A modified β-amyloid hypothesis: intraneuronal accumulation of the β-amyloid peptide – the first step of a fatal cascade


Address correspondence and reprint requests to O. Wirths, Department of Psychiatry, Division of Neurobiology, University of the Saarland Medical Center, Building 90, 66421 Homburg/Saar, Germany.
E-mail: oliver.wirths@uniklinik-saarland.de


Accumulating evidence points to an important role of intraneuronal Aβ as a trigger of the pathological cascade of events leading to neurodegeneration and eventually to Alzheimer's disease (AD) with its typical clinical symptoms, like memory impairment and change in personality. In the present article, we review recent findings on intracellular monomeric and oligomeric β-amyloid (Aβ) generation and its pathological function in cell culture, transgenic AD mouse models and post mortem brain tissue of AD and Down syndrome patients, as well as its interaction with oxidative stress and its relevance in apoptotic cell death. Based on these results, a modified Aβ hypothesis is formulated, that integrates biochemical, neuropathological and genetic observations with AD-typical neuron loss and plaque formation.

Abbreviations used



Alzheimer's disease


amyloid precursor protein


β-secretase cleavage


99-residue C-terminal fragment


endoplasmic reticulum


multivesicular bodies


paired helical filaments


trans-Golgi network

Alzheimer's disease (AD), the most frequent form of senile dementia, is characterized by extracellular senile plaques, neurofibrillary tangles, as well as vascular amyloid and progressive neurodegeneration. The extracellular plaques mainly contain β-amyloid (Aβ) peptides (Glenner and Wong 1984), which are derived by two proteolytic cleavages from the larger amyloid precursor protein (APP). In the first step, a 99-residue C-terminal fragment (C99) is generated by β-secretase cleavage (BACE1). This product is further processed by γ-secretase activity generating Aβ peptides. An alternative cleavage pathway by α- and subsequent γ-cleavage precludes the generation of Aβ and leads to the generation of the 16 amino acids shorter peptide termed p3 (for a review, see Bayer et al. 2001). Mutations in the APP gene and the presenilins (PS-1, PS-2) account for most of the familial early onset cases of AD either by enhancing the production of pathological Aβ or the 42-amino acid form, which easily aggregates (Suzuki et al. 1994). In addition to the deposition of Aβ peptides into extracellular plaques, there is increasing evidence that Aβ occurs intracellularly and that it is initially involved in the disease process. The concept of a Aβ cascade has been valid for more than 10 years (Hardy and Allsop 1991), and provided a reasonable basis for AD therapeutic strategies (Fig. 1). However, the Aβ hypothesis has been controversial, as the plaque load in AD brain does not correlate with the disease state in contrast to the amount of tau neurofibrillary tangles (Arnold et al. 1991; Braak and Braak 1991) reflecting very well the clinical phenotype (Morrison and Hof 1997; Nagy et al. 1999). The current point of view is that neuronal loss occurs in AD, but during normal aging there is no evidence for neuron death in the hippocampus and neocortex (Morrison and Hof 1997). The hippocampal formation and adjacent entorhinal cortex are particularly affected and have been shown to be the primary site for neuron death in the brain of patients suffering from AD. In this article we review the challenging findings of intraneuronal Aβ, a pathological feature of AD that has long been neglected and is turning out to be the key factor in the pathogenesis prior to the extracellular Aβ deposition.

Figure 1.

Scheme showing the current β-amyloid hypothesis, which is based on biochemical, neuropathological, molecular-biological and genetic evidence.

Sites of intracellular Aβ generation

Today it is a matter of common knowledge that the extracellular Aβ aggregates are of neuronal origin and are secreted as soluble peptides. First evidence for intracellular Aβ was reported by Wertkin et al. who observed intracellular Aβ in a differentiated neuronal cell line (Wertkin et al. 1993). A more detailed analysis of NT2N neurons showed substantial amounts of Aβ40 and Aβ42 at a ratio of 3 : 1 in cell lysates, whereas analysis of secreted forms yielded a ratio of 20 : 1 for Aβ40:Aβ42 (Turner et al. 1996).

