Introduction
The β-amyloid (Aβ) peptide, the major component of amyloid plaques, has a crucial early role in Alzheimer’s disease (AD) pathogenesis [
58]. The membrane aspartic protease β-site amyloid precursor protein (APP) cleaving enzyme 1 (BACE1) is the major β-secretase enzyme that initiates the production of Aβ from APP [
20,
54,
66,
79]. Mutations in APP at the BACE1 cleavage site cause (K670N/M671L [
38], A673V [
8]) or prevent (A673T [
24]) AD by increasing or decreasing BACE1 cleavage of APP and Aβ production, respectively. Also, a mutation at the β′-site (E682K [
87]) causes AD by shifting BACE1 cleavage toward β-site processing of APP and Aβ generation. These mutations and other evidence strongly support therapeutic BACE1 inhibition for AD [
67]. BACE1 inhibitor drugs are in clinical trials; however,
the safety and efficacy of these drugs are unknown. BACE1 null mice have multiple neurological phenotypes indicating BACE1 inhibitor drugs could have mechanism-based toxicities [
67]. BACE1 levels are elevated in AD brain [
14,
18,
31,
80,
86] potentially necessitating high doses of BACE1 inhibitor drugs, thus increasing the risk of side effects.
Lowering/normalizing BACE1 levels in AD brain rather than direct inhibition of enzyme activity offers an alternative therapeutic approach that could avoid BACE1 inhibitor side effects. Elucidating the mechanism of BACE1 elevation in AD is essential for developing BACE1 lowering strategies.
While the mechanism of BACE1 elevation in brains of AD patients or mouse models of AD is not yet clear, recent data indicate that BACE1 levels are upregulated during stresses associated with AD risk, such as energy deprivation [
39,
68], hypoxia and stroke [
56,
71,
84], oxidative stress [
61], and traumatic brain injury [
2,
65]. A large number of molecular pathways have been proposed to increase BACE1 levels: increased caspase 3 activity leading to impaired lysosomal degradation [
26,
60], Cdk5 phosphorylation of transcription factor Stat3 [
72], altered microRNAs [
3,
12,
17,
69,
88], transcription factor HIF1α activity, [
84], elevated phosphorylation of the elongation initiation factor eIF2α [
39]. Thus, BACE1 appears to be a stress-response protein that can be regulated via diverse molecular pathways, making it challenging to identify the precise mechanism(s) involved in AD-relevant BACE1 elevation.
Insights into the mechanism of BACE1 elevation in AD have come from analysis of the localization pattern of increased BACE1 in the brains of AD patients and transgenic mouse models of amyloid pathology. Importantly, BACE1 elevation is not uniform throughout the brain, but is concentrated in presynaptic dystrophic neurites that surround amyloid plaques [
25,
86] where it potentially spurs Aβ generation and plaque growth. Close proximity to plaques is associated with presynaptic dystrophy and Aβ42 oligomers increase BACE1 levels in cultured neurons [
48,
49], thus implicating Aβ neurotoxicity in these processes. Reticulon 3 is involved in dystrophic neurite formation [
19,
52,
53], but the role of Aβ is poorly understood. We recently determined that BACE1-YFP expressed from a doxycycline-inducible transgene lacking the endogenous 5′ UTR that controls BACE1 translation accumulates around plaques in an APP transgenic mouse similar to that observed in AD [
48]. These results suggest that BACE1 elevation in AD occurs via a post-translational mechanism involving Aβ neurotoxicity that is closely associated with amyloid plaques, and does not appear to involve transcriptional or translational regulation.
Here, we show by live-cell imaging that Aβ42 oligomers cause microtubule disruption and neuritic beading. In BACE1-positive dystrophic neurites surrounding amyloid plaques of AD and the 5XFAD transgenic mouse model, tubulin isoforms are mis-localized, often forming aberrant accumulations or voids. By EM, 5XFAD dystrophic axons appear distended with multi-lamellar vesicles, but notably lack intact microtubules. This observation, together with aberrant localization of microtubule motor proteins and other neuronal proteins, and evidence of reduced lysosomal function and autophagic intermediate accumulation, suggests that microtubule-based transport is impaired in dystrophic neurites surrounding amyloid plaques. Most importantly, BACE1 and APP accumulate in peri-plaque dystrophies to very high levels and lead to increased generation of BACE1-cleaved APP products, including Aβ42 that may exacerbate plaque growth. Taken together, our results suggest that amyloid plaques cause a local toxic effect, possibly mediated by soluble Aβ42 oligomers, that generates presynaptic dystrophic neurites by disrupting microtubles and impairing transport. As a result, peri-plaque dystrophies accumulate BACE1, APP, and γ-secretase, further contributing to Aβ generation and plaque growth in a feed-forward mechanism.
