Most neurodegenerative diseases are proteinopathies, which are characterized by the aggregation of misfolded proteins. Although many proteins have an intrinsic propensity to aggregate, particularly when cellular clearance systems start to fail in the context of ageing, only a few form fibrillar aggregates. In Alzheimer disease, the peptide amyloid-β (Aβ) and the protein tau aggregate to form plaques and tangles, respectively, which comprise the histopathological hallmarks of this disease. This Review discusses the complexity of Aβ biogenesis, trafficking, post-translational modifications and aggregation states. Tau and its various isoforms, which are subject to a vast array of post-translational modifications, are also explored. The methodological advances that revealed this complexity are described. Finally, the toxic effects of distinct species of tau and Aβ are discussed, as well as the concept of protein 'strains', and how this knowledge can facilitate the development of early disease biomarkers for stratifying patients and validating new therapies. By targeting distinct species of Aβ and tau for therapeutic intervention, the way might be paved for personalized medicine and more-targeted treatment strategies.
Alzheimer disease belongs to the group of proteinopathies; it is characterized by deposition of the peptide amyloid-β (Aβ) as amyloid plaques and of the protein tau as neurofibrillary tangles
Aβ neurotoxicity is attributable to specific types of Aβ, generated as a consequence of proteolytic cleavage and post-translational modifications, which are susceptible to aggregation into different assembly states
Similar to Aβ, tau exists as multiple brain isoforms that undergo aggregation and is subject to a host of post-translational modifications, including phosphorylation and acetylation
Impairment of multiple cellular functions by Aβ and tau has been demonstrated in cellular and transgenic models, and crosstalk between these molecules has been demonstrated, particularly at the synapse
Assessment of Aβ and tau pathology is being facilitated by increasingly sensitive methods, which have clinical relevance in diagnosis and in the validation of therapeutic interventions in disease
A prerequisite for personalized medicine is the identification of distinct Aβ and tau species that are suitable for use as biomarkers in cerebrospinal fluid, blood, urine and saliva
Life expectancy has increased substantially in past decades, owing to general improvements in lifestyle and medication; however, the ensuing prominent demographic upward shift in age distribution has led to an increased prevalence of diseases such as cancer and dementia. Many treatment options, ranging from surgery and radiotherapy to tailored pharmacological and hormonal treatments, are now available for cancer. However, the treatment of dementia has remained symptomatic, and despite a large number of costly trials, no single drug or treatment strategy has been approved. Several reasons have been put forward to explain this failure: selection of inappropriate populations of patients; suboptimal dosing or drug exposure ('too little, too late'); the challenge of defining suitable primary end points for such trials, which reflects the poor sensitivity of clinical instruments; and the lack of a detailed understanding of the causes of dementia — more specifically, an incomplete understanding of the role that distinct forms of amyloid-β (Aβ) and tau have in neurodegenerative disease and dementia — the major topic of this Review.
Dementia is marked by memory disorder, personality changes and impaired reasoning. Clinically, dementia can have many causes, but its most prevalent form by far is Alzheimer disease (AD), which affects an estimated 47 million people worldwide and accounts for 60–80% of all patients with dementia. By 2050, the prevalence of AD is expected to quadruple, and 43% of these patients will need high levels of care1. Other prevalent forms of dementia are dementia with Lewy bodies (DLB) and frontotemporal dementia (FTD, the pathological hallmark of which is frontotemporal lobar degeneration (FTLD)), a progressive neurodegenerative disorder that includes semantic dementia, progressive nonfluent aphasia and behavioural variant subtypes within its spectrum. Dementia is also present in movement disorders such as Parkinson disease (Table 1). The overarching feature of all these conditions is the accumulation of insoluble protein deposits in the brain. Most current therapeutic strategies for these diseases are based on the assumption that these proteins and their aggregation do not constitute an epiphenomenon but in fact are causal and drive its progression2.
Characteristic of these disorders is the propensity of key proteins to form oligomers that act as a template or nucleus for the conversion and co-aggregation of endogenous proteins, which eventually form fibrils3. The two key molecules implicated in AD are Aβ (Fig. 1) and tau (Fig. 2). Aβ is a small peptide derived by proteolytic cleavage from amyloid precursor protein (APP); aggregates of Aβ form histological lesions known as amyloid plaques. Amyloid plaques are also prevalent in patients with DLB. Tau is a microtubule-associated protein (MAP) that forms neurofibrillary tangles. Tau pathology is found not only in AD but also in many other diseases, collectively termed tauopathies. The tauopathies include an important subset of FTLD, termed FTLD-tau. Although tau is mainly a neuronal protein, patients with corticobasal degeneration and progressive supranuclear palsy also show prominent glial tau pathology. All these diseases can be either sporadic or familial4.
The vast majority of AD cases are sporadic. In familial AD, disease-causing mutations have been identified in several genes linked to Aβ formation, including APP itself, as well as PSEN1 and PSEN2 (encoding presenilin 1 and presenilin 2, respectively), whereas in familial FTLD-tau, disease-causing mutations have been identified in the tau-encoding MAPT gene5. Together, this genetic and histological evidence has established a role for both tau and Aβ in AD, leading to the generation of transgenic animal models that reproduce important aspects of the human pathology and have been used to study therapeutic strategies6. Because sporadic and familial AD do not differ histologically or clinically (except for the earlier age of onset in familial AD), research has been guided by the assumption that a therapeutic strategy validated in models of familial AD would also be of value in sporadic AD.
In this Review, we explore what is known about the physiological roles of Aβ and tau and describe how things go wrong in disease. We focus on the increasing understanding that these molecules come in a vast variety of forms and assembly states, not all of which are toxic. We explain how advancing our knowledge of distinct Aβ and tau species and their biological effects could facilitate the development of improved biomarkers and targeted therapies. Finally, we describe refined methods for detecting proteinopathies and review what AD has in common with other proteinopathies.
Physiological roles of APP, Aβ and tau
APP is a type I transmembrane glycoprotein composed of a large metal-binding and heparin-binding ectodomain, a single membrane-spanning domain and a short cytoplasmic tail. Two paralogues exist, amyloid-like proteins 1 and 2 (APLP1 and APLP2, respectively)7, which, unlike APP, lack the Aβ sequence. Mice deficient in any single APP family member are viable; however, double-knockout mice (either APP−/−APLP2−/− or APLP1−/−APLP2−/−) die early in postnatal life, suggesting redundancy between APLP2 and the other two family members8. Although APP is also expressed in many peripheral organs, most studies have been conducted in the brain9, where APP and its cleavage products have important roles in neurogenesis, plasticity and synaptic function as well as the cellular stress response10,11,12,13. Under conditions of metabolic stress, such as acute hypoxia, release of the APP extracellular domain, soluble APPα (sAPPα), contributes to calcium homeostasis in a process involving voltage-gated calcium channels14, whereas another APP cleavage product, AICD (APP intracellular domain), has a role in lipid biosynthesis15 (Fig. 1). Importantly, Aβ production is a normal physiological process that is enhanced by synaptic activation and plasticity16,17. Aβ also contributes to lipid homeostasis by directly binding to and transporting cholesterol18, but the vast majority of studies have focused on its toxic properties.
