Alzheimer's disease: β-Amyloid protein and tau

Abstract

Research on the molecular pathogenesis of Alzheimer's disease (AD) has made great strides over the last decade. This progress is the result of protein chemical analysis of two extracellular and intracellular fibrillary lesions in AD brain conducted during the 1980s, which identified β-amyloid protein (Aβ) and tau as their major components, respectively. Linkage analysis of familial AD identified four responsible genes: three causative genes (β-amyloid precursor protein, presenilin 1, and presenilin 2) and one susceptibility gene (apolipoprotein E ϵ4). All those genes causing and predisposing to AD exhibit a common phenotype: an increased production of Aβ42, a longer, more amyloidogenic Aβ species, and/or its enhanced deposition. This observation was substantiated when presenilins were shown to be directly involved in Aβ production. Whereas Aβ deposition is relatively specific for AD, tau deposition is observed in various neurodegenerative diseases and is assumed to be intimately associated with neuronal loss. The genetic analysis of frontotemporal dementia and parkinsonism linked to chromosome 17 (FTDP-17) revealed the presence of mutations in the tau gene in affected members. Thus, tau can lead to intracellular tau deposits and neuronal loss, although the mechanism remains to be clarified. Taken together, Aβ might exert neurotoxicity through tau, leading to neuronal loss in the AD brain. © 2002 Wiley-Liss, Inc.

Alzheimer's disease (AD) is the most common cause of dementia among elderly people. Its prevalence steeply increases at the age of 80 years and above, a characteristic epidemiological finding shared in well-developed countries. The number of AD patients is constantly increasing in these countries, largely because the population older than 80 years is expanding, and this particular population makes the greatest contribution to AD's prevalence. Thus, a dramatic rise in the life span over the last century has forced us to face this devastating disease, which was not such a serious problem 100 years ago. The pathological hallmarks of AD brains include cortical atrophy, most predominant in the medial temporal lobe and, microscopically, extensive neuronal loss and intracellular and extracellular abnormal fibrillar deposits; termed neurofibrillary tangles and senile plaques, respectively. Since the first description in 1907 by Alois Alzheimer, a Bavarian psychiatrist, of a woman showing progressive dementia when aged in her early 50s, several neuropathologists and neurologists had been involved in the elucidation of this demented disorder, especially of the filamentous structures, but progress had been very slow. Presumably, the discovery of Congo red birefringence of the abnormal fibrils and their ultrastructural identities are the two major milestones in the research on AD conducted for a period of 60 years after the first description. The greatest strides in research on the molecular mechanisms of AD occurred mostly over the last 20 years, especially over the last decade.

The first advance was made through painstaking protein chemical characterization of senile plaques and neurofibrillary tangles. Although these lesions are highly resistant to various kinds of detergents or denaturants, Glenner successfully purified cerebrovascular amyloid from AD brains in 1984 and determined the first 24 amino acids of its major, ∼4 kDa constituent (Glenner and Wong, 1984). This small protein is designated as β-amyloid protein (Aβ), after the β-pleated sheet structure. Beyreuther and Masters extended this work and showed that the same protein (Aβ) is also the major constituent of amyloid cores in the parenchyma of AD brains (Masters et al., 1985). In 1987, cDNA encoding β-amyloid precursor protein (APP) was cloned (Kang et al., 1987). This was the beginning of the subsequent intensive molecular and cell biological studies of AD. At first, Aβ was thought to be a pathological molecule generated by abnormal cleavage of APP, which was observed only in AD brains. One good reason for this speculation was that APP must be cleaved in the middle of the membrane to generate Aβ. However, in 1992 Aβ was found to be secreted constitutively from various types of cells (Haass et al., 1992; Shoji et al., 1992).

With regard to neurofibrillary tangles, by 1988, immunochemistry and protein chemistry approaches used by several groups had identified tau, a microtubule (MT)-associated phosphoprotein, as their major constituent (Nukina and Ihara, 1986; Grundke-Iqbal et al., 1986; Wischik et al., 1988; Kondo et al., 1988). Ubiquitin was identified as another major component of the tangles (Mori et al., 1987). Within the tangles, N-terminally processed tau was found to be ubiquitinated (Morishima-Kawashima et al., 1993). These observations were the first showing that the degeneration-associated intracellular deposits are ubiquitinated, probably because of the failure to degrade through proteasomes. Similar ubiquitination of intraneuronal inclusions, including Lewy bodies, Pick bodies, and intranuclear aggregates in polyglutamine diseases, were then reported.

