gene is positively controlled by the LXRs and retinoid X receptors (RXRs), with oxysterols and retinoid as the main activating ligands; these ligands act in a synergistic manner [for details, see (35)]. In addition, the Abca1 gene is negatively regulated by several microRNAs (miRs), including miR-106b, miR-758, and miR-33, which are expressed abundantly in the brain [reviewed in (36)].

Lipid raft domains
The lipid and protein compositions in various membrane organelles are highly heterogeneous; however, they all contain cholesterol as a ubiquitous lipid ingredient. The cholesterol that binds tightly with sphingolipids forms a sterol-rich sphingolipid-rich domain, referred to as the lipid raft domain. These domains are present in various cell membranes, and serve as platforms to host various membrane proteins involved in many cell-signaling processes (37). Lipid-based platforms that are independent of cholesterol are also present in cell membranes (38). Within a given membrane, whether the distribution of cholesterol across the lipid bilayer is asymmetrical or not remains unclear. For example, by using different approaches, different investigators reported that cholesterol is either enriched at the inner leaflet (39,40) or enriched at the outer leaflet of the PM (41).

The concept of "active cholesterol"
Within the membrane, cholesterol interacts with sphingolipids as well as phospholipids to form complexes. The cholesterol content in excess of that that forms complexes with phospholipids/sphingolipids can be considered as active cholesterol; active cholesterol has a higher tendency to move away from the membrane than the "bound" cholesterol (42). Cholesterol is practically insoluble in water; various transport mechanisms are needed in order to catalyze the movement and the recycling of cholesterol within the cell.

Nonvesicular lipid transport
The transfer and/or exchange of cholesterol and other lipids between two adjacent subcellular membrane organelles are facilitated mostly via nonvesicular mechanisms. Unlike vesicular transport [that accounts for the vast majority of intracellular protein transport (43,44)], the nonvesicular transport occurs without involving membrane fusion. This process is governed by various lipid-binding proteins (LBPs) through membrane contact sites (MCSs). For example, ER membranes form MCSs with the membranes of mitochondria, endosomes, the PM, and the Golgi (45). MCSs also exist between other membrane pairs. Various tethering factors are involved in producing MCSs (46)(47)(48). Two classes of LBPs, the steroidogenic acute regulatory protein (StAR)-related lipid transfer domain protein family (STARDs), and the oxysterol binding protein-related protein (ORP) family, have been studied extensively. STARDs (which include 15 members in mammals) (49,50), including StAR as its first member (51), are involved in transferring cholesterol and/or other lipids (52)(53)(54). The ORP family, including its first member, oxysterol binding protein (OSBP) (55), has ten members in mammals (56). ORPs contain multiple domains that enable them to transfer different classes of lipids between two adjacent membranes in asymmetric manner (57)(58)(59)(60).