Tau belongs to the MAP family, which also includes MAP2 and MAP4. Unlike MAP4, which is expressed in many tissues, MAP2 and tau are predominantly found in neurons. In the adult brain, MAP2 is mainly localized to cell bodies and dendrites, whereas tau is abundant in axons, although it is (albeit at much lower levels) also found in the soma and dendrites. A sole MAP homologue, the microtubule-associated protein PTL-1 (protein with tau-like repeats), is present in the roundworm Caenorhabditis elegans, where it regulates ageing of the organism and its nervous system19. In humans, tau is encoded by a single gene on chromosome 17q21.31 that spans 16 exons. Alternative splicing gives rise to six major brain isoforms, which differ with regard to the number of insertions in the projection domain (0N, 1N and 2N) within the amino-terminal half of the peptide, and the presence of either three or four highly conserved microtubule-binding repeat domains (3R or 4R) in its carboxy-terminal half (Table 2). Natively unfolded in solution, tau is post-translationally modified at many sites under physiological conditions, primarily via phosphorylation20, which causes tau to adopt a paper-clip conformation and bind to microtubules21. Increased phosphorylation reduces the affinity of tau for microtubules (Fig. 2g), and these changes provide tau with the dynamic properties essential for neuronal plasticity. Interestingly (and reminiscent of Aβ), endogenous tau is released in response to neuronal activity22,23,24, but the physiological relevance of this observation remains elusive. Tau has also been localized to the nucleus, where it might have a role in maintaining DNA integrity25 (Fig. 2a). Although tau has been largely perceived as a microtubule-stabilizing protein, when tau is genetically knocked out in mouse models, microtubules do not destabilize, possibly because of compensatory mechanisms involving other MAP family members26. Moreover, given the lack of overt neurological impairment in tau-knockout mice (unless the mice reach an advanced age), reducing tau levels has been proposed as a suitable treatment strategy for AD20,27. Therefore, perhaps the best descriptor of tau is that its various isoforms act as versatile scaffolding proteins28.
Aβ species and assembly states
Formation and intracellular location. Several APP isoforms exist: the predominant neuronal form is APP695, whereas astrocytes and microglia also express APP751 and APP770 (Ref 29). Most aspects of APP processing have been studied in non-neuronal cells, which lack the compartmentalization and long processes of neurons30,31. However, APP processing occurs in many neuronal compartments, including axons, nerve terminals and dendrites, giving rise to a variety of biologically active fragments (Fig. 1; Table 2).
In the nonamyloidogenic (that is, non-Aβ-forming) APP-processing pathway, which is stimulated by synaptic activity32, surface APP is mainly cleaved by the α-secretase disintegrin and metalloproteinase domain-containing protein 10 (ADAM10)33,34. Upon release of sAPPα into the extracellular milieu, a short piece of APP (α-secretase C-terminal fragment (α-CTF), also known as C83) remains inserted in the membrane (Fig. 1). This fragment is recognized by the γ-secretase complex, which performs an endopeptidase-like ε-cleavage that leads to the intracellular release of AICD and a carboxypeptidase-like γ-cleavage that generates the p3 fragment35. Because the initial α-cleavage occurs within the Aβ region, Aβ formation is precluded.
The canonical amyloidogenic pathway (which leads to Aβ formation) posits that surface APP that has not undergone α-cleavage is internalized into endosomes, where it is cleaved by β-secretases 1 and 2 (BACE1 and BACE2, respectively) at a site that becomes the amino terminus of Aβ36. The long β-secretase C-terminal fragment (β-CTF, also known as C99) remains tethered to the membrane, whereas soluble APPβ (sAPPβ) is released (Fig. 1). The γ-secretase complex initially cleaves β-CTF via endoproteolytic ε-cuts, generating AICD and the two species Aβ48 and Aβ49. These two peptides are further processed by γ-secretase, which makes exoproteolytic γ-cuts at every three to four residues within the hydrophobic sequence, forming shorter peptides: Aβ48 gives rise to Aβ45, Aβ42 and Aβ38, whereas Aβ49 gives rise to Aβ46, Aβ43 and Aβ40 (Refs 37,38). Other even shorter (rare) cleavage products have also been identified37,38.
APP processing is tightly regulated, and secretases are trafficked together with APP through the secretory pathway along dendrites and axons and into presynaptic boutons39. The interaction between BACE1 and APP occurs in all these compartments, and neuronal activity potentiates the convergence of these enzymes and their substrates in the same recycling endosomes40. On the other hand, active γ-secretase resides in late recycling endosomes, lysosomes, the trans-Golgi network and plasma membrane41 (Fig. 1). Similar to BACE1, the γ-secretase complex undergoes exocytosis to reach the plasma membrane, followed by endocytosis and retrograde trafficking to reach its final active compartments, predominantly the trans-Golgi network42. The mitochondria-associated endoplasmic reticulum membrane has received increasing attention, as the γ-secretase complex is highly active in this compartment43 (Fig. 1) and the function of this membrane is perturbed in cells from patients with AD44. Although Aβ can be secreted both presynaptically and postsynaptically45, the majority of axonally secreted fragments are endocytosed in the soma, processed and then transported to the presynapse46. Other noncanonical (δ-secretase, meprin-β, η-secretase and caspase) APP-processing pathways are beginning to be explored, but their influence on disease is not yet clear47,48.
