What Controls Tau Spread in Alzheimer's Disease: Plaques, Brain Circuits, or Nearby Cells?

Brains of AD patients exhibit characteristic atrophy of the parahippocampal gyrus (including the entorhinal and transentorhinal areas) and of the hippocampus. In advanced stages, diffuse cortical atrophy becomes evident (Figure 1). Microscopically, AD pathology is characterized by a combination of neurofibrillary tangles (NFTs) and senile plaques. The current diagnostic criteria require the severity of both NFTs and Aβ deposits to be above cutoff, and for Aβ, the presence of neuritic plaques on silver impregnation or thioflavin is mandatory [16]. This combined assessment of NFT and senile plaques arises from the fact that neither NFTs nor senile plaques alone can reliably distinguish AD from normal aging. Although Braak and colleagues established immunohistochemical staging of NFTs on the basis of a large autopsy sample size, their study did not include clinicopathological correlations [17]. Similarly, while the burden of senile plaques correlates with AD pathology, it often lacks a distinct boundary separating AD from age-related changes [18].

NFTs are flame-shaped, filamentous intraneuronal inclusions which are strongly labeled with silver impregnation and thioflavin (Figure 2). NFTs are dense aggregations of hyperphosphorylated tau that are composed of fibrils, ultrastructurally termed paired-helical filaments [19]. Early-stage lesions often exhibit diffuse and amorphous tau immunopositivity in the neuronal cytoplasm but lack a defined tangle structure or argyrophilicity, known as "pre-tangles". Pre-tangles are considered to represent the early phase of tau aggregation, dominated by tau oligomers rather than mature fibrils [20,21] (Figure 2). Tau also aggregates within the neurites, forming neuropil threads, which can be identified in macroscopic observations of tau-immunostained preparations in severely affected regions (Figure 2).

Tau is encoded by the microtubule-associated protein tau (MAPT) gene, and alternative splicing of exons 2, 3, and 10 yields six isoforms of tau [22]. These tau isoforms are largely classified into two groups: three-repeat (3R) isoforms lacking exon 10, and four-repeat (4R) isoforms containing exon 10 [23]. Both 3R and 4R isoforms are present in the NFTs of AD; pure 4R aggregates are observed in progressive supranuclear palsy (PSP) and corticobasal degeneration (CBD), whereas pure three-repeat aggregates are typical of Pick body disease [24,25]. The hierarchical distribution of tau pathology progression, as shown by Braak et al., demonstrates initial involvement of the transentorhinal area of the parahippocampal gyrus, and spreads to the entorhinal area and hippocampus (Braak stages I–II), then lateral and superior temporal gyri (stages III–IV), followed by widespread neocortices (stages V–VI) [12,17].

Senile plaque is characterized by the deposition of Aβ in the neuropil, in the form of extracellular deposits that presumably interact closely with the neuronal and glial membranes [26]. Although the classification of senile plaques has been complicated because of methodological differences (e.g., anti-Aβ immunohistochemistry, thioflavin, or silver impregnations), anti-Aβ immunostaining is largely classified into two major types: neuritic plaques and diffuse plaques (Figure 3). Neuritic plaques comprise a dense amyloid core surrounded by a fuzzy halo. Argyrophilic and thioflavin-positive dystrophic neurites form a crown around the amyloid core, and some of them contain tau aggregates. The core and crown can be identified by hematoxylin-eosin (H&E) staining. In contrast, diffuse plaques are fuzzy deposits, lacking distinct core and crown structures. Diffuse plaques are hardly detectable on H&E staining and show weak or no positivity on silver impregnation, although the Campbell–Switzer method recognizes diffuse plaques well. Familial AD associated with presenilin-1 (PSEN1) or amyloid precursor protein (APP) mutations may also contain cotton wool plaques that are characterized with a clear boundary but lack core/crown structures. Cotton wool plaques are quite exceptional among sporadic cases [27]. APP is cleaved by β-secretase at its N-terminus and by γ-secretase (which includes PSEN1) at its C-terminus, generating several Aβ species of different molecular weights. Aβ₄₀ predominates in neuritic plaques, particularly within the amyloid core, whereas Aβ₄₂ is enriched in diffuse plaques [28].