The three cholesterol trafficking routes in a mammalian cell
The transport and movement of cholesterol through various subcellular compartments can occur by at least three distinct metabolic pathways, resulting in three major cholesterol pools not in rapid equilibration with one another. This concept is outlined in Fig. 2, and is elaborated as follows.
First, exogenous cholesterol is supplied to the recipient cells mainly by cholesterol-rich lipoproteins, which enter the cell by endocytosis; the best example being the clathrin-mediated endocytosis of LDL via the LDLR (20). (In the CNS, ApoE, instead of LDL, is the major cholesterol carrier in the interstitial fluid and cerebrospinal fluid.) After endocytosis, LDL first enters a distinct early endocytic compartment enriched in the acid lipase that catalyzes the hydrolysis of the CEs (61). Cholesterol released from LDLderived CEs then moves to the endosomes that contain a pair of cholesterol binding proteins designated as Niemann-Pick type C (NPC)1 and NPC2. NPC2, a soluble protein present in the luminal region of the endo/lysosomes (62), binds to cholesterol and transfers it to NPC1 (63, 64), a protein that contains multiple transmembrane domains, including the "sterol-sensing domain" (65). The sterolsensing domain is also present in HMGR and in SCAP, as well as in several other membrane proteins involved in sterol dependent cell signaling processes. NPC1 then exports cholesterol to the exterior of the late endo/lysosome, in manners that are not well understood at present. Cholesterol exiting from the late endosomes (LEs) arrives at other membrane compartments, via various transport mechanisms. These recipient membrane compartments include the PM (66,67), the ER (68-70), the TGN (71-73), the mitochondria (described in the next section), and the peroxisomes (described in the next section). Mutations in NPC1 or in NPC2 can cause the NPC disease in humans and in certain other animal species. Children affected with NPC disease almost invariably die before reaching adulthood. 95% of NPC cases are caused by mutations in the Npc1 gene, with the rest caused by mutations of the Npc2 gene (5%). Currently, there is no cure for this disease. In NPC patients, unesterified cholesterol accumulate within the endo/lysosomes of various organs, including the liver, the spleen (74), and various regions of the brain [ (75)(76)(77); reviewed in (78)].
Second, in addition to exogenous uptake, cells also acquire cholesterol through endogenous biosynthesis de novo. The majority of sterols, including cholesterol, lanosterol, and multiple other precursor sterols newly synthesized at the ER, quickly move to the PM within a few minutes; this movement is not yet well-understood at the molecular level, but is independent of NPC1 (79,80). In yeast, several StAR-like proteins have been shown to participate in the sterol movement from the ER to the PM (50). Upon arriving at the PM, a significant amount of the newly synthesized sterols, especially lanosterol, is released to the cell exterior by the previously described sterol efflux process that depends on ABCA1 and helical apolipoproteins (34). Cholesterol, lanosterol, and other precursor sterols remaining at the PM recycle between the PM and various internal compartments, including the endo/lysosomes (81,82). Interestingly and unexpectedly, in the absence of sterol acceptors in the media, deficiency in ABCA1 delays sterol sensing and delays sterol esterification at the ER (83). This result implicates that the retrograde movement of cholesterol and other sterols from the PM to the ER partially depends on ABCA1. The molecular nature of the ABCA1-assisted retrograde sterol movement is not clear at present, but this process requires the intrinsic ATPase activity present in ABCA1 (83) and involves clathrin-independent endocytosis (83,84). ABCA1 may be located within the lipid raft domain, as shown in Fig. 2; however, this assignment is only tentative (34). In addition to ABCA1, two soluble LBPs, oxysterol binding protein-related protein (ORP)1 and ORP2, have been implicated in the cholesterol movement from the PM to the ER (58). In addition, the StAR-like lipid transfer domain protein 4 (STARD4) has also been implicated in moving cholesterol from the PM to various destinations including the ER (53,85).
Third, cholesterol that builds up at the ER gets esterified for storage by the enzyme, ACAT1 SOAT1 . Both cholesterol derived from lipoprotein and cholesterol synthesized endogenously serve as the substrate for ACAT1 SOAT1 . Cholesterol delivery to ACAT1 SOAT1 is partially dependent on NPC1/NPC2 (86,87) and partially dependent on the ABCA1 (83). CEs sequester as cytoplasmic lipid droplets. These lipid droplets are subject to hydrolysis by enzymes that are collectively designated as CE hydrolases (CEHs). The expressions of CEHs are tissue and cell specific (88). Neutral cholesteryl esterase and hormone-sensitive lipase play key roles in macrophages (89), while in rat hepatocytes, a member of the carboxylesterase family, called ES-4, accounts for the majority of CE hydrolysis (90). Under cholesterol-rich conditions, a cholesterol-CE cycle occurs continuously (91); the majority of cholesterol in this cycle originates from sterols synthesized endogenously, rather than derived from lipoproteins (92). When ACAT1 SOAT1 is inhibited, the cholesterol pool destined for storage rapidly reaches the PM to serve as a substrate for ABCA1-mediated lipid efflux (93). There are at least three different cholesterol pools at the PM (67); one of these pools has been recently shown to be highly enriched at the microvilli region of the PM (94). The relationship between the PM cholesterol pools and the three cholesterol trafficking routes described here requires further investigation.
Here we discuss cholesterol flux in mitochondria and in peroxisomes. Mitochondria produce ATP to meet the cellular demand for energy. In steroidogenic cells, cholesterol in the inner membranes of mitochondria serves as the precursor for biosynthesis of pregnenolone. In nonsteroidogenic cells, cholesterol in mitochondria serves as the precursor for biosynthesis of 27-hydroxycholesterol, which is the most abundant oxysterol in the blood. Cholesterol in mitochondria comes from at least three sources: 1) The PM. The molecular nature of the PM-mitochondria cholesterol movement process is not clear at present, but this process does not require NPC2 (95). At the mitochondria, the transfer of cholesterol from the outer membrane to the inner membrane is mainly mediated by the protein, StAR (96). 2) In cells that express low level of StAR, a StAR-like protein, MLN64 (also called STARD3), present in the LEs (97,98) works along with the NPC2 protein to transport cholesterol from the LE/lysosome to the mitochondria (99,100). This process explains the unusual finding that, in mutant NPC1 cells, cholesterol overloading occurs in mitochondria (101,102).
3) The third source of mitochondrial cholesterol comes from a specialized membrane region designated as the mitochondrial-associated membranes, which are part of the ER membranes in close physical contact with the mitochondrial membranes (103). These membranes are enriched in enzymes that biosynthesize phosphatidyl serine (104), and in the protein content of ACAT1 SOAT1 (105). Mitochondria associated membranes are also rich in cholesterol and in the simple sphingolipid ceramide (106). An outer mitochondrial membrane adaptor protein, hypoxia-upregulated mitochondrial movement regulator (HUMMR), increases the ER-mitochondria MCSs to facilitate cholesterol flux to mitochondria (107). Interestingly, studies in a mouse model for Alzheimer's disease (AD) showed that, in the early stage of AD development, synaptic mitochondria exhibit significant functional deficits; the degree of deficits correlates positively with amyloid- (A) peptide accumulation (108). In the future, it would be interesting to test to determine whether mitochondrial cholesterol overload might execrate A toxicity in synaptic mitochondria. Peroxisomes play important roles in the biosynthesis and catabolism of various lipids. Disrupting the contact sites between peroxisome and lysosome impairs the LDL-dependent decrease in SREBP processing and increase in cholesterol esterification; these findings implicate peroxisomal cholesterol transport as an important intermediate step in delivering LDL-derived cholesterol from the late endo/lysosome to the ER (109).