Post-translational modifications. Aβ isolated from AD brains shows a variety of post-translational modifications at the amino terminus, including oxidation, pyroglutamylation, phosphorylation, nitration, racemization, isomerization and glycosylation, which can all modify the oligomerization and fibril-forming properties of the peptide49. Table 2 summarizes these different species, and Box 1 presents selected modes of toxicity. Oxidation of the sulfur atom of Met35 to sulfoxide facilitates the formation of ion-channel-like complexes of Aβ in lipid membranes50, which might not then be available for protofibril and fibril formation51. Pyroglutamate modification of Aβ has been reported in vivo at positions 3 and 11 (3 pR-Aβ and 11 pE-Aβ, respectively)52,53 and increases the propensity of Aβ to aggregate in vitro54,55. Formation of 11 pE-Aβ requires the enzyme glutaminyl cyclase, expression of which is increased in AD56. Phosphorylation of Aβ at Ser8 favours the formation of oligomeric aggregates with increased resistance to degradation, which have been found in AD mouse models as well as in patients with AD57,58. Nitrated Aβ is found in the core of amyloid plaques and might initiate plaque formation59. Finally, Aβ O-glycopeptides have been identified in the cerebrospinal fluid (CSF) of patients with AD; however, only short Aβ peptides were glycosylated. No modified Aβ40 or modified Aβ42 was detected, suggesting that O-glycosylation affects the ability of γ-secretase to cleave APP60.
Assembly states. Adding to this complexity, several Aβ assemblies, ranging from dimers to trimers, dodecamers and multimers, have been identified in vitro and in vivo61,62,63 and broadly categorized into two types64 (Table 2). Type 1 oligomers do not develop into larger assemblies62; these include dodecamers (previously described as Aβ*56 on the basis of their 56 kDa size). Type 1 oligomers appear early in AD, whereas type 2 oligomers appear only after plaques have formed; at the plaque surface, type 2 oligomers self-organize into small, stable structures with a parallel β-sheet architecture. Conformation-specific anti-Aβ antibodies (A11 and OC, respectively) have facilitated the identification of type 1 and type 2 oligomers65.
The specific process that leads to plaque formation is only incompletely understood. When Aβ is secreted into the extracellular space, it aggregates to form amyloid plaques (Fig. 1). Plaques can also be formed inside cells upon internalization of Aβ; formation of intracellular plaque is associated with a disturbance of multivesicular bodies that precedes cell death66. Overall, the existence of such a broad variety of Aβ species and aggregation states suggests that patient-specific differences affect disease progression and therapeutic outcomes.
How Aβ impairs neuronal function
In this section, we discuss cellular functions that are impaired by distinct Aβ species and assemblies as well as the cellular compartments where such disturbance occurs (Table 2). Early efforts aimed to determine whether amyloid plaques (the end-stage lesions of AD) were toxic and to discriminate between monomeric, oligomeric and fibrillar forms of Aβ. Initially, plaque burden (as distinct from neurofibrillary tangles) was thought not to correlate with AD severity67; however, when all plaques (rather than just limbic and neocortical ones) are considered, amyloid deposition that has progressed to involve the striatum is highly predictive of dementia68,69. Amyloid plaques are not inert, as surrounding neurons exhibit clear symptoms of toxicity, including elevated calcium levels, dystrophic neurites and synaptic loss70. The concept of soluble toxic oligomers (also termed Aβ-derived diffusive ligands) has gained broad acceptance71. Type 1 oligomers can travel through the brain, whereas type 2 oligomers are confined to the vicinity of plaques. Thus, type 1 oligomers are more likely than type 2 oligomers to interfere with synaptic function and cause cognitive deficits (Table 2). Plaques might also act as a reservoir of neurotoxic Aβ species that coexist in a dynamic equilibrium with potentially inert fibrils (Fig. 1). Levels of the type 1 oligomer Aβ*56 in mouse models and in human samples correlate with cognitive impairment and pathological tau accumulation72.
Of the different Aβ species, Aβ42 is much more toxic than Aβ40, possibly because of the stronger tendency of Aβ42 to aggregate73. Whereas Aβ40 can inhibit Aβ42 oligomerization74, both species are thought to drive plaque formation and neurotoxicity (Table 2). Nitrated Aβ can suppress long-term potentiation to a greater degree than unmodified Aβ75,76; however, the toxicity of any Aβ species is ultimately related to its concentration. Aspartate residues in Aβ can spontaneously isomerize, forming isoaspartate, which is particularly resistant to enzymatic degradation77. This feature might explain why levels of isoaspartyl-modified Aβ are higher in older plaques than in younger ones and exhibit a positive correlation with dementia severity78. Little is known about the toxicity of other Aβ species, with the exception of pyroglutamylated Aβ. When glutaminyl cyclase (which is required to generate 11 pE-Aβ) was downregulated in mouse models of AD, Aβ40 and Aβ42 levels, plaque burden and inflammatory reactions were all reduced, concomitant with an improvement in learning79 (Table 2). Thus, different Aβ species have seemingly different effects.
As we discuss further below, the AD field has received a lot of input from work on prions, the protein agents that cause transmissible spongiform encephalopathies80. One of the most intriguing findings is the existence of different prion strains with unique biochemical profiles and clinical features. Although the proteins implicated in AD have so far not been convincingly demonstrated to be infectious in the same way that prions are, the evidence of their unique biochemical profiles has lent support to use of the same terminology. By 2005, different Aβ40 fibril morphologies had already been linked to distinct toxicities in cell culture81. Furthermore, in a study of Aβ40 extracted from two different patients, each Aβ40 'strain' was found to have a single predominant fibril structure, which also suggested that certain structures are more pathogenic than others82. Nuclear magnetic resonance measurements of Aβ40 and Aβ42 fibrils prepared by seeded growth from AD brain extracts suggested that a rapidly progressive form of AD is related to a specific fibrillar Aβ structure83. Similarly, an earlier study had reported increased levels of distinctly structured Aβ42 particles composed of 30–100 monomers, and reduced levels of particles composed of <30 monomers, in brain samples from patients with rapidly progressive AD84. However, what leads to these different profiles and how they might affect the development or progression of AD is not really understood.
Synaptic toxicity. The synapse is a major compartment where Aβ exerts its toxicity, and because synaptic loss is a better correlate of cognitive impairment in AD than plaque burden, AD has been termed 'a synapse failure' (Ref. 85). Synaptic plasticity involves AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid) and NMDA (N-methyl-d-aspartate) receptors (AMPARs and NMDARs, respectively), which induce long-term potentiation. Both synthetic and naturally secreted Aβ oligomers, but not fibrils, reduce long-term potentiation in brain slices and in vivo86, although the underlying mechanisms are not fully understood. Aβ can indirectly activate and directly bind to NMDARs and AMPARs and decrease levels of the adaptor protein postsynaptic density protein 95 (PSD95, also known as DLG4), which further negatively regulates these receptors through endocytosis and a reduction in the expression of receptor subunits87,88. Long-term potentiation also generally involves post-translational modifications, including sumoylation, and this process is impaired by oligomeric Aβ42 (Ref. 89).