Discussion
In this study, we provide evidence that Aβ mediates microtubule disruption and microtubule-based transport impairment leading to dystrophic neurite formation, BACE1 accumulation, increased BACE1 cleavage of APP and Aβ production. Primary neurons treated with Aβ42 oligomers exhibited disrupted microtubules, neuritic beading, and reduced BACE1-YFP trafficking after only a short Aβ42 exposure in vitro. In the 5XFAD mouse model of amyloid pathology, presynaptic dystrophic neurites surrounding plaques showed BACE1 accumulation that correlated with aberrant localization patterns of tubulins, microtubule motor proteins, and synaptic and cell body proteins. Tubulin was largely absent from human AD and 5XFAD BACE1-positive peri-plaque dystrophies and showed increased BACE1:tubulin ratios and inverse relationships between BACE1 and tubulin. Although 5XFAD peri-plaque dystrophic neurites accumulated the lysosomal protease cathepsin D, it largely appeared to be immature pro-cathepsin D, suggesting that impaired lysosomal maturation correlates with BACE1 accumulation. By EM, peri-plaque dystrophic axons lacked intact microtubules, consistent with reduced tubulin by immunofluorescence microscopy. Remarkably, the toxic effects of amyloid on dystrophies were highly localized to the immediate vicinity of the plaque, with neighboring axons and terminals appearing normal. Multi-photon confocal microscopy 3D-reconstructions of live 5XFAD;BACE1-YFP brain slices invariably showed physical contact between BACE1-YFP-positive dystrophies and amyloid deposits. Most importantly, peri-plaque dystrophies displayed BACE1 and APP accumulation, increased BACE1 cleavage of APP, and elevated Aβ42 generation.
Taken together, our results suggest the following working hypothesis of presynaptic dystrophic neurite formation and plaque progression (Fig. S6). During early stages of amyloid deposition, a plaque nidus forms stochastically in the parenchyma and is too small to be significantly toxic to axons and terminals. Following further Aβ addition, the plaque grows large enough to come into close proximity to a nearby axon or terminal and causes neurotoxicity derived from high local concentrations of either soluble Aβ oligomers or insoluble Aβ fibrils. An as yet undefined cascade is triggered that leads to local destabilization and/or depolymerization of microtubules in the axon; microtubules in more distant regions of the axon are unaffected. Microtubule-dependent axonal transport is impaired locally in the region of the axon nearest the plaque in which microtubules are disrupted, although transport is normal in segments further away from the plaque. Vesicles proximal and distal to the plaque undergoing anterograde and retrograde transport, respectively, detach from the ends of disrupted microtubules and begin to accumulate causing swelling and axonal dystrophy nearest the plaque. The accumulation of vesicles containing APP, BACE1, and γ-secretase leads to increased APP processing and Aβ generation in the dystrophic region of the axon and may accelerate growth and development of the nearby plaque. This in turn results in a feed-forward mechanism of increased axonal dystrophy, BACE1 and APP accumulation, and Aβ generation. Together with dysfunction associated with axonal swelling and impaired transport, this cascade is likely to exacerbate downstream neurodegeneration leading to cognitive deficits. Despite being likely of presynaptic origin, peri-plaque dystrophies lack a structural protein of active zones, bassoon, suggesting that they are not functional in synaptic transmission. Importantly, this cascade would disrupt vesicular trafficking from the soma to the terminal and vice versa. Protein turnover could be inhibited resulting in accumulation of certain proteins like BACE1, and loss of others according to the specific cellular mechanisms by which their transport and degradation are controlled. In future work, it will be important to determine the potential roles of other major players in dystrophic neurite generation, such as Reticulon 3, which has been implicated in dystrophic neurite formation [
19,
52,
53], or the recently discovered η-secretase processing of APP [
73] (see Supplementary Text, Fig. S7), in the processes that we describe here.
Our data suggest that BACE1-positive peri-plaque dystrophic neurites are axonal or terminal in origin due to the presence of the predominantly axonal protein neurofilament NF-M [
25,
74], synaptic vesicle protein synaptophysin (Fig.
6b) [
25,
85,
86] and myelination (Fig.
4), and absence of the somatodendritic protein MAP2 (Fig.