There is a hypothesis that diffuse and neuritic plaques are on a pathological continuum, corresponding to disease progression in AD. Neuropathological studies have reported that preclinical or early AD cases are associated with a diffuse plaque-dominant state, whereas advanced AD cases are associated with a neuritic plaque-dominant state [29,30]. Supportively, a basic study using APP-transgenic (tg) mice has clarified that Aβ deposition precedes neuritic swelling, and the neuritic changes were ameliorated after anti-Aβ immunotherapy [31]. However, controversies may exist in the interpretation of plaque subtypes; an exogeneous Aβ-injection into APP-tg models has revealed that the morphological subtypes of Aβ deposits depend on both the host and source of the agent, suggesting the presence of distinct Aβ strains with variable biological activities [32].

It is also known that neuritic plaques often attract glial cells, particularly microglia, whereas diffuse plaques do not [33] (Figure 3). These findings emphasize the importance of neuroinflammation in Aβ-related pathology in AD. Disease-associated microglia may act as sensors of neuronal damage, initiating inflammatory cascades [34]. Nevertheless, the question of whether plaque-associated microglia are merely reactive or actively drive plaque-induced neurodegeneration remains controversial [35].

The severity and distribution of tau pathology are more strongly correlated with brain atrophy and the clinical presentation of AD than Aβ deposition, though Aβ pathology may have synergistic effects with tau-derived neuronal damage [5,32,36,37]. Tau and Aβ PET scan studies have confirmed that tau pathology has a higher correlation than Aβ pathology with cognitive status and its evolution [38]. Basic studies also revealed that a reduction in endogenous tau resulted in rescuing neurodegenerative phenotypes in an APP-mutant model, highlighting the pivotal role of tau in AD pathophysiology [39,40].

The concept of the amyloid cascade is widely accepted as an explanation of AD pathogenesis. In brief, it proposes that an increased Aβ₄₂/Aβ₄₀ ratio and subsequent plaque formation are crucial for AD pathogenesis. In this theory, tau aggregation is considered to be a downstream event of Aβ deposition. Indeed, clinical and neuropathological findings support this hypothesis. Mutations in PSEN and APP are causative of familial AD, and importantly, these mutations are associated not only with Aβ deposits but also with severe NFT pathology. Actually, the original case described by Alzheimer also had a mutation in PSEN1 [41]. Moreover, APP is located on chromosome 21, and Down syndrome (21 trisomy) is linked to a high prevalence and early onset of AD; both Aβ and tau aggregation are facilitated in this setting [42] (Figure 4). The extent of Aβ deposition is correlated with that of tau aggregates among AD cases and an aged population [13] or in AD model mice with APP/MAPT-tg [43]. By contrast, mutations of the MAPT gene, encoding tau, are causative of frontotemporal dementia with parkinsonism (FTDP), demonstrating marked tau aggregation but little or no Aβ deposition [44].

However, the direct pathway from Aβ-deposition to tau-aggregation remains largely unknown and has proven difficult to reproduce experimentally. The iatrogenic transmission of Aβ pathology does not induce a tau pathology. It has been reported that patient-derived Aβ has a prion-like property of transmission, whose aggregative seed demonstrates transmission across individuals or species [45]. However, when Aβ is transmitted in vivo, Aβ deposition alone does not seem to cause AD-type tau pathology (3R and 4R mixed tauopathy and formation of NFTs). An example is the human transmission of cerebral amyloid angiopathy, often associated with parenchymal Aβ deposits, through grafts of dura mater, without or with minimal development of a tauopathy, even after decades [46]. Another example is the iatrogenic transmission of Aβ via the injection of cadaver-extracted human growth hormone (crGH). These batches of crGH were contaminated with prions, Aβ, and tau. Postmortem studies of the recipients revealed Aβ deposits, including senile plaques and CAA, in combination with Creutzfeldt–Jakob disease (CJD). Tau aggregates were occasionally detected in these cases, mainly showing dots or neuritic patterns near the plaques and very occasional NFTs; the prevalence, morphology, severity, and distribution of tau pathology differed from and was far smaller than those in AD [47,48,49]. Intracerebral injection of patient-derived Aβ into APP-tg mice did not result in tau aggregation [50]. Another study reported that the induction of patient-derived Aβ dimers in primary hippocampal neurons produced neuritic dystrophy and facilitated the phosphorylation of tau within dystrophic neurites, but did not lead to the formation of NFTs [51]. Although injection of Aβ in wild-type mice does not induce the formation of NFT, the injection of Aβ in mutant-MAPT Tg mice (which develop NFTs) increases the formation of tau aggregates [52] and crossing APP mice with mutant MAPT-Tg mice also greatly increases the formation of tau aggregates [53,54]. The injection of human tau seeds also induces more neuritic tau pathology in the presence of amyloid plaque [55,56]. These facts highlight the accelerating role of Aβ in the induction of more NFT, when NFTs are already present.