At that time, small numbers of investigators had already started to take an entirely different approach, based on inheritance alone rather than pathology: linkage analysis and positional cloning. This approach met with great success over the next 10 years. One could call the last decade of the 20th century the era of genetics. The molecular cloning of the cDNA of APP in 1987 caused great excitement, because its gene was mapped on chromosome 21q, and patients with Down's syndrome (trisomy 21) were known to invariably develop AD neuropathology (senile plaques and neurofibrillary tangles) in earlier life, their 20s to 40s. After some confusion regarding the pathogenic significance of chromosome 21, a missense mutation at codon 717 of APP was identified in the two families affected by autosomal dominantly inherited familial AD (FAD; Goate et al., 1991). Thus, the APP gene was the first identified causative gene for FAD (Table I). This established a view that AD is not a single-gene disease but rather a genetically heterogeneous disorder exhibiting similar clinical and pathological phenotypes (in other words, AD syndrome; St. George-Hyslop et al., 1990). This discovery boosted an extensive search for other loci associated with the autosomal dominant FAD. In 1995, two causative genes for FAD, presenilin (PS)1 on chromosome 14 (Sherrington et al., 1995) and PS2 on chromosome 1 (Levy-Lahad et al., 1995; Rogaev et al., 1995), were successively identified. On the other hand, the polymorphic allele of apolipoprotein E, ϵ4, on chromosome 19 was identified as a major susceptibility gene in late-onset AD (Corder et al., 1993). The gene ϵ4 increases the risk of developing AD in a gene-dosage-dependent manner. The search for other genetic factors associated with AD is ongoing. At least two unidentified susceptibility genes have been localized to chromosomes 10 (Bertram et al., 2000; Ertekin-Taner et al., 2000; Myers et al., 2000) and 12 (Pericak-Vance et al., 1997), but they have not yet been cloned.

Table I. Causative and Susceptibility Genes for AD

Gene	Chromosome	Onset	Aβ phenotype
β-Amyloid precursor protein (APP)	21q21	Early	Production of all Aβ species or Aβ42 ↑
Presenilin 1	14q24.3	Early	Production of Aβ42 ↑
Presenilin 2	1q31-42	Early (∼late)	Production of Aβ42 ↑
Apolipoprotein E polymorphism (ε4)	19q13.2	(Early∼) late	Brain accumulation of Aβ ↑

Phenotypic analysis of these genes revealed, importantly, that the four genes so far identified, as well as an unidentified gene on chromosome 10, are all associated with increased Aβ42 production or its enhanced deposition in the brain (Suzuki et al., 1994; Scheuner et al., 1996; Borchelt et al., 1996; Morishima-Kawashima et al., 2000; Ertekin-Taner et al., 2000; Table I). This strengthened the “amyloid cascade hypothesis” proposed by Hardy and Higgins in 1992. It was demonstrated in 1998 that PS plays a critical role in Aβ production (De Strooper et al., 1998), and many people now agree that Aβ, especially Aβ42, is in the closest position in the etiology of AD.

Linkage analysis has also shown that tau is an essential molecule in the pathway to neuronal loss. In 1998, mutations in the tau gene were found in several families affected by frontotemporal dementia with parkinsonism linked to chromosome 17 (FTDP-17), which exhibits extensive neuronal loss and tau deposition (Poorkaj et al., 1998; Hutton et al., 1998). This implies that mutations of tau alone are sufficient to cause both neuronal loss and tau deposition, the former of which may have a dominant role in causing dementia.

Thus, Aβ and tau, the molecules composing the framework of the two extracellular and intracellular fibrillar structures, respectively, were originally considered simply to represent consequences of neurodegeneration, but they have turned out to be the critical molecules in the pathogenesis of AD. This brief review focuses on these two molecules.

β-AMYLOID

Generation of Aβ From APP

Aβ is a 39–43-residue protein with a molecular weight of ∼4 kDa. The major Aβ species found in vivo are Aβ40, which ends at Val40, and Aβ42, which has two additional hydrophobic residues, Ile and Ala. As a result, Aβ42 is more hydrophobic and has higher aggregation potential (and thus is more amyloidogenic).