6c) [
25,
85,
86]. Moreover, endogenous BACE1 is highly concentrated in vesicles of presynaptic terminals [
25]. Consistent with our results, other APP transgenic mice have peri-plaque APP-positive dystrophies that stain for synaptophysin and the presynaptic vesicular glutamate transporter VGLUT1 but lack MAP2, and also have dystrophies that are myelinated and accumulate autophagic intermediates by EM [
50]. VGLUT1 and other presynaptic markers including growth-associated protein 43 (GAP43), glutamic acid decarboxylase 67 (GAD67), and choline acetyltransferase (ChAT) have also been shown to co-localize with BACE1 in dystrophic neurites, supporting their presynaptic/axonal origin [
85]. Although lack of BACE1 and MAP2 co-localization does not guarantee that BACE1-positive dystrophies are not dendritic, the robust co-localization of BACE1 with five presynaptic markers strongly suggests presynaptic origin. The absence of synaptic active zone scaffold protein bassoon in BACE1-positive dystrophies (Fig.
6a) could be caused by loss of normal synaptic structure. The post-synaptic density protein PSD95 is also reduced around plaques [
29], supporting this hypothesis. Thus, the majority of evidence suggests that BACE1-positive peri-plaque dystrophies are presynaptic in origin, although we cannot exclude the possibility that some BACE1-positive dystrophic neurites are post-synaptic/dendritic in nature.
There has been much debate about whether plaques are a toxic agent in Alzheimer’s disease, or a mechanism for the brain to sequester amyloid in a less harmful form. Early reports indicated that plaque load did not correlate well with cognitive impairment [
1,
59]. However, recent work has shown that cerebral amyloid deposition identified by amyloid-PET imaging in cognitively normal and mildly impaired individuals predicts conversion to AD [
22]. Furthermore, numerous studies have documented toxic effects such as mitochondrial loss [
77], oxidative damage [
78], synapse loss [
29], and spine loss [
75] in the immediate vicinity of plaques. In our study, we found that dystrophic neurites appear to be in contact with amyloid plaques at the ultrastructural level (Fig.
4), by immunofluorescence microscopy (Figs.
3,
5,
6,
9, S3, S4) and in live slices by multi-photon confocal microscopy (Fig. S5; Videos S1, S2), whereas areas of neuropil slightly further away are morphologically normal. Interestingly, a study using an antibody specific to oligomeric Aβ (NAB61) determined that a halo of oligomeric Aβ extends ~6.5 μm beyond the edge of the fibrillar Aβ core [
29]. These results suggest that plaques are sources of soluble Aβ oligomers, widely thought to be the toxic species in AD, that are in dynamic equilibrium with insoluble Aβ deposits, thus creating high local concentrations of neurotoxic Aβ.
Previous studies have reported that soluble Aβ levels in the brain are in the picomolar range [
33,
76]. However, this estimate is based on biochemical isolation of Aβ from post-mortem brain lysates and thus is the
average concentration of soluble Aβ in the brain, but it may not reflect the
local soluble Aβ concentration near the plaque. To our knowledge,
absolute soluble Aβ concentrations in the immediate vicinity of the plaque have not been quantified. Although a halo of Aβ oligomers emanating from individual plaques has been demonstrated [
29], the
absolute quantification of soluble peri-plaque Aβ concentrations was not performed in this study. Thus, we suggest that
local peri-plaque concentrations of soluble Aβ may be much higher than
average Aβ concentrations in the brain. If so, high
local soluble Aβ concentrations may cause microtubule disruption and neuritic swelling in the portion of the neurite that is very near the plaque, but parts of the neurite that are farther away would be exposed to lower soluble Aβ concentrations that may not induce neuritic dystrophy. Such a localized effect on only the peri-plaque portion of the neurite may not induce cell death processes in the distant cell body, at least not initially. This scenario is consistent with both our in vitro and in vivo results.
Although it is challenging to measure Aβ oligomer concentrations in the halo surrounding the amyloid deposit, the local concentrations of Aβ may reach very high values approaching the plaque. We used relatively high concentrations of Aβ42 oligomers (1–10 μM) in our primary neuron experiments to model the high local oligomeric Aβ42 concentration in the immediate vicinity of the amyloid deposit. It is unlikely that microtubule disruption (Figs.