Aβ peptides are generated at different subcellular sites. Whereas Aβ40 is generated solely in the trans-Golgi network (TGN) (Hartmann et al. 1997; Xu et al. 1997; Greenfield et al. 1999), Aβ42 is generated in the endoplasmic reticulum (ER) (Greenfield et al. 1999) as well as Golgi compartments. This has been shown by the retention of APP in the endoplasmic reticulum/intermediate compartment (ER/IC), which leads to an elimination of intracellular Aβ40, whereas the synthesis of intracellular Aβ42 was not affected (Cook et al. 1997). A blockade of APP transport within Golgi compartments causes enhanced levels of N-terminally truncated Aβx-40 and Aβx-42, indicating that γ-secretase cleavage can occur in the ER and later secretory pathway compartments like Golgi (Wild-Bode et al. 1997). However, there is also data indicating that Aβ1–42 is generated in distal Golgi compartments, whereas Aβx-42 is generated in cis- or medial Golgi (Sudoh et al. 1998). In addition to the normal APP processing machinery, another γ-secretase activity seems to exist that competes with the proteasome in the ER/IC. This secretase activity is responsible for the generation of intracellular Aβ42, which accumulates in the ER/IC as an insoluble pool (Skovronsky et al. 1998). Further evidence for a third pathway leading to Aβ42 generation in the early secretory compartments is gained by the observation that intracellular Aβ42 in the ER/IC is unaffected in PS-1/PS-2-deficient cells, whereas the production of secretory Aβ species is almost completely abolished (Wilson et al. 2002). The detection of secreted and intracellular Aβ-species in presenilin-deficient mouse fibroblasts provides further evidence for a γ-secretase activity unrelated to the presenilins (Armogida et al. 2001), however, these data have been doubted by others (Grimm et al. 2002; Nyabi et al. 2002).

Intraneuronal Aβ pathology in AD and Down syndrome patients

Masters et al. reported already in 1985 that amyloid is first deposited in the neuron, and later in the extracellular space (Masters et al. 1985). Recent reports have shown that intraneuronal Aβ42 accumulates in vulnerable brain regions in AD patients and might reflect an early event in the pathological process. The intraneuronal Aβ staining was most evident in pyramidal neurons of the hippocampus and entorhinal cortex, regions in which the first extracellular pathological AD alterations are evident. Interestingly, with increasing cognitive dysfunction and plaque deposition, intraneuronal Aβ42 immunoreactivity seemed to be attenuated (Gouras et al. 2000). At the same time, Mochizuki et al. reported that cells, which were immunoreactive for Aβ42 colocalize with or are adjacent to amyloid plaques in sporadic AD cases. Neither astrocytic, nor microglia markers costained the Aβ42-positive cells, which led the authors to conclude that they were non-pyramidal neurons (Mochizuki et al. 2000). In a very recent report, AD brains with different degrees of severity were studied by immunohistochemical methods, using antibodies against Aβ and paired helical filaments (PHF). Intracellular Aβ deposition was detected prior to the appearance of PHF-positive structures, indicating that it is one of the first neurodegenerative alterations in the AD brain (Fernandez-Vizarra et al. 2004).

It has further been shown that Aβ42 accumulates in discrete granules in the perikaryon of pyramidal cells, which appeared to be cathepsin-D-positive, implying that the lysosomal system is involved. Therefore, it was hypothesized that Aβ42-burdened neurons can undergo lysis, as nuclear remnants were often found at the cores of amyloid plaques (D'Andrea et al. 2001). Despite of a plethora of immunohistochemical studies on AD brains, which all showed consistent Aβ-staining in diffuse or mature plaques, there are scarce papers on intraneuronal Aβ immunoreactivity. This may be due to different techniques of immunohistochemical stainings, as it was shown that prominent intracellular Aβ42 in pyramidal neurons can be detected by heat-induced antigen retrieval but not just by enzymatic pretreatment (D'Andrea et al. 2002). Whereas treatment with formic acid is well suited for the detection of extracellular amyloid plaques, the enhancement of Aβ-immunoreactivity by heat-induced methods is more appropriate for the detection of intraneuronal Aβ (D'Andrea et al. 2003).