Aβ is produced from APP through sequential proteolytic processing. APP is a type-I membrane protein, and Aβ consists of its ectodomain just outside the membrane (28 residues) and the luminal half of the transmembrane domain (12 or 14 residues; see Fig. 1). APP undergoes N-linked and O-linked glycosylation in the endoplasmic reticulum and Golgi apparatus, respectively, and is transported via secretory vesicles to the plasma membrane. A fraction of the APP molecules is then internalized through a clathrin-dependent endocytic pathway. During this trafficking, APP undergoes various proteolytic processing steps (Fig. 2). The cleavage of APP by β-secretase produces a soluble form of APP (APPsβ), which is shed into the lumen, and a membrane-bound C-terminal fragment (CTF, or C99). The subsequent cleavage of C99 by γ-secretase releases Aβ1–40 and Aβ1–42. There is another pathway, in which APP is cleaved by α-secretase between Lys-16 and Leu-17 in the middle of the Aβ luminal region. This pathway generates p3 (Aβ17–40 and Aβ17–42) and another CTF, C83, and precludes the production of amyloidogenic Aβ.

Figure 1. Illustration of the structures of APP and Aβ. The amino acid sequence of APP, including Aβ and its flanking regions, is shown. The sequence Aβ1–42 is boxed, and the predicted transmembrane domain is underscored. The cleavage sites for α-, β-, and γ-secretases and the missense mutations causing FAD or related disorders are indicated by arrows. The amino acid numbering follows the APP770 isoform. KPI and CHO represent a Kunitz-type serine protease inhibitor domain and a glycosylation site, respectively.

Figure 2. Degradation pathway of APP. The two proteolytic processing pathways are shown (large arrows). APP undergoes cleavage by α-secretase, which results in release of APPsα and retention of the 83-residue CTF (C83) in the membrane (upper large arrow). The subsequent cleavage of C83 by γ-secretase liberates p3 beginning at the Aβ17 position. In an alternative amyloidogenic pathway, APP undergoes cleavage by β-secretase, resulting in release of APPsβ and retention of the 99-residue CTF (C99; lower large arrow). C99 undergoes cleavage by γ-secretase and produces Aβ.

There is still disagreement about where Aβ is produced within the cell. According to our current knowledge, Aβ appears to be produced early in the secretory pathway and/or the endosomal pathway. Most Aβ, in particular Aβ 40, is secreted from the cells. Consistently with this, Aβ can be detected in the cerebrospinal fluid (CSF) as well as in the plasma in normal subjects.

Neuropathological Aspects of Aβ

Aβ is deposited as senile plaques in the extracellular space of the brain parenchyma. Senile plaques are subdivided into neuritic plaques, primitive plaques, and diffuse plaques. The former two consist of Aβ fibrils and accompany dystrophic neurite and glial reactions (reactive astrocytes and activated microglia). Diffuse plaques consist of amorphous granular Aβ rather than fibrils.

Senile plaques are observed in brains in a limited number of diseases, including AD, and appear to be more specific for AD than neurofibrillary tangles. However, they are also observed in most brains of aged, nondemented subjects. Indeed, Aβ accumulates in the insoluble fraction of the normal (nondemented) human brain during aging (Funato et al., 1998; Morishima-Kawashima et al., 2000). Insoluble Aβ is present in the brain even in normal, young subjects. Aβ accumulation starts in some brains by the late 40s. The proportion of subjects showing significant Aβ deposition increases in an age-dependent manner. The accumulation of Aβ in these subjects appears to increase exponentially during the next 20 years (50–70 years), reaching a plateau at about 70 years of age. Aβ42 is the first species to be deposited, followed by Aβ40, but the levels of Aβ40 increase in a more gentle fashion.