Lipoproteins in the blood versus lipoproteins in the CNS
Lipoproteins transport lipids and fat-soluble vitamins to various cells for utilization and for storage. In the plasma, lipoproteins include chylomicrons, VLDLs, LDLs, and HDLs. These lipoproteins differ in apolipoprotein/lipid composition, in size and density, and in function. Chylomicrons deliver dietary triglycerides to body cells; they contain ApoB48 as the major apolipoprotein. VLDLs deliver triglycerides synthesized in the liver to body cells, and contain ApoB100 as the major apolipoprotein. LDLs deliver cholesterol to body cells, and contain ApoB100 as the only apolipoprotein. HDLs play important roles in cellular efflux of cholesterol and phospholipids; they contain ApoA1 and ApoAII as the major apolipoproteins. Chylomicrons, VLDLs, and HDLs also contain ApoE and ApoCs as minor apolipoproteins. To address a key function of ApoE in lipoprotein metabolism in systemic tissues: chylomicrons circulating in the blood are rapidly catabolized by lipoprotein lipase and converted to remnant particles of much smaller sizes, designated as chylomicron remnants. ApoE that is present in chylomicron remnants serves as the key ligand, such that the remnant particles can be recognized by the receptor present at the cell surface, the chylomicron remnant receptor, also called LDL-related protein 1 (LRP1). LRP1 is highly expressed in several tissues, including hepatocytes, macrophages, and adipose tissue [reviewed in (110,111)]. Through LRP1, chylomicron remnants undergo endocytosis to deliver lipids for utilization and storage in various tissues. To address the function of ApoE in the CNS: Due to the BBB, the vast majority of plasma lipoproteins cannot readily enter the brain interior. In the brain, ApoE is the major apolipoprotein and is produced within the brain. The ApoE is present as lipid-rich ApoE particles in the interstitial fluid and cerebrospinal fluid, and exhibits density and size similar to those of plasma HDL. Similar to the ApoE present in chylomicron remnants, the lipid-rich ApoE also uses LRP1 as one of the main receptors for binding and endocytosis, and LRP1 is highly expressed in various brain cells within the CNS. Humans contain three slightly different alleles of the Apoe gene, e2, e3, and e4. The most common allele is e3, and it is found in more than half of the population. ApoE2 binds poorly to chylomicron remnant receptor/LRP1. Humans homozygous for ApoE2 are deficient in the clearance of chylomicron remnants, and tend to have hypercholesterolemia and premature atherosclerosis. Instead, the ApoE4 isoform confers significantly increased susceptibility to lateonset AD (LOAD) (see below). In addition to producing ApoE, brain cells produce a different apolipoprotein, ApoJ (also called clusterin); polymorphism in the Apoj gene also confers a risk factor for LOAD [reviewed in (112)]. Cells in the brain do not produce ApoB48, ApoB100, or ApoA1. However, the cerebrospinal fluid (that circulates between the subarachnoid space of the brain and the spinal cord) can acquire a significant amount of ApoA1 from the blood via unknown mechanism(s) [reviewed in (113)]. Under the conditions of certain chronic vascular diseases, such as hypertension, hypercholesterolemia, and diabetes, the BBB becomes partially leaky and may allow plasma lipoproteins (as well as other macromolecular components present in the blood) to become more readily accessible to the brain interior [reviewed in (113)]. Recently, Zlokovic and colleagues (113a) showed that unlike normal mice, the ApoE KO mouse exhibits a leaky BBB phenotype; expressing human ApoE2 or human ApoE3, but not human ApoE4, rescued the leaky BBB phenotype of the mouse. In addition, in a cultured BBB model, Michikawa and colleagues showed that ApoE plays an important role in controlling the integrity of tight junction in an isoform-specific manner (114). Together, these results support the hypothesis that the inability of ApoE4 to maintain the integrity of the BBB could be a causative factor that leads to various neurodegenerative diseases, including AD (115).

