A Theoretical Analysis of the Synergy of Amyloid and Tau in Alzheimer’s Disease

INTRODUCTION

Since the initial description of Alzheimer’s disease (AD) pathology in 1906 by Alois Alzheimer, extracellular plaques and intracellular neurofibrillary tangles (NFT) have been identified as two paramount pathological features. The essential component in extracellular plaques is now understood to be contributed by amyloid-β (Aβ) proteins, which are abnormally truncated at abnormal sites. The intracellular NFT are contributed by Tau protein, which is hyperphosphorylated, truncated, and aggregated. The amyloid hypothesis suggests that Aβ elevation is the triggering factor, which in turn sequentially initializes Tau protein hyperphosphorylation, synaptic loss, and cell death. The strongest supporting evidence for this model are (1) overexpression of amyloid precursor protein (APP) in Down’s syndrome sufficiently causes AD; and (2) overexpression of human mutated APP in mice sufficiently causes neurodegeneration and impairs cognitive function even without NFT. Therefore, to cause AD, APP mutation is not necessary as long as the overall APP concentration is high, as seen in Down’s syndrome; and NFT is not necessary, at least in animal models of AD, as long as Aβ is strong. Hence, the elevation of Aβ is the most upstream pathogenic factor leading to a cascade of downstream events (Fig. 1A).

This serial model well explains many clinicopathological observations. For example, the reduction of cognitive function is most closely correlated with cell degeneration and with Tau phosphorylation and NFT. It correlates only fairly with amyloid plaques. This association makes sense because the proximal factor (cell degeneration) correlates the most, whereas the remote upstream factor (amyloid) correlates the least. In addition, if a factor other than Aβ could affect a downstream target, it would precipitate cognitive dysfunction even without the involvement of Aβ. An example of this Aβ independent factor is chronic brain injury, which induces Tau phosphorylation and NFT, resulting in dementia similar to AD but without paramount Aβ deposition. Furthermore, this serial model suggests a direct action of Aβ on Tau phosphorylation, which has been demonstrated by various in vitro studies [1, 2]. Thus, the amyloid cascade model is supported by most experimental evidence.

Recent clinical trials targeting Aβ have produced less than expected cognitive benefits, although they improved Aβ-associated biomarkers [3 –6]. However, these clinical trials did reveal that Aβ reduction halted Tau phosphorylation, as indicated by the cerebrospinal fluid (CSF) p-Tau marker. It is possible that the reduction in p-Tau is not large enough to produce cognitive benefit. However, these results may not necessarily discredit the amyloid cascade model but instead indicate that this model could be refined to emphasize the importance of Tau phosphorylation.

Another dual-pathway model has been proposed as an alternative to the amyloid cascade model [7]. It contests that Aβ and Tau phosphorylation are two independent pathogenic factors that work in parallel (Fig. 1B). These two factors converge to cause synaptic dysfunction and cell loss. This model is equally valid in explaining non-AD tauopathy. It also explains that Aβ by itself could lead to neurodegeneration without NFT, as seen in APP transgenic models. Strong evidence supporting this model is the independent origin of Aβ and NFT. Aβ plaques start to appear at multiple locations across frontal, parietal, and temporal cortices, and then they involve entorhinal, hippocampal, and other subcortical areas [8]. By contrast, NFT starts from transentorhinal to hippocampal and finally populates the cortex [9, 10]. Thus, the Aβ plaques progress in an “outside-to-inside” or “top-down” fashion, whereas NFT progress in an “inside-out” or “bottom-up” fashion. However, this parallel model did not take into account the effects of Aβ on Tau phosphorylation.

We hypothesize a refined Aβ-Tau interactive model (Fig. 1C). Our model does not address whether Aβ and Tau are mutually independent factors, but instead emphasizes that Aβ and Tau synergistically precipitate pathological changes in AD. Their interaction is essential to predict cognitive outcome and the likelihood of AD. We conducted mathematic modeling on reported CSF and imaging biomarker studies, combined with an analysis of our clinicopathological data. Furthermore, we reviewed experimental data in the medical literature that agree or disagree with our model.