Mitochondrial dysregulation. Mitochondria, which regulate both energy metabolism and apoptotic pathways, are another important target of Aβ toxicity. High numbers of mitochondria are present in neurons, and they are particularly enriched in synapses. Owing to their limited glycolytic capacity, neurons are highly dependent on mitochondrial energy production90. All four major aspects of mitochondrial function are compromised in AD: oxidative phosphorylation; bioenergetics; mitochondrial dynamics (transport, fission and fusion); and mitophagy (removal of damaged mitochondria by autophagy). Dysregulation of mitochondrial function leads to disrupted synaptic transmission, apoptosis and, ultimately, neurodegeneration. Aβ and tau both have discrete roles in this mitochondrial dysfunction91 (Fig. 3). For example, in a transgenic mouse model of AD, Aβ specifically impairs complex IV of the oxidative phosphorylation system, whereas tau impairs complex I (Ref. 92). Mitochondrial impairment can be caused by both oligomeric and fibrillar Aβ, whereas monomeric Aβ has no effect93.
Mitochondria are the major producers of damaging reactive oxygen species (ROS) as well as targets of ROS toxicity94. However, how oxidative stress is caused by Aβ is only incompletely understood. One proposed mechanism involves the ability of Aβ to bind metal ions (in particular Cu2+ and Fe3+) and reduce them into species (notably Cu+ and Fe2+) that eventually promote lipid and protein peroxidation95. Aβ can also form pores in lipid membranes96, facilitating the unregulated influx of Ca2+ ions and causing upregulation of gating of various Ca2+ release channels, ultimately leading to cell death97,98. This pore-forming mechanism of action is similar to that of some antibacterial agents. Specific blockade of these Aβ pores can reduce both Ca2+ influx and neuronal damage99.
Effects on microglia. Finally, glia have a role in Aβ toxicity. Aβ oligomers and amyloid plaques initiate inflammatory responses in the brain, leading to activation of astrocytes and recruitment of microglia, processes that have been widely reported in AD and related tauopathies100. So far, a consensus view has not been reached on the effect of specific Aβ species on microglial reactivity (and thus on microglia-elicited neurotoxicity), but oligomers and fibrils apparently stimulate different microglial activation pathways. Specifically, oligomeric Aβ might stimulate microglial cytokine and chemokine production while decreasing their phagocytic capacity101,102. However, in a 2016 study, microglia engulfed synaptic material in adult mouse brains when exposed to soluble Aβ oligomers, and this process was dependent on complement receptor 3 (Ref. 103). These inflammatory responses seem to be downstream events in the pathogenic cascade in AD.
Tau species and assembly states
Formation and intracellular location. In the human brain, tau exists as six major isoforms, with an equimolar ratio of the 3R and 4R isoforms (Table 2). Distortion of the 3R:4R tau isoform ratio in either direction is well known to be associated with distinct clinical manifestations of tauopathies; for example, corticobasal degeneration is associated with >4R tau, whereas Pick disease is associated with >3R tau104. Why this is the case is still not really understood. However, changes in the 3R:4R ratio can impair multiple cellular functions, including axonal transport of APP105, and a loss of splicing factors (implicated in FTLD) results in changes in the tau isoform ratio106. These observations highlight an interesting crosstalk between Aβ and tau pathologies, which is discussed in more detail below.
Tau is generally perceived as an axonal protein; tau toxicity occurs because pathological forms of tau accumulate in compartments where tau levels are normally low, such as the soma. These abnormal deposits of tau can sequester other proteins and prevent them from executing their physiological function. For example, in tau transgenic mice, hyperphosphorylated soluble tau impaired the axonal transport of various cargoes, including mitochondria, by trapping the kinesin motor adaptor protein Jun-amino-terminal kinase-interacting protein 1 (JIP1) in the cell body107 (Fig. 2c). The question of how the 'axonal' protein tau accumulates in the somatodendritic compartment in AD has not yet been definitively answered, but hyperphosphorylated tau in the axon is generally assumed to detach from the microtubules and thereby become capable of passing through the axon initial segment (which serves as a diffusion barrier for physiologically phosphorylated tau), enabling its accumulation in the cell body and dendrites. As discussed below, an alternative, more cogent mechanism has been discovered involving de novo protein synthesis of tau in the soma108.
Post-translational modifications. Tau undergoes several types of post-translational modifications, including arginine monomethylation, lysine acetylation, lysine monomethylation and dimethylation, lysine ubiquitylation, serine O-linked N-acetyl-glucosamination (O-GlcNAc) and serine, threonine or tyrosine phosphorylation20, which open new avenues to link distinct tau species to particular modes of toxicity. Of these modifications, phosphorylation and acetylation are by far the most common.
The longest human tau isoform contains 45 serine, 35 threonine and 5 tyrosine residues, any of which can be potentially phosphorylated. Under physiological conditions, a mole of tau contains on average two to three moles of phosphate; under pathological conditions, tau can contain as many as seven to eight moles of phosphate109. In this situation, termed hyperphosphorylation, some residues are phosphorylated to a higher degree than in the healthy brain, and others are de novo phosphorylated. In cell lines, recombinant tau derived from healthy human brain promotes the assembly and bundling of microtubules, whereas hyperphosphorylated tau (isolated from AD brain cytosol) inhibits microtubule assembly and disrupts existing microtubule networks by sequestering normal brain tau and MAP2 (Ref. 110). For many years, the focus of tau research has been on the phosphorylation of serine and threonine, which is common in pathologically hyperphosphorylated tau20.
Hyperphosphorylation, brought about by the increased activity of kinases and decreased activity of phosphatases, is thought to be a prerequisite for the accumulation of fibrillar tau in the somatodendritic domain. Phosphorylation of distinct serine and threonine epitopes also drives tau localization to spines111,112. Tyrosine phosphorylation is mediated by several SRC kinases, in particular tyrosine-protein kinase FYN113, which interacts via its SH3 domain with SH3-binding sites (PXXP motifs) in tau and phosphorylates tau at Tyr18 (Refs 113,114). Tyr18 phosphorylation of tau by FYN facilitates its direct interaction with the SH2 domain of FYN115.
In AD, tau also undergoes acetylation (discussed below), ubiquitylation, methylation, glycation, nitration and truncation116,117,118,119,120. Unfortunately, the investigation of human AD tissue or that of transgenic animals typically involves only a small number of antibodies, each of which detects a distinct post-translationally modified epitope in tau; for example, the AT8 antibody recognizes tau phosphorylated at both Ser202 and Thr205. By necessity, such experiments do not provide a comprehensive map of tau modifications.
