Amyloid beta-protein and the genetics of Alzheimer's disease.

The effort to decipher the mechanism of Alzheimer’s disease (AD) has attracted the interest of investigators from diverse biological disciplines, including biochemistry, cell biology, molecular genetics, neuroscience, and structural biology. The eclectic nature of research approaches to AD and the intensity of scientific interest in the problem have made it increasingly likely that AD will become a premier example of the successful application of biological chemistry to the identification of rational therapeutic targets in a major human disease. Much of the recent progress in elucidating the pathogenesis of AD has centered on the apparent role of the 40–42-residue amyloid b-protein (Ab) (1) as a unifying pathological feature of the genetically diverse forms of this complex disorder.

uble neuronal protein into an insoluble filamentous polymer seems to involve a disregulation of cytoplasmic phosphorylation/dephosphorylation cascades, but the factors that trigger this imbalance are poorly understood. Importantly, neurofibrillary tangles composed of paired helical filaments containing hyperphosphorylated tau molecules are found in a variety of etiologically diverse neurological diseases besides AD, strongly suggesting that this cytoskeletal alteration can develop as a secondary (albeit important) response to a variety of cerebral insults. In accord with this view, the gene on human chromosome 17 that encodes the tau polypeptides is not known to be the site of disease-causing mutations in familial forms of AD.
In contrast, studies of A␤, which is the subunit of the amyloid fibrils found in neuritic plaques and in some meningeal and cerebral blood vessels, resulted in the identification of the first specific genetic cause of AD. The purification of both plaque and vascular amyloid deposits and the isolation of their ϳ40-residue constituent peptide (A␤) led to the cloning of the type 1 integral membrane glycoprotein from which A␤ is proteolytically derived, namely the ␤-amyloid precursor protein (␤APP) (5). The localization of the ␤APP gene to chromosome 21q appeared to explain the observation that patients with trisomy 21 (Down's syndrome) incur ␤-amyloid deposits in late childhood or young adulthood and subsequently develop the classical neuropathological features of AD in their forties (6 -9). This realization led in turn to a specific search for families with autosomal dominant AD who had genetic linkage to chromosome 21, resulting ultimately in the identification of six different missense mutations in ␤APP, five associated with familial AD (10 -14) and one with the neuropathologically related syndrome of hereditary cerebral hemorrhage with amyloidosis of the Dutch type (15). Although ␤APP mutations, all of which are clustered in the A␤ region of the precursor, have proven to be a very rare cause of familial AD, representing less than 1% of such cases, they are important for understanding the mechanism of the progressive cerebral A␤ deposition that occurs in all cases of AD.
Most of the ␤APP missense mutations have now been examined in transfected human cell lines, and their phenotypic effects have also been studied directly in primary skin fibroblasts and plasma obtained from subjects harboring these mutations. In each example studied to date, the missense mutations have been shown to alter the proteolytic processing of the precursor (see below) in a way that results in increased production of A␤ peptides, particularly of the highly hydrophobic (and thus amyloidogenic) 42-residue form of A␤ (A␤ 42 ) (16 -20). For example, one of the disease-causing missense mutations occurs immediately N-terminal to the start of the A␤ region, i.e. at the P1Ј position for the protease (called "␤-secretase"), which generates the N terminus of A␤ (Fig. 1). Other mutations occur 4 residues C-terminal to the end of the A␤ region and lead to increased cleavage of ␤APP by the protease (called "␥-secretase") that generates the C terminus of the A␤ 42 peptide (Fig. 1). The elucidation of the effects of the ␤APP mutations on cellular A␤ production was made possible by the discovery in 1992 that A␤ is proteolytically generated from ␤APP under normal metabolic conditions and is constitutively secreted by essentially all neural and non-neural cells that express the precursor (21)(22)(23). The heightened cerebral deposition of A␤ that results from the effects of the ␤APP missense mutations can be viewed as loosely analogous to the excessive tissue deposition of another normal metabolite, cholesterol, in subjects with various forms of familial hypercholesterolemia. It remains unclear as to why A␤, although secreted by many cell types throughout the body, is overwhelmingly deposited in the brain in AD, with only miniscule amounts of non-fibrillar A␤ deposits observed in some peripheral tissues (24 -26).
predispose to the premature development of the disease. The second gene to be specifically implicated in familial AD was that encoding the cholesterol transport protein, apolipoprotein E (apoE). Using synthetic A␤ as an immobilized ligand, Strittmatter et al. (27) searched for proteins in human cerebrospinal fluid that were capable of binding to A␤ and isolated apoE. The fact that the gene encoding apoE was on chromosome 19q in a region that had previously been shown by these investigators to be linked to lateonset AD led to the discovery that the inheritance of one or two ⑀4 alleles of the apoE gene confers both a substantially higher likelihood and an earlier age of onset of the late-onset form of AD (28), whereas inheritance of the ⑀2 allele appears to do the opposite (29). Therefore, the normally occurring polymorphism of the apoE protein represents a major genetic risk factor for the development of AD. Because many E4 carriers do not develop AD and, conversely, many AD patients do not carry the E4 allele, apoE4 is a risk factor, not a causative mutation.
