The Role of Tau in Neurodegenerative Diseases and Its Potential as a Therapeutic Target

The abnormal deposition of proteins in and around neurons is a common pathological feature of many neurodegenerative diseases. Among these pathological proteins, the microtubule-associated protein tau forms intraneuronal filaments in a spectrum of neurological disorders. The discovery that dominant mutations in the MAPT gene encoding tau are associated with familial frontotemporal dementia strongly supports abnormal tau protein as directly involved in disease pathogenesis. This and other evidence suggest that tau is a worthwhile target for the prevention or treatment of tau-associated neurodegenerative diseases, collectively called tauopathies. However, it is critical to understand the normal biological roles of tau, the specific molecular events that induce tau to become neurotoxic, the biochemical nature of pathogenic tau, the means by which pathogenic tau exerts neurotoxicity, and how tau pathology propagates. Based on known differences between normal and abnormal tau, a number of approaches have been taken toward the discovery of potential therapeutics. Key questions still remain open, such as the nature of the connection between the amyloid-β protein of Alzheimer's disease and tau pathology. Answers to these questions should help better understand the nature of tauopathies and may also reveal new therapeutic targets and strategies.


Protein Aggregation in Neurodegenerative Diseases.
Cells and organisms devote considerable time and energy toward the proper folding of proteins upon biosynthesis from ribosomes and prevention of inappropriate protein aggregation [1,2]. Nevertheless, misfoldings and aggregations do occur, with some proteins being more susceptible than others. Such events would quickly cause problems without means of identifying misfolded and aggregated proteins and clearing them out. Despite the multiple means of chaperoning and correctly folding proteins and clearing out misfolded and aggregated ones, a variety of protein folding disorders can develop in humans as well as other organisms. One example is sickle cell anemia, in which a single missense mutation in the -globin chain of haemoglobin leads to noncovalent polymerization and dysfunctional red blood cells [3]. Another is transthyretin amyloidosis, in which mutant transthyretin, a protein that normally transports thyroid hormone and vitamin A, aggregates in the form of amyloid deposits in the heart, kidney, nerves and intestines [4].
Neurons can be particularly sensitive to the effects of misfolded, aggregated proteins, because these cells are oen in place for life aer they become fully differentiated and postmitotic. Furthermore, their narrow axonal projections can be quite long, providing many opportunities to clog up the system and interfere with essential transport of biomolecules and organelles. Once axons become blocked by aggregated proteins, neurons can then degenerate from axon to cell body. us, the unique biology of neurons is likely the reason why abnormal protein deposition is a common feature in a variety of neurodegenerative diseases [5]. ese include Alzheimer's disease (AD), characterized by the extraneuronal deposition of the amyloid -protein (A ) in the form of plaques and the intraneuronal deposition of the microtubule-associated protein tau in the form of �laments; Parkinson's disease (PD), with neuronal deposits of -synuclein; Huntington's disease (HD), with cytoplasmic and nuclear deposition of the huntingtin protein and fragments thereof; amyolateral sclerosis (ALS), characterized by TAR DNA-binding protein 43 (TDP-43) deposition in ubiquitin-positive inclusions; the prion diseases, such as Creutzfeldt-Jakob disease (CJD) in humans, mad cow disease in bovines, and scrapie in sheep, which are characterized by the aggregation of extraneuronal plaques of the prion protein.
ere has been considerable debate about the pathogenic nature of the protein deposits per se. Indeed, dissociation between protein deposition and neurodegeneration appears to be the case with HD [6] and prion diseases [7] and may well be true of other neurodegenerative diseases [5]. Nevertheless, it is now clear that abnormal forms of these proteins, if not plaques and other microscopically observable deposits, are directly involved in initiating neurodegeneration. is is because missense mutations in the genes encoding each of these proteins (or their precursors) are associated with dominant, inherited forms of their respective diseases. In all of these genetic forms of the disease, mutation of the protein leads to the characteristic pathology, neurodegeneration, and, ultimately, disease onset and progression. Why disease onset takes decades is unclear but may be due to a reduced ability to clear out misfolded and aggregated proteins with age. is might explain why sporadic Alzheimer's disease oen does not present until aer age 70, and its prevalence is as high as 50% among those over age 85.

Tau as a Common Pathological Marker in Many Neurodegenerative Diseases.
Tau deposition in the form of intraneuronal �laments is found not only in AD, but in a variety of other neurodegenerative diseases collectively referred to as tauopathies [8,9]. ese include a spectrum of diseases now placed under the umbrella name of frontotemporal dementias (FTD) (e.g., Pick's disease, corticobasal degeneration, progressive supranuclear palsy, frontotemporal lobar dementia with parkinsonism linked to chromosome 17 , and dementia pugilistica). FTD is the second most common form of presenile dementia aer AD. Tau pathology is also oen seen in some forms of PD, ALS, and prion diseases, although it is not characteristically the most common feature of these other neurodegenerative disorders.
At the microscopic level, tau deposition is readily observed post mortem through silver staining of brain slices [10]. Filaments are found in neuronal extensions, leading to abnormal processes referred to as neuro�brillary tangles (NFTs) by Alois Alzheimer, who �rst described this type of pathology in an AD brain in 1906. Although tau is normally found primarily in axons, with smaller levels in dendrites, it can form �laments in dendrites and in the neuronal cell body, suggesting missorting. Although both amyloid plaques (of A ) and neuro�brillary tangles (of tau) are both common pathological features in AD, the latter correlate more closely with neurodegeneration. is observation, as well as others described later, suggests that aberrant tau might mediate A pathogenicity and be more proximal to neurotoxicity in the pathogenic pathway leading to AD. In genetic forms of FTD, mutations in tau clearly implicate abnormal tau as the initiator of neurodegeneration and disease [11].
e particular type of neurodegeneration that manifests itself is determined by the neurons and networks of neurons in which misfolded tau accumulates and aggregates. In AD, tau pathology apparently �rst occurs in the entorhinal cortex, and then spreading to the hippocampus and cerebral cortex [12][13][14][15][16]. As these brain regions play critical roles in learning and memory, the primary problem in AD is cognitive de�cits. In FTD, tau generally deposits in the frontal and temporal lobes [8], regions of the brain that are important for executive functions and behavior. us, the �rst signs of FTD are typically irrational decisions and socially inappropriate behavior. However, tau pathology can also be found in other brain areas (e.g., basal ganglia and subthalamus) depending on the speci�c type of FTD (e.g., progressive supranuclear palsy).
Why particular neurons are susceptible to the buildup of misfolded tau and tau aggregation is unknown. For wild-type tau, stochastic events may trigger misfolding and aggregation in a given neuron or set of neurons. However, certain mutations of tau are associated with speci�c forms of FTD [8], suggesting that speci�c neurons and neuronal networks are especially susceptible to those particular mutations. On the other hand, the same mutation (P301L) can apparently lead to corticobasal degeneration or FTDP-17 in the same family, suggesting that other genetic, epigenetic, or environmental factors may in�uence which neurons are affected. Regardless of why speci�c neurons are affected, it is critical to understand the normal role of tau and how it becomes pathogenic in order to consider strategies for prevention and treatment.