AD and the involvement of cholesterol and ApoE4
First, the hallmark of AD consists of extracellular amyloid plaques, mainly composed of A peptides (especially A1-42) and neurofibrillary tangles, mainly composed of hyper-phosphorylated tau, accompanied by chronic neuroinflammation. The current review focuses only on the link between cholesterol and A. A peptides are mainly produced in neurons, and are derived from the amyloid precursor protein (APP) by sequential proteolytic cleavages; with time, A oligomerizes to form the insoluble and aggregated amyloid plaques in neurons. Deposition of A peptides also occurs within the smooth muscle cell layer of the arterial blood vessels within the brain, producing a pathological condition designated as cerebral amyloid angiopathy [reviewed in (116,117)]. Defects in the clearance of A from the brain through cellular degradation mechanism(s) and/or by transport pathway(s) across the BBB underlie many cases of LOAD [reviewed in (115,118)]. Tau is a soluble microtubule binding protein produced abundantly in neurons. Various forms of tau (i.e., soluble, misfolded, mislocalized, and/or hyperphosphorylated forms) may eventually aggregate and form neurofibrillary tangles [reviewed in (119)]. A peptides and tau can interact with each other synergistically to cause neuronal dysfunctions and trigger the disease [reviewed in (120)]. Dysfunctional neurons and misfolded proteins can cause activations in microglia and in astrocytes that lead to neuroinflammation [reviewed in (35,121)].
Second, the brain is an organ very rich in cholesterol; it contains 23% of the body's total cholesterol, though it constitutes only 2.1% of total body weight (122). Due to the BBB, cholesterol within the brain does not readily equilibrate with cholesterol bound to lipoproteins in the blood. Thus, essentially all the cholesterol in the brain is derived from biosynthesis within the brain. Both astrocytes and neurons have a high demand for cholesterol, and both cell types can synthesize cholesterol, as discussed in detail by Pfrieger and Ungerer (123). Studies in mice show that neurons need to synthesize their own sterols during the prenatal stage (124), but may be able to acquire enough sterols from other sources during adult life (125). In the CNS, cholesterol transport between different cell types occurs, and the lipid-rich ApoE (ApoE/PL/C) plays the major role in transporting cholesterol (21). ApoE is a 36,000 Da protein present in the plasma and in the brain. As mentioned earlier (in the section, Lipoproteins in the blood versus lipoproteins in the CNS), in humans, ApoE is polymorphic, with three major alleles, ApoE2 (cys112, cys158), ApoE3 (cys112, arg158), and ApoE4 (arg112, arg158). For LOAD, besides aging, the ApoE4 allele is the most important risk factor (126,127). Conversely, the ApoE2 variant appears to have a protective role in LOAD. The human ApoE-targeted knock-in mice have been used extensively to investigate the role of ApoE isoforms in the onset and progression of LOAD [reviewed in (112) and (159)]. ApoE affects A aggregation and clearance in the brain in an isoform-specific manner (128,129). ApoE in the CNS is mainly produced by astrocytes (130). Under brain injury conditions, neurons also produce ApoE (131). Once synthesized, ApoE is secreted as a complex of ApoE/PL/C. The formation of the ApoE/PL/C depends on ABCA1 (132)(133)(134). Once secreted, a portion of cholesterol in ApoE/PL/C is esterified by the enzyme, LCAT, in an ABCA1-and ApoE-dependent manner (135). ApoE/PL/C delivers lipids from the astrocytes to the neurons and other CNS cells through internalization after binding to various ApoE receptors, including LRP1, LDLR, ApoE receptor 2, and VLDL receptor [reviewed in (111,136)]. ApoE consists of a receptor-binding region and a lipid-binding region; the different ApoE isoforms have different lipidation states, which influence their binding properties to the cognate receptors (129,137). ApoE4 is less lipidated than ApoE2 or ApoE3, and undergoes more rapid degradation within the CNS (138). In neurons, once internalized, ApoE is thought to participate in regulation of lipid metabolism, including phospholipid metabolism (139), intracellular cholesterol transport, and lipid efflux (140), and cholesterol esterification (141), in an ApoE isoform-specific manner. In neurons, cholesterol can be esterified by ACAT-1 SOAT1 (142); it can also be converted to the major oxysterol in the brain, 24S-OH (143). The lipidated ApoE also participates in the efflux of phospholipids and cholesterol in an isoform-specific manner; this process also depends on ABCA1 (144,145). A related protein, ABCG1, works in concert with ABCA1 to control the effluxes of cholesterol and 24S-OH (146,147). At present, the mechanisms that govern the fate of ApoE-derived cholesterol in neurons and in astrocytes are not well-understood. To stimulate further investigation, here we provide a working model that depicts ApoE-mediated cholesterol homeostasis in astrocytes and neurons (Fig. 3). This model is based largely on the results described in Refs. 134, 135, 137-145. In this model, we also hypothesize that 24S-OH, secreted by neurons in an ApoEand ABCA1-dependent manner (146,147), may reach astrocytes to downregulate sterol biosynthesis and to activate gene expressions of ApoE, ABCA1, and other key proteins involved in lipid homeostasis in astrocytes. The validity of the model depicted in Fig. 3, especially at the in vivo level, requires further investigation. This topic is also reviewed in detail by Rebeck (112).
Third, in addition to ABCA1 and ABCG1, a separate ABC family member, called ABCA7, has been identified as one of the AD susceptibility genes (148). Deleting Abca7 increases A accumulation in a mouse model for AD (149). Although the precise function of ABCA7 is not well-understood, it is probably involved in lipid transport. A role of ABCA7 related to A biogenesis in the brain has been reported (150).
Fourth, in addition to affecting cholesterol homeostasis in astrocytes and neurons, ApoE also affects the functions of various receptors present in the membranes of postsynaptic neurons in an isoform-specific manner. For an example, the N-methyl-D-aspartate receptor (NMDAR) is a glutamate receptor and an ion channel protein. This receptor is important for controlling synaptic plasticity and memory formation [reviewed in (151)]. Studies in cell culture show that ApoE4 causes malfunction of the NMDARmediated signaling in the hippocampus and in the cortex (152)(153)(154) [reviewed in (136)]. NMDAR is concentrated at the PMs of postsynaptic neurons; its function is very sensitive to cholesterol and to other lipids (155). Many other membrane receptors, channels, and transporters are also significantly affected by cholesterol content in the neuronal membrane [reviewed in (42)]. In the future, it would be interesting to test whether ApoE affects the activities of membrane receptors/channels/transporters and different signaling pathways through its ability to control lipid homeostasis.
Fifth, studies in vivo have shown that human ApoE4 causes age-dependent learning and memory impairment in mice without amyloidopathy (156,157) [reviewed in (112)]. It also exacerbates neuroinflammation [reviewed in (112,158,159)].