Besides AD patients, individuals suffering from Down syndrome displayed similar immunohistochemical findings. In a study with 14 Down syndrome patients ranging from 11 months to 56 years, all showed intraneuronal Aβ immunostaining of a vesicular pattern, which was strongest in the perikaryon, but also found in neuronal processes. Large pyramidal neurons in the hippocampus and entorhinal cortex were prominently stained even in very young patients of about one year of age (Gyure et al. 2001). This finding was corroborated in a further study in Down syndrome patients, which showed that all tested Aβ42-antibodies revealed strong intraneuronal immunoreactivity, whereas Aβ40-antibodies only detected mature plaques. It was clearly shown that immunoreactivity was consistently strong in very young patients, but declined with the deposition and maturation of extracellular Aβ plaques (Mori et al. 2002). Analyses of cultured Down syndrome cortical astrocytes, as well as Down syndrome brains, revealed intracellular Aβ42 accumulation; this was detected in Down syndrome brains throughout the neuronal cytoplasm, including primary and in some cases also secondary dendrites, as well as in the cytoplasm and processes of cortical astrocytes. This study further showed that intraneuronal Aβ42 in DS astrocytes was detected in the detergent-insoluble pellet, which suggests that Aβ is retained inside the cell and accumulates in the form of insoluble aggregates (Busciglio et al. 2002).

Intraneuronal Aβ in APP-transgenic mice

Mice transgenic for APP have been proven valuable model systems for AD research. In a few studies early pathological changes, like deficits in synaptic transmission (Hsia et al. 1999) or changes in behavior, differential glutamate responses and deficits in long-term potentiation were reported (Moechars et al. 1999). In addition, a reduced spontaneous alternation performance in a ‘Y’-maze was evident in another transgenic mouse model (Holcomb et al. 1998). In all these model systems, deficits occurred before plaque deposition became prominent and therefore may reflect early pathological changes, probably induced by intraneuronal APP/Aβ mistrafficking or intraneuronal Aβ accumulation. Alternatively, it cannot be ruled out entirely that a transgene integration effect or other by-products of APP, like C-terminal fragments, are responsible for these early phenotypic changes. In accordance with these findings we have previously shown that intraneuronal Aβ accumulation precedes plaque formation in transgenic mice expressing mutant APP695 with the Swedish, Dutch and London mutations in combination with mutant presenilin-1. These mice displayed abundant intraneuronal Aβ immunoreactivity in hippocampal and cortical pyramidal neurons (Wirths et al. 2001). This was even more pronounced in a transgenic mouse model expressing Swedish and London mutant APP751 together with mutant PS-1 (Blanchard et al. 2003). In young mice, a strong intraneuronal Aβ staining was detected in vesicular structures in somatodendritic and axonal compartments of pyramidal neurons and an attenuated neuronal immunoreactivity with increasing age. The intraneuronal immunoreactivity declined with increased plaque accumulation (Wirths et al. 2002), a finding that was also reported in Down syndrome patients, where the youngest patients displayed the strongest immunoreactivity (Mori et al. 2002). The latter mouse model is of particular interest, as it is the first model that shows substantial, age-related neuron loss in the pyramidal layer of the hippocampus, a finding that has long been proposed by the assumption that intraneuronal Aβ could be toxic. Moreover, this neuronal loss did not correlate with the amount of extracellular deposited Aβ, suggesting a novel pathological observation characterized by high levels of intraneuronal Aβ (Schmitz et al. 2004). The critical point is the ratio of Aβ42 to the total amount of Aβ, which is increased in those mouse lines with neuronal death. As neurodegeneration has been shown in mice overexpressing Aβ42 (LaFerla et al. 1995), data on substantial neuronal loss in APP-transgenic mice have been a frequent subject of scientific discussions.