1,
2) and axonal transport impairment (Fig. S2) in Aβ42-treated primary neurons were the result of cell death processes, since propidium iodide staining showed no increase in dead cells in the Aβ42-treated cultures (Fig. S1). Additionally, we previously showed that 5 days of treatment with 1 or 2 μM Aβ42 oligomers caused no significant increase of activated caspase 3 in primary neurons [
49]. Microtubule disruption may occur at concentrations lower than 1 μM Aβ42, although we have not tested this yet. However, a previous study has shown that 0.5 μM Aβ42 oligomers does not appear to disrupt microtubules in primary neurons [
7], suggesting that 1 μM Aβ42 may be a threshold concentration for microtubule disruption. The relationship between our in vitro and in vivo results still requires further investigation, but even in aged 5XFAD mice with high levels of soluble Aβ42, we observe microtubule disruption and aberrant tubulin localization only in the immediate vicinity of amyloid deposits, suggesting that high Aβ42 concentrations are necessary for these effects. When conditional expression of APP in transgenic mouse brain is reduced by over 90 % with doxycycline treatment, Aβ40 and Aβ42 levels in interstitial fluid drop by about 70 %, suggesting that while plaques are fairly stable over time, they do contribute to soluble Aβ in the brain [
13]. The hypothesis has been advanced that the toxic species of Aβ is generated in plaques, as the Aβ bound there over time becomes modified, gaining neurotoxicity [
83]. Alternatively, insoluble fibrillar Aβ in plaques could be directly neurotoxic, or indirectly induce toxicity through secondary mechanisms such as neuroinflammation. Much future research remains to precisely define the exact neurotoxic species of Aβ in AD.
The cellular and molecular mechanisms leading to amyloid-induced microtubule disruption and axonal dystrophy are not yet clear. Recent studies investigating the effects of Aβ42 oligomers on Tau mis-sorting indicate that the microtubule-severing enzyme spastin mediates microtubule breakdown in dendrites [
82]; however, a role for this mechanism in axons is unclear. It is unlikely that Tau is directly involved in the generation of dystrophic neurites, as hAPP; Tau−/− mice have the same plaque area and same percentage of plaques with dystrophic neurites as hAPP; Tau+/+ mice despite having improved survival and cognitive function [
47]. Importantly, the absence of Tau did not affect the generation of APP-positive dystrophies that, based upon our results, were also likely BACE1-positive (Fig.
9e). These results suggest that Tau has an important role in Aβ-related cognitive deficits, but Tau does not appear central to the effects of amyloid deposits on dystrophic neurite formation. It is more likely that depolymerization of microtubules affects Tau phosphorylation, mis-sorting, and aggregation.
Aβ42 oligomers have been proposed to mediate toxicity through a number of extracellular receptors such as glutamate receptors, p75 neurotrophin receptors, nicotinic acetylcholine receptors, amylin receptors, the receptor for advanced glycation end products (RAGE), insulin and IGF receptors, and others (reviewed in [
42]). Activation of these receptors leads to changes in LTP, modulation of various signaling cascades, including Jun kinase, p38, MAPKs and p53, as well as caspase activation (reviewed in [
42]), some of which can cause apoptosis and cell death. Oligomeric Aβ42 leads to synaptic dysfunction that correlates with cognitive decline, perhaps mediated through effects on NMDA receptors [
63]. One study showed that NMDA receptor antagonists and GSK3β inhibitors prevented Aβ oligomer-induced disruption of axonal trafficking, suggesting that these pathways are involved [
7].
The identity of BACE1-positive vesicles in peri-plaque dystrophic axons is not yet clear. Our previous EM analysis indicates that the dystrophies are heterogeneous, some containing smaller, clear vesicles, possibly endosomes, and others containing larger, multi-lamellar, electron-dense vesicles that may be autophagic/lysosomal intermediates [
25]. By immuno-EM, BACE1 is predominantly found in dystrophies with small, electron-translucent vesicles. Recent work indicates that large numbers of Lamp1-positive immature lysosomes accumulate around plaques, and that BACE1 co-localizes with Lamp1, suggesting BACE1 accumulation in immature lysosomes [
15]. Although we have observed Lamp1 [
25] and cathepsin D (this study) accumulation in peri-plaque dystrophies, co-localization of these lysosomal proteins with BACE1 is limited, suggesting that BACE1 trafficking to lysosomes is impaired. Additionally, our data showing elevated immature proteolytically inactive pro-cathepsin D in 5XFAD brains support the hypothesis that lysosomal maturation is impaired, which could reduce the degradation of BACE1 and cause it to accumulate. Further studies are needed to better characterize the heterogeneity of peri-plaque dystrophies, and determine its causes and consequences.
While we clearly observed Aβ42 in presynaptic dystrophic neurites around plaques, future investigation must determine whether these dystrophies are a significant source of Aβ42 for plaque growth. Local Aβ generation should be higher in plaque-rich than in plaque-poor brain regions, and in older APP transgenic mice with elevated BACE1, compared to young animals. If so, these data could indicate that stabilization of microtubules might be beneficial in slowing pathology and plaque growth, and perhaps preserve neuronal and cognitive function for the treatment of AD.