|Home | About | Journals | Submit | Contact Us | Français|
The aberrant accumulation of aggregated β-amyloid peptides (Aβ) as plaques is a hallmark of Alzheimer’s disease (AD) neuropathology and reduction of Aβ has become a leading direction of emerging experimental therapies for the disease. The mechanism(s) whereby Aβ is involved in the pathophysiology of the disease remain(s) poorly understood. Initially fibrils, and subsequently oligomers of extracellular Aβ have been viewed as the most important pathogenic form of Aβ in AD. More recently, the intraneuronal accumulation of Aβ has been described in the brain, although technical considerations and its relevance in AD have made this a controversial topic. Here we review the emerging evidence linking intraneuronal Aβ accumulation to the development of synaptic pathology and plaques in AD, and discuss the implications of intraneuronal β-amyloid for AD pathology, biology, diagnosis and therapy.
The anatomically selective and differential accumulation of Aβ plaques and neurofibrillary tangles (NFTs) are neuropathological hallmarks of AD [193, 207]. Brains of those afflicted by AD are additionally characterized by aging-related cerebral atrophy, loss of neurons, inflammation and typically also amyloid angiopathy. Although not routinely analyzed in clinical neuropathological examination of AD brains, research studies have reported that loss of synaptic markers, specifically of the pre-synaptic protein synaptophysin, is the best brain correlate of cognitive decline in AD [40, 42, 168, 184, 192]. Over the past two to three decades significant progress has been made in understanding the genetics, biology and pathology of AD [11, 44, 65, 148]. The short, hydrophobic, and up to 42/43 amino acid long Aβ peptide has taken center stage in AD research, with the majority of emerging experimental therapies directed at reduction of cerebral Aβ. Although traditionally viewed as causing only extracellular pathology, the past decade has provided increasing evidence for a critical role for the accumulation of Aβ peptides within neurons.
The Aβ peptide moved to the center of AD research when converging genetic, biological and pathological clues pointed to its importance. Specifically, all autosomal dominant mutations associated with rare familial forms of early onset AD in the amyloid precursor protein (APP) and presenilin (PS) 1 and 2 were shown to increase the proportion of Aβ42 to Aβ40 peptides. In addition, the added copy of APP in trisomy 21 was known to be associated with the early and invariable development of AD pathology in Down syndrome . Moreover, familial AD (FAD) associated with triplication of wild type APP was reported . Current evidence argues against another APP cleavage product being more important than specifically the Aβ42 isoform in AD. First, the location of FAD mutations in APP point to the role of Aβ, since they localize either to the β- or γ-cleavage sites in APP involved in Aβ generation or within the Aβ domain itself (Fig. 1). Second, changes in other APP metabolites, such as the APP intracellular domain (AICD) and APP C-terminal fragments (CTFs), are not as consistent as the increased Aβ42 to Aβ40 ratio in FAD . At the same time, the most important genetic risk factor for the development of AD is apolipoprotein E (apoE) [30, 166]. Present in humans as a combination of two apolipoprotein ε2, ε3 and/or ε4 alleles, it is specifically the ε4 allele that markedly increases the relative risk for the development of AD. Although the biological mechanism(s) of apoE’s involvement in the disease process is at a relatively early stage of understanding [17, 111], apoEε4 has been consistently associated with increased cerebral Aβ load .
Notably, the slightly longer Aβ42 species that is specifically linked with FAD is also the first Aβ species deposited in cerebral plaques . Interestingly, the earliest Aβ42 in plaques appears to be N-truncated . It is known that progressive N-truncation increases the propensity of aggregation [146, 165] and toxicity , although the most important N-terminus of highly aggregable Aβx-42 peptides in plaque formation remains unclear. In fact, a diverse set of N- and C-terminally truncated Aβ peptides exist in the brain . In addition, β-cleavage of APP normally generates both N-terminal AβAsp1 and Glu11 peptides [21, 59, 197]. Overall, data suggest that pyroglutamate Aβ3-42 [68, 157, 206] and Aβ11-42 [131, 155], among others, might be particularly important in AD.
Additional support for the importance of Aβ has come from therapeutic studies on AD transgenic mouse models. Experimental therapies, such as Aβ immunotherapy [8, 53, 138, 164, 178], were not only shown to reduce cerebral Aβ, but even more importantly, to improve behavior in transgenic mouse models of AD-like cerebral β-amyloidosis. Whether such Aβ-targeted therapies can also be effective in human AD, not consistently linked with elevated Aβ production as AD transgenic mice, remains less clear [75, 176].
At the same time, one needs to be vigilant about assigning too much of a role to one small peptide. Aging is considered the most important risk factor driving the development of AD. There are many hypotheses for aging that are also active areas of investigation in the field of AD research. Oxidative stress and mitochondrial dysfunction are considered to play major roles in aging and to promote the development of age-related neurodegenerative diseases [102, 104, 109, 144, 183]. Inflammation [128, 208] and the immune system  are also thought to modulate the disease process and are under investigation for their role in AD.
The main constituent of NFTs, the microtubule-associated protein tau, has received somewhat less attention compared to Aβ, mainly because tau pathology is less specific to AD than Aβ pathology, and because research has shown that in human AD and transgenic mouse models of AD, tau follows Aβ pathology rather than vice versa [60, 138]. Moreover, mutations in tau, rather than associated with AD, have been linked with familial forms of frontotemporal dementia (FTD; ), a less common form of dementia with a different clinical presentation and neuropathology that is not characterized by amyloid plaques. Evidence in experimental AD mouse models supports that Aβ acts via tau, particularly in promoting neurodegeneration [56, 101, 140, 151, 160].
Current data suggest a hierarchy in the development of AD with aging, that emerging evidence increasingly supports begins with declining levels of Aβ42 in cerebrospinal fluid (CSF) and accumulation of intraneuronal Aβ42, although the precise steps linking Aβ to synaptic dysfunction, cognitive decline and neurodegeneration remain to be determined.