Some of the insoluble Aβ molecules in the brain can be fractionated into the low-density membrane fraction by sucrose density-gradient centrifugation (Lee et al., 1998; Morishima-Kawashima et al., 1998). Mutant but not wild-type PS2 enhances the levels of Aβ42 in this fraction (Sawamura et al., 2000). Moreover, the amounts of Aβ in the low-density membrane fraction are well correlated with the extent of extracellular Aβ deposition in human brains as well as APP transgenic (PDAPP) mouse brains (Oshima et al., 2001). This may imply that the Aβ42 associated with low-density membrane domains is tightly coupled to the process of extracellular Aβ deposition. Consistently with this assumption, electron microscopic observations show that Aβ accumulates in juxtaplasma membranes at the earlier stage of Aβ deposition (Yamaguchi et al., 2000). These low-density membranes are enriched in cholesterol and sphingolipids and are known to include particular membrane microdomains, the so-called caveolae and rafts. Aβ binds to GM1 ganglioside, an interaction that is enhanced by cholesterol and helps Aβ molecules to be converted to β-sheets, leading to fibril formation (Choo-Smith et al., 1997; Kakio et al., 2001). The low-buoyant-density membranes contain high levels of γ-secretase activity, which depend on cholesterol levels in the membranes (Wahrle et al., 2002). It is also known that alterations in membrane lipid composition and fluidity are associated with aging. Thus, during aging, these distinct mechanisms may contribute to Aβ accumulation in specific membrane domains and therefore in the brain parenchyma.

FAD Genes Associated With Aβ42 Production

Several mutations of APP have been known to segregate with FAD. Most of them are located in the flanking regions of the Aβ molecule (Fig. 1), and one can speculate that these locations affect either β-secretase or γ-secretase, thereby increasing total Aβ or Aβ42 production. Some APP mutations that are located in the middle of the Aβ molecule may enhance fibril or protofibril formation of Aβ, causing Aβ deposition much earlier in life. Seeking the secretases has been one of the hottest topics in AD research. BACE1 (β-site APP cleaving enzyme; also termed Asp2 or memapsin 2) was identified as the long-sought β-secretase in 1999, based on the cleavage site in APP and enhanced cleavage in a Swedish mutation of APP (Vassar et al., 1999; Yan et al., 1999; Sinha et al., 1999; Hussain et al., 1999). BACE1 is a type-I membrane-bound aspartyl protease belonging to the pepsin protease family. In the cell, BACE1 is preferentially located in the Golgi apparatus and endosomes and might cleave APP at the β-site in these organelles. On the other hand, the identity of γ-secretase, which cleaves the C-terminus of Aβ either at Val-40 or at Ala-42 within the transmembrane domain, is still ambiguous. The “PS is the γ-secretase” hypothesis that was proposed in 1998 led to great excitement among AD investigators, but its validity remains to be established (see below).

PS1 and PS2 (PSs) are highly homologous multispanning membrane proteins: an eight-span transmembrane (TM) model is most preferred. More than 70 mutations in the PS1 gene and six mutations in the PS2 gene have been found, and these FAD mutations invariably cause enhanced production of Aβ42 (Scheuner et al., 1996; Borchelt et al., 1996). PSs are cleaved in a large cytoplasmic loop between TM domains 6 and 7, generating an ∼30 kDa N-terminal fragment (NTF) and an ∼20 kDa CTF. The two fragments form a heterodimer and are incorporated into a 150–250 kDa high-molecular-weight complex (Yu et al., 1998; Capell et al., 1998), presumably a functional unit of the PS molecule. PSs play a critical role in γ-cleavage of APP. The cells from PS1-deficient mice (De Strooper et al., 1998) and the cells from PS1 and PS2 doubly deficient mice (Herreman et al., 2000; Zhang et al., 2000) show markedly reduced and almost undetectable Aβ production, respectively, which accompanies a remarkable accumulation of APP CTFs, C83 and C99, the immediate substrates for γ-secretase. This indicates that PS is an essential component for γ-secretase. Using peptidomimeric transition state analogue inhibitors for γ-secretase, which were designed to mimic the C-terminus of Aβ, Wolfe and his colleagues assumed that γ-secretase has the properties of aspartyl protease (Wolfe et al., 1998, 1999b). They further showed that mutations of two conservative Asp in TM6 and TM7 of PS (Asp-257 and Asp-385 for PS1, respectively) led to a marked reduction in Aβ production and an accumulation of APP CTFs, as had been observed in the PS-deficient cells (Wolfe et al., 1999a). They proposed, in an analogy with human immunodeficiency virus (HIV) protease, that PS may be a diaspartyl protease executing γ-cleavage. This hypothesis was substantiated by a further observation, that the transition state analogue γ-secretase inhibitors specifically affinity labeled PS fragments (Li et al., 2000; Esler et al., 2000). Nevertheless, there is still skepticism about the interpretation of these results (see Sisodia et al., 2001). An alternative interpretation would be that PS functions as a cofactor for γ-secretase or as a modifier of trafficking of APP derivatives and γ-secretase. Aβ generation in reconstituted systems from known components should be required to prove the “PS is γ-secretase” hypothesis.