The mechanism by which the apoE4 protein predisposes subjects to AD remains unsettled. The most consistent phenotypic clue to the action of apoE4 is the finding that AD patients harboring one or two ⑀4 alleles have a significantly higher density of both plaque and vascular A␤ deposits in their brains than do AD subjects lacking this allele (30 -34). This well confirmed observation has led to several in vitro studies in which purified apoE4 (generally in a non-lipidated state and at supraphysiological concentrations) has produced enhanced aggregation of synthetic A␤ peptides under cell-free conditions (e.g. Refs. 35 and 36). Although these studies suggest a role of the apoE4 protein as a facilitator of A␤ aggregation, an alternate explanation for the phenotypic effect could be a decreased efficiency of clearance of the peptide from the extracellular space in the presence of apoE4. Regarding A␤ production, evidence obtained in stably transfected cells co-secreting A␤ and each of the apoE isoforms at physiological levels suggests that apoE4 does not lead to increased generation of A␤ (37), in contrast to the mechanism of the ␤APP missense mutations.

Missense Mutations in the presenilin 1 and presenilin 2 Genes Cause Many Cases of Early-onset Familial AD
The discoveries that the ␤APP and apoE genes were implicated in the early-onset and late-onset forms of familial AD, respectively, left open the question of which other genes could explain the large majority of early-onset AD cases. Linkage analysis followed by positional cloning led to the identification in 1995 of two highly homologous familial AD genes, currently termed presenilin 1 and presenilin 2 (40 -42). These genes encode 467-and 448-residue polypeptides whose sequences and hydropathy profiles suggest that they contain 7-9 transmembrane domains. The normal functions of these widely expressed proteins are not yet known. Accordingly, the mechanism by which missense mutations, of which 27 have already been reported (25 in presenilin 1 and 2 in presenilin 2), cause early-onset familial AD remains to be elucidated by studies in transfected cell lines and transgenic animals. However, a major clue to their pathogenic effect has come from recent analyses of A␤ levels in plasma and the conditioned media of skin fibroblasts obtained from carriers of mutant presenilin genes. Sensitive enzyme-linked immunosorbent assays show that A␤ peptides ending at residue 42 (A␤ 42 ) are selectively augmented by at least some missense mutations in presenilin 1 and 2, whereas levels of the major form of A␤ (A␤ 40 ) are largely unchanged (43). A␤ 42 peptides have been found to have a markedly enhanced rate of aggregation into amyloid fibrils in vitro, compared to that of A␤ 40 peptides (44). Moreover, immunocytochemical studies using antibodies specific to the two A␤ C termini reveal that A␤ 42 peptides are the initially deposited species in the earliest ("diffuse") plaques in both AD and Down's syndrome brains (8,9,45,46).
In summary, four genes have been implicated to date in familial forms of AD: three that, when mutant, cause autosomal dominant forms of the disease (␤APP, presenilin 1, and presenilin 2) and one in which a naturally occurring polymorphism (apoE4) represents a major genetic risk factor for the development of the disease (Table  I). Available evidence strongly suggests that each of these four genes predisposes to the AD phenotype by enhancing the production and/or the deposition of A␤ peptides (or in the case of apoE4, perhaps by decreasing its clearance from tissue).