Assembly states. The critical steps leading to tau aggregation in AD as a consequence of post-translational modifications are not fully understood. As cytoplasmic tau levels increase, tau aggregates and eventually forms insoluble filaments that fill the entire soma, leading to neurofibrillary tangles and neuropil threads. Fibril formation itself occurs as a result of transition from a random-coil architecture to the β-sheet structure typical of all amyloid fibrils121.
The microtubule-binding repeat domain in tau — more specifically the PHF6 hexapeptide (VQIVYK) motif located at the beginning of the third microtubule-binding repeat in all tau isoforms — is pivotal for tau nucleation in vitro122. Surprisingly, a 31-residue peptide containing the VQIVYK motif was able to induce tau aggregation in HEK-293 cells expressing a fluorescence-tagged version of tau123. This hexapeptide was hidden (that is, conformationally inaccessible) in monomeric tau that lacked seeding capacity but was exposed in tau molecules with the ability to seed and self-assemble124. Therefore, antibodies that target the VQIVYK motif might prevent tau aggregation. Cysteine residues in tau can form intramolecular disulfide bonds, which also have a critical role in tau aggregation125. Consequently, small molecules that bind to cysteine and block disulfide bond formation might prevent tau oligomerization and consequently the formation of insoluble tau aggregates. These molecules provide a promising new therapeutic option in the management of tauopathies125.
Tau initially forms oligomers, similarly to Aβ126, and oligomeric tau induces synaptic dysfunction and memory loss before fibril formation127. Such a role is supported by the fact that the anti-tau oligomer-specific monoclonal antibody, TOMA, both prevents and reverses the cognitive deficits associated with tau pathology, again without affecting neurofibrillary tangle pathology128. Aggregation of tau into fibrils is a defining feature of all tauopathies. Even if we assume that intraneuronal neurofibrillary tangles are biologically inert cytoplasmic lesions, they might still interfere with vital physiological functions simply because they occupy space (Fig. 2). In fact, an analysis of 20,000 dendritic spines after intracellular injections of Lucifer yellow revealed that the early diffuse accumulation of phosphorylated tau (before the development of mature neurofibrillary tangles) did not induce dendritic changes. However, the accumulation of tau neurofibrillary tangles was associated with a progressive loss of dendritic spines, morphological changes and dendritic atrophy depending on the degree of tangle pathology129. On the other hand, in an inducible tau model, reduction of tau expression still improved cognition, although neurofibrillary tangles continued to accumulate130. This observation suggested that soluble, nonfibrillar tau was the species that induced neuronal dysfunction. To resolve this discrepancy, we argue that the toxicity of different forms of tau might depend on the particular stage of disease; thus, fibrillar tau might become toxic only when it starts to occupy a substantial amount of cellular space (Table 2).
How tau impairs neuronal function
Tau toxicity is highly dependent on post-translational modifications, and a large number of studies have revealed that hyperphosphorylated tau impairs neuronal function. How easily tau can be phosphorylated or dephosphorylated depends on its conformation. For example, the cis-conformation of tau has been identified as an early pathogenic driver of AD and other tauopathies because it has increased resistance to protein-phosphatase-2A-mediated dephosphorylation and degradation131,132. Ageing also has a role in accelerating pathological tau phosphorylation: when transgenic mice expressing human tau were repeatedly backcrossed onto a senescence-accelerated SAMP8 background, mice of the new hybrid strain showed an age-related increase in pathological tau phosphorylation at specific residues (namely, Ser202, Thr205 and Ser235) compared with the parent transgenic mice133.
A completely new approach to tau toxicity reflects the concept that the progression of tau pathology in AD, as categorized by Braak and colleagues into six stages on the basis of neuronal location and clinical severity134, might in fact reflect cell-to-cell propagation of the disease, achieved by the release of tau into the extracellular space and reuptake by recipient neurons (seeding and spreading), rather than the successive involvement of different neuronal subpopulations with varying vulnerabilities to the disease. Extracellular tau aggregates are now well accepted to be capable of trans-synaptic spread, thereby causing tau pathology in recipient neurons135,136 (Fig. 2f). Seeding-competent tau aggregates have been described with diverse assembly states ranging from trimers137,138 through low-molecular-mass aggregates and short fibrils139,140 to high-molecular-mass species141, although small assemblies of at least hexamer size are the most efficiently internalized variety and have a higher seeding capacity than either larger or smaller species139,140,141. One way of releasing tau is by encapsulation within exosomes142,143; in fact, it is tempting to speculate that tau seeds encapsulated by lipid membranes are the physiological way in which tau aggregates are secreted and passed on to neighbouring cells. In 2017, spreading of tau down neuronal circuits was found to precede synaptic and neuronal degeneration in an AD mouse model with spatially confined expression of tau144.
Interestingly, tau monomers derived from aggregated tau can themselves induce tau aggregation, owing to the conformational exposure of a VQIVYK motif that is normally inaccessible in physiological tau monomers124. Hyperphosphorylation of tau might have a critical role in the exposure of this motif145. However, seeding does not occur in a unified way. Rather, various 'prion-like' strains of tau, each with unique biochemical properties, induce diverse pathological phenotypes of aggregation in vitro and in vivo and have the ability to propagate tau pathology at different rates depending on the targeted brain region146,147. These strain-like features of tau were revealed by injecting brain extracts from humans who had died with various tauopathies into the hippocampus and cerebral cortex of transgenic mice that express human tau. Argyrophilic tau inclusions formed in all recipient mice, and, following the injection of the corresponding brain extracts, the hallmark lesions of distinct tauopathies such as progressive supranuclear palsy and corticobasal degeneration were recapitulated148. Thus, tau strain variation could account for the diversity of human tauopathies.
Presynaptic and postsynaptic effects. Hyperphosphorylated tau also has pronounced effects in the presynaptic and postsynaptic regions, as shown by the presence of presynaptic deficits in the entorhinal cortex of transgenic mice that overexpress a mutant form of tau linked to familial FTD149 (Fig. 2b,e). Pathogenic tau can also restrict synaptic vesicle mobilization at presynaptic terminals150. Postsynaptic impairments are discussed in more detail below.
Historically, many studies have investigated synaptic impairments, but the role of tau in neuronal excitability has remained largely unexplored. However, in 2017, hyperphosphorylated tau was shown to reduce hippocampal excitability by relocating the axon initial segment further down the axon151 (Fig. 2e). Antibody-binding studies showed that phosphorylation of tau at specific residues (namely, those recognized by AT180 (Thr231 and Ser235) or 12E8 (Ser262 and Ser356), but not PHF-1 (Ser396 and Ser404)) was necessary for this pathological relocalization151. This observation points towards a role for distinct species of phosphorylated tau in the regulation of neuronal action potential generation. Yet, tau phosphorylation is not purely pathogenic. Site-specific tau phosphorylation at Ser396 is required for hippocampal long-term depression152, whereas tau phosphorylation at Thr205, mediated by mitogen-activated protein kinase 12 (also known as mitogen-activated protein kinase p38γ), alleviates Aβ-induced excitotoxicity by interfering with postsynaptic excitotoxic signalling153.