A few intriguing early studies using newly developed antibodies directed at Aβ peptides had noted immuno-labeling of some neurons and NFTs in the brain [4, 62, 80, 115]. It is important to understand that these pan-Aβ antibodies directed at the Aβ domain detect not only Aβ peptides but also the normally much more abundant full-length APP, and additionally, other Aβ-containing APP fragments such as β-secretase cleaved APP βCTFs (Fig. 1). Hereafter, pan-Aβ antibodies will be referred to as Aβ/APP antibodies. This reactivity of Aβ/APP antibodies to APP has contributed to considerable misunderstanding with regards to the presence of intraneuronal Aβ in the field. Widely used Aβ/APP antibodies for detection of plaques should label all cells, since APP, in contrast to Aβ, is normally present at relatively high levels in cells. Since cell biological studies in the late 1980s and early 90s did not support the presence of intraneuronal Aβ, labeling of neurons and tangles with Aβ/APP antibodies was eventually ascribed to cross-reactivity with lipofuscin  or to an artifactual shared epitope with tau . Furthermore, the wide use of well-established Aβ/APP antibodies (such as 4G8 and 6E10) was optimized for plaque detection in diagnostic neuropathology, where short 3,3'-Diaminobenzidine (DAB)/peroxidase reaction times typically reveal minimal labeling next to plaques in AD brain. The subsequent development of antibodies directed at the free C-terminal ends of Aβ40 or Aβ42  led to the ability to differentiate Aβ from APP labeling, since such antibodies now did not label the Aβ domain within APP as Aβ/APP antibodies had done (Fig. 1). In contrast, while antibodies to the free Asp1 N-terminus of Aβ1-x do not cross-react with full-length APP, these antibodies (for example antibody 3D6) do still detect the 1st product of amyloidogenic APP processing, β-cleaved βCTFs, which has the same N-terminus as Aβ1-x peptides . Confocal microscopy of mouse  and human (Fig. 2) primary neurons in culture underscores the complexity and differences in subcellular labeling of APP and its various metabolites, including Aβ.
In 1993 the first biological paper appeared on the presence of Aβ within cells . Subsequent work showed that particularly the disease-linked Aβ42 peptide, rather than the more abundantly secreted Aβ40 peptide, increased with differentiation and time in culture within clonal NT2 cells that can develop a neuron-like phenotype [174, 195]. In addition, marked up-regulation of newly generated intracellular Aβ42 was reported following treatment of cells with extracellular Aβ1-42 . AD transgenic mice showed physiological [32, 74, 77, 124] and neuritic alterations [22, 125] prior to plaques, shifting the emphasis from fibrillar Aβ to soluble, and potentially intracellular, Aβ42 oligomers. Similarly, biochemical studies on human AD brain indicated that the soluble pool of Aβ42 correlated best with degree of cognitive loss [108, 118, 130]. Intriguing in vitro and in vivo studies showed that intraneuronal Aβ1-42 was particularly neurotoxic , although lack of plaque formation in intraneuronal Aβ42 accumulating transgenic mice with concomitant neurodegeneration [36, 93] limited conclusions that could be drawn to the human disease, since pathological diagnosis of AD requires Aβ plaques. An intracellular environment should be more favorable for allowing Aβ aggregation, since intracellular organelles can have a lower pH, higher metal ion concentrations and provide a limited space. These and other considerations led to the hypothesis that the intracellular pool of Aβ may be quite important in the disease process [69, 203].
Two publications appeared in January 2000 on the intraneuronal accumulation of particularly Aβ42 by light microscopy within neurons of human brain. One of these  showed Aβ42 positive non-pyramidal neurons near plaques in advanced AD brain, although on closer inspection nuclei are not evident in these neuron-like structures that more resemble NFTs. The other publication  studied intraneuronal Aβ40 and Aβ42 in Down syndrome, non-neurological controls, mild cognitive impairment (MCI) and mild, moderate and severe AD. This publication described prominent early Aβ42 labeling of AD vulnerable neuron cell bodies, but not their nuclei, in pre-plaque Down syndrome brains and also non-Down syndrome brains, particularly those with early AD neuropathological changes. This publication noted a tendency for increased intraneuronal Aβ42 labeling with increasing age in Down syndrome prior to plaques and also in normal aging; the most marked intraneuronal Aβ42 accumulation was noted with early AD plaque and tangle pathology in patients with MCI or mild dementia. Intriguingly, this and subsequent studies noted an apparent decline in intraneuronal Aβ42 labeling with advancing plaque pathology [95, 139]. Intraneuronal Aβ42 accumulation has since been reported by many groups in brain tissue of human AD [24, 38, 45, 64, 126, 141, 185, 204] and Down syndrome [19, 24, 64, 126], and virtually all transgenic mouse [47, 89, 106, 114, 137, 140, 172, 182, 188, 196, 205] and rat  models of AD.
Major support for the importance of intraneuronal Aβ came with the development of new transgenic mouse models. In particular, the triple transgenic mouse harboring mutations in APP, PS1 and tau was shown to develop prominent intraneuronal Aβ accumulation that coincided with the emergence of behavioral, physiological and synaptic abnormalities several months prior to Aβ plaques [12, 140]. Additionally, PS1/APP mutant transgenic mice were shown to have prominent intraneuronal Aβ accumulation associated with neuron cell death [23, 35, 167]. Another transgenic mouse over-expressing human APP harboring 5 different FAD mutations was shown to have intraneuronal Aβ42 fibrilization, indicated by thioflavin S staining, associated with neuronal death . Confirmation for the presence of intraneuronal Aβ was provided by a study that showed the disappearance of intraneuronal Aβ in AD transgenic mice when Aβ levels were reduced either by treatment with γ-secretase inhibitor or by crossing the mice with BACE knockout mice .