The Expression and Post-translational Processing of ␤APP
Growing evidence implicating extracellular A␤ 42 accumulation and resultant amyloid plaque formation as early and necessary steps in the pathogenesis of the known forms of heritable AD has heightened interest in understanding the details of the trafficking and proteolytic processing of ␤APP (reviewed in Ref. 47). This protein occurs in numerous different isoforms, which arise from alternative splicing of a single gene. The shortest of the major isoforms (695 amino acids) is expressed almost exclusively in neurons, whereas the other two common forms (751 and 770 amino acids, respectively) are expressed both in neural and non-neural cells. Additional heterogeneity of the ␤APP polypeptides arises from their complex post-translational modifications, including sulfation, phosphorylation, and both N-and O-linked glycosylation (e.g. Refs. 48 -50). These modifications occur during the trafficking of the protein through the secretory pathway. ␤APP is co-translationally translocated into the endoplasmic reticulum via its signal peptide and then matured during passage through the Golgi by acquiring sulfate, phosphate, and sugar groups, following which a minor percentage of mature molecules is transported to the plasma membrane via secretory vesicles (47). At the cell surface, some ␤APP molecules undergo proteolysis by an unidentified protease designated "␣-secretase," which cleaves between lysine 687 and leucine 688, i.e. between residues 16 and 17 of the A␤ region, releasing the large, soluble ectodomain (referred to as ␣-APP s ) into the medium (51) (Fig. 1). Alternatively, uncleaved surface ␤APP   (52), following which the full-length precursor is trafficked to late endosomes and lysosomes for apparent degradation (53,54) or is rapidly recycled within early endosomes to the cell surface (55). The latter pathway has been shown to be a principal site for the two proteolytic cleavages that generate the A␤ peptide (56). A protease termed "␤-secretase" initiates A␤ generation by cleaving ␤APP after methionine 671, creating a 99-residue (ϳ12 kDa) membrane-retained C-terminal fragment having residue 1 (aspartate) of A␤ as its N terminus (57), and this results in the secretion of a truncated APP s molecule, called ␤-APP s , into the medium (58) (Fig. 1). The 12-kDa fragment may then undergo mechanistically enigmatic "␥-secretase" cleavages within the hydrophobic transmembrane domain at either valine 711 or isoleucine 713 that release the 40-or 42-residue A␤ peptides, respectively, into the medium. It appears that only a minority of all biosynthesized ␤APP molecules undergoes either the ␣-secretory or the ␤-secretory fate; many full-length precursor molecules remain inserted into internal membranes, particularly in the Golgi. The diverse metabolic fates of ␤APP just summarized are under complex regulation. For example, several first messengers, including cholinergic agonists and other neurotransmitters that can activate the phospholipase C/protein kinase C-dependent pathway, can enhance ␣-secretase cleavage of ␤APP and the consequent release of ␣-APP s into the extracellular fluid (e.g. Refs. 59 and 60). The mechanism by which this enhancement occurs is unclear. It does not involve a direct change in the phosphorylation state of ␤APP (50) and may involve instead the phosphorylation of ␣-secretase or an enhancement of the trafficking of Golgi-derived vesicles containing ␤APP to the cell surface (61), where ␣-secretase is known to be active (62). In addition to the regulated processing of ␤APP through the ␣-secretory pathway, the amyloidogenic processing of ␤APP (i.e. ␤followed by ␥-secretase cleavages) can be enhanced, for example by increases in intracellular free calcium levels (63).

␤APP Function and Dysfunction
Although the foregoing summary demonstrates that there has been considerable progress in delineating factors that regulate the processing of ␤APP, the functional implications of these varied effects remain unclear. The physiological consequences of the enhanced secretion of either ␣-APP s or A␤ have not been defined. In particular, a specific receptor for ␣-APP s is not known, although this major secreted derivative has been shown to stimulate a mitogen-activated protein kinase-mediated p21 ras -dependent signal transduction cascade in cultured cells (65). It has been surmised from the amino acid sequence of exon 7 of ␤APP and from studies of its function in cultured cells that this region acts as a serine protease inhibitor of the Kunitz family. ␣-APP s molecules derived from ␤APP 751 and ␤APP 770 (which contain this alternatively spliced exon) have been shown to inhibit several different serine proteases in vitro (66) and to occur in human plasma as an inhibitor of factor XIa of the coagulation cascade (67,68). Other functions of ␣-APP s and/or cell surface holo-␤APP that have been postulated on the basis of cell culture experiments include as a trophic (69) or neuroprotective (70) molecule and as a mediator of cell-cell (71) and cell-substrate (72,73) interactions.
With regard to ␤APP function in vivo, a mouse in which both ␤APP alleles have been deleted appears to develop and reproduce normally and has shown minimal phenotypic alterations to date (74). One explanation for the largely innocuous phenotype resulting from knocking out ␤APP may be the fact that it is one member of a conserved gene family that includes the substantially homologous amyloid precursor-like protein-1 and -2 in mammals (75,76) as well as related homologues in invertebrates (77,78).