Neuropathological studies have revealed that NFT can arise in the absence of or with sparse Aβ deposition among older people, termed primary aging-related tauopathy (PART). The morphology and topographic distribution of NFTs in PART do not differ from those in early AD [57]; a subset of PART may be symptomatic, termed senile dementia with NFT (SDNFT) [58]. As mentioned above, MAPT mutations cause FTDP; however, certain variants such as V337M and R406W yield AD-type tau filaments containing both 3R- and 4R-tau isoforms even in the absence of Aβ deposition [59,60]. Collectively, these findings indicate that tau aggregation can occur without Aβ deposits. Tau aggregates in AD might be regulated by multiple facilitating or preventive factors [61], even though Aβ deposition is a powerful driver of tau accumulation in AD.

Two questions, however, remain unresolved in this theory. First, it is highly controversial whether PART is really on a continuum of tau aggregation of AD or a distinct entity from AD. It has been described that conformation-dependent tau antibodies distinguish pathological tau in AD from other tauopathies but do not distinguish AD from PART [62], and tau seeding activity from the transentorhinal/entorhinal cortices (where tau pathology starts early) does not differ between AD and PART [63]. PART filaments are identical to those of AD and made of 3R and 4R tau isoforms [64]. However, the opposite findings are also reported; the seeding activity of neocortical tau in PART differs from that of early AD brains containing frequent neuritic plaques [65]. This result indicates that biochemical properties of insoluble tau in NFTs differ between AD and PART in the neocortex. Second, NFTs never progress beyond Braak stage IV in the absence of Aβ [16], arguing against a simple pathological continuum between PART and AD. At least, these findings suggest that Aβ deposition is necessary for the full progression of tau pathology characteristic of definite AD and that PART might play a role in the initiation of AD-related tau pathology, leading to its Aβ-induced progression. However, the precise molecular mechanisms linking Aβ to tau aggregation and the factors modulating this relationship remain incompletely understood.

The spaces between the amyloid core and halo contain the crown, composed of swollen neurites, which are usually negative according to Aβ immunostaining. Neuropathological studies have revealed that the swollen neurites involved in neuritic plaques fundamentally express epitopes of axons and axon terminals rather than dendritic epitopes [66,67]. This finding, together with the sparse presentation of senile plaques in white matter, indicates that neuritic plaques prominently involve the distal portion of the axon (Figure 5).

The swollen axons within the neuritic plaques abundantly contain hyperphosphorylated tau aggregates (Figure 5). Studies using multiple dissections of autopsied AD tissues cleared with the CLARITY protocol revealed that intra-axonal tau aggregates of the fornix were continuously extended to the mamillary body and made direct contact with senile plaques in the mammillary body [68,69]. Dystrophic neurites in senile plaques in the neocortex contain tau aggregates when NFTs are present in the same areas but not when NFTs are absent, indicating a close spatial relation between local neurons containing NFTs and plaque-associated tau aggregates [70], although the intra-axonal tau aggregates do not always connect to NFTs [71]. Interestingly, for PART, a recent study suggested an opposite sequence, in which tau aggregation in the entorhinal neurons (pre-α) starts from the dendrites and then spreads to the cytoplasm and axons [72]. Overall, neuritic tau aggregation in the senile plaque is strongly correlated with cytoplasmic NFTs, although it remains unclear whether the former always retrogradely develop the latter.