PS1 has an important role in facilitating the Notch1 signaling pathway. A morphological analysis of PS1-deficient mice suggested a similarity with Notch1 knockout mice (see Selkoe, 2000). Then, investigations using PS1-deficient cells and flies demonstrated that PS1 is essential for the intramembranous site-3 cleavage of Notch1 (De Strooper et al., 1999; Struhl and Greenwald, 1999). This cleavage liberates the Notch intracellular domain, termed NICD, which translocates to the nucleus, activates transcription, and transduces Notch signaling. Thus, PS1-dependent γ-secretase activity is believed to process both APP and Notch in a similar fashion. However, several paradoxes emerge if the same enzyme is involved in these cleavages seen on APP and Notch1. One important point is the intramembranous topography of the cleavage sites: The γ-cleavage site on APP is present in the middle of the membrane, whereas site-3 cleavage of Notch1 occurs just four residues in from the cytoplasmic boundary of the membrane (Fig. 3). Recently, another cleavage site was found on the APP molecule (Gu et al., 2001; Sastre et al., 2001; Fig. 3). The major site was the N-terminus of Val-50 (according to Aβ numbering), and a minor site was that of Leu-49. Similar cleavage was also observed for other members of the APP family, β-amyloid precursor-like protein (APLP)-1 and APLP2 (Gu et al., 2001). This new type of cleavage shares several characteristics with Notch1 site-3 cleavage: 1) The cleavage site is located only several residues inside the membrane cytoplasmic boundary; 2) the residue at the P1′ position is a bulky, hydrophobic residue such as Val and Leu; 3) the cleavage is PS-dependent; and 4) γ-inhibitors suppress this cleavage. Thus, it is more likely that the Notch1 site-3 cleavage corresponds to this distinct type of cleavage rather than γ-cleavage itself. Hereafter, this cleavage will be called ϵ-cleavage, according to the German group (Weidemann et al., 2002). It should be noted that this cleavage in the APP family has only loose specificity for amino acid sequence compared with Notch1. Other substrates for PS-dependent γ-secretase-like activity include ErbB4 (Ni et al., 2001) and CD44 (Okamoto et al., 2001), and certainly more substrates will be added to the list in the near future (Fig. 3). Those substrates are all type-I membrane proteins, as noted above, cleaved following ectodomain shedding, and have an obvious stop-transfer signal. Thus, PS-dependent intramembranous cleavage near the membrane cytoplasmic boundary is likely to be shared by various kinds of the type-I membrane proteins and thus represents a particular degradation step in the general membrane protein metabolism. It is also possible that γ-cleavage, the cleavage in the middle of the transmembrane domain, proceeds at the same time as ϵ-cleavage, so Notch 1 might undergo γ-cleavage in addition to ϵ-cleavage (Zhang et al., 2002). Although some mutations in PS1 and newly developed inhibitors are claimed to distinguish between Aβ production and Notch1 site-3 cleavage (Capell et al., 2000; Petit et al., 2001), we currently do not know whether these apparently separate types of cleavages (γ and ϵ) of APP can be distinguished. It might be better to develop an inhibitor specific for either γ- or ϵ-cleavage. Furthermore, the relationship between those two types of cleavages in terms of Aβ40 and Aβ42 production should be clarified. In particular, determining the relationship between Aβ40 and CTF50–99 and between Aβ42 and CTF49–99 would be interesting.

Figure 3. Sites of γ- and ϵ-cleavage. The PS-dependent intramembranous cleavage sites are compared with one another among the substrates so far identified. Predicted transmembrane domains are boxed. Large and small arrowheads indicate major and minor cleavage sites on APP by γ-secretase, which generate Aβ40 and Aβ42, respectively. Large and small arrows represent major and minor ϵ-cleavage sites on each molecule by a PS-dependent γ-secretase-like protease, respectively. The predicted site of ErbB4 is indicated by a question mark.