␤APP will continue to be the subject of intensive study not only because of its role in AD pathogenesis but also because it is an example of an integral membrane polypeptide that is expressed both intracellularly and at the cell surface and can be converted at both of these sites to several different secreted derivatives. The functional consequences of this highly complex proteolytic processing are important to elucidate. However, there is currently no compelling evidence that ␤APP is not functioning in patients with Alzheimer's disease, even in individuals carrying ␤APP missense mutations. Rather, a large body of data strongly suggests that ␤APP plays a role in both familial and "sporadic" AD via its A␤ fragment. Increased production (or decreased clearance) of A␤, particularly of the A␤ 42 form, appears to lead gradually to its aggregation in the extracellular space of the brain and its microvessels. That this process can occur very early in the development of AD, probably decades prior to the onset of clinical symptoms, is supported by the finding that patients with Down's syndrome show many A␤ 42 -containing diffuse plaques as early as age 12 years (9). Such individuals do not develop "mature" A␤ plaques containing dystrophic neurites, activated microglia, and reactive astrocytes until some two decades later, and the appearance of this surrounding cytopathology is generally associated with the accrual in the plaques of the more abundantly produced A␤ 40 peptide (9). The morphological evidence of an association of aggregated, fibrillar A␤ with local cytotoxicity that has long been known from light and electron microscopic studies of AD brains is consistent with many in vitro studies during the last few years that indicate that aggregated but not monomeric A␤ peptides can reproducibly exert toxicity on neurons (e.g. Refs. 79 and 80), astrocytes (81), microglia (82), and endothelial cells (83) in culture. However, such in vitro experiments necessarily employ short-term exposure of cells to supraphysiological doses of synthetic A␤ 1-40 , whereas ␤-amyloid deposition in vivo evolves very slowly from peptides with heterogeneous N and C termini that are secreted at picomolar to low nanomolar concentrations by neurons, astrocytes, microglia, and endothelial and smooth muscle cells in the brain.
A more compelling model of the consequences of excessive A␤ accumulation has come from the production of a transgenic mouse that markedly overexpresses an AD-linked mutant form of ␤APP in selected neurons (84). These animals secrete large amounts of both A␤ 40 and A␤ 42 peptides into the extracellular fluid of brain and begin to accumulate diffuse and compacted (fibrillar) A␤ deposits after about age 5-6 months. Many of these amyloid plaques are intimately associated with dystrophic neurites, activated microglia, and reactive astrocytes. Neurofibrillary tangles have not been described to date, but some cerebral neurons develop immunoreactivity for phosphorylated tau and neurofilament proteins, suggesting that they are undergoing AD-like cytoskeletal alterations (85). The substantial similarity of the neuropathology in this model to that of AD strongly supports the hypothesis that excessive accumulation of first soluble and then aggregated A␤ 42 and A␤ 40 peptides can initiate Alzheimer-type neuronal and glial changes. This and other transgenic mouse models should allow a much finer dissection of the temporal course of the macromolecular, cellular, and behavioral changes that occur in AD. Most importantly, these animals can be used for the preclinical evaluation of compounds that interfere with one or another step in the pathogenic mechanism.

Conclusion
The concept that alterations in several distinct genes (four of which have been identified to date) can lead by different mechanisms to a chronic imbalance between A␤ production and clearance that results in aggregation of first the 42-residue and then the 40-residue peptide into cytotoxic plaques is now supported by multiple lines of evidence. Exactly how aggregated A␤ and the locally secreted and blood-borne proteins that become associated with it in plaques (e.g. heparan sulfate proteoglycan (86), ␣ 1 -antichymotrypsin (87), apolipoprotein E (88), complement components (89,90), serum amyloid P component (91), and cytokines (92)) exert toxic effects on surrounding cells is an area of intensive study. It appears that aggregated but not monomeric A␤ peptides can induce cell dysfunction and death in vitro by a range of presumably interrelated mechanisms that include oxidative injury (83,93), alterations in intracellular calcium homeostasis (38,39), and cytoskeletal reorganization (64). Sufficient knowledge of some of the principal elements of the amyloid-induced cascade has emerged that the process of identifying small molecules which could inhibit one or another step is now well under way. Of particular therapeutic interest are attempts to decrease the secretion of A␤ peptides from neuronal and glial cells, something which may be possible to accomplish prior to knowing the identities of the requisite proteases. Interfering with the aggregation of A␤ 42 and A␤ 40 peptides or inhibiting the toxicity that these extracellular aggregates produce on neurons, their processes, and glial cells are also of great therapeutic relevance. Finally, controlling the specialized inflammatory response that appears to be triggered by aggregated A␤ (including microglial stimulation, activation of the classical complement cascade, cytokine release, and reactive astrocytosis) should also prove to be of benefit to patients with this disease. Given the accelerating pace of progress, there can be little doubt that further biochemical and pharmacological research should lead to a range of therapeutic options during the next several years.