Intraneuronal Aβ deposition in the hippocampus was described in transgenic mice expressing APP with the Swedish mutation (Shie et al. 2003). Takahashi et al. used immunoelectron microscopy to determine the subcellular localization of Aβ. They could show that intraneuronal Aβ42 in APP-transgenic mice accumulated predominantly in multivesicular bodies (MVBs) within presynaptic and postsynaptic compartments, as well as in human AD brains. This accumulation was associated with an altered synaptic morphology, which preceded extracellular amyloid plaque deposition (Takahashi et al. 2002). Increased intracellular Aβ peptides were reported to be associated with neuroinflammation. Treatment of APP-transgenic mice with lipopolysaccharide resulted in an altered APP processing, as well as a significant increased number of F4/80-immunoreactive microglia in the proximity of Aβ-bearing neurons (Sheng et al. 2003).

A recent report described a new triple-transgenic mouse model expressing mutant APP in combination with mutant PS-1 and mutant tau-protein. These mice displayed early synaptic dysfunction before plaque or tangle deposition was evident, together with early intraneuronal Aβ immunoreactivity preceding plaque deposition. Interestingly, tau and Aβ immunoreactivity colocalized in hippocampal neurons, implying that early intraneuronal Aβ accumulation affects tau pathology (Oddo et al. 2003). This is supported by the observation that Aβ42-specific immunoreactivity was associated with neurofibrillary tangles at an early stage in patients with AD and other neurodegenerative diseases, as well as in elderly controls (Schwab and McGeer 2000).

Aβ oligomers

There is increasing evidence that Aβ oligomers contribute significantly to the pathological alterations underlying memory failure in AD. This oligomerization was observed to occur intracellularly (Walsh et al. 2000) and Aβ42 oligomers turned out to be potent neurotoxins in neuronal organotypic cultures at nanomolar concentrations (Lambert et al. 1998). These oligomers (also referred to as ADDLs, for Aβ-derived diffusible ligands) are further able to inhibit hippocampal long-term potentiation and disrupt synaptic plasticity (Lambert et al. 1998; Walsh et al. 2002a). Using antibodies against synthetic Aβ oligomers, these ADDLs have been found in brains of AD patients in concentrations up to 70-fold higher than in control brains (Gong et al. 2003). In a very recent report it was shown that in primary neuronal cultures from APP transgenic mice, Aβ42 redistributes from the outer membrane of MVBs to the inner membranes of abnormal appearing endosomal organelles and microtubules, as it aggregates from mono- to oligomeric state. These Aβ42 oligomers were observed within abnormal processes and synaptic compartments in human AD brains (Takahashi et al. 2004). Dimeric Aβ peptides further rapidly accumulate in lipid rafts and were detected as sodium dodecyl sulfate-soluble forms in brain tissue of APP transgenic mice. This begins as early as memory impairments become obvious, providing a further link to the assumption that Aβ oligomers are tightly linked to memory dysfunction in AD (Kawarabayashi et al. 2004). Moreover, biochemical data implicate that Aβ-dimers can easily aggregate into fibrils and may have originated from APP-homodimers by proteolytic processing (Schmechel et al. 2003). The physiological relevance of the existing Aβ dimers has been demonstrated in vivo. The presence of Aβ dimers in the cortex of patients with AD has been suggested to initiate the accumulation of Aβ in the human brain (Enya et al. 1999). In APP transgenic mouse brain two forms of sodium dodecyl sulfate-stable Aβ homodimers exist, species ending at Aβ40 and Aβ42 (Schmechel et al. 2003). Spectroscopic analyses showed that engineered dimeric peptides ending at residue 42 displayed a much more pronounced β-structural transition than corresponding monomers (Schmechel et al. 2003). In contrast to Aβ42, the β-sheet content of the α- and γ-secretase-generated p3 fragments did not necessarily correlate with the tendency to form fibrils. Nonfibrillar sodium dodecyl sulfate-stable dimers have been characterized as neurotoxic species (Lambert et al. 1998) and selectively blocked hippocampal long-term potentiation in the absence of monomers, protofibrils, or fibrils (Walsh et al. 2002b).