Immuno-electron microscopy (immuno-EM) was utilized to determine the subcellular localization of Aβ in neurons of the brain. Remarkably, neurons in normal wild-type mouse, rat and human brains showed localization of Aβ42 primarily associated with endosomal organelles, often identifiable on electron micrographs as late endosomal multivesicular bodies (MVBs; ). Importantly, Aβ42 labeling was shown to be absent in neurons of APP knockout mouse brains and the monoclonal Aβ42 antibody used (MBC42) was determined by Western blot to bind Aβ42, but not Aβ40, full-length APP or APP C-terminal fragments (CTFs). In addition, dual-immuno-EM further indicated that APP has a different subcellular localization (primarily Golgi apparatus) compared to Aβ42 (primarily endosomes). Endosomal Aβ42 labeling by immuno-EM has since been confirmed by other groups [49, 95, 143]. In addition, Aβ42 labeling in endosomes was reported in neurons in culture [2, 154] and brain [24, 196].
Consecutive immuno-EM for Aβ42 in aging, Aβ plaque-forming AD transgenic mice (Tg2576 harboring the FAD Swedish mutation ) during a time period known to show marked increases in brain Aβ  but prior to plaques, revealed that Aβ42 accumulated particularly in endosomes within distal axons and dendrites (neurites), and their pre- and post-synaptic compartments, where with marked accumulation it was, at times, directly associated with intraneuronal pathology . Confirmatory data were obtained in human AD brain tissue. Moreover, a subsequent immuno-EM study demonstrated that intraneuronal accumulation of higher molecular weight Aβ42 oligomers (antibody M16; ) was invariably associated with neuritic degeneration, even prior to plaques, in vulnerable distal neurites and synaptic compartments in brains of AD transgenic mice and human AD . These immuno-EM studies were the first to directly associate Aβ with synaptic and neuritic damage in the brain.
Although the presence of intraneuronal Aβ accumulation has now been widely detected in brains of human Down syndrome and AD, and transgenic models of AD, skepticism persists. A major initial concern was that intraneuronal Aβ42 labeling in the brain was an artifact. There are different aspects to this concern. Many investigators, including clinical neuropathologists, did not appreciate intraneuronal Aβ immuno-labeling in postmortem AD brains. Not seeing intraneuronal Aβ in brain can be due to many factors other than its absence. In diagnostic neuropathology, typically brief development times for DAB/peroxidase reactions are used with Aβ/APP antibodies. This leads to ”clean” plaque staining without “background”, although as noted above, one should be aware that APP, in contrast to Aβ, normally is an abundant protein in cells. Therefore, neurons should, and eventually do, stain with Aβ/APP antibodies that also detect APP, if one uses appropriate immuno-histochemical procedures, including allowing for longer reaction times.
As noted above, the use of Aβ/APP antibodies has been an issue in the acceptance of intraneuronal Aβ. Some investigators inadvertently led to confusion by using such antibodies to describe intraneuronal Aβ labeling in the brain. In the case of the triple transgenic mouse, the first transgenic mouse shown to develop both neuropathological hallmarks of the human disease, staining with an Aβ/APP antibody was compared to that of an APP C-terminal antibody with an epitope outside of the Aβ domain . Since the APP antibody did not show similar intracellular accumulation, it was logical to conclude that it was Aβ rather than APP that was accumulating. It is preferable to directly use Aβ end-specific antibodies for determination of intraneuronal Aβ, although it is important to be aware that not all C-terminal end-specific Aβ40 and Aβ42 antibodies are optimal. The quality of antibodies is of major importance, particularly for Aβ detection. For example, it is not common knowledge that commercial antibodies can change. Particularly polyclonal antibodies can have batch to batch variations that can significantly affect interpretation. A significant development that further complicated interpretation of results with Aβ antibodies was that of Aβ antibody conformational specificity . As Aβ aggregates from monomers to oligomers and then fibrils, the conformation and antibody accessibility of peptide epitopes within Aβ can change. Diverse new antibodies have been developed against different conformations of Aβ. For example, C-terminal specific Aβ42 antibodies typically are best at detecting Aβ42 monomers, and have minimal reactivity against oligomers larger than trimers . One also needs to be cautious when interpreting immunohistochemistry with antibodies against the N-terminus of Aβ1-x peptides. Although these antibodies do not cross-react with full-length APP, they do also react against the same N-terminus on APP βCTFs. N-terminal Aβ1-x/βCTF-specific antibodies also fail to label N-truncated Aβ42 peptides that appear to be particularly important in AD [60, 98, 157, 158, 206].
Thus, the above caveats further underscore that the absence of labeling does not have to reflect absence of Aβ within neurons. Detection of intraneuronal Aβ is even more challenging in human brain, where in contrast to AD transgenic mice, APP is not over-expressed and FAD mutations that elevate Aβ42 are rare. Moreover, intraneuronal Aβ is best appreciated at early stages of pathology in AD transgenic mouse and human AD brains, while postmortem human AD brains are more often examined at end stages of disease. It should be noted that there also is variability in the degree of intraneuronal Aβ labeling , which is more pronounced in human compared to AD transgenic mouse brains. AD transgenic mice allow for considerably less variability, since not only can genetically identical mice be studied, but time to, and length of, fixation can be carefully controlled. Overall, detection of intraneuronal Aβ can be challenging, and the use of complementary methods and APP knockout mice  are helpful controls. While there appeared to be an increase in intraneuronal Aβ42 labeling with aging in AD vulnerable neurons in normal human brains, individual differences in neuroanatomical involvement and relative intensities of Aβ40 compared to Aβ42 labeling were noted . For example, particularly prominent intraneuronal Aβ42 labeling was detected in the CA1 hippocampus in one young case of Down syndrome (3 year old; ) with concurrent tetralogy of Fallot (personal observations). Given the influence of, among others, head injury on APP metabolism and Aβ deposition, coincident illness and agonal state likely contribute to variability of intraneuronal Aβ labeling in human brain.
Additional concerns for the acceptance of intraneuronal Aβ are applicable to all immuno-histochemical data. Postmortem interval, pre-treatments, counter-stains, fixation and post-fixation, detergent, and blocking solutions, all can influence interpretation of immuno-histochemistry, and also that of intraneuronal Aβ [34, 39]. For a detailed protocol for Aβ immunohistochemistry and immuno-electron microscopy see Gouras and Takahashi, 2005 .