Tau acetylation at lysine residues (such as Lys280 and Lys174) is attracting increasing attention, as this post-translational modification can inhibit tau degradation154, impair the interaction of tau with microtubules, promote pathological tau aggregation155, disrupt synaptic signalling (by reducing levels of postsynaptic kidney and brain protein (KIBRA), a memory-associated protein) and ultimately drive cognitive deficits155,156,157,158 (Fig. 2b).
Mitochondrial and nuclear effects. Distinct truncated forms of tau, generated by caspases and asparagine endopeptidase, contribute to accelerated neurofibrillary tangle formation and impaired memory function119,159 and elicit mitochondrial dysfunction160,161. Indeed, pathological forms of tau impair all major aspects of mitochondrial function, although mitophagy has not been studied extensively91. Additional impairments have been reported in the nucleus. For example, in a transgenic Drosophila model, pathogenic tau promotes neurodegeneration through global chromatin relaxation162. In primary neurons, tau directly regulates the integrity of pericentromeric heterochromatin, which seems to be disrupted in AD neurons163 (Fig. 2a). Thus, pathological tau impairs functions throughout neurons.
Effects on microglia. With regards to glia, microglial activation can precede neurofibrillary tangle formation, as revealed in tau transgenic mice164, whereas in a tau transgenic strain lacking the microglial fractalkine receptor CX3C-chemokine receptor 1 (CX3CR1), reactive microglia seemed to drive tau pathology and contribute to the spreading of pathological tau165. A role for soluble hyperphosphorylated tau in driving microglial degeneration has been demonstrated in tau transgenic mice166. However, exactly how glia in the AD brain respond to the different pathogenic forms of tau and Aβ still needs to be determined.
Crosstalk between tau and Aβ pathology
The existence of crosstalk between Aβ and tau in the dendritic postsynapse was established by the finding that dendritic tau mediates Aβ toxicity by targeting FYN into dendrites and spines. FYN then phosphorylates the NMDAR subunit NR2B (also known as glutamate receptor ionotropic, NMDA 2B), which facilitates recruitment of PSD95 to form an excitotoxic complex. Aβ then signals through this complex to mediate excitotoxicity. Interestingly, in the presence of elevated levels of phosphorylated tau (as occur in FTD), FYN levels are increased in the spines, which augments Aβ toxicity114 (Fig. 3a). Nonetheless, tau also has a physiological role in this neuronal compartment, as demonstrated by the activity-dependent translocation of tau into spines, which is disrupted by oligomeric Aβ167.
Accumulation of tau in the cell body and dendrites is partly mediated by Aβ168,169. However, whether an Aβ-mediated mechanism other than the relocalization of tau accounts for the massive accumulation of tau in the somatodendritic compartment is not yet known. In 2017, a cogent mechanism was demonstrated to involve Aβ-mediated local production of tau in the somatodendritic domain108. In this study (which involved cell lines, primary neuronal cultures and several in vivo tauopathy models), oligomeric Aβ caused de novo protein synthesis of tau in the somatodendritic compartment, mediated by FYN, extracellular signal-regulated kinase 1 (ERK1; also known as MAPK3), ERK2 (also known as MAPK1) and ribosomal protein S6. This increase in tau protein levels was associated with increased phosphorylation of tau at tyrosine, threonine and serine residues108. This novel pathological mechanism implies that, in AD, lowering overall tau levels might be a better therapeutic strategy than blocking only serine-directed or threonine-directed tau phosphorylation. The new data also suggest that FYN is a suitable drug target, because it regulates not only tyrosine-directed phosphorylation, but also serine-directed and threonine-directed phosphorylation170.
Additional pathways have been suggested to mediate the synaptotoxic effects of Aβ through tau, including the calcium/calmodulin-dependent protein kinase kinase 2 (CaMKK2)–AMP kinase pathway for oligomers171. The type I oligomer Aβ*56 can form a complex with synaptic NMDARs, resulting in aberrant increases in intracellular calcium levels and activation of calcium/calmodulin-dependent protein kinase type II subunit-α (CaMKIIα), which then pathologically phosphorylates tau172 (Fig. 3a). Tau phosphorylation caused by these signalling cascades was formerly assumed to facilitate the formation of an excitotoxic complex, until a mechanism of increased Thr205 phosphorylation mediated by p38γ was identified in 2016 and shown to promote dissociation of this complex, thereby inhibiting Aβ-induced excitotoxicity153. That tau is essential for the synaptic toxicity of Aβ was demonstrated by the finding that hippocampal slices from tau-knockout mice are resistant to the impairment of long-term potentiation induced by oligomeric Aβ173. Aβ oligomers can also cause synaptic impairment (by a tau-dependent mechanism that involves severing of microtubules) via the proteins spastin and tubulin polyglutamylase TTLL6 (also known as tubulin–tyrosine ligase-like protein 6)174 (Fig. 3b). Together, these observations demonstrate a multifaceted interplay of Aβ and tau at the synapse; however, more work is required to complete this picture and to determine the relative contributions of these convergent pathways.
Finally, crosstalk also exists at the level of mitochondrial function, because combining Aβ and tau pathologies in transgenic mouse models leads to synergistic reductions in mitochondrial membrane potential, ATP synthesis and respiration as well as synergistic increases in levels of ROS92. Other aspects of mitochondrial function are also impaired by the complicated crosstalk between Aβ and tau, including mitochondrial dynamics175,176, transport177 and mitophagy161 (Fig. 3c). Together, these studies paint a complex picture of Aβ and tau interactions.
Assessment of Aβ and tau pathology
Validated Aβ and tau biomarkers are required to improve the staging and differential diagnosis of AD and for use in clinical trials (especially studies of short-term treatments)178. However, the field of Aβ and tau detection presents specific impediments. Whereas the detection of particular Aβ species is challenging owing to their low concentrations179, tau is generally investigated as though it were a single protein, ignoring differences in isoforms, subcellular localization and a wide variety of poorly understood post-translational modification fingerprints, which only increase in complexity under pathological conditions21,180.