If we resort to the amyloid cascade hypothesis, it can be expected that NFTs should be enriched in neurons whose distal axons are involved in neuritic plaques. This speculation seems to be partially reasonable because Aβ-vulnerable regions often receive innervation from areas that develop early NFT pathology. For example, the upper molecular layer of the hippocampal dentate gyrus preferentially exhibits Aβ deposition and receives abundant glutamatergic innervation from the entorhinal cortex, one of the early regions affected by NFTs. Correlations between senile plaque burden in the dentate gyrus and NFT burden in the entorhinal cortex have been reported [73,74]. Similarly, early Aβ deposition in the cerebral neocortices corresponds to cholinergic innervation from the basal forebrain, which also develops NFTs during the early phase of AD [75].

However, mismatches between the Aβ and tau pathologies are also noted despite these anatomical correspondences. As discussed in the next chapter, tau aggregation in AD spreads laterally through adjacent cortical areas in a cortico-cortical manner with a smooth gradient, starting from the transentorhinal cortex. It is questionable whether such a highly hierarchical, lateral spreading of tau aggregation arises from distant effects by Aβ deposition, whose spread may start from any cerebral cortices varying among individuals. Moreover, certain regions show discordant pathology: the locus ceruleus is one of the earliest sites of NFT formation and abundantly projects to the cerebellum. Nevertheless, the cerebellum shows Aβ deposition at the latest stage of AD pathology. Similarly, the mamillary bodies and thalamus, which receive abundant innervation from the hippocampus, show relatively sparse Aβ accumulation. A recent study on a large sample size has addressed a mismatch between the Aβ and tau pathologies in subsets of AD cases that display severe Aβ deposition (including extensive neuritic plaques) but unexpectedly sparse NFTs [76].

Overall, Aβ seems to be a powerful facilitator of tau aggregates at a cellular level. However, spatial distribution and temporal spreading of tau aggregates demonstrate some discrepancies from what is expected from Aβ deposits in the context of cytoplasm–axon relationships.

From the viewpoint of tauopathy, AD can be defined as a form of selective 'system degeneration' primarily affecting the memory network [77,78]. This notion arises from the observation that tau aggregation begins in the transentorhinal cortex and extends to the hippocampus and associative neocortices [79], and this spreading course contains some critical nodes of the limbic memory system, including the parahippocampal–hippocampal loop and Papez circuit [80], in AD brains. The selective and highly hierarchical spreading of tau aggregates proposes a tau-autonomous mechanism for tau pathology in AD.

Neuropathological studies of various neurodegenerative disorders have revealed that atrophy, neuronal loss, and accumulation of neuron-aggregative proteins tend to develop along specific neuroanatomical systems that are distantly connected with long tracts beyond the white matter, a process termed 'system degeneration'. Representative examples include amyotrophic lateral sclerosis (ALS) and frontotemporal lobar degeneration (FTLD). In TAR DNA-binding protein 43kDa (TDP-43)-related ALS (ALS-TDP), upper and lower motor neurons that are distantly connected by the pyramidal tract are affected, showing neuronal loss and TDP-43 aggregation in both areas, resulting in a system degeneration of motor neurons [81,82]. In addition, TDP-43-related FTLD (FTLD-TDP) demonstrates system degeneration along corticostriatal circuits, which are critical for executive functions, mood regulation, and language generation [83,84]. A loss of projection neurons and their axon terminals, combined with TDP-43 aggregates, has been observed across the cerebral cortex, neostriatum, and globus pallidus or substantia nigra [85,86].

In addition to ALS/FTLD, various neurodegenerative disorders including PSP (dentato-rubral or pallido-nigral/luysian degeneration), CBD (cortico-striato-nigral degeneration), and multiple system atrophy (MSA) (olivo/ponto-cerebellar or striato-nigral degeneration) demonstrate system degeneration, showing selective, longitudinal, and poly-nodal impairment along these functional systems [87].