The relative dearth of recent biochemical evidence on intraneuronal Aβ has contributed in delaying the acceptance of intraneuronal Aβ. The aggregation-prone nature of Aβ makes the accurate biochemical analysis of intraneuronal Aβ particularly challenging. The limitation of a standard and widely used Aβ assay was highlighted in a study using an enzyme-linked immunosorbent assay (ELISA) . When the same amounts of Aβ1-42 were incubated over time, levels of Aβ42 assayed by a standard non-denaturing ELISA appeared to decline as a function of time, while denaturing Western blotting indicated that levels were actually unchanged; this occurred because ongoing Aβ aggregation progressively blocked access of the Aβ42 antibody in the ELISA. Furthermore, other biochemical Aβ detection methods, such as immunoprecipitation, can vastly underestimate total Aβ levels, since spiking of samples with a defined amount of Aβ1-42 reveals that only a very small fraction of the added Aβ can subsequently be detected (personal observations).
Biochemical determination of Aβ levels in brain can also show considerable variation between studies. For example, the predominance of Aβ40 compared to Aβ42 in brain can vary, with some studies showing more and others less Aβ42 than Aβ40 in normal brain [131, 191]. It is particularly challenging to determine intra- compared to extra-cellular Aβ in the brain using biochemical methods. Studies have increasingly differentiated between soluble and insoluble pools of Aβ within the brain, where it is the increase in the soluble pool of Aβ42 that tracks best with cognitive decline in AD [108, 118, 130]. Attempts have been made at further separating the soluble pool of Aβ in the brain as soluble intracellular-enriched, membrane-associated or extracellular-enriched pools , although the feasibility of separating these pools of Aβ after brain homogenization, which breaks apart cell membranes, is questionable. It is intriguing to note that a recent study indicated that cognitive decline in AD tracked best with the membranous pool of Aβ42 in brain tissue from human AD ; it was concluded that this membrane-associated Aβ42 was cell associated and reflected intraneuronal Aβ42. Very interesting recent biochemical studies used laser capture microdissection combined with sensitive Aβ ELISA to demonstrate elevated levels of Aβ42 and an increased Aβ42 to Aβ40 ratio in both AD vulnerable CA1 hippocampal pyramidal and relatively AD resistant cerebellar Purkinje neurons in both familial and sporadic AD compared to controls, although the amount of Aβ42 was markedly higher in CA1 pyramidal compared to cerebellar Purkinje neurons in AD [5, 70].
Another challenge for the acceptance of intraneuronal Aβ accumulation were subsequent reports for different subcellular locations of accumulating Aβ than endosomes, which included the plasma membrane [136, 209] and mitochondria . In particular Aβ localization to mitochondria has received considerable recent attention. There is a large group of investigators who have for years emphasized the importance of energy metabolism, oxidative stress and mitochondria in aging and AD. Thus, the localization of Aβ to mitochondria was an exciting new link between their compelling prior work on the role of mitochondria in aging and AD with the peptide most linked with AD, and questions of how Aβ might traffic to mitochondria were of less concern. APP trafficking and processing are not typically thought to involve mitochondria [159, 175, 194]. Overall, the data for the early accumulation of Aβ in mitochondria appear weaker compared to that for endosomes. In fact, the initial study on Aβ localization to mitochondrial  looked at much older AD transgenic mice with abundant plaques compared to studies on Aβ in endosomes . Nevertheless, Aβ localization to mitochondria may still be quite important in AD pathogenesis. Thus, while aberrant Aβ42 accumulation begins in endosomes, mitochondrial Aβ may be a critical contributor in the pathological cascade leading to neuronal dysfunction and pathology in AD [33, 103, 149]. In fact, we have noted the presence, at times, of Aβ-accumulating endosomes adjacent to mitochondria by immuno-EM (personal observations), which could suggest a link between endosomal and mitochondrial Aβ . At the same time, the role of mitochondria and oxidative stress in AD might be even more important upstream, rather than downstream, of Aβ, since aging is the most important risk factor for Aβ accumulation and AD. Indeed, elevated oxidative stress was reported to promote Aβ plaques in the brains of AD transgenic mice .
It is widely thought that FAD mutations in APP or the presenilins cause increased secretion of Aβ, although the common denominator among FAD mutations is not an increase in Aβ secretion, but an increase in the Aβ42 to Aβ40 ratio . It is less widely known that FAD mutations actually cause a relatively greater increase in intracellular compared to secreted Aβ (, and unpublished observations). FAD mutations can even decrease secretion of the most abundantly released Aβ isoform, Aβ40, while still increasing the Aβ42 to Aβ40 ratio [90, 134, 173]. Emerging data from AD transgenic mice support that Aβ40 might even be protective against cerebral amyloid . The Arctic FAD mutation within the mid-region of Aβ  primarily increase the propensity for Aβ aggregation rather than secretion. In fact, FAD transgenic mouse models with the highest level of amyloidogenic Aβ display particularly prominent early intraneuronal Aβ accumulation [89, 106, 196]. Remarkably, a recent E693Delta mutation in APP was reported to be associated with an AD clinical phenotype in the absence of plaques; cells harboring this mutation were shown to have reduced Aβ secretion and increased intracellular Aβ accumulation and oligomerization .