TAU

Neurofibrillary Tangles and Tau

Tau is a microtubule (MT)-associated protein of ∼55 kDa and is expressed abundantly in the brain. In the human brain, six isoforms are produced from a single gene on chromosome 17q by alternative mRNA splicing. In its C-terminal portion, three or four repeats composing the MT-binding domain are located in tandem (Fig. 4). Tau binds to tubulin through the repeats and promotes MT assembly. The splicing out of the second 31-residue repeat, encoded by exon 10, gives rise to three-repeat isoforms. Embryonic brains exclusively express three-repeat tau, which could easily destabilize MTs and affect the ability of neurons to extend or retract their processes.

Figure 4. Molecular structure of tau and mutations in the tau gene identified in patients affected by FTDP-17. The shaded boxes represent the regions regulated by alternative splicing. Tau has three or four repeats (indicated by solid boxes) in its C-terminal portion. Splicing out of the second repeat, which is encoded by exon 10 (the rightmost shaded box in the upper diagram), gives rise to three-repeat tau. Otherwise, four-repeat tau is produced. Because of combination with the two alternatively spliced exons in the N-terminal portion, six isoforms are expressed in human adult brain. The upper diagram represents the exonic mutations in tau in individuals with FTDP-17. They are clustered in the MT-binding domain and its flanking region. The amino acid numbering is according to the 441-residue human tau isoform. The lower diagram represents the intronic mutations, all of which are present in the splicing donor site of exon 10, destabilize the predicted RNA stem loop, and lead to altered mRNA splicing to increase four-repeat tau.

In AD brains, tau may dissociate from MTs, possibly because of hyperphosphorylation or for other reasons, and accumulates in the neuronal perikarya and their processes, especially in dendrites, as paired helical filaments (PHFs), the unit fibrils of neurofibrillary tangles. Indeed, MTs are not observed in the region where PHFs are present within the neuron. Whereas all six full-length isoforms are known to compose the framework of PHF (PHF-tau; Lee et al., 1991), the C-terminal half containing the MT-binding domain appears to constitute the core of PHF that is insoluble in SDS (Kondo et al., 1988). The most remarkable characteristic of tau in PHFs is its hyperphosphorylation. No fewer than 25 phosphorylation sites have been identified in PHF-tau purified from AD brains (Morishima-Kawashima et al., 1995; Hanger et al., 1998). Most of the phosphorylation sites are clustered in the flanking region of the MT-binding domain. The extent of phosphorylation as well as the numbers of phosphorylation sites far exceed those of tau in the normal brain. According to the motifs for protein kinases, multiple protein kinases, including GSK3β, cdk5, MAPK, and PKA, appear to be involved in hyperphosphorylation of tau (Morishima-Kawashima et al., 1995). This indicates that multiple phosphorylation cascades may be activated within tangle-bearing neurons.

In contrast to senile plaques, neurofibrillary tangle formation is a late event and is observed in brains affected by various neurodegenerative diseases as well as AD. Such diseases that present with intracellular tau inclusions are called tauopathies. It is known for individuals with AD that the areas forming neurofibrillary tangles precisely match those areas exhibiting neuronal loss and that the abundance of tangles correlates well with the extent of neuronal loss. Moreover, the abundance of tau inclusions or the extent of neuronal loss also correlates well with the degree of dementia. However, the application of an unbiased stereological method for AD brains revealed that the numbers of lost neurons greatly exceed those of neurofibrillary tangles, indicating that neuronal loss may occur without neurofibrillary tangles. Thus, neuronal loss might not be a consequence of tangle formation and might proceed rather independently. It is possible that the same cascade leads to both neuronal loss and tangle formation. Discovery and establishment of FTDP-17 (Poorkaj et al., 1998; Hutton et al., 1998) led to a consensus that tau mutation is sufficient for both neurofibrillary tangles and neuronal loss. This in turn led us to speculate that a similar cascade leading to neuronal death is triggered by Aβ deposition in AD.