The A/lipid rafts connection
In AD, oligomeric forms of A peptides are probably the most toxic molecular species that cause synaptic loss (160,161). The A monomer/oligomer conversion is affected by ApoE in an isoform-specific manner; A peptides can enter the cells via multiple mechanisms; some of them depend on ApoE and the ApoE receptors, especially the LDLR and LRP1. The interactions between A peptides and ApoE/PL/C are complex [reviewed in (117,129)]. Inside the neurons and other cell types, the oligomeric A peptides are reported to cause numerous functional disturbances, such as: alterations in mitochondrial morphology and oxidative stress (162)(163)(164); alterations in Golgi morphology, causing fragmentation and malfunctions (165,166); alterations in mitochondria-associated membranes (ER/mitochondria contact sites) (167,168); alterations in cellular cholesterol metabolism (169); and alterations in synaptic organelle transport, including the transport of recycling endosomes, mitochondria, etc. (170). A peptides are highly concentrated at the lipid raft region (171). Specifically, A peptides preferentially bind to the monosialotetrahexosylganglioside (GM1) (a prototype of gangliosides) in membranes (172); the interaction between A peptides and GM1 is much enhanced when cholesterol is present (173,174). In addition, A peptides may also bind to cholesterol directly (175). Thus, it has been proposed by several investigators that the toxicity of A may be produced, in part, by disturbing the lipid raft domains present in various membrane organelles. While A can affect the structure and function of raft domains, the converse is also true: the enzymes involved in the APP processing pathway [i.e., the , , and  secretases (176)(177)(178)(179), as well as APP itself (180)] have all been reported to be affected by the composition of the lipid raft domains in which they reside [Reviewed in (191)]. The fate of APP itself is a complicated matter; in addition to its cleavages by the secretases, a significant fraction of APP and its C-terminal fragments can also be degraded by lysosomal hydrolases (181,182). To summarize, the A/lipid raft connections are complex and need further investigation.