Aβ uptake from extracellular space

Besides intraneuronal Aβ production, cellular uptake of Aβ from the environment is a second mechanism that contributes to intracellular Aβ accumulation. Selective intracellular accumulation of Aβ42 was reported in cells treated with synthetic Aβ-peptides. This internalization could be prevented under conditions that do not allow endocytosis (Knauer et al. 1992). At least some of the internalized Aβ accumulated in insoluble fractions and was resistant to degradation for several days in cell cultures (Ida et al. 1996). This accumulation and degradation resistance seemed to be specific for Aβ42, as internalized Aβ40 and shorter peptides were eliminated with a half-life of about 1 h (Burdick et al. 1997). The subcellular localization of internalized Aβ42 seemed to be the endosomal/lysosomal system, where Aβ42 induced a loss of lysosomal impermeability with concomitant membrane damage (Yang et al. 1998; Ditaranto et al. 2001). Studies of rat organotypic slice cultures, which were treated with various Aβ-peptides, revealed a selective Aβ42 internalization within CA1 subfield neurons, whereas CA3 and dentate gyrus remained almost unaffected (Bahr et al. 1998).

Intracellular Aβ and apoptosis

One of the mechanisms leading to cell death and neurodegeneration in AD has been suggested to be apoptotic cell death. It was reported that intracellular Aβ deposits correlated with cell damage and apoptotic cell death in brains of AD patients, as shown by terminal deoxynucleotidyl transferase-biotin dUTP nick-end labeling (TUNEL) staining (Chui et al. 2001). An earlier study demonstrated numerous damaged cells with apoptotic appearance that contained Aβ-like immunoreactivity, as well as intracellular accumulation of ApoE (LaFerla et al. 1997). Transgenic mice expressing Aβ42 in neurons, showed extensive neurodegeneration and substantial evidence for Aβ toxicity in vivo and apoptosis (LaFerla et al. 1995). Further evidence that intraneuronal Aβ may lead to neuronal apoptosis resulted from cell culture experiments. Adenoviral infection of rat cortical neurons with human APP lead to accumulation of intracellular Aβ42, associated with a reduced neuronal survival rate and appearance of apoptotic nuclei. Treatment with a functional γ-secretase inhibitor, strongly reduced the production of intracellular Aβ42, and was associated with a significant recovery of cell survival (Kienlen-Campard et al. 2002). Microinjections of Aβ42 into human primary neurons induced cell death, whereas the injection of Aβ40 or various antisense peptides showed no toxic effects. The addition of Bax-neutralizing antibodies, a p53 dominant-negative mutant and caspase inhibitors prevented cell death, indicating that apoptotic mechanisms are involved in the cascade leading to cytotoxicity (Zhang et al. 2002). Very recently, an essential serine protease with proapoptotic activity, named HtrA2/Omi, was identified to interact with Aβ in yeast two-hybrid assays. Co-immunoprecipitation assays further confirmed the interaction of Aβ and the protease in cell culture experiments (Park et al. 2004). Treatment of cell cultures with apoptosis inducing substances like etoposide or melphalan decreased secreted Aβ levels but led to an increase in cellular Aβ42-levels in damaged neurons. Sodium azide, which induces necrosis, did not affect intracellular Aβ42-levels (Ohyagi et al. 2000). Okadaic acid, a substance that serves as a potent inducer of apoptosis (Benito et al. 1997), significantly increases intracellular Aβ in CHO cells, as shown by Aβ sandwich ELISA (Sun et al. 2002). Moreover, APP/PS-1 double-transgenic mice with a 25% neuron loss in CA1-3 also showed rare TUNEL-positive neurons (Schmitz et al. 2004). On the other hand, a recent review discussed in great detail that very complex events may contribute to neuronal death along with possible repair mechanisms, and challenged the idea that apoptosis is the sole cause for neurotoxicity in AD (Jellinger 2001).