Tau Mutations in FTDP-17

FTDP-17 is a familial neurological disorder characterized genetically by autosomal dominant inheritance, clinically by behavioral abnormalities and parkinsonism, and neuropathologically by tauopathy. Intracellular deposits of hyperphosphorylated tau are observed within neurons as well as glial cells in the affected brains. More than 25 exonic and intronic mutations in the tau gene have thus far been identified (Fig. 4). The exonic mutations include missense mutations, deletion mutations, and silent mutations. Most of the exonic mutations are located within or close to the MT-binding domain, except for one site that is located in the N-terminus of tau. The intronic mutations are all located in the 5′ splice site of exon 10. Those mutations are considered to modify tau functions either through a reduced ability of MT assembly or through altered ratios of three- to four-repeat tau.

Most of the exonic missense mutations slightly or significantly decrease MT-promoting activity (Hasegawa et al., 1998; Hong et al., 1998). The extent of this reduction is assumed to be parallel to the deposition of tau. Using site-specific antibodies that distinguish between wild-type and mutant tau, we analyzed molecular species of tau in the soluble and insoluble fractions of the brain affected by two FTDP-17 mutations. In brains with an aggressive mutation, P301L, mutant tau was preferentially deposited in the insoluble fraction, whereas protein levels of mutant tau in the soluble fraction were selectively decreased despite there being no detectable decrease in its mRNA levels (Miyasaka et al., 2001b). In contrast, in brains with a less aggressive mutation, R406W, almost equal amounts of wild-type and mutant tau were deposited in the insoluble fraction, and their levels in the soluble fraction did not differ from each other (Miyasaka et al., 2001a). P301L tau has a greatly reduced ability for MT assembly compared with wild-type tau, whereas R406W tau shows only a slightly reduced ability (Hasegawa et al., 1998). Thus, one can speculate that most exonic mutations reduce the affinity of tau for MTs to a certain extent, leading to destabilization of the MTs, and that the resultant cytosolic free tau, in which the proportion of mutant tau is determined by its MT-promoting ability, may become highly phosphorylated and aggregate into PHF-like fibrils, which may in turn exert neurotoxicity.

On the other hand, all the intronic mutations and some exonic mutations (S305S, S305N) destabilize a predicted RNA stem-loop in the splicing donor site of exon 10 and alter mRNA splicing in such a direction that it causes an increase in four-repeat tau (Hutton et al., 1998). These findings present us with an enigma, in that four-repeat tau has a greater ability for MT assembly than three-repeat tau. Thus, tau deposition and neuronal loss/neurodegeneration induced by the FTDP-17 mutations cannot be explained simply by reduced MT-promoting activity.

Tau deposited in FTDP-17 brains is hyperphosphorylated to an extent similar to that of PHF-tau in AD brain. Although dozens of reports have claimed that such high phosphorylation of tau is associated with the degenerative process, its significance is still unclear. In vitro formation of PHF-like fibrils from recombinant tau shows that phosphorylation is not necessary for PHF formation (Goedert et al., 1996). Moreover, it is not known whether the FTDP-17 mutations induce elevated tau phosphorylation. Instead, R406W tau is less phosphorylated than wild-type tau in the soluble fraction of the affected brains as well as in the stably transfected cells (Matsumura et al., 1999; Miyasaka et al., 2001a). Despite having a lesser extent of phosphorylation in the soluble fraction, insoluble R406W tau was phosphorylated to an extent similar to phosphorylation in insoluble wild-type tau (Miyasaka et al., 2001a), indicating that the environment of tangles predisposes the mutant tau to hyperphosphorylation. It is not known whether the enhanced phosphorylation cascade targets only tau. When phosphorylated, tau becomes more resistant to proteases, stays within the cell longer, and readily aggregates into PHF, whereas other proteins, when phosphorylated, may become more susceptible to proteases. Thus, the hyperphosphorylation might be not a causative in but a later event associated with neurodegeneration. Nevertheless, once phosphorylated tau is deposited, it may affect the physiological functions of MTs, such as axonal flow, thereby accelerating neurodegeneration.

The most important implication derived from studies on individuals with FTDP-17 is that the mutation of tau gene itself is sufficient to cause formation of intracellular tau deposits and neuronal loss. In other words, tau is causative for these two lesions. This may be applied to other tauopathies, including AD. Because causative and susceptibility genes for AD are tightly coupled to Aβ deposition in earlier ages, as described above, tau lesions are likely to be downstream of Aβ lesions in the development of AD. This assumption is consistent with recent observations that deposition of filamentous tau is greatly enhanced in FTDP-17 tau transgenic mice by crossing with APP transgenic mice that form senile plaques (Lewis et al., 2001) or by injecting Aβ in the brain region where axons are projecting (Gotz et al., 2001). Thus, Aβ may exert its neurotoxicity via tau lesions in individuals with AD.