LIPID-BASED POTENTIAL TARGETS TO TREAT AD
Here we discuss targets and candidate drugs being tested at the clinical stage first, and then discuss those being considered at the preclinical stage.

Statins and HMGR
The statin drugs have been successful in treating patients with dyslipidemic cardiovascular diseases (183). They work by inhibiting the key enzyme in endogenous sterol biosynthesis, HMGR (184,185). In general, statin drugs inhibit HMGR with an inhibitor constant K i at less than 1 nM (185). When used at super-high concentrations (1,000 times higher than the K i value), statins have been shown to exert offtarget effects that are independent of HMGR inhibition; for example, at 2-10 M, statins can act as PPAR ligands (186). Here we focus on the effects of statins on inhibiting HMGR. Studies in vitro and in animal models showed that statins strongly reduce the levels of the A peptides, A42 and A40 (187,188). Thus, several clinical trials have been conducted to determine whether the statin drugs might benefit AD patients. Unfortunately, the results have been inconclusive (189). To speculate about the actions of statins on AD patients: HMGR produces mevalonate as its enzymatic reaction product. Mevalonate is an essential precursor for the biosyntheses of sterols as well as several nonsterol metabolites, including coenzyme Q, dolichol, etc. These metabolites are essential for cell growth and maintenance. Mevalonate is also needed for the enzymatic modification (by prenylation) of numerous proteins; these proteins depend on prenylation for their biological functions (190). Statins used at low dosage may reduce cholesterol content present in the lipid raft, thereby decreasing the interaction between the lipid raft and A, and diminishing the toxicity produced by A; on the other hand, statins may affect various channels/receptors that depend on optimal membrane cholesterol content to maintain their functionalities, especially those channels/receptors involved in the learning and memory process. Statins used at high dose may significantly decrease the levels of mevalonate-derived nonsterol metabolites, in addition to decreasing the cholesterol content present in the lipid raft; the long-term consequence of treating AD patients with high concentrations of statins cannot be predicted at present.

Nuclear receptor agonists
The nuclear receptor agonists are promising agents to treat AD. These agonists activate several ligand-dependent transcription factors, including LXRs, RXRs, and PPARs, to induce the gene expressions of ApoE, ABCA1, and ABCG1. This topic is reviewed in detail by Moutinho and Landreth (35).

Omega-3 fatty acids and other nutritional supplements
Omega-3 fatty acids are essential fatty acids, and include DHA and EPA. They are richly present in deep-sea fish oil, but absent in plant oil. They are being tested as an AD prevention strategy in humans. This topic is reviewed in detail in (191).

Cyclodextrin
Cyclodextrins (CDs) are water-soluble oligomers of glucose. They do not elicit immune responses and have low toxicities in animals and humans. CDs can form water-soluble inclusion complexes with hydrophobic small molecules, such as cholesterol and other lipid molecules (192). Hydroxypropyl--CD (HP--CD) is a Food and Drug Administration-approved drug delivery vehicle for various pharmaceutical purposes. In NPC disease, treating animal models for NPC disease with HP--CD overcame the cholesterol transport defect caused by mutations in NPC in various organs and produced significant improvement in delaying the disease onset, in ameliorating the disease progression, and in prolonging the life span (193)(194)(195)(196) [reviewed in (78)]. Currently, HP--CD is under clinical trial to treat children affected with NPC disease (78,197). CD enters the cell interior rapidly and acts by mobilizing cholesterol within the endo/lysosomal compartment, thus facilitating the cholesterol transfer from the endo/lysosomes to other cellular compartments, including the ER and PM (198,199). Based on the results of using CD to treat NPC disease, CD has been tested in a mouse model for AD and the results were interesting: beginning at postnatal day 7, continuous intravenous injections of CD for 4 months led to reduced A accumulation, diminished tau immunoreactive dystrophic neuritis, and rescued cognitive deficits; HP--CD may act by increasing the APP processing and by increasing the gene expression of ABCA1 (200). One needs to be aware that when CD is used to treat animal models for NPC disease, it is most effective when administered at a very young age. In addition, using CD at high concentrations causes hearing loss in cats (201) and in mice (202).