Antibody-based detection. Studies of post-mortem brain tissue routinely use classic biochemical methods, such as thioflavin S, Congo red or Gallyas staining, or histological detection methods that often rely on a small set of antibodies. Unfortunately, over time, these antibodies can undergo changes in affinity and specificity, and these changes (together with the lack of consensus regarding which antibodies should be used) often hinder the comparison of different studies and prevent firm conclusions from being drawn. Furthermore, some frequently used Aβ antibodies (such as 4G8 and 6E10) also detect APP because of the presence of a shared sequence181. Other antibodies are specific for Aβ, such as 3D6, or plaques (such as mE8-IgG2a, which detects pyroglutamylated Aβ (3 pR-Aβ1–42))182. Many modification-specific anti-tau antibodies became available in the early days of tau research and enabled the identification of modified tau under pathological conditions20,183. Other useful anti-tau antibodies are pan-specific or isoform-specific or are able to discriminate between human and mouse tau, which is critical when working with transgenic animal models184,185.
Imaging studies. In living patients, visualization of Aβ and tau pathology became available with the development of Aβ PET tracers. The first such tracer, 11C-Pittsburgh compound B, targets Aβ plaques and has a half-life of just 20 min186, whereas 18F-based Aβ PET tracers (such as florbetapir, florbetaben and flutemetamol) have the advantage of half-lives measured in hours187. As a result of the advent of PET tracers for tau such as 18F-FDDNP, 11C-PBB3 and 18F-AV-1451 and members of the 18F-THK arylquinoline series, which are currently in clinical development, longitudinal studies of the evolution of tau and Aβ pathology and how this correlates with cognitive impairment during disease progression are now possible188. However, these scans require specialized and expensive equipment and are not widely available.
Currently available biomarkers. Other methods of investigating Aβ or tau as pathological markers have focused on their detection and quantification in body fluids such as CSF and blood189,190,191 (Table 3). Aβ42 levels in CSF seem to be able to distinguish AD from FTD, but not from other non-AD dementias (possibly because of the presence of mixed pathologies)192, whereas the Aβ42:tau ratio in CSF as determined by enzyme-linked immunosorbent assay (ELISA) is a good predictor of the development of an AD-type dementia193. However, CSF analysis is impractical as a large-scale screening method. Blood could potentially serve as a superior source, as it is easily accessible and requires less-invasive sampling techniques194,195. Indeed, a meta-analysis published in 2016 confirmed a strong association between AD and the core CSF biomarkers of neurodegeneration (total tau, phosphorylated tau and Aβ42) as well as with the CSF biomarker neurofilament light chain and the plasma biomarker total tau196. Additionally, Aβ can now be detected in saliva in concentrations as low as 20 pg/ml with the use of a nanoparticle-based assay197, although an earlier study employing a Luminex assay could detect tau, but failed to detect Aβ42, in saliva198. Salivary lactoferrin levels can discriminate patients with mild cognitive impairment and AD from healthy controls199, but salivary levels of Aβ and tau still need to be established as useful biomarkers200. Even less work has gone into detecting Aβ and tau in urine and exploring their usefulness as biomarkers201.
Advanced assessment methods. A few emerging highly sensitive techniques can detect traces of Aβ and tau in various body fluids; however, these methods have not yet been fully developed as diagnostic tools. One such method employs voltametric detection of Aβ oligomers in the 100 pM ranges; signals are generated from a complex formed by anti-Aβ antibodies and DNA–peptide aptamers combined with gold particles and thionine groups202. In another study, a four-electrode electrochemical biosensor detected full-length 2N4R tau in serum with exceptionally high sensitivity (lower limit of detection 0.03 pM), opening up the possibility that extracellular tau could be used as a biomarker in AD203.
Several cutting-edge biochemical techniques have been employed to study protein aggregation. For example, fluorescence correlation spectroscopy has been used to study the turnover of intermediates during Aβ aggregation, revealing aggregates that ranged from 260 kDa to >106 kDa in size204. Researchers using this same technique have also observed an interaction of Aβ with apolipoprotein E (ApoE, an AD risk protein) at a single-molecule resolution and revealed that ApoE3 (unlike ApoE4, which is also associated with an increased risk of AD) delayed the oligomerization of synthetic Aβ in solution, suggesting that this process also occurs in vivo205. Mass spectrometry has also been employed206,207 but is not likely to be used for routine diagnostic purposes. Surface-enhanced Raman spectroscopy, which uses laser nanotextured substrates as the capture surface for Aβ oligomers, has proven useful as a fast and label-free method for testing AD biomarkers208.
With the advent of powerful super-resolution microscopy techniques, it is now possible to deepen our insight into the cellular processes and compartmentalization of Aβ and tau processing, modification and aggregation209. For example, stimulated emission depletion microscopy has been used to analyse the structural organization of Aβ and tau in the CSF of patients with AD210. Extended-focus Fourier domain optical coherence microscopy has been used to generate stain-free, high-resolution 3D images of Aβ plaques, raising the possibility of performing minimally invasive Aβ studies in vivo211. Confocal microscopy of APP tagged with a photo-activatable fragment of green fluorescent protein revealed fast APP transport from the trans-Golgi network to lysosomes, where APP is enzymatically processed to generate Aβ fragments212. In conjunction with the newly available transcription activator-like effector nuclease (TALEN) genome-editing tool, a photo-inducible human tau transgenic mouse line was generated to study the mobility of endogenous tau in response to various stimuli without the confounding effects of overexpression213. These microscopy methods are expected to become increasingly important for understanding the subcellular compartmentalization of Aβ and tau toxicity.
Oligomerization states remain difficult to evaluate by microscopy, owing to technical limitations. Consequently, spatial intensity distribution analysis was developed, which can measure oligomerization states in different subcellular compartments in live cells214; however, to our knowledge, this method has not yet been applied to tau or Aβ. In silico modelling of the spontaneous aggregation of Aβ has demonstrated that monomers tend to aggregate into stable globular-like oligomers in a process of initial collapse followed by slow relaxation215. We anticipate that such computer simulation models will not only help to elucidate pathogenic mechanisms of protein aggregation but will also facilitate therapeutic interventions216. In 2017, cryo-electron microscopy structures of tau filaments were published at an impressive 3.5 Å resolution217, and this structural information might be useful for drug design.
We expect that these new techniques will help to uncover the pathogenic events that initiate misfolding of Aβ and tau and lead to the formation of aggregated species as well as add to the repertoire of biomarkers by increasing the specificity and sensitivity of detection methods aimed at differentially diagnosing AD at its insidious stage.