A concern that has been raised is that even if intraneuronal Aβ42 accumulation does occur, it may be irrelevant to the disease. Emerging data, such as FAD mutations noted above, and the direct association of intraneuronal Aβ42 with subcellular pathology [187, 188], cell death [23, 167], and behavioral and physiological abnormalities, even prior to plaques, in AD transgenic mouse models , are providing increasing evidence for the importance of intraneuronal Aβ. On the other hand, the apparent decrease in intraneuronal Aβ42 with advancing plaque pathology noted in several of the initial reports on intraneuronal Aβ accumulation [60, 64, 140, 205], might suggest that while intraneuronal Aβ accumulation is an early event, it unlikely is the driving force in AD pathology, since in that case it should be progressive and correlate with cognitive decline. One possible explanation for this concern is that Aβ aggregation leads to conformational changes and thereby the reduced ability of Aβ antibodies in detecting their antigens. In fact, as noted above, standard Aβ42 C-terminal end-specific antibodies typically are only optimal at detecting Aβ42 monomers, and then lose their reactivity after dimers and trimers . Another challenge is that the most marked accumulation of intraneuronal Aβ42 occurs in distal neurites and synaptic compartments as determined by immuno-EM  that cannot be visualized effectively by light or standard immuno-fluorescence microscopy. It remains unclear whether and, if indeed the case, why cell body Aβ42 by light microscopy seems to decline with AD progression in AD vulnerable neurons such as CA1 pyramidal neurons, while by immuno-EM it continues to increase particularly within neurites and synapses . It could, among other possibilities, be a combination of loss of Aβ epitopes in cell bodies as Aβ42 aggregates, increased synaptic Aβ as cell body Aβ declines, and even increased ability to degrade cell body compared to neurite-associated Aβ42.
Substantial questions remain for the role of intraneuronal Aβ, and also for Aβ in general, in AD pathogenesis. Notably, a subset of patients from the aborted Wyeth/ELAN active Aβ vaccine trial that were studied at autopsy showed Aβ plaque removal in the absence of cognitive improvement . This study has been highlighted as evidence against the importance of Aβ in AD. Additionally, it remains unclear whether the decline in CSF Aβ42 seen early in the conversion from normal to MCI and AD [48, 66], continues with progressive cognitive decline in AD. At the same time, increases in soluble Aβ42 levels in the brain have been shown to track with decline in AD [108, 118, 130]. The combination of increasing soluble Aβ42 in brain, declining CSF Aβ42, and immuno-EM evidence of concomitant intraneuronal Aβ42 accumulation, seem to strongly suggest that the soluble pool of Aβ42 in the brain that correlates with disease progression mainly reflects intraneuronal Aβ42. We note that one of the biochemical studies on soluble Aβ42 levels correlating with AD progression  studied the same postmortem human brains analyzed in the initial immuno-histochemical study on intraneuronal Aβ42 accumulation . It will be important for future clinical-pathological studies to investigate for a correlation of biochemical and immuno-histochemical data on intraneuronal Aβ with cognitive decline. Additional correlation with levels of CSF Aβ42 will also be important, particularly in individual AD patients over time, since population based comparisons might fail to appreciate progressive changes, because of, for example, person-to-person differences in baseline levels of CSF Aβ42.
A central question is why intraneuronal Aβ42 increases in the first place. A more precise understanding of how aging and other injurious processes, such as cardiovascular disease or traumatic brain injury, can modulate Aβ levels will be important. The mechanism whereby intraneuronal Aβ42 first adversely affects the biology of neurons is a central unanswered question. At the most basic level it seems that Aβ42 could alter the normal function of endosomes near synapses. In one study, Aβ-induced abnormalities in MVB endocytic sorting were demonstrated in APP mutant neurons, which appeared to be due to impairment of the ubiquitin proteasome system (UPS; ). The role of extracellular Aβ on intracellular Aβ accumulation and plaque formation also remains unclear, as well as what proportion of intraneuronal Aβ is derived from the inside compared to the outside of neurons.
Since accumulation of Aβ42 within AD vulnerable neurons has been associated with physiological, pathological and behavioral abnormalities even prior to plaques [12, 140, 187, 188], the evidence is mounting that intraneuronal Aβ42 is the driving force in Aβ-related AD pathogenesis [9, 58, 92]. Cumulative pathological data on intraneuronal Aβ accumulation support a scenario of increased Aβ42 in vulnerable distal neurites and synapses leading to their destruction from within (Fig. 3). The subsequent formation of more amorphous plaques reflect remnants of this destruction further shaped by activated inflammatory cells. Reactive inflammatory cells and plaques may have both protective and detrimental roles that contribute to AD pathogenesis. It is possible that the localized and suddenly high levels of extracellular Aβ42 following neuritic degeneration from within can then propagate Aβ pathology to neighboring neurites via a prion-like ability of extracellular Aβ42  to up-regulate intracellular Aβ42 [51, 58]. The lack of correlation between Aβ plaques and cognitive decline in AD may relate to person-to-person differences in the ability of inflammatory cells to effectively remove Aβ plaques from the brain. This scenario of Aβ-related pathogenesis makes the cell biological study of the molecular and biological mechanism(s) whereby intraneuronal Aβ accumulation leads to synaptic dysfunction of central importance. Intraneuronal Aβ accumulation can also illuminate many ongoing questions and paradoxes in the field.
The implications for intraneuronal Aβ in AD include providing new insights into a link between Aβ and tau pathologies that were formerly considered to be separate and how these two pathologies relate to synapse pathology . By beginning within neurons, Aβ is more directly accessible to altering tau, the principle component of the other neuropathological hallmark of AD. Specifically, immuno-histochemical data support that neuron cell bodies with early intraneuronal Aβ42 accumulation are those that go on to develop NFTs [60, 140], while their terminal fields are sites of early plaque pathology. An example of this is the early accumulation of Aβ42 within layer II neuron cell bodies of the entorhinal cortex and early plaques at their axonal terminal fields in the outer molecular layer of the dentate in the hippocampus . A more recent study investigated the subcellular link between intracellular Aβ and tau pathologies . It had been known that even in APP mutant transgenic mice hyperphosphorylated tau can be seen within some dystrophic neurites around plaques . Since Aβ42 accumulates within dystrophic neuritis , Aβ42 and phospho-tau were examined by dual-immunofluoerscence and -EM in FAD mutant transgenic mouse and human AD brains. Remarkably, dystrophic neurites around plaques showed co-localization of phospho-tau with Aβ42 in AD transgenic mice and human AD . Since the triple transgenic mouse expressing mutant APP, PS1 and tau develops early Aβ accumulation in CA1 hippocampal neurons and later tau paired helical filaments , the stratum lacunosum-moleculare (SLM) of hippocampus containing the CA1 distal apical dendrites was selectively examined by immuno-EM. Remarkably, the SLM showed not only prominent early Aβ42 increases but also the earliest labeling with antibodies against phospho-tau, such as the well-established antibody AT8. By dual-immuno-EM phospho-tau was specifically evident in markedly Aβ42 accumulating, distal CA1 apical dendrites and their post-synaptic compartments in the SLM . Thus, tau, normally localized to axons, showed early mislocalization specifically in aberrant Aβ42 accumulating distal dendrites. These data build on previous studies to provide an early temporal and spatial relationship of the main constituents of the two neuropathological hallmarks of AD and thereby also link them with synapse pathology in AD.