MiR-33
MicroRNAs (miRNAs)are short noncoding regulatory RNAs. They specifically bind to various target mRNAs, repressing the expression of the corresponding target genes through translational repression and/or through mRNA decay. In animal studies, a specific miRNA, miR-33, present within intron 2 of the gene that encodes SREBP2, downregulates the gene expression of ABCA1 and causes a decrease of the HDL level in vivo (203). Interestingly, in mouse brain, overexpression of miR-33 suppresses ABCA1 expression and causes impaired cellular lipid efflux, as well as increased extracellular A accumulation; conversely, inhibition of miR-33 induces ABCA1 and causes increased lipidation of ApoE and reduced A levels (204). Furthermore, pharmacological inhibition of miR-33 by using an antisense oligonucleotide specifically in the brain markedly decreased A levels in the cortex of a mouse model for AD (204). In the brain, ABCA1 is redundantly targeted by miR-106b, miR-758, and miR-33 (36). These results suggest that inhibition of these miRNAs may provide a novel therapeutic strategy to treat AD.

ApoA1 mimetic peptides
In cell culture and in systemic tissues, the synthetic ApoA1 mimetic peptides are known to mimic the effects of ApoA1 and ApoE in stimulating ABCA1-dependent cellular cholesterol efflux, as reviewed in (205). Michaelson and colleagues injected one of these mimetic peptides, CS-6253, directly into the brains of young ApoE4 knock-in (KI) mice, and showed that this peptide increased the lipidation of the ApoE4-associated lipoproteins (206). CS-6253 also reversed much of the ApoE4-associated pathology, including A accumulation and tau hyperphosphorylation in hippocampal neurons, as well as synaptic impairments and cognitive deficits. These results show that increasing the lipidation of ApoE4-associated lipoproteins is a promising strategy to combat AD.

Cyp46a1 and Cyp46a1 activators
The enzyme, Cyp46A1, also called 24S-hydroxylase (207), converts cholesterol to 24S-OH, which is the most abundant oxysterol in the brain (208). A study in mice showed that Cyp46a1 gene KO reduced cholesterol excretion from the brain by more than 50% and reduced the cholesterol biosynthesis rate in the brain by 40%, without altering the overall brain cholesterol content. These results show that Cyp46A1 is responsible for the turnover of at least 40% of brain cholesterol; in the absence of Cyp46a1, synthesis of cholesterol de novo is reduced in order to maintain cholesterol homeostasis (143). The Cyp46a1 KO mouse was used to test the effect of reducing 24S-OH in AD pathology. The result showed that Cyp46a1 gene KO did not affect the amount of the insoluble amyloid plaques (209). The AD mice employed in this study had a shorter life span (for unknown reasons); Cyp46a1 gene KO prolonged the life span of these AD mice, but did not affect the life span of the non-AD mice (209). Because multiple sterols and oxysterols have been shown to exert similar and/or redundant regulatory functions, the results of the single gene [Cyp46a1] KO study cannot predict the outcome of increasing 24S-hydroxycholesterol in the AD mouse brain. To test the effect of increasing 24S-OH in the AD mouse brain, the Cyp46a1 overexpression experiments were carried out next. The results showed that in two different mouse models for AD, overexpressing Cyp46a1 ameliorated amyloid pathology (210) or ameliorated tauopathy (211). In addition, overexpressing Cyp46a1 in a mouse model for Huntington's disease decreased neuronal atrophy and improved motor neuron deficits (212). These results showed that overexpressing Cyp46a1 in brains with certain neurological diseases is neuroprotective and suggest that specific Cyp46A1 activators may provide a novel therapeutic strategy to treat AD and other related neurodegenerative diseases. The mechanistic basis for the neuroprotective effect(s) of increasing 24S-OH remains to be clarified.