Tau Biology
2.1. e Tau Gene and Haplotypes. MAPT, the gene encoding the tau protein, located on chromosome 17q21.3, spans ∼150 kb and consists of 16 exons, 11 of which are expressed in the central nervous system [17]. Two main haplotypes exist, H1 and H2, resulting from an ancient inversion of a ∼900 kb region that includes the entire MAPT gene. Although the amino acid sequence between H1 and H2 tau is unchanged, these haplotypes do differ in a set of de�ned single nucleotide polymorphisms (SNPs) and a 238 bp deletion in intron 9 in H2. Interestingly, the H1 haplotype is much more diverse than is H2, which is relatively invariant and found almost entirely in those of European ancestry. While the H1 and H2 inversion is estimated to have occurred ∼3 million years ago, the H2 haplotype was apparently reintroduced in the European population some 10,000-30,000 years ago. Since Homo sapiens and Homo neanderthalensis coexisted in Europe during that time, it was suggested that Neanderthals may have reintroduced H2 into the Homo sapien genome [18]. Recent determination of the Neanderthal genome, however, does not support this contention [19].
Although H1 has been reported to be overrepresented in Caucasian cases of progressive supranuclear palsy [20] and sporadic PD [21], the association with MAPT per se is unclear. Considering the diversity of the H1 haplotype, more re�ned analyses support a particular subhaplotype, H1c, being highly associated with PSP and CBD, with H2 being strongly negatively associated [22]. e H1c subhaplotype has also been reported to be associated with sporadic AD [23], although this has been controversial. Since the H1 and H2 haplotypes do not differ with respect to tau protein Scienti�ca 3 sequence, any pathogenic effects associated with a particular haplotype or subhaplotype may instead be due to differences in transcription, splicing or other posttranscriptional modi-�cations, transcript stability or localization, or rates of translation. It is already clear that silent or intronic mutations that shi tau pre-mRNA alternative splicing can cause familial FTD (see below), and recent evidence suggests that the H1c subhaplotype is associated with a similar alteration in splicing [24,25].

Protein Expression, Sequence, and Splice Isoforms.
Tau protein is expressed in a variety of mammalian tissues, being found at relatively high levels in the heart, skeletal muscle, lung, kidney, and testis and at low levels in the adrenal gland, stomach, and liver [26]. However, it is most highly enriched in neurons in the central nervous system. is makes sense, as tau is a microtubule-associated protein [27,28], and microtubules are especially important to neuronal structure and function. e tau protein is primarily found within axons [29], although it is also found at low levels normally in dendrites [30]. Abnormal accumulation of tau in somatodendritic compartments is associated with Alzheimer's and other tauopathies [31].
e expression of the different tau isoforms in the brain is under developmental control, suggesting that the regulation of tau isoforms is important during formation of the brain. Human fetal tau is almost exclusively 0N/3R tau, while adults express all six isoforms [38,40]. Disruption of the normal 4R-to-3R ratio is associated with neurodegeneration (see the following). Consistent with an important role of tau in brain development, microdeletions in MAPT are associated with mental retardation [41,42], and knockdown of tau can result in reduced neuronal connectivity in the brain [43]. Despite these observations, complete knockout of tau in mice yields viable and fertile animals with grossly normal brain anatomy and no evidence of neurodegeneration [44][45][46][47][48], although this may be due to functional redundancy with other microtubule-associated proteins (MAPs). Of four independent tau knockout lines that have been established so far, only one has been reported to have motor de�cits and hyperactivity, with debatable learning impairments [46]. No de�cits of any kind have been reported in hemizygous knockout mice.

2.3.
Phosphorylation. e tau protein can undergo a variety of phosphorylation events at its many serine, threonine, and tyrosine residues, although most occur at Ser/r-Pro sites [49]. Some 30 or more possible phosphorylation sites are found in every tau isoform. ese phosphorylations can control the normal biological functions of tau, such as its role in microtubule stability, as well as its pathological functions, such as its ability to self-assemble into neuronal �laments found in neurodegenerative diseases. e functions, if any, either physiological and pathological, of many of these phosphorylations are unknown, and in some cases it is unclear if they even occur in vivo. Moreover, these phosphorylations can be mediated by numerous candidate kinases in vitro, but the ability of many of them to carry out the speci�c phosphorylations of tau in cells and in vivo at physiological levels and locations is likewise unclear. A diagram showing sites on tau that have been reported to be phosphorylated is depicted in Figure 2.
Some of these phosphorylations, however, do appear to be particularly relevant. Among these are Ser202, r205, Ser396, Ser404, as well as repeated KXGS motifs located within each microtubule binding domain. e microtubuleaffinity-regulating kinase (MARK) selectively phosphorylates the KXGS motifs, particularly the one including Ser262 [50,51]. Phosphorylations at these sites reduce the affinity of tau for microtubules and may thereby regulate microtubule formation and function [52,53]. e glycogen synthase kinase 3 (GSK3 ) and cyclin-dependent kinase 5 (cdk5) are also associated with many of tau phosphorylations, both in vitro and in vivo, although the ability of cdk5 to do so directly in vivo is unclear [54,55]. Cdk5 is activated by the noncyclin regulatory proteins p39 and p35 in postmitotic cells; however, proteolysis of p39 and p35 by calpain to p29 and p25, respectively, results in prolonged half-life of these regulatory proteins and, therefore, prolonged cdk5 activity and increased tau phosphorylation. Protein kinase A is also especially implicated in tau phosphorylations [56,57]. Although tau is hyperphosphorylated in AD, it is unclear which particular sites are relevant, or even whether hyperphophorylation occurs before or aer �lament formation. Tau can also undergo dephosphorylation by phosphatases such as PP1 and PP2A. e balance between kinase activity and phosphatase activity is critical to the phosphorylation state of tau and therefore to tau function and dysfunction.

Role of Tau in Microtubule Formation, Stability, and
Dynamics. e most studied cellular role of tau is in its interaction with and effects on microtubules. While microtubules are most noted in cell biology for the formation of the mitotic spindle during cell division, these polymers of tubulin are essential to postmitotic cells as well. Microtubules are particularly important for neuronal integrity and function, providing structure to the axonal processes that extend from the neuronal cell body and a path for the transport of materials to and from the oen relatively distant synapses. Tau is involved in the assembly and stability of microtubules [27,28], roles that have been extensively studied; however, in vivo, tau may be more involved in microtubule dynamics, in the assembly/disassembly process critical to axonal transport. In addition, the gradient of tau along the axon, with levels highest at the synapse [58], likewise suggests a role in axonal transport, and tau can block the binding of motor proteins that move materials along microtubules [59,60]. us, the high levels of microtubule-bound tau near synapses may favor the release of motor proteins and their cargo where needed.
As mentioned earlier, despite these seemingly critical functions of tau, tau knockout mice are viable, fertile, and relatively normal, with no signs of neurodegeneration. Furthermore, knockdown of tau with siRNA does not kill primary neurons in culture or prevent axon formation [61]. us, tau is not essential to neurons or to microtubules, likely due to compensation by and/or redundant functions of other proteins. Indeed, the microtubule-associated protein 1B (MAP1B), also found primarily in axons, apparently plays such roles. MAP1B is likewise involved in microtubule stability, and knockout of MAP1B results in abnormal brain development and early death [62]. Knockout of both tau and MAP1B worsens this phenotype [62]. However, to determine if MAP1B or other factors compensate for the loss of tau during development, conditional tau knockout mice should be generated in order to abrogate tau in the adult mouse brain. Addressing this issue is critical for determining how misfolded or aggregated tau might cause neurodegeneration (discussed in more detail later).