Synapses have long been viewed as critical sites of AD pathology [6, 28, 112, 119, 168] and diverse evidence had supported that synapses are major sites of Aβ pathology and also generation. Lesioning of the perforant pathway originating in the early AD prone entorhinal cortex leads to reduced plaque pathology at terminal synaptic fields in the hippocampus [96, 171]. In contrast, lesioning of another projection pathway, that from the locus coeruleus, was reported to increase plaque pathology in terminal fields of the cerebral cortex . APP is transported down both axons and dendrites and Aβ appears to be generated preferentially at synapses [25, 54, 85]. In fact synaptic activity was recently shown to promote APP transport down neurites to synapses . Furthermore, extracellular Aβ1-42 oligomers preferentially target synapses when added to neurons [43, 91]. Immuno-EM for PS1 localized this critical component of the γ-secretase complex particularly to endosomes in distal neurites, including in both pre- and post-synaptic compartments . Indirect support for synapse-specific Aβ generation also comes from PS1 conditional knockout mice, which showed aberrant accumulation of APP βCTFs in distal axons/pre-synaptic compartments  and also some anatomically selective accumulation in distal dendrites/post-synaptic compartments (personal observations).
Overall, emerging work on intraneuronal Aβ is most consistent with the scenario of aging, and other genetic and environmental risk factors, such as apoEε4 and head injury, respectively, leading to up-regulation and then oligomerization of intraneuronal Aβ42 in distal neurites and synapses, which then leads directly to destruction of neurites and synapses from within (Fig. 3). Immuno-EM images of completely abnormal Aβ42 accumulating dystrophic neurites [187, 188], lacking the normal microtubule tracks required for organelle trafficking, secretion and exocytosis, argue against the active secretion of aggregating Aβ42 from neuritis prior to the dissolution of the plasma membrane . Thus, the nidus of Aβ plaques develops from within neurites at a time course that could also fit that of plaque formation determined by multiphoton microscopy (MPM) occurring within a day . Remarkably, this MPM study of AD transgenic mouse brain in vivo also described the emergence of dystrophic neurites in areas distant from plaques that appeared and, at times, even disappeared, although intraneuronal Aβ was not examined in these neurites. Plaque development in terminal processes can help explain the lack of neuron cell death in most AD transgenic mice in the presence of considerable plaque pathology, since neuron cell bodies can survive following loss of some neurites and synapses. At the same time, a subset of plaques appear to emerge directly from intracellular Aβ accumulating neuron cell bodies [38, 60]. While Aβ immuno-EM provides evidence for all forms of plaques, including diffuse plaques, representing destruction of neurites, there are several possible reasons for why degree of plaque pathology is not as good a measure of cognitive dysfunction compared to soluble, and we hypothesize intraneuronal, Aβ42. Degree of intraneuronal Aβ42 might tightly correlate with synaptic dysfunction, whereas synapse-independent factors might dissociate plaque burden from cognitive decline, for example, genetic heterogeneity may exist in the ability of inflammatory cells to remove plaques.
Intraneuronal Aβ can provide new insights into head trauma as a risk factor for AD [116, 127]. Individuals, including children, who died following coma not long after severe head trauma were shown to develop amorphous cerebral Aβ plaques . At first this seems difficult to reconcile with recent data showing that patients in coma have marked reductions in extracellular Aβ42, which rises and falls with cognitive improvement and decline, respectively . We have observed marked up-regulation of intraneuronal Aβ42 in surrounding neurites and neuron cell bodies near needle track injuries in the rodent brains 48 to 72 hours following brain injury (Fig. 4a). By immuno-EM these increases in Aβ42 occurred predominantly in endosomes (Fig. 4b). Thus, head trauma provides an example of extracellular–appearing amyloid developing in the setting of what has been shown to be a state of reduced interstitial, extracellular Aβ and, as experimental evidence now indicates, concomitant intraneuronal Aβ42 accumulation. In addition, dementia pugilistica that develops in professional boxers , while characterized by significant tauopathy and comparably less amyloid pathology, limited mainly to diffuse plaques , also shows considerable intraneuronal Aβ42 (Fig. 4c).
Many questions remain for the role of intraneuronal Aβ in AD pathogenesis. It remains unclear whether and how Aβ accumulation and pathology might differ in axons and pre-synaptic compartments compared to dendrites and post-synaptic compartments. Overall, more prominent increases in Aβ42 accumulation were observed in distal dendrites compared to axons , although there are anatomical differences, and it remains possible that either axons or dendrites are the initial site of intraneuronal Aβ pathology. Immuno-histochemical and –EM observations support that both distal and proximal dendritic arborizations are important sites of Aβ42 accumulation and plaque formation (Fig. 3).
What determinants are involved in the selective vulnerability of certain neurons and their processes to Aβ42 accumulation is another unanswered question. Differential anatomical levels of secretases  and chronic sites of synaptic activity  may be important in the selective vulnerability of AD. The cause of differences in plaque morphology and the involvement of inflammatory cells in plaque formation are also not fully clear . In particular dense core plaques appear to be derived from neuron cell bodies, although even here it is possible that dendritic Aβ42 accumulations are the inciting event. A recent report on in vitro plaque formation in cultured microglia  suggests the possibility that similar intracellular Aβ accumulation and pathology as described in neurons might also occur in inflammatory cells in AD and thereby further contribute to plaque formation.