MATERIALS AND METHODS

Postmortem brain tissue and CSF were obtained from the Brain and Body Donation program of the Banner Sun Health Research Institute, Sun City, Arizona. The approval for the study was provided by the Western Institutional Review Board. Written informed consent had been provided by the families of donors. All AD cases were selected as being intermediate or high probability for AD according to the Reagan criteria of the National Institute on Aging and the Alzheimer Association (National Institute on Aging–Alzheimer Association criteria [11]). We did not include samples containing pathological evidence for other neurodegenerative disorders, such as vascular dementia, Parkinson disease, dementia with Lewy bodies, frontotemporal dementia, hippocampal sclerosis, progressive supranuclear palsy, dementia lacking distinctive histologic features, multiple system atrophy, motor neuron disease with dementia, and corticobasal degeneration. Mild cognitive impairment (MCI) was diagnosed by consensus as a syndrome of cognitive impairment beyond age-adjusted norms that is not severe enough to impair daily function or fulfill clinical criteria for dementia.

These patients and cognitively normal (CN) subjects were recruited from the Arizona Alzheimer’s Disease Center (ADC). This research registry includes annual longitudinal assessments, data, and consensus diagnosis as part of the Uniform Data Set contributed to the National Alzheimer Coordinating Center, and it constitutes our discovery cohort. We have standardized patient recruitment, sample collection, and storage procedures adapted from the ADC and Alzheimer’s Disease Neuroimaging Initiative (ADNI) protocols. All assays were performed in a double-blinded manner. The last cognitive scores (Mini-Mental State Examination [MMSE], Dementia Rating Scale [DRS], Stroop Color and Word Test, and Auditory Verbal Learning Test long-term memory [AVLT-LTM]) reported in this study were obtained within the year preceding the death of patients.

Histopathologic analysis and quantification were evaluated by Dr. Thomas G. Beach, neuropathologist with the Brain and Body Donation Program of the Banner Sun Health Research Institute, Sun City, Arizona. Amyloid plaque density was evaluated on the basis of the CERAD (Consortium to Establish a Registry for Alzheimer’s Disease) algorithm (0, none; 1, sparse; 2, moderate; 3, frequent) [11]. Tangle density was also evaluated on the basis of the CERAD template (0, none; 1, sparse; 2, moderate; 3 frequent). For the frontal, temporal, parietal, and occipital lobes, a mean density score was assigned to each lobe. A global density score was averaged from the density of all four cerebral lobes.

Total Tau protein from CSF was measured using a commercial ELISA (enzyme-linked immunosorbent assay) kit (KHB0041, Life Technologies, NY). Briefly, CSF samples were pre-diluted in a 1 : 20 ratio and incubated with capturing antibody in a 96-well plate for 2 h. Then after rigorous washing with PBS for 4 times, we incubated the samples with detection antibody for 1 h and subsequently applied chromogen. The final results were read at wavelength 450 nm. Standard Tau protein with a serial dilution was measured in parallel to construct a standard curve. The CSF Tau concentration was calculated as pg/ml.

In this study, we obtained pathological data on 12 CN, 15 MCI, and 26 AD subjects. We also examined CSF samples from a subset of these subjects (8 CN, 12 MCI, and 19 AD). DRS scores obtained within one year of death were available for 8 CN, 8 MCI, and 24 AD subjects. Stroop Color and Word Test scores obtained within one year of death were available for 8 CN, 10 MCI, and 20 AD subjects. AVLT-LTM test scores obtained within one year of death were available for 9 CN, 11 MCI, and 22 AD subjects. Demographics and cognitive scores of these subjects are summarized in Table 1.

RESULTS

DISCUSSION

In parallel with our hypothesis based on mathematic modeling, several lines of evidence emerged in these in vitro experiments, strongly indicating that Aβ induces Tau phosphorylation, whereas Tau protein may enhance the production of Aβ.

When the neuroblastoma SH-SY5Y cell line was cultured in media supplemented with Aβ₄₂, Tau phosphorylation increased at the proline-rich domain (S199/S202), the repeat sequence domain (S262), and the carboxyl terminus (S422) [1]. This increase in Tau phosphorylation led to accelerated Tau aggregation [1, 2]. However, Aβ₄₀ was not able to induce Tau hyperphosphorylation or aggregation [2]. Although mouse Tau aggregate did not result in NFT, mouse Tau phosphorylation and aggregation did show a significant increase under Aβ₄₂ [21]. Injecting Aβ₄₂ into zebrafish brain also induced Tau phosphorylation [22]. Aβ₄₂-induced Tau phosphorylation is a universal phenomenon across different species. Aβ₄₂ impairs synaptic function and reduces long-term potentiation (LTP). However, this Aβ₄₂-impaired LTP was not seen in Tau (–/–) mice or in wild-type mice injected with GSK-3 inhibitor [23], which suggests that Tau phosphorylation is an essential intermediary linking Aβ₄₂ and impaired LTP. Aβ₄₂-induced Tau phosphorylation is most likely mediated by GSK-3β [24], but other kinases and phosphatase may also participate. Tau hyperphosphorylation may lead to redistribution of Tau protein from axon to somatodendritic compartment [25]. In conjunction with FYN and PSD95, the Tau protein is structurally linked to N-methyl-d-aspartate receptors and mediates Aβ-induced excitotoxicity (reviewed in [26]).