Other Tau Functions.
Tau has been reported to interact with a number of other proteins besides tubulin/microtubules [63], although the biological relevance of many of these interactions is not clear. However, in recent years evidence has been mounting that the interaction between tau and the protein kinase Fyn may be important for modulating neuronal cell signaling through Fyn [30]. As mentioned earlier, tau is found normally at low levels in dendrites, colocalizing with Fyn and the postsynaptic density protein 95 (PSD95). ese three proteins can be coimmunoprecipitated, suggesting that tau may be serving as a scaffolding protein along with PSD95 to localize Fyn at synapses, enabling Fyn activation through NMDA receptors. Indeed, tau is required for phosphorylation of NMDA receptor subunit NR2B in dendrites. e tau-Fyn interaction may also play a role in process extension in oligodendrites [64].

Alzheimer's Disease
3.1.1. Amyloid and Tau. e AD brain is characterized by extraneuronal deposits of A and intraneuronal �laments of tau [10]. A is produced through proteolytic processing of the type I integral membrane protein amyloid -protein precursor (APP) [65] (Figure 3). e membrane-tethered -secretase sheds the APP ectodomain [66], followed by cleavage of the 12-kDa remnant within its transmembrane domain by the -secretase complex [67]. Dominant missense mutations that cause early-onset, familial AD are found in the APP gene, in and around the small (4 kDa) A region of the much larger APP, and these mutations affect the level, type, or aggregation properties of A produced [68]. Dominant missense mutations are also found in Presenilin, the protein of which is the catalytic component of -secretase, and these mutations also change the type of A , increasing the proportion of longer, more aggregation-prone forms of the protein [68]. No mutations in MAPT have been found associated with AD. Rather, such mutations are associated with FTD (discussed later). us, human genetics supports A as the initiator of AD pathogenesis [68]. However, tau is a common feature of a spectrum of neurodegenerative diseases, and A may be one of many possible ways of triggering tau pathology [9]. Further evidence that tau is downstream of A but critical to AD pathogenesis comes from mouse studies. Mice that are doubly transgenic for human APP and presenilin-1 carrying familial AD mutations display amyloid plaques pathology and learning and memory de�cits [69]. Crossing these mice with tau knockout mice revealed that loss of tau did not affect APP expression, A production, or plaque formation [47]. However, the cognitive de�cits were rescued in a genedosage-dependent manner, with loss of one tau allele providing partial rescue. In addition, a seizure phenotype observed in the transgenic mice was likewise rescued. Surprisingly though, in another report, crossing a tau knockout with a human mutant APP mouse led to increased axonal degeneration [70]. Mice transgenic for human tau can show tau pathology and cognitive and behavioral de�cits [9]. However, crossing these mice with the APP/Presenilin-1 mice resulted in triple transgenic mice with increased tau pathology, along with amyloid pathology, and more severe learning and memory de�cits [71]. Treatment of the triple transgenic mice with anti-A monoclonal antibodies attenuated the tau pathology as well as the cognitive de�cits [72]. Overall, these �ndings suggest that tau is part of the pathogenic process subsequent to A and might be a worthwhile target for therapeutic intervention.
Cellular studies also support the ability of A to affect tau. Treatment of neuronal cultures with A can lead to hyperphosphorylation of tau [73,74], although this has not been observed in APP transgenic mice with endogenous tau expression. Perhaps due to the increase in tau phosphorylation and reduced affinity of tau for microtubules, A treatment can also induce the inappropriate increase of tau in dendritic spines, a phenomenon that correlates with a degree of spine degeneration [74]. Most recently, dimeric A isolated from postmortem human AD brains has been reported to cause tau hyperphosphorylation in cultured rat hippocampal neurons, release of tau from microtubules, and neurite degeneration [75]. Moreover, A oligomers containing a truncated, N-terminally pyroglutaminylated form of A implicated in AD pathogenesis were found to be highly toxic to cultured neurons in a tau-dependent manner [76]. However, despite the possibility that A may affect tau phosphorylation, mislocalization, and neuronal health, the ability of A to cause assembly of endogenous tau into oligomers or �laments has not been observed.
How A might cause these changes in tau is not clear. However, one possible connection is through the protein tyrosine kinase Fyn, which, as mentioned earlier, can interact with tau and PSD95 in dendrites and thereby trigger phosphorylation of the NMDA receptor subunit NR2B. e cellular prion protein (PrPc) has been recently reported as a receptor for soluble A oligomers [77], forms of A more implicated in AD pathogenesis than amyloid plaques. is interaction can lead to increased Fyn activity and NR2B phosphorylation in neurites. Whether knockout or knockdown of tau can alter this affect of A remains to be tested, but as pointed out previously, tau is required for NR2B phosphorylation by Fyn in dendrites. Further evidence for an A -Fyn-tau pathogenic pathway in AD: Fyn overexpression worsens the neuronal and cognitive phenotype of APP transgenic mice [78,79], and this effect is attenuated by knockout of tau [80].

Loss versus Gain of Function
. e question of whether aberrant tau is pathogenic due to a loss of function or a gain of function has been debated considerably over the years. On the one hand, hyperphosphorylation of tau and tau aggregation may lead to less tau that is capable of facilitating microtubule formation, stability, and dynamics. As microtubules are especially important for neuronal health, particularly in axons, reduction of functional microtubules would lead to disruption in the transport of molecules and organelles to and from the synapse, thereby resulting in synaptic dysfunction and degeneration. On the other hand, hyperphosphorylated and/or aggregated forms of tau may result in a toxic gain of function, in which the aberrant species per se cause neuronal dysfunction and degeneration. e aggregated forms, for instance, may block the axon to prevent proper transfer of materials along microtubules. It seems clear that pathogenesis is not strictly due to a loss of function; as the tau knockout mice do not display neurodegeneration, tau can be knocked down in neuronal cultures without causing cell death or preventing axon formation, and none of the FTD-causing MAPT mutations cause deletion or truncation of the protein. ese facts, however, do not preclude that a reduction of microtubule-associated tau, due to tau hyperphosphorylation and/or aggregation, may be a secondary contributor to neuronal dysfunction and death along with the primary contributor, an aberrant gain of toxic function.