Implications for treatment
The amyloid cascade hypothesis (which guided the first clinical trials in AD) led to a toxic role being assigned specifically to soluble Aβ42 species, which were considered the preferred therapeutic target. Nonetheless, clinical trials of humanized monoclonal antibodies such as solanezumab, which binds to small soluble Aβ species, and the small-molecule verubecestat, which reduces Aβ production by inhibiting BACE1, have all failed218. However, given the complexity and diversity of Aβ species, we can reasonably assume that a given monoclonal antibody will recognize only a subset of toxic Aβ molecules and will leave other toxic species untouched, a factor that might account for these treatment failures. By contrast, a phase Ib clinical trial of the anti-Aβ antibody aducanumab (derived from a cognitively healthy aged individual) yielded promising results219, which prompted the launch of a phase III clinical trial. The positive results of the phase Ib study underscore the theme that it is important to understand the variability of the disease-related proteins and interacting therapeutic agents, while also implying that a fairly straightforward approach to lowering abnormal Aβ levels could be broadly effective against AD, particularly if administered in the early stages of the Aβ cascade. This possibility is reinforced by the discovery of the A673T mutation in APP, which both lowers the production of Aβ and reduces the risk of developing AD220.
The failed clinical trials of Aβ-targeted agents also led to increased enthusiasm for anti-tau approaches. Tau aggregation inhibitors and anti-tau vaccines have been tested in clinical trials, as have indirect strategies targeting enzymes such as glycogen synthase kinase 3 (GSK3), serine/threonine-protein phosphatase 2A (PP2A) and histone acetyltransferase p300. Many trials of anti-tau agents (including vaccination strategies targeting distinct phosphorylated tau species) are still ongoing, although a shift towards pan-tau-directed agents is evident; however, the tau field has also seen its failures, as illustrated by the aggregation blocker methylthioninium chloride and GSK3 inhibitors. The finding that not all phosphorylation events of tau are toxic adds an additional layer of complexity to the efforts to target tau therapeutically152,153. Considering that Aβ and tau both have important physiological roles, precision medicine will be needed to remove only the toxic species and to target distinct forms of Aβ and tau while maintaining normal levels of nontoxic tau and APP species200.
The current consensus is that therapies for AD should ideally be started before the onset of symptoms; consequently, several current trials are aiming to prevent or slow the progression of AD in people at risk of developing the disease221. Improved therapeutic strategies are also on the horizon. These include new antibody-based techniques, such as bispecific antibodies222, gene-therapy approaches, such as anti-Aβ-specific and anti-tau-specific nanobodies (which can also be employed for diagnostic purposes223,224), and novel methods that increase brain uptake of therapeutic agents generally. For example, therapeutic ultrasound to transiently open the blood–brain barrier might turn out to be effective225. Therapeutic ultrasound applied to APP mutant mice not only effectively clears a range of Aβ species, ranging from monomers to oligomers and high-molecular-mass species, but also restores memory functions226,227. Interestingly, studies in a mouse model of tauopathy show that tau aggregates can also be partially cleared with this approach228, and its applicability to proteinopathies in general can be envisaged.
To date, AD can be accurately diagnosed only post-mortem, although as the disease process becomes better understood, diagnosis and staging of AD in living patients could become possible. Increased understanding of AD will be facilitated by identifying relevant Aβ and tau species, characterizing their functional relevance and employing novel PET, CSF and blood biomarkers that capture the different forms of Aβ and tau and the distinct protein complexes in which they are involved. Aβ and tau are often used as singular descriptors for proteins that in fact encompass a large range of molecules and assembly states. This diversity (and how it is generated) is not only interesting from a pure research perspective, as it informs our understanding of pathogenic processes, but also has clinical implications for the design of therapies and the development of highly sensitive and specific biomarkers.
The failed clinical trials of Aβ-targeted therapies have raised concerns about the validity of the amyloid cascade hypothesis and whether Aβ really has a direct role in AD pathogenesis. Several lines of evidence suggest that distinct soluble species of Aβ and tau, rather than the insoluble end-stage lesions, amyloid plaques and neurofibrillary tangles, exert the majority of the observed toxic effects of Aβ aggregates (Table 2). To treat AD specifically and at an early (presymptomatic) stage, biomarkers and technologies are required that enable the detection and quantification of minute amounts of soluble oligomeric forms of Aβ and tau, as well as the modifications they have undergone, in CSF — and even better, in blood samples. For example, improved understanding of the mechanisms of neuronal spread could lead to approaches to halt AD at its inception.
Because of the multifactorial nature of AD, which is caused by the combination of a host of environmental and genetic factors, preventive or therapeutic approaches might need to be personalized. A prerequisite for such precision medicine is that the characteristic pattern of risk factors and biological dysfunctions can be comprehensively determined for each patient, as reflected by genomic and genetic variants, neuroimaging indicators (structural, functional and metabolic) and fluid-based biomarkers in CSF, blood, urine and saliva. The different types of Aβ and the proteins it forms complexes with could serve as excellent markers. Tau could potentially have a crucial role; in clinically different tauopathies, such as progressive supranuclear palsy and corticobasal degeneration, differences in the tau isoform ratio and post-translational modification fingerprints of the histological lesions are probably reflected by different marker profiles in CSF and blood. Given the complexity of tau species and their pervasive effect on neuronal physiology, combination therapies might be required to simultaneously target tau and Aβ. Moreover, such therapies might need to target more than one toxic Aβ or tau species. Even removing two or three toxic molecular species might not be sufficient to achieve therapeutic outcomes because we currently do not fully understand their interactions and the potential hierarchy of selected targets. Thus, reducing tau and APP levels generally might be an efficient strategy for eliminating several toxic tau and Aβ species at once.
Together, these factors indicate that approaches to improve drug delivery to the brain, in conjunction with combinations of targeted agents, is probably the best strategy to achieve cost-effective therapeutic outcomes. It is further reasonable to assume that similar complexity exists for the signature molecules of other proteinopathies and that this complexity could equally be exploited for the early diagnosis and staging of these diseases.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
J.G. is supported by the Estate of Clem Jones AO, the Australian Research Council (grant DP160103812) and the National Health and Medical Research Council of Australia (NHMRC; grants GNT1037746 and GNT1127999). F.A.M. is supported by the Australian Research Council (grants DP170100125, LE0882864 and LE130100078) and the NHMRC (grant GNT1058769 and NHMRC Senior Research Fellowship GNT1060075). L.-G.B. is supported by the Peter Hilton Fellowship. The authors thank R. Tweedale for critically reading the manuscript.
- Long-term potentiation
A cellular mechanism underlying learning and memory that involves a persistent increase in synaptic strength following high-frequency stimulation.
A post-translational modification involving conjugation with small ubiquitin-like modifiers (SUMOs).