THERAPEUTIC STRATEGIES AND PERSPECTIVES

Several therapeutic approaches to AD have recently emerged, and in most cases Aβ is a therapeutic target. Their objectives include enhanced Aβ clearance, inhibition of Aβ fibrillogenesis, and reduced Aβ production.

A report on the effect of Aβ vaccination in PDAPP mice was most exciting (Schenk et al., 1999). Immunization of Aβ in the APP transgenic mice caused a substantial reduction in the amount of Aβ deposition in the brain. Moreover, long-immunized mice appeared to clear preexisting Aβ depositions. The mechanisms underlying those observations were explained as follows: A trace amount of the Aβ antibody enters the brain through the blood–brain barrier, activates microglia, and hence induces inflammatory reactions (Bard et al., 2000). An alternative explanation is that the antibody traps Aβ in plasma, so Aβ, accumulated in the brain, flows out into plasma to maintain an equilibrium (DeMattos et al., 2001). Active immunization and passive immunization are both effective. These experiments were carried out in mice, which express human APP and exhibit senile plaque pathology but not neurofibrillary tangles or neuronal loss. Because the “amyloid cascade” hypothesis assumes that Aβ deposition causes all of the subsequent events, including neurofibrillary tangles and neuronal loss, and because neuronal loss is tightly associated with the degree of dementia, it is reasonable to speculate that prevention or elimination of Aβ deposition would reduce further neuronal loss and thus prevent the progression of dementia. To address this issue, one has to proceed to clinical trials. Based on the results obtained in the phase 1 studies, Elan Corporation and Wyeth-Ayerst Laboratories initiated a multicenter exploratory phase 2A clinical trial with hundreds of patients with mild to moderate AD. Quite recently, it was announced that four patients have experienced clinical signs consistent with “inflammation in CNS” and that additional patients have presented with similar symptoms (see http://www.elan.com). Because of these unexpected consequences, clinical trials for Aβ vaccination were suspended. This adverse effect may be meningoencephalitis, and there appears to be no correlation between the titer of anti-Aβ antibody and its consequence. This suggests that the encephalitis may be due to T-cell-mediated processes.

Nevertheless, the above-mentioned potential immunotherapy has opened a new research area, communication between Aβ in the brain and Aβ in the plasma, which has been largely overlooked in the past. The “peripheral” Aβ in the plasma might be in equilibrium with Aβ in the CSF, which in turn is in equilibrium with Aβ in the brain parenchyma. This indicates that the peripheral Aβ can cross the blood–brain barrier and contribute to brain Aβ deposition. Conversely, a specific ligand for Aβ, such as an antibody, if administered in the systemic circulation, would accelerate clearance of Aβ by forming complexes, shift the equilibrium, and thereby reduce brain Aβ deposition. If a ligand—other than an antibody— can be developed that is highly specific for Aβ, facilitates Aβ clearance, does not penetrate the blood–brain barrier, and has little adverse effect, such a ligand would not cause inflammation in the CNS and could replace the antibody used for immunotherapy.

Reducing Aβ production is another therapeutic target. Various kinds of β-secretase or γ-secretase inhibitors are being developed. However, less information is available from these clinical studies. With regard to γ-secretase inhibitors, adverse effects are possible, because they inhibit lymphocyte maturation (Doerfler et al., 2001) and because γ-secretase activity is likely to be involved in processing of many as-yet-unidentified membrane proteins. Epidemiological workers have pointed out that the use of nonsteroidal antiinflammatory drugs (NSAID) may reduce the prevalence of AD, but a recent report suggested the unexpected mechanism; a subset of NSAID reduces the production of Aβ42, which is not mediated through cyclooxygenases 1 and 2 (Weggen et al., 2001). Thus, this effect is distinct from the antiinflammatory actions of NSAID and could be another approach to reducing Aβ production. Several epidemiological reports have suggested that the cholesterol-lowering reagents statins could lower the prevalence of AD (see Wolozin, 2001). Although membrane lipids, including cholesterol, have been shown to affect Aβ production, the precise mechanisms have not yet been clarified.