e Spread of Tauopathy.
Tau pathology in AD is �rst observed in the entorhinal cortex [12]. is region of the brain, located in the caudal end of the temporal lobe, is connected to the hippocampus and the neocortex and critical for memory formation and consolidation. From the entorhinal cortex, tau pathology spreads, apparently from neuron to neuron via synaptic connections, to the hippocampus, and then to the neocortex [12]. Studies in cell culture have shown that extracellular aggregates of tau can be taken up by cells through endocytosis and cause aggregation of intracellular tau that then spreads throughout the culture [81][82][83]. Similar experiments have been performed in vivo, in which brain extracts containing aggregated tau were injected into the brains of mice transgenic for human tau, leading to aggregation of tau expressed from the transgene and the spreading of tau pathology into connected regions of the brain [84].
Most recently, mice were engineered to express human P301L tau speci�cally in the entorhinal cortex in response to tetracycline treatment [15,16]. is mutation causes familial frontotemporal dementia with typical tau pathology. Induction of mutant tau expression in the entorhinal cortex in adult mice resulted in tau aggregation and neurodegeneration in this speci�c region. From there, the pathology spread to the hippocampus and neighboring regions, areas in which only endogenous mouse tau was expressed at normal physiological levels. With the caveat that very low levels of the mutant transgenic tau might have been expressed in areas other than the entorhinal cortex, the results suggest that pathological tau can induce similar aggregated states in normal tau in neighboring or connected neurons. Such a process is reminiscent of the mechanism by which prion proteins propagate: a pathological conformation that leads to aggregation can induce normal forms of the protein to take on the same pathological conformation [86]. While tau is not infectious like pathological prion proteins, it can apparently propagate a pathological conformation from neuron to neuron in a similar manner.
How aberrant tau might spread from one neuron to another though is unclear. One possibility is that misfolded and aggregated tau is released upon neuronal cell death, whereupon it is taken up by neighboring neurons via endocytosis. Another is that tau is somehow secreted from neurons, even though it is not the kind of protein that generally enters the secretory pathway. A particularly intriguing alternative is that tau is contained within exosomes [87], microvesicles that are released from the cell when multivesicular bodies fuse with the plasma membrane. Subsequent fusion with the membrane of a neighboring neuron results in uptake. e fact that tau is released, whether by dead neurons or live, by secretion or in exosomes, is clear: tau is detectable in human cerebral spinal �uid �CSF�, and an increase in CSF tau correlates with AD. Indeed, CSF tau or phosphorylated tau is a leading biomarker for prodromal AD [88].

Oligomers versus Aggregates.
For both amyloid plaques and tau �laments, a major question for many decades has been whether these microscopic deposits are involved in the pathogenesis of AD, are markers where neuronal cell death and destruction have already occurred, or are otherwise tightly coupled to the actual pathogenic forms of the proteins. e formation of insoluble aggregates of A or tau could even be a protective response, reducing the levels of the real neurotoxic species and allowing the clearance of these proteins through processes such as phagocytosis or autophagy. For A , this may well be the case, as amyloid deposits per se do not correlate well with neurodegeneration [89]. Tau �lament formation and neuro�brillary tangles are more closely correlated with neurodegeneration than amyloid plaques are. For A , soluble oligomeric forms of the protein, rather than insoluble deposits, have been especially implicated in AD pathogenesis [90]. As for tau, soluble oligomers at the very least appear to be on the pathway to NFTs [91][92][93] and may well be pathogenic themselves.
Neuronal cell death and synaptic defects have been seen independent of tau-containing NFTs in mice transgenic for wild-type human tau [94,95]. Loss of synapses in the hippocampus, impaired synaptic function, and microgliosis also occurred before formation of NFTs in an FTDmutant human tau mouse model [96], observations that have been reported in other tau animal models as well [97,98]. Intriguingly, a mouse line with inducible FTDmutant tau recovered in cognitive function and stabilized in number of neurons upon lowering transgenic expression, even though NFT formation continued [99]. Moreover, accumulation of oligomeric tau correlated better than NFTs with neurodegeneration and behavioral de�cits in two different tau transgenic mouse models [100,101]. In human AD, neuronal loss can apparently precede NFT formation [13], and granular tau oligomers have been detected and isolated at very early stages of the disease [91,102]. Taken together, these results suggest that tau oligomer accumulation, neuronal loss, and behavioral de�cits occur before the formation of NFTs. Nevertheless, it seems difficult to envision that accumulation of tau �laments in neuronal soma and processes can be consistent with a healthy, functioning neuron. Indeed, extraneuronal tau �laments are found in the shape of neuronal cell bodies, and hence their name "ghost tangles" [103]. NFTs are not likely harmless, although perhaps they are less deleterious than soluble tau oligomers. However, few studies have examined tau oligomers in AD patients. e development of tau oligomer-speci�c antibodies [93] may offer an important new tool to elucidate the role of soluble tau aggregates in AD pathogenesis. at least in vitro [104]. Among these proteases, caspases are perhaps the most validated. Caspases are cysteine proteases that cleavage substrates with aspartate residues in position P1. ese proteases are produced as inactive zymogens, and activation of certain caspases is a key step in programmed cell death or apoptosis. Several caspases are capable of proteolyzing tau at D421 (2N/4R isoform numbering), close to the C-terminus, to generate a fragment ∼5 kDa smaller. is truncation of tau has been validated both in vitro and in vivo and is found in NFTs in the AD brain, although its role and that of caspase-3 in pathogenesis are unclear [105][106][107][108][109][110].
Interestingly, treatment of cultured cells with A leads to tau cleaved at D421, and such truncated tau can facilitate aggregation [105,111]. us, a pathway has been suggested in which neurotoxic forms of A can induce caspase cleavage of tau to the D421-truncated form, providing seeds for the �lament formation of full-length tau. However, such aggregation of endogenous tau upon expression of truncated tau in cells has not been observed, except when coexpressed with GSK3 [112]. e meaning of this is unclear, as pseudophosphorylation of S422 inhibits tau cleavage at D421 in vitro [106]. e role of caspase cleavage of tau in neurodegeneration is further complicated by a recent in vivo brain imaging study in tau transgenic mice [113]. In this study, neurons with both activated caspases and NFTs were generally not undergoing apoptosis, suggesting that tau truncation by caspases may correlate with tau �lament formation but not neurodegeneration. Caspase cleavage preceded NFT formation, consistent with, although not proving, a cause-and-effect relationship. Interestingly, most NFT-containing neurons did not contain active caspase.
Calpains, calcium-activated cysteine proteases, have also been particularly implicated in tau cleavage. Calpain are abnormally activated in the AD brain, and levels of endogenous calpain inhibitor calpastatin are reduced [114,115]. e formation of a 17 kDa fragment of tau is seen both in vivo and in vitro [116,117] and is triggered in neuronal cell culture by treatment with aggregated A [118]. e generation of this 17-kDa tau fragment has been attributed to calpain, as its appearance is blocked by calpain inhibitor [118]. Although the exact identity of the fragment is unknown, expression of an approximate recombinant version, tau , can lead to apoptosis in cell culture [118]. However, the presence of the 17-kDa tau fragment has not been detected in human AD brains, raising concerns about its relevance to pathogenesis. Other proteases suggested to cleave tau but with less validation include thrombin, various cathepsins, and puromycinsensitive aminopeptidase [104]. e proteasome has also been implicated and may be particularly important for the degradation of misfolded tau [119,120].