Intracellular Aβ and oxidative stress

Increased oxidative stress is an early event in AD that decreases with disease progression and formation of characteristic lesions (Nunomura et al. 2001). NT2 differentiated cells that were incubated with low concentrations of 4-hydroxy-2,3-nonenal, a product of lipid peroxidation, showed a significant increase of intracellular Aβ production, as well as activation of different protein kinase C isoforms (Paola et al. 2000). Induction of oxidative stress by treatment with H2O2 caused a significant increase of intracellular Aβ levels, together with a decrease in full-length APP and C-terminal APP fragment (CTF) levels, whereas APP gene expression remained unimpaired (Misonou et al. 2000). These results suggest that oxidative stress fosters intracellular Aβ accumulation due to enhanced amyloidogenic APP processing.

Mutations in AD-related genes and their impact on intracellular Aβ-levels

Mutations in APP, as well as in the presenilins, influence intracellular Aβ levels. The Swedish mutation of APP results in increased intracellular Aβ levels, which were two- to threefold higher in Swedish-APP expressing cells, compared with cells expressing the respective wild-type APP (Takeda et al. 2004). Furthermore, intracellular Aβ peptides seem to be produced by a slightly different mechanism than secreted Aβ (Martin et al. 1995). The ratio of intracellular Aβ42/Aβ40 increased markedly in cells coexpressing APP and either mutant PS-1 or PS-2, compared to the situation in APP-transfected cells expressing only wild-type presenilins. The highest intracellular Aβ42 levels were detected in cells expressing mutant PS-2 proteins (Takeda et al. 2004). This finding was corroborated by a different study, showing that several PS-2 and few PS-1 mutations lead to enhanced intracellular Aβ42-levels with a concomitant decrease in intracellular Aβ40 (Qi et al. 2003). Analysis of C99-constructs with point mutations in the proximity of the γ-cleavage site revealed only little effects on the ratio of intracellular Aβ42/Aβ40 (I45F, V50F). This is of particular interest, as it has been shown that one of these mutations (I45F) increased the ratio of secreted Aβ42/40 dramatically (Grimm et al. 2003).

The modified Aβ cascade hypothesis

Overall, there is accumulating evidence to suggest that intraneuronal Aβ42 is a major risk factor for neuron loss and a trigger for the Aβ cascade of pathological events. Extracellular Aβ deposition has long been challenged to be a correlate for the striking region specific neuron loss, like the layer two pyramidal neurons in the entorhinal cortex and the CA1 neurons in the hippocampus (reviewed in Morrison and Hof 1997). Interestingly, a link between the Aβ hypothesis and neuron loss in the hippocampus has recently been demonstrated in a bigenic mouse model expressing familial APP and PS-1 variants. While these mice do not exhibit neurofibrillary tangles, a high level of intracellular Aβ42 to Aβ ratio has been demonstrated (Blanchard et al. 2003). At 17 months of age, these mice exhibited a 25% neuron loss in CA1-3, which did not correlate with amyloid plaque load and site of amyloid deposition, suggesting intraneuronal Aβ42 toxicity. Therefore, a modified Aβ hypothesis is formulated on the basis of the recent findings in transgenic mouse models and earlier evidence from cell culture, biochemical and human genetic data (Fig. 2).

Figure 2.

Scheme showing the modified β-amyloid hypothesis, which is based on novel findings on the role of intraneuronal Aβ42. Due to ageing, Down syndrome and mutant APP, PS-1 or PS-2, intraneuronal Aβ42 levels increase and accumulate, leading to synaptic and neuronal dysfunction and degeneration in brain areas responsible for distinct memory functions. In parallel, increased secretion and deposition of Aβ42 leads to extracellular plaque formation. As Aβ plaques are dynamic structures, Aβ can also be internalized again, thereby increasing the intraneuronal pool of Aβ42, which in turn increases the neurotoxic intracellular events.


This study was supported by the Fritz Thyssen Foundation (to TAB).