The involvement of the more abundantly secreted Aβ40 in intraneuronal Aβ accumulation and plaque formation is less clear. Although not apparently present in early Aβ42 plaques as determined in Down syndrome , Aβ40 typically is the predominant peptide in amyloid angiopathy. An intriguing study showed that shifting the ratio of Aβ42 to Aβ40 in APP mutant transgenic mice under a neuron-specific promoter shifted the preponderance of parenychmal Aβ42 plaques to vascular Aβ40 amyloid . This supports that amyloid angiopathy derives from perivascular drainage of Aβ40 generated by neurons [61, 200]. Likewise, more Aβ42 over-producing mice, by retaining Aβ42 within neurites, can therefore be hypothesized to have less amyloid angiopathy.
Future comprehensive clinical-pathological studies that carefully compare and examine various intraneuronal Aβ40 and Aβ42 isoforms in different brain regions from cases with normal aging, mild cognitive impairment and different stages of AD, according to apoE genotype and intercurrent illness, and also in relation to neuropsychological testing, positron emission tomography (PET) amyloid and functional brain imaging, and CSF Aβ levels, will be important to help clarify the role of intraneuronal Aβ accumulation in AD .
Aberrant accumulation of Aβ within neurons in AD underscores the importance of uncovering the cellular and molecular mechanisms by which this pool of Aβ leads to synaptic dysfunction. That this pathological accumulation is most prominent in distal neurites and synapses of AD transgenic mice provided the first direct association between Aβ and synapse pathologies in the brain . Since accumulating Aβ42 localizes particularly to the outer limiting membrane of endosomes, and endosomes are known to reside near synapses , altered function of these organelles  might be important for Aβ-induced synaptic dysfunction. Fortuitously, it was possible to model intraneuronal Aβ accumulation in cultured primary neurons derived from AD transgenic mice . Comparing AD transgenic to wild-type neurons allowed for biological studies to begin elucidating early intraneuronal Aβ-related abnormalities in neurons. Remarkably, APP mutant neurons showed differential alterations in synaptic proteins  that resembled those that have been described in AD brain, such as early loss of the pre-synaptic protein synaptophysin [113, 163, 192] and the post-synaptic protein PSD-95 . It is of interest that among the earliest and most significantly Aβ-altered synaptic proteins are AMPA glutamate receptor subunits [3, 79], that are particularly important for synaptic plasticity.
In an effort to elucidate dysfunction of Aβ-accumulating endosomes, a dynamic cell biological approach was used to compare APP mutant to wild-type neurons . Using the well established endocytosis of the epidermal growth factor receptor (EGFR), this study demonstrated that specifically the MVB sorting pathway, and not internalization or the recycling pathway, was altered in Aβ accumulating neurons. The study went on to provide evidence that abnormalities in MVB sorting were due to Aβ-dependent alterations in the ubiquitin-proteasome system (UPS). The UPS is increasingly being linked to neurodegenerative diseases of aging . Best known for its role in degrading cytosolic proteins, the UPS also has important roles in synaptic plasticity  and lysosome-dependent degradation of trans-membrane proteins .
An area of considerable interest has been how extracellular Aβ alters synapses. Many in vitro and in vivo studies have shown that exogenous application of Aβ alters synaptic plasticity [170, 198] and important synaptic proteins [3, 79, 177], although how precisely this occurs remains unclear. Remarkably, it was shown that extracellular Aβ1-42 was not toxic to APP knockout neurons  and that even point mutations in the NPXY motif in the C-terminus of APP block toxicity . In addition, alterations of extracellular Aβ on selective synaptic proteins mirrored those from intraneuronal Aβ accumulation . Recently, additional evidence pointed to the importance of amyloidogenic processing of APP, and therefore potentially intraneuronal Aβ, in extracellular Aβ-induced synapse dysfunction . Work in the 1990s had shown that extracellular Aβ1-42 led to marked up-regulation of endogenously generated Aβ42 [51, 211], which also occurs in neurons . Although it is thought that only a small amount of extracellular can be internalized into neurons [156, 211], Aβ42 appears to a have a prion-like ability of transferring an altered aggregation state from the outside to the inside of cells . It is possible that small amounts of extracellular Aβ are internalized into neurons via APP at synapses, where extracellular Aβ preferentially binds  and APP is enriched, particularly with synaptic activation .
We are at an early stage of understanding the molecular mechanism(s) whereby accumulating Aβ42 can alter normal synaptic function. Many questions remain next to figuring out the critical first step in intraneuronal Aβ-induced synaptic dysfunction. Elucidating the normal function of APP [199, 212] and probably even of Aβ  are additionally of major interest.
Recently it is becoming apparent that the earliest diagnostic change indicating that an individual without known cognitive deficits will go on to decline cognitively and develop AD is a drop in levels of Aβ42 in the CSF [15, 48]. That the best predictor for conversion to MCI and then AD is a decrease in the levels of extracellular Aβ42 might have seemed surprising to proponents of the prevalent extracellular Aβ cascade hypothesis. The explanation put forward to keep these data in line with the extracellular Aβ hypothesis was that high levels of extracellular Aβ lead to progressive oligomerization and eventually plaque formation, with plaques then sequestering secreted Aβ. There is a problem with this argument. It seems the extracellular Aβ hypothesis wants it both ways. On one hand, toxic extracellular Aβ oligomers are thought to be released from plaques to damage nearby, although still physically removed, neurites, while on the other hand, extracellular Aβ levels are supposed to be reduced because of increased sequestration to plaques. A more parsimonious and novel explanation, that also is in line with the Aβ42 immuno-EM neuropathological data in brain and new insights relating to head injury described above, is that extracellular Aβ42 declines because secretion of Aβ42 is reduced as it accumulates instead within neurons.