Tau Mutations in FTD.
In 1998 mutations in the tau gene were discovered to be associated with frontotemporal dementia with parkinsonism linked to chromosome 17 (and speci�cally characterized by tau pathology, now called FTDP-17T) [121][122][123]. Initially identi�ed mutations cluster in and around the regions encoding the microtubule binding domains, suggesting that perturbed ability to bind microtubules might be involved in neuronal destruction and death in these families. e FTDP-17T mutations provided clear evidence that alterations in tau alone are sufficient to cause neurodegenerative disease and strongly suggested that aberrant tau plays a pathogenic role in other tauopathies, including AD.
To date nearly 40 mutations associated with FTDP-17T or related disorders have been identi�ed (reviewed in [124]) ( Figure 4). ese are all either missense mutations, mutations that affect splicing, or both, and almost all cluster in the portion encoding the C-terminal region or in an intervening sequence near exon 10. e C-terminal missense mutations all appear to impair tau binding to microtubules and the ability of tau to promote microtubule assembly, while a disease-associated mutation of a conserved arginine in the N-terminal region has been found to disrupt binding to the p150 subunit of the dynactin complex [125]. Most of the silent mutations increase the 4R-to-3R ratio by modulating alternative splicing of exon 10. Missense mutations found within exon 10 only affect the three 4R isoforms, while those found outside this region affect all six tau isoforms. Along with the consistent tau deposition in these familial cases and the observation that complete knockout of tau in mice does not lead to neurodegeneration, the fact that all but one of the disease-associated mutations [85] are dominant strongly suggests a gain of a toxic function: one normal copy remains, and in the case of exon 10 missense mutations, even the 3R tau translated from the disease allele is normal. No disease-associated mutations have been identi�ed that lead to either a truncated protein or the nonsense-mediated decay of message.
Differences in the clinical and pathological phenotypes are seen between the various tau mutations (reviewed in [8,124]). While many mutations lead to a phenotype resembling FTDP-17T, others lead to phenotypes overlapping or identical to Pick's disease, corticobasal degeneration, progressive supranuclear palsy, or AD. Some mutations lead to tau pathology in both neurons and glial cells, while others lead to tau pathology primarily or strictly in neurons. However, rather than suggesting speci�c phenotypic differences between tau mutations, many of these differences may re�ect the different genetic and environmental contexts in which these mutations happen to reside. Ultrastructural differences in tau �laments in brain tissue are also observed between different tau mutations, with some correlating with twisted helical �laments, some with paired helical �laments (similar to what is seen in AD) and still others with straight �laments [124]. How these mutations may lead to differently assembled tau �laments is unclear.

Effects of Mutations on Protein Aggregation.
In general, missense mutations apparently affect tau's role in microtubule assembly. ese mutations are primarily found in and around the microtubule binding domains and reduce the ability of tau to promote microtubule assembly from tubulin [126,127]. Two exceptions are S305N and Q336R, which enhance the ability of tau to facilitate microtubule assembly [128,129]. e S305N mutation alters exon 10 Some of these are speci�c to the 4R isoforms, as they are encoded in exon 10. Mutations in red lead to increased 4R tau relative to 3R tau. e asterisk near S352L denotes a single example of a lethal recessive tau mutation, which presented as respiratory hypoventilation and displayed neuronal tau pathology [85]. (b) Other mutations are found in the intron following exon 10, and these disrupt a hairpin structure in the pre-mRNA that regulates exon 10 inclusion/exclusion. splicing [130]; thus, distinguishing the mechanism by which this mutation exerts pathogenicity is difficult. However, the ability of the Q336R mutation to enhance the microtubule binding role of tau again suggests that the balance of tau proteins capable of interacting with microtubules may be a critical factor ultimately dictating whether tau will selfassemble into �laments or not. Other evidence suggests that missense mutations may directly confer the ability of tau to form �laments. Some studies, for instance, show that a wide range of missense mutants (R5L, K257T, I260V, G272V, ΔK280, P301L, P301S, G335V, Q336R, V337 M, and R406W) promote in vitro formation of �laments in the presence of polymerization-inducing agents such as heparin and arachidonic acid [129,[131][132][133][134][135][136][137]. Filament formation has also been observed in neuroglioma cells upon expression of mutant tau [138].
Phosphorylation is also thought to be critical to the pathogenicity of tau (reviewed in [49]). Tau is phosphorylated in multiple sites by multiple kinases, but some sites tend to be more phosphorylated in tau pathology. Studying the role of these phosphorylations in tauopathies is fraught with difficulties however. In vitro phosphorylations by particular kinases are typically challenging to con�rm in vivo, and the order of events (phosphorylation vis-à-vis microtubule dissociation or �lament formation) is still unclear. Some kinase inhibitors have been shown to dramatically reduce tau pathology in transgenic mice [139,140], but such agents notoriously lack speci�city. Nevertheless, some evidence has been reported that disease-associated tau missense mutations can lead to enhanced phosphorylation [141], and hyperphosphorylation can induce tau self-assembly into �laments [142]. Other studies show that tau mutants bind less to phosphatase 2A [143], the principal phosphatase in the brain, which associates with the microtubule binding repeat domains.

Effects of Mutations on Splicing.
Roughly half of the identi�ed tau mutations associated with FT�P-17 affect the alternative splicing of exon 10 (reviewed in [8,124]) ( Figure  4). While a number of these are intronic mutations near the exon 10 5 ′ splice site, others are mutations in the coding region of exon 10. Virtually all of these mutations lead to an increased inclusion of exon 10 and therefore an increase in the 4R-to-3R ratio.
e silent and intronic mutations that increase in 4R tau would be expected to have the opposite effect on microtubule binding to most of the missense mutations: the former would lead to increased binding while the latter decrease binding. How then can both types of mutations lead to �laments of hyperphosphorylated tau? One idea is that the increase in 4R tau saturates the binding sites on microtubules and thereby results in unassociated 4R tau that is more prone to phosphorylation and self-assembly [124]. Another possibility is that higher association with microtubules may create higher local concentrations of 4R tau and therefore a better microenvironment for assembly. e silent and intronic mutations near the exon 10 5 ′ splice site enhance exon 10 inclusion through two general mechanisms: altering linear cis splicing elements or destabilizing a stem loop structure at the exon-intron junction. e stem loop structure was hypothesized [121,122,144] upon the initial discovery of FTDP-17 mutations in the tau gene, noting the apparent self-complementarity in this region. Subsequent determination of the solution structure of an oligonucleotide based on this exon-intron junction by nuclear magnetic resonance spectroscopy [145] led to re�nement of the stem loop model to seven speci�c base pairs (Figure 4), with an adenosine bulge between the sixth and seventh base pair. e structure further shows that this unpaired purine ring is intercalated back into the Aform RNA duplex. Disease-associated mutations would be predicted to destabilize the stem loop and make this site more available to splicing factors (speci�cally the U1 snRNP that interacts with 5 ′ splice sites; see Figure 4). ermal stability studies of oligonucleotides demonstrated that disease-associated mutations within the putative stem loop lower the melting temperature of the RNA duplex (i.e., where the double-stranded RNA dissociates to the single-stranded form) [145,146]. A minigene construct encoding exon 9-11 recapitulates normal tau exon 10 splicing for the wild-type sequence and increased exon 10 inclusion for disease-causing mutations [147]. is minigene has been used to demonstrate that other mutations speci�cally designed to enhance stability of the stem loop (and located distal to the U1 snRNP binding site) reduce exon 10 inclusion to decrease 4R/3R as predicted [146].
Other evidence has been taken to suggest that the stem loop is not a biologically relevant structure and that these intronic mutations instead either enhance the complementarity of the 5 ′ splice site with the U1 SNP or affect an intron splicing silencer (ISS) and the binding of a repressor protein [130,148,149]. However, U1 SNP complementarity changes cannot explain the effects of all the mutations in this region, an ISS cannot explain the effects of stem-loop stabilizing mutations, and no protein factor involved in ISS recognition at this site has been identi�ed.
Other disease-causing mutations outside the stem loop can also increase exon 10 splicing, likely due to strengthening of an exon splicing enhancer (ESE) or weakening of an ISS or an exon splicing silencer (ESS). Putative ESE regions have been identi�ed in exon 10 [148,150], with one being purinerich ESE and where mutations N279 K and ΔK280 reside. e splicing factor Tra2 , part of a class of RNA-binding proteins that contain serine-and arginine-rich (SR) domains, binds to this purine-rich ESE to enhance exon 10 inclusion [150]. In addition, an intron silencer modulator sequence element has been identi�ed just downstream of the stem loop [149].