Importantly, the extension pattern of AD-related tau aggregation depends on spatial proximity, with lateral spreading toward adjacent areas resembling the 'spreading of stain'. When the concept of a 'system' is considered, the propagation of tau aggregates in AD tends to terminate in a mono-nodal manner (Figure 6). For example, the Papez circuit, a classical memory system comprising the hippocampus, fornix, mammillary body, anterior nucleus of thalamus, cingulate gyrus, and parahippocampal gyrus, is known to be impaired in AD [80]. Indeed, this system contains AD-vulnerable nodes such as the hippocampus and the parahippocampal gyrus. However, the gradients of tau aggregates and neuronal loss do not necessarily follow the anatomical connection of the circuit; for example, the mammillary body, the next node after the hippocampus, is not preferentially involved in NFTs and neuronal loss [88].

A similar phenomenon is observed in the hippocampal loop [89]. The glutamatergic memory circuit originating from the entorhinal cortex mainly innervates the granule cells of the hippocampal dentate gyrus via the perforant path, followed by CA3 and CA4 via mossy fibers, and finally reaches CA1 via Schaffer collaterals. In this system, the entorhinal cortex is highly vulnerable to tau aggregation, whereas the dentate gyrus and CA3/CA4 are relatively spared; subsequently, CA1 and the subiculum are again vulnerable (Figure 6). Thus, although the memory circuits contain tau-vulnerable regions, neuronal loss and tau aggregation cannot be aligned along a single gradient within these systems. These findings support the view that tau aggregates in AD may spread along short-range neural connections, such as cortico-cortical projections, rather than long tract-dependent circuits [90].

A remarkable supporting finding comes from an autopsied AD patient who had undergone brain surgery 27 years earlier [91]. During surgery, a portion of the frontal cortex was dissected together with a tumor. At autopsy, an island-like cortical area was found to be isolated from the surrounding frontal cortex due to the surgical dissection, remaining connected only through long tracts in the deep white matter. Interestingly, the isolated cortex demonstrated diffuse plaques but not NFTs and neuritic plaques, although these were abundant in other cortical areas. This finding suggests that AD-related tau aggregation preferentially involves cortico-cortical tracts and their associated neurons rather than long-tract projections.

Experimental studies have clarified the propagative properties of tau aggregates derived from patients with AD, PSP, or CBD. Prion-like mechanisms of extracellular release and the propagation of tau aggregates along neuroanatomically connected areas might be central to this process [92,93,94]. Neuronal uptake of induced tau and seed-dependent aggregation have been reported using both in vitro and in vivo models expressing mutant [95,96] or wild-type [97,98] human tau. The hypothesis that tau seeds propagate via short cortico-cortical tracts in the AD brain, resulting in lateral spreading of tau aggregates, is therefore compelling.

However, questions remain regarding tau pathology in AD. As addressed at the beginning of this article, tau aggregates may mature from pretangles to NFTs and ultimately remain as ghost tangles, which lack recognizable cellular structures of the cytoplasm and nucleus. Importantly, it has been reported that tau isoforms change from a 4R-tau-rich state (pretangles and early NFTs) toward a 3R-tau-rich state (mature NFTs and ghost tangles) during the maturation of tau aggregates [24]. Upon a cross-sectional observation, the 3R-tau/4R-tau ratio correlates with the severity of tau aggregation, showing an increasing gradient from mildly affected to severely affected areas [24].

The tau-propagation theory may be compatible with the spread of 4R-tau in PSP and CBD but hardly explains the isoform shift observed in AD, because these isoforms are produced through alternative splicing of the MAPT gene. Moreover, events related to MAPT-splicing may not explain the isoform shift in tau aggregates in AD. Although several factors, including RNA-binding proteins [99], ribonucleoproteins [100], and splicing factors [101,102,103], have been reported to influence MAPT splicing, their roles remain mostly unknown in AD brains. It is known that mRNA levels of total or isoform-specific MAPT vary across studies and may not be in line with the isoform shift in tau in AD [104]. Indeed, a study reported that the 3R/4R mRNA ratio tended to be lower in advanced AD cases compared with normal controls or those with mild cognitive impairment [105]. While the intracellular growth of tau aggregates can be explained by seeding ability or post-translational events, the mechanisms underlying the 3R/4R isoform shift remain to be elucidated in AD.