On the other hand, Tau molecules induce Aβ production. Bright et al. [27] knocked down MAPT expression using siRNA, resulting in a concurrent reduction of Tau, Aβ₄₀, and Aβ₄₂. They further identified the Tau protein as being secreted from human-induced pluripotent stem cells into the conditioned media. Notably, the secreted Tau protein (eTau) is truncated Tau411. The synthetic eTau was added into media to culture human fetal primary neurons and significantly increased Aβ₄₀ and Aβ₄₂ secretion but decreased sAPPα levels. Although not addressed in this study, it is likely that eTau affects the relative activity of α- and β-secretases.

In addition to the mutually enhanced production, existing Aβ and Tau molecules synergistically exert pathological function. APP intracellular domain (AICD) by itself induces 8% cell death, whereas Tau by itself induces 17% cell death; however, the combined effect of AICD and Tau reaches 45% [28]. This combined outcome is much higher than a simple additive effect (8% +17%  = 25%), which suggests an intrinsic cooperative biological interaction between APP and Tau. The physical interaction was seen in Hirano bodies, which is a third pathological marker for AD in addition to plaques and NFT [29]. Hirano bodies contain Tau and APP C-terminal fragments [30, 31], but do not seem to promote Tau-APP interaction and instead may trap both components and reduce their interaction [28]. The AICD-Tau interaction may be mediated by an adapter protein such as Fe65 [32]. The PTB1 domain of Fe65 physically binds to the N-terminal of the Tau protein and links to APP [32]. Clinical evidence suggests a strong correlation between Fe65 and NFT in AD [33]. The in vivo evidence of the APP-Tau interaction was supported by 3×Tg APP/PS1/TauP301L mice. Aβ clearly accelerates Tau propagation and toxicity in this AD model [34]. A caveat is that the TauP301L mutation is not seen in human AD, thus the 3×Tg mice model does not represent the majority of human AD cases.

In summary, Aβ₄₂ enhances Tau phosphorylation, truncation, and aggregation, whereas Tau further induces the production of Aβ species. Their synergistic interaction accelerates AD pathogenesis. The in vitro and in vivo studies are largely qualitative, thus difficult to unify with a quantitative model. A systemic quantitative evaluation of the interaction between Aβ₄₂ and Tau is indispensable to establishing a useful model (Fig. 1C). Nevertheless, the available in vitro data indicate a mutual interaction between Aβ₄₂ and Tau phosphorylation occurring in multiple experimental conditions.

While this theoretical analysis reconfirmed the interaction between Aβ and the Tau protein, it certainly has intrinsic limitations. For example, we did not include CSF p-Tau in this theoretical analysis for three reasons. First, p-Tau is linearly related to CSF total Tau [35], therefore from a modeling standpoint, it is not an independent factor. Second, the ratio of p-Tau/t-Tau increased 3-fold in AD compared to the cognitive normal control in some studies [36], but was unchanged in other studies [37]. These incongruent results from different studies prevent us from making a confident assumption from our modeling. Third, our model is based on the ATI, which includes only total Tau and not p-Tau. In addition, CSF p-Tau 181 is the phosphorylated site used in most clinical AD CSF biomarker studies, but the complete profile of Tau phosphorylation could be far more complicated. In fact, p-Tau species could be toxic or protective, depending on the specific phosphorylation sites [38]. Another limitation is that our clinicopathological data are from a small sample size derived from a single center. The ultimate proof of this hypothesis relies on a multicenter large population PET-imaging study with dual markers, preferably imaging plaques and NFT, simultaneously. The future drug discovery for AD may need to simultaneously target both amyloid and Tau, or else aim to reduce their interactions.