Parkinson's and Others.
Tau pathology and tau mutations are found in certain forms of parkinsonism, which is a PD-like movement disorder without all the de�ning pathological features of PD per se [151]. One example is FTDP-17T, mentioned earlier. However, recent genomewide association studies (��AS) have identi�ed the MAPT gene as a risk factor for sporadic PD [152][153][154][155]. is is especially surprising as PD typically does not display NFT pathology. In contrast, MAPT has not been identi�ed as a genetic risk factor for AD, although AD is de�ned by the presence of tau pathology. Perhaps this is yet another piece of evidence that NFTs are not the form of tau that is most pathogenic. Interestingly, tau exon 10 splicing has been reported to be altered in PD, with an increase in 4R isoforms [156], similar to that observed with dominant FTD-associated MAPT mutations.
Tau pathology is also observed in chronic traumatic encephalopathy (CTE), in which multiple concussions or other head injuries elicit neurodegeneration with dementia and PD-like symptoms [157]. e disease is commonly found in athletes involved in contact sports and in military personnel exposed to blasts. One subtype of CTE is dementia pugilistica, named for its association with professional boxers, among whom Muhammad Ali is perhaps the most wellknown case. e role of NFTs in pathogenesis is unclear in CTE, as it is in other tauopathies. However, the presence of similar pathology to that seen with tau mutations that cause other neurodegenerative diseases suggests that some form or forms of aberrant tau are responsible.

Animal Models.
A number of transgenic mice expressing tau have been developed in order to generate disease models for tauopathies, and some show tau pathology and neurodegeneration as well as behavioral de�cits (reviewed in [9]). ese mice have provided critical in vivo models for determining the role of aberrant tau in neurogeneration. For instance, evidence suggests that microgliosis and synaptic pathology may be the earliest manifestation of neurodegenerative tauopathies and that abrogation of tau-induce microglial activation may be therapeutically bene�cial [96]. Crossing these mice with those that overproduce A has provided important models for AD that have shed light on the role of tau with respect to A [72]. A mouse line transgenic for human genomic MAPT expresses all six brain isoforms of tau and allows the study of the splicing of human tau in vivo [158]. Another mouse model likewise provides alternative splicing of the human transgene [159]. Moreover, knockout of endogenous mouse tau alleviates A -induced memory de�cits [47], suggesting that targeting tau may be a worthwhile AD therapeutic strategy, although another report suggests the opposite [70].
Questions remain regarding the pathological role of tau �laments and neurodegeneration. �hile most transgenic tau mice suggest that �lament formation is connected to neurodegeneration (see [160]), results with a mouse line containing an inducible tau gene showed that suppressing tau improved memory even though neuro�brillary tangles continued to grow [99]. One important caveat to this study: tau levels under suppressed conditions in these mice were still dramatically higher than wild-type mice. Other evidence in worms, �ies, and zebra�sh suggest that tau can cause neurodegeneration in the absence of �lament formation [97,98,161]. Nevertheless, it would be surprising if tau �laments within neuronal cell bodies and along axonal and dendritic projections are compatible with normal neuronal function and health.

Tau Kinases.
As hyperphosphorylated tau is associated with NFTs and tau-related neurodegenerative diseases, inhibition of the responsible kinases has been a top strategy for targeting tau for the prevention or treatment of AD and other tauopathies [162]. Inhibitors of GSK3 and CDK5 have been of particular interest. Treatment with the GSK3 inhibitor LiCl induces a reduction of tau phosphorylation at Ser-202 and Ser-396/404 but not at Ser-262 or Ser-422 and has been shown to decrease tau aggregation and reduce axonal degeneration in tau transgenic mice [139].
LiCl is not selective for GSK3 versus GSK3 and can have many other off-target effects, making assessment of its ability to inhibit tau phosphorylation via GSK3 problematic. However, an apparently selective GSK3 inhibitor AR-A014418 was also found to reduce tau phosphorylation, �lamentous tau, and axonal degeneration aer oral administration for one month in tau transgenic mice [139]. An analog of AR-A014418 entered into clinical trials for AD in 2007, but its development has since been discontinued. Another selective GSK3 inhibitor, Nypta (Tideglusib), is noncompetitive for ATP [163] and appears to be well tolerated and to improve cognitive performance in a 4-6-week clinical trial [164]. Although the study was not large enough to be statistically signi�cant and target engagement was not clear, Tideglusib has entered Phase IIb trials.
While such results seem promising in many ways, GSK3 is important for normal human physiology; as its name implies, it is critical to proper glycogen metabolism. Moreover, a selective GSK3 inhibitor was found to cause neurodegeneration in healthy wild-type mice [165], suggesting that at least some level of GSK3 activity is essential to neuronal function. erefore, although abnormally high GSK3 activity is associated with several different diseases (diabetes, cancer, and bipolar disorder) and GSK3 appears to be a promising target, complete inhibition is apparently not desirable.
Cdk5 has also been considered a potentially worthwhile target, especially as this kinase is not involved in the cell cycle and is active in postmitotic cells [166]. However, the development of speci�c Cdk5 inhibitors has proven challenging. Other cdks as well as GSK3 are typically affected as well. Dual Cdk5/GSK3 inhibitors may be desirable, but other cdks should be avoided so as not to interfere with cell replication. Relatively nonspeci�c kinase inhibitors have also been considered, with the idea that multiple kinases phosphorylate tau, and hitting many of these partially would reduce tau phosphorylation while minimizing toxicity. Indeed, one such compound, SRN-003-556, which inhibits GSK3, cdk1, PKA, PKC, and ERK2 to similar degrees, reduced tau phosphorylation and reduced motor de�cits in a tau transgenic mouse model but without affecting NFT count [140], further evidence that NFTs per se may not be pathogenic. Nevertheless, great care should be taken to de�ne the range of targets of this compound and to pro�le its toxicology.