ACAT1 SOAT1 inhibitors
In the disease, atherosclerosis, CEs produced by ACAT-SOAT accumulate in macrophages and smooth muscle cells, causing them to become foamy. For this and other reasons, ACAT SOAT inhibitors were produced for anti-atherosclerotic purposes. Several ACAT SOAT inhibitors, including CI1011 [an inhibitor that inhibits ACAT1 SOAT1 and ACAT-2 SOAT2 at equal potency (213)] and K604 [an ACAT1 SOAT1specific inhibitor (214)], advanced to stage 2 or stage 3 clinical trials. Due to a lack of efficacy, none of them became a medicine. Regarding the CE levels in mouse and human brains, in normal states, the values are very low, making up less than 1% of the free unesterified cholesterol. However, in the vulnerable (entorhinal cortex) regions of brain samples from AD patients, CE levels increase by 1.8-fold (215). In the brains of three different AD mouse models (that express mutant human APP or mutant APP and mutant presenilin 1), the CE levels rose to values 3-to 11-fold higher than those in the control mice (215,216). In addition, under high-fat diet, the brain CE content in ApoE4 KI mice is significantly higher than that in ApoE3 KI mice (217). Together, these results suggest that increases in CE content correlate positively with AD development. In mouse models for AD, both the pharmacological approach (218,219) and the molecular genetic approach (142,220) showed that inhibiting ACAT1 SOAT1 significantly reduced amyloid plague load and restored cognitive deficits. The mechanism(s) underlying the beneficial effects seen with blocking ACAT1 SOAT1 , as summarized in Fig. 4, include: In cell culture, blocking ACAT1 SOAT1 increases autophagy-mediated lysosomal biogenesis, the capacity to degrade oligomeric A in microglia (221), and the capacity to degrade the soluble form of mutant tau in neurons (222). In vivo, blocking ACAT1 SOAT1 increases the content of the major oxysterol 24S-OH and decreases the content of the full-length human APP in the brain of AD mice (142). In addition, when the isoform-non-specific ACAT SOAT inhibitors, CP-113,818 (218) or CI-1011 (223), were employed, the results showed that APP maturation was inhibited, resulting in reduced A production. Together, these results suggest that the ACAT SOAT inhibitors can provide benefits to AD through multiple mechanisms. A detailed review on this topic is available (224). It is known that, similar to the brains of AD patients, increases in amyloid plaques have been observed in aging normal brains, though the accumulations are much less than those in AD patients (116). In addition, dysfunctions in microglia have been linked with AD (225) and with aging (226). Thus, it is tempting to speculate that inhibiting ACAT1 SOAT1 in myeloid cells may benefit aging, in addition to benefiting AD. Would inhibiting ACAT1 SOAT1 affect the progression of other human diseases? Recent results in mouse models show that ACAT1 SOAT1 is also a potential target for treating various forms of cancer (227)(228)(229). In addition, in a mouse model for atherosclerosis, a recent study showed that, in contrast to the result of knocking out ACAT1 SOAT1 in the whole body, knocking out ACAT1 SOAT1 in the myeloid cell lineage (including monocytes/macrophages, neutrophils, and eosinophils) actually reduces atherosclerotic lesions (230). Thus, ACAT1 SOAT1 may be a potential target to treat multiple diseases. Regarding the toxicity issue, feeding certain very hydrophobic and highly potent ACAT SOAT inhibitors, such as CP-113,818 or ATR101 (231), to guinea pigs or to dogs caused ER-stress that led to cell apoptosis in the adrenals; the toxicity was restricted to the adrenals and did not occur in macrophages or in other tissues examined. The ACAT SOAT inhibitors that passed clinical phase 2, including CI 1011 and K604 described earlier, are less hydrophobic than ATR101; these inhibitors do not cause adrenal toxicity. Thus, the adrenal toxicity seems to be caused by using extremely hydrophobic ACAT SOAT inhibitors at high doses.

FUTURE PERSPECTIVES
LOAD is a disease with complex etiology. We consider AD as a special lipid disease. In this review, we chose topics that we are familiar with for in depth discussions. Regretfully, a number of important research topics were left with little or with no discussion. Here, we recommend nine research areas that relate cholesterol metabolism with AD for further investigations: 1) The roles of cholesterol and other lipids in affecting the integrity and function of the BBB.

2)
The in vivo significance of the ApoE-mediated cholesterol homeostasis in the CNS, in a cell type-specific manner (depicted in Fig. 3 as a working model).
3) The specific functions of ApoJ, ABCA7, ORPs, or STARDs in affecting cellular cholesterol homeostasis in the CNS, in a cell typespecific manner. 4) The in vivo significance of APP/Amediated disturbance in cellular cholesterol homeostasis and in membrane biology. 5) The possible link between cholesterol overload, lipid raft domain, and oligomeric A accumulation at the synapse mitochondria. 6) The roles of various oxysterols in controlling brain cholesterol metabolism in vivo. 7) The roles of neurosteroids in affecting brain cholesterol metabolism and in affecting cognition and behavior. 8) The in vivo significance of the nonvesicular and vesicular cholesterol movements in affecting the lipid raft domains in various cells of the CNS. 9) The possible link between tau and cellular cholesterol homeostasis in the CNS.
AD is a disease that affects the CNS. To be effective as a primary therapy to treat AD, the candidate drug needs to enter the brain interior. Therefore, we also recommend the development of methods for facile CNS drug delivery as a top priority research endeavor. Fig. 4. Beneficial effects of ACAT1 SOAT1 blockage on amyloidopathy and on tauopathy. In neurons, ACAT1 SOAT1 blockage results in reduced full-length human APP content that leads to less A production; ACAT1 SOAT1 blockage also results in reduced protein content of the non-hyper-phosphorylated tau. In microglia, ACAT1 SOAT1 blockage increases A clearance. See text and (224) for details.