At the same time, postmortem neuropathological studies and PET amyloid imaging indicate that individuals without known cognitive dysfunction can nevertheless have considerable plaque pathology. This has been used to argue that Aβ and plaques are not themselves important for neurodegeneration . One needs to be cautious with this interpretation, since considerable brain pathology can occur without detectable cognitive deficits. For example, relatively large strokes or tumors in strategic locations can fail to show deficits on bedside neurological examination. In fact, it is well known that most elderly patients will have signs of microvascular ischemic changes/damage on brain magnetic resonance imaging (MRI) even in the absence of known neurological deficits. Thus, microscopic Aβ plaque pathology should also be able to go unnoticed, at least for a while. Furthermore, a recent imaging study supports that the presence of plaques by Pittsburgh compound B (PiB) PET amyloid imaging even in those diagnosed as cognitively normal does reflect deficits on functional imaging . It needs to be stressed again that by immuno-EM high molecular weight Aβ42 oligomers are consistently associated with pathology within neurites even prior to plaques . Intraneuronal Aβ42 accumulation therefore clearly can be detrimental already prior to plaques and obvious cognitive decline. One can even speculate that intracellular Aβ contributes to the well-known normal age-related decline in cognitive function that is accounted for by declining normative scores with age in neuropsychological testing. Similarly, intraneuronal Aβ accumulation with advancing age may be important for what is termed benign forgetfulness of aging. The triple transgenic mouse supports the notion that elevated intraneuronal Aβ can cause cognitive decline prior to plaques and tangles .
Most clinical trials now in progress, and even more that are planned, are targeted at reduction of brain Aβ. Pharmacological inhibition of Aβ generation by inhibiting β- or γ-secretase has turned out to be more challenging than was originally hoped [13, 41]. Although reduction of the neuropathological hallmark and peptide most linked with the disease, Aβ plaques and Aβ peptides, is reasonable, a misunderstanding into the pathological role of Aβ can have negative repercussions. For example, screening for small molecule chemical inhibitors of specifically Aβ secretion could select compounds that are ineffective, and potentially even detrimental, for AD, since Aβ secretion may not be harmful. The neuropathological and biological evidence on intraneuronal Aβ accumulation increasingly suggest that this is the critical pool of Aβ that needs to be contained. A recent neurobiological study showed that it was reduction of intraneuronal rather than extracellular Aβ that correlated with synaptic improvement . Remarkably, it was demonstrated that Aβ antibodies can reduce the intraneuronal pool of Aβ in vivo  and in vitro , which for the latter was shown to required Aβ antibody internalization. It is possible that the Wyeth/ELAN Aβ vaccine trial failed precisely because it was not able to effectively reduce this critical pool of intraneuronal Aβ. One can hypothesize that the abundant Aβ in cerebral plaques might have sequestered the Aβ antibodies and thereby prevented removal of the important intraneuronal pool of Aβ. The concept that extracellular antibodies can be internalized and specifically alter intracellular proteins is not fully appreciated in clinical medicine and has therapeutic implications beyond AD.
Intraneuronal Aβ can also illuminate a major current paradox in the field regarding the role of synaptic activity in AD. Synaptic activity has been shown to increase the secretion of Aβ in hippocampal slices and living rodent brain [26, 85]. Furthermore, Aβ secretion is also increased with cognitive activity, since patients in coma showed elevations in Aβ with cognitive improvement; a subsequent decline in cognitive function correlated with decline in interstitial/extracellular Aβ levels . Hyperexcitability is present in AD brain  and epilepsy was shown to promote plaque pathology . Accordingly, dampening of synaptic activity is under consideration for the treatment of AD. The Aβ/synaptic activity paradox arises when one considers that cognitive activity is encouraged as potentially protective for patients with MCI and AD, and environmental enrichment was shown to reduce plaques in AD transgenic mice [55, 97]. Moreover, synaptic activity was recently shown to reduce the intraneuronal pool of Aβ in vitro and in vivo . Furthermore, functional brain imaging shows reduced resting brain activity in subjects with AD, and even in genetically at-risk subjects decades prior to predicted cognitive symptoms . Thus, reduced cerebral activity with AD should imply reduced Aβ secretion in AD brain. The more widespread cortical activity seen by functional brain imaging in AD patients performing a specific task  could be used to argue for anatomical spread of AD along these now more hyperactive cortical areas then secreting more Aβ with disease progression. Nevertheless, such task-specific spread in cortical activation is seen as a compensatory response with other brain injuries, and patients with AD would still be expected to have reduced cerebral activity most of the time. Another explanation for the potential Aβ/synaptic activity paradox is that sites of high cerebral activity, in the default system, are indeed sites of early AD neuropathology [18, 179], but not necessarily because of increased Aβ secretion. Rather these sites of high metabolic activity also generate high levels of reactive oxygen species that might contribute to a progressively decreased ability to effectively degrade intraneuronal Aβ42 with aging. The protease shown to most efficiently degrade Aβ, neprilysin , was reported to be critical in synaptic activity-induced intraneuronal Aβ42 reduction . Given that neprilysin localizes to synapses and declines with aging [20, 50], declining neprilysin may be important in the age-related intraneuronal accumulation of Aβ in synapses.
Reviews on AD have for years typically started with the statement: “Alzheimer’s disease is characterized by extracellular β-amyloid plaques and intracellular neurofibrillary tangles”. Emerging evidence on intraneuronal Aβ strongly argues that this needs to change. Similar to tau aggregation developing within neurons but then as ghost tangles no longer within living neurons, Aβ accumulation also begins intracellularly and ends up extracellular following neuritic degeneration. The intraneuronal origin of Aβ plaques makes the pursuit of the neurobiology of Aβ within neurons all the more important. AD is a tremendously complex disease that incorporates some of the most intricate processes that we know of: aging, memory and synapses. Finding more effective therapy for AD sooner may hinge on over-coming, but not just replacing, our own obstacles and dogmas, and pressing on with work to elucidate the molecular neuropathology and neurobiology of this most common neurodegenerative disease of aging.
The support of National Institutes of Health grants AG028174 and AG027140, and Alzheimer’s Association New Investigator award (D.T.) and Zenith award (G.K.G.). We thank Xun (Julie) Lian for helpful technical assistance.
Conflict of interest statement: The authors declare that they have no commercial or financial relationship that could be construed as a potential conflict of interest.