Tau
Aggregation. e assembly of tau into neurotoxic oligomers or �brils makes blocking tau-tau interactions a reasonable strategy for therapeutic intervention. A number of screens in search of compounds that inhibit tau selfassembly have been carried out, and a variety of compound types have been identi�ed, including phenothiazines, porphyrins, polyphenols, cyanine dyes, aminothienopyrazines, anthraquinones, phenothiazolyl hydrazides, and rhodanines [167]. Many of these screens were carried out using the binding of thio�avin T as a readout. io�avin T binds to cross--�bril structures of many assemble proteins, including A and tau, and such binding results in �uorescence. Compounds identi�ed in such screens may inhibit �bril formation but not soluble oligomers and other assemblies. Indeed, the potential to increase soluble oligomers exists. erefore, counterscreens are needed to ensure that compounds block early steps in the aggregation process. One example is to tag tau with a �uorescent moiety and use �uorescence polarization as a readout, as increased size of the assemblies (e.g., monomer to dimer to trimer, etc.) results in an increased signal [168].
e most advanced tau assembly inhibitor is the phenothiazine dye methylene blue (Rember). In 2008, this compound was reported to slow cognitive decline in a small Phase II clinical trial over almost one year, although this result was complicated by the fact that the highest dose led to lower than expected drug exposures, and no evidence for target engagement was reported [169]. A study in a tau transgenic mouse model, however, showed that very high doses that lowered soluble tau levels also resulted in cognitive improvement [170]. While a Phase III trial would help clarify the efficacy of methylene blue in humans, no such trial has yet been initiated. However, a second-generation methylene blue analog, LMTX, is allegedly in line for a Phase III trial. No other tau aggregation inhibitors are in clinical trials at this time.
Among the more promising new approaches to the search for effective tau assembly inhibitors is a structurebased design strategy, targeting the cross--sheet structure of aggregated tau. While the detailed structure of tau �brils is unknown, the structure of �brils of the tau peptide VQIVYK, residues 306-311, was deduced by x-ray microcrystallography [171] and found to be a "steric zipper" of -sheets. is short tau peptide is critical for tau assembly in the context of the full-length protein and on its own assemble into �brils similar to that formed by tau itself. Dpeptides were computationally designed to strongly interact with this structure and prevent �bril elongation [172]. One such peptide was found to delay �bril formation, even at substoichiometric concentrations, and could also inhibit �bril elongation. Small changes to the structure of this Dpeptide rendered it ineffective, and the active peptide was found to have no effect on A �bril formation, a remarkable demonstration of speci�city. �hile this D-peptide itself is not a good drug candidate (peptides, even D-peptides, rarely are), it may be a reasonable starting point for drug design and validates the structure-based design approach.
Despite some promising advances in tau aggregation inhibitors, a number of issues will need to be addressed to facilitate translation into the clinic. First, the stoichiometry required to effectively engage tau and show efficacy should be understood for a given agent. In most cases, tau aggregation inhibitors need to be at least equimolar with tau protein in order to have the desired effects. Secondly, the speci�city of the compounds must be established to ensure that normal protein-protein interactions are not unduly affected. ird, the steps in tau assembly that are affected by the compound should be elucidated. As mentioned earlier, �brils per se may not be the primary neurotoxic entity, and blocking �bril formation alone may lead to the buildup of more deleterious soluble oligomeric assemblies. Fourth, the ability of the compounds to interact with free tau versus microtubule-bound tau should be determined or otherwise address whether there is interference with normal tau functions. Although the knockout of tau in mice leads to viable, fertile, and (for the most part) healthy mice, the effect of conditionally knocking out tau is presently unknown.

Tau Clearance.
Stimulating the natural ability of cells to clear away misfolded or aggregated proteins has been another strategy for targeting aberrant tau. One speci�c approach is by enhancing degradation of tau by the ubiquitin-proteasome system. Misfolded proteins are tagged with chains of ubiquitin that enable interaction with and breakdown by the proteasome. is process can apparently be stimulated with inhibitors of the ATPase activity of heat shock protein (HSP) 90, a molecular chaperone involved in protein refolding. Inhibition of this HSP90 activity results in increased expression of HSP70, which together with CHIP (the carboxyterminus of HSP70-interacting protein) is involved in the ubiquitination of misfolded proteins [173]. Knockout of CHIP in mice results in elevated phosphorylated tau, suggesting that HSP70/CHIP is involved in the degradation of this particular form of tau [174]. Consistent with this notion, HSP90 inhibitors increase proteasomal degradation of phosphorylated tau through HSP70/CHIP in cell culture [175,176] and decrease hyperphosphorylated tau in tau transgenic mice [176,177].
Toward the identi�cation of small-molecule stimulators of the ubiquitin-proteasome system to enhance the clearance of deleterious proteins, the discovery that a proteasomeassociated deubiquitinating enzyme, USP14, inhibits proteasomal degradation of proteins by trimming off ubiquitin chains inspired a high-throughput screen to �nd inhibitors of USP14 activity [120]. e hit compound that emerged from this screen, dubbed IU1, reversibly inhibited USP14 and did so selectively over other deubiquitinating enzymes. IU1 could also stimulate the degradation of tau in a concentrationdependent manner in cell culture, and this tau degradation required USP14 and active proteasome complexes. e degradation of another protein implicated in neurodegeneration, TDP-43, was likewise enhanced by IU1, while a ubiquitin-independent substrate of the proteasome was not affected.
Monomeric, misfolded tau can be degraded by the proteasome, but aggregated forms of tau cannot, as such aggregates cannot pass through the proteasome pore. However, protein aggregates as well as whole intracellular organelles can be cleared through a process called autophagy (literally, selfeating) [178]. In one type of autophagy (macroautophagy) protein aggregates or organelles are enveloped into vesicles that then fuse with lysosomes for degradation. Some evidence suggests that reduction of this process can lead to neurodegeneration [179,180] and that enhancement of autophagy can help clear out aggregated proteins associated with neurodegeneration [181]. e immunosuppressant drug rapamycin can enhance macroautophagy and prevent plaques, NFTs, and cognitive de�cits in an AD mouse model [182]. However, the safety of this agent for long-term treatment of AD in humans is a serious concern. Recent results suggest that trehalose also activates macroautophagy, reduces insoluble tau, and is neuroprotective in a tau transgenic mouse model [183], although trehalose has multiple biological effects, and the safety of this agent at chronic pharmacological doses is likewise unclear. Other means of pharmacologically increasing autophagy that are more speci�c may be required, but even then, the consequences of chronic stimulation of autophagy are unclear.

Antibodies.
Active and passive immunization targeting tau is another promising approach for the prevention and treatment of AD and other tauopathies [184]. e impetus for this approach was promising �ndings on the use of anti-A antibodies in transgenic mouse models, demonstrating that A plaques could be prevented in younger mice and even cleared in older mice and that cognitive de�ciencies in these mice could be prevented or reversed, even upon peripheral administration. ese results have led to a number of clinical trials for the treatment of AD. Despite the preconceived notion that antibodies cannot cross the blood-brain barrier, studies show that low levels of anti-A antibodies do access the brain and elicit clearance of amyloid plaques. However, these plaques are extracellular, while tau and tau pathology is largely intraneuronal, suggesting that anti-tau antibodies may not be capable of engaging their target. Nevertheless, initial evidence shows that tau pathology can be reduced and behavioural phenotypes slowed by active immunization with antiphosphotau peptide in tau transgenic mice [185][186][187][188]. Passive immunization has likewise shown promise [189,190].
How might antibodies prevent tau pathology? One possibility is that they enter neurons through endocytosis, although there is no evidence that this occurs, and even if it did, the antibodies would have to somehow escape the endocytic vesicles to gain access to tau in the cytoplasm. Another possibility is that tau is secreted from neurons to some degree and that this allows the spread of tau pathology from neuron to neuron, a mechanism discussed earlier. In this scenario, anti-tau antibodies would interact with tau at synapses and intercept misfolded or aggregated tau before it is taken up by the neighboring neuron. Consistent with this idea is the fact that tau is found in the CSF and is considered a promising biomarker for AD [88].

Tau RNA.
Because of the difficulty in targeting the tau protein directly (it is not obviously "druggable, " as it has no natural "ligands" per se) and the uncertainty of targeting enzymes that modulate tau (e.g., kinases), some have considered targeting the tau mRNA to either lower overall tau protein levels or to shi splicing in a potentially therapeutically productive way. e idea of knocking down the tau message with siRNA or reducing tau protein translation via antisense oligonucleotides is based on the �ndings in mice that knockout of the MAPT gene attenuates the learning and memory de�cits of FAD-mutant APP/presenilin transgenic mice [47], which suggests that tau is critical to the pathogenic process initiated by aggregated forms of A . Although most reports on MAPT knockout suggest that the mice are normal, fertile, and healthy, the possibility of compensation (e.g., by MAP1) cannot be excluded. As no conditional MAPT knockout mice have yet been reported, it is unclear if tau is critical for normal neuronal function in vivo, so complete ablation of tau by siRNA or antisense oligonucleotides may not be desirable. Also, another report describes crossing a tau knockout with a human mutant APP mouse leading to increased axonal degeneration [70].
Shiing the splicing of tau pre-mRNA is a more subtle strategy that should not affect overall levels of tau protein. e rationale for this approach is the �nding that nearly half of the ∼40 mutations in the MAPT gene that are associated with FTD alter the balance of tau exon 10 splicing to increase the ratio or 4R to 3R tau isoforms. is skewing of tau splicing, with no change in the encoded wild-type protein sequence, is sufficient for tau pathology, neurodegeneration, and dementia. erefore, use of agents that can shi the splicing back toward a more normal 1 : 1 physiological ratio, (or even in favor of 3R) may be bene�cial in treating these FTDs. Such agents might also be therapeutic for AD as well, although there is little evidence that the 4R-to-3R ratio is altered in AD. Antisense oligonucleotides that target the exon10-intron10 boundary can shi splicing in this potentially bene�cial way. ese include antisense directed to the linear sequence at this boundary [191] as well as bipartite oligonucleotides [192] directed to the 5 ′ and 3 ′ regions that �ank the hairpin structure at this boundary, which prevent access by the U1 snRNP. e search for small, drug-like molecules that bind and stabilize this hairpin has also identi�ed compounds such as neomycin and mitoxantrone [193][194][195][196][197]. However, these compounds lack speci�city for the tau RNA hairpin structure.

Other Strategies.
Several other approaches to targeting tau or compensating for potential loss of tau function have been explored. One of these is to target the prolylisomerase PIN1, which switches speci�c tau p-Ser/r-Pro motifs between trans-and cis-rotamer forms [198]. Recent evidence using tau pT231-P232 rotamer-speci�c antibodies [199] suggests that the cis-, but not the trans-, rotamer is an early conformer associated with AD progression; that PIN1 restores the ability of cis-pT231-Tau to promote microtubule assemble; that cis-pT231-Tau is more stable and more prone to aggregation than trans-pT231-Tau; that cis-, but not trans-, pT231-Tau correlates with NFTs in the AD hippocampus; that PIN1 promotes cis-to-trans isomerization of pT231-Tau in AD mouse models. us, increasing PIN1 activity may be a worthwhile strategy for therapeutic intervention for AD at the level of tau. How to do this, however, is not obvious. Alternatively, immunotherapy with the conformer-speci�c antibodies might block the pathogenic functions of cis-pT231-Tau.
Reducing the acetylation of tau may be another reasonable approach for lowering pathogenic forms of tau [200]. Acetylation at lysines oen competes with ubiquitination at the same sites. us, acetylation can prevent the degradation of misfolded tau by the ubiquitin-proteasome system. Tau is acetylated by the lysine acetyltransferase p300 and deacetylated by the lysine deacetylase SIRT1 [201]. us, inhibition of p300 activity or stimulation of SIRT1 activity might facilitate ubiquitination and proteasomal degradation of misfolded tau. Indeed, a small molecule inhibitor of p300 has been shown to prevent neuronal accumulation of phosphorylated tau [201]. However, p300 and SIRT1 have many substrates, raising questions about the consequences of altering their levels of activity.
Another approach is to pharmacologically rescue the potential loss of function of tau in stabilizing microtubules. While the knockout of tau does not result in neurodegeneration and none of the FTD-associated MAPT mutations result in truncation or deletion of large regions of the protein, it remains possible that misfolded or aggregated tau might lead to insufficient levels of tau protein for microtubule stabilization. For this reason, antimitotic agents that act as microtubule stabilizers have been explored for their ability to rescue this tau function. One such agent, paclitaxel, prevented axonal transport de�cits and motor impairments in a tau transgenic mouse [202], and epothilone D improved microtubule density, axonal integrity, and cognition of another tau mouse model [203], even in aged mice that already had pathology [204]. An epothilone analog has likewise shown bene�t [205]. A peptide called NAP also stabilizes microtubules and (surprisingly) reduces tau hyperphosphorylation and ameliorates cognitive de�cits in tau mouse models [206][207][208]. Intranasal delivery of NAP has shown some promise in a phase II clinical trial.

Conclusions and Perspective
Substantial evidence supports the tau protein as a key contributor to the pathogenesis of AD. However, the form of tau that is neurotoxic, the means by which this toxicity is transmitted, and the role of A in triggering aberrant tau are all unresolved issues. Misfolded and/or aggregated tau protein appears to be responsible, but the exact nature of the pathogenic form(s) (e.g., oligomeric state and degree of phosphorylation) remains unclear. Despite this uncertainty, genetic studies in mice suggest that reducing tau expression should be therapeutic for AD, and FTD-causing mutations in the MAPT gene itself indicate a pathogenic role of aberrant tau. Moreover, as tau pathology correlates better with neurodegeneration than amyloid plaques, pharmacological intervention at the tau level may be efficacious later in the disease process than it would be at the A level, which may require presymptomatic diagnosis and treatment.
Many different strategies have been pursued in the hopes of intervening therapeutically at the level of tau ( Figure  5). Nevertheless, tau remains relatively underdeveloped as a therapeutic target, especially compared with A . is is due to the unresolved issues mentioned previously along with the fact that the tau protein itself is not a typical "druggable" target, with endogenous small molecule ligands or otherwise containing pockets to which drug molecules might obviously bind in a pharmacologically meaningful way. Together, these challenges make it difficult to decide on a suitable strategy for drug discovery. More basic research is needed to understand the role of tau in neurodegenerative diseases to identify new targets and approaches for pharmacological intervention. In addition, the delivery of tau-targeting agents to the brain is a major impediment; novel delivery approaches to the brain, such as nanoparticles [209], should be explored in conjunction with drug discovery.
In the meantime, however, one or more of the current approaches may eventually bear fruit in the form of an approved drug for AD and/or related disorders, so these efforts should continue. e medical need is simply too great to halt the ongoing search for an effective means of targeting tau. Only one or a few of the many reasonable approaches need to effectively prevent or delay the onset or progression of tauopathies, and at this point it is impossible to say which of these approaches might be successful. us, all reasonable strategies should be pursued in the hope of �nding diseasemodifying therapies for these devastating brain disorders.