A Decade of Tau Transgenic Animal Models and Beyond

Abstract

The first tau transgenic mouse model was established more than a decade ago. Since then, much has been learned about the role of tau in Alzheimer’s disease and related disorders. Animal models, both in vertebrates and invertebrates, were significantly improved and refined as a result of the identification of pathogenic mutations in Tau in human cases of frontotemporal dementia. They have been instrumental for dissecting the cross‐talk between tau and the second hallmark lesion of Alzheimer’s disease, the Aβ peptide‐containing amyloid plaque. We discuss how the tau models have been used to unravel the pathophysiology of Alzheimer’s disease, to search for disease modifiers and to develop novel treatment strategies. While tau has received less attention than Aβ, it is rapidly acquiring a more prominent position and the emerging view is one of a synergistic action of Aβ and tau in Alzheimer’s disease. Moreover, the existence of a number of neurodegenerative diseases with tau pathology in the absence of extracellular deposits underscores the relevance of research on tau.

INTRODUCTION

Histopathologically, the Alzheimer’s disease (AD) brain is characterized by abundant amyloid plaques, neurofibrillary lesions and the loss of nerve cells and synapses. This review focuses on tau, a microtubule‐associated protein (MAP) and the principal component of the neurofibrillary lesions (41). They are found in nerve cell bodies and apical dendrites as neurofibrillary tangles (NFTs), in distal dendrites as neuropil threads and in the abnormal neurites that are associated with some amyloid plaques (neuritic plaques). In the absence of plaques, tau inclusions are abundant in a range of neurodegenerative diseases, which include Pick’s disease (PiD), progressive supranuclear palsy (PSP), corticobasal degeneration (CBD), argyrophilic grain disease, sporadic frontotemporal dementia (FTD) and the inherited FTD and Parkinsonism linked to chromosome 17 (FTDP‐17) (48, 86).

Tau is expressed predominantly in neurons and at lower levels in astrocytes and oligodendrocytes (143). Moreover, in some diseases, tau also forms aggregates in glial cells and these can outnumber neurons with aggregates (48). Tau contains a particularly high content of serines and threonines, many of which are phosphorylated under physiological conditions (42). Under pathological conditions, tau becomes hyperphosphorylated, which means a higher degree of phosphorylation at physiological sites, as well as de novo phosphorylation at additional sites (15, 43). Phosphorylation decreases the binding of tau to microtubules. This increases the pool of soluble tau and is thought to trigger the disintegration of microtubules (43). In addition to phosphorylation, tau is subject to ubiquitination, nitration, truncation, prolyl isomerization, association with heparan sulphate proteoglycans, glycosylation, glycation and modification by advanced glycation end‐products (AGEs) (20).

Mutations in Tau have not been found in AD. Instead, mutations have been identified in the amyloid precursor protein (APP) gene, from which Aβ is derived by proteolytic cleavage, and in the presenilin‐1 (PS1) and presenilin‐2 (PS2) genes, both of which encode proteins involved in Aβ formation. These pathogenic mutations account for less than 1% of AD cases (27). In 1998, research into tau was spurred on by the identification of exonic and intronic mutations in the Tau gene in FTDP‐17, a familial dementia related to AD (69, 119, 137). These findings established that dysfunction of tau can cause neurodegeneration and lead to dementia. In addition to FTD, Tau mutations can give rise to clinical syndromes that resemble PSP, CBD, PiD and progressive subcortical gliosis (16, 44, 86, 153). So far, 39 different Tau mutations have been identified in over 100 families.

ANIMAL MODELS

Over the past decade, many tau transgenic animal models have been developed that reproduce aspects of AD and FTD (Figure 1). As outlined below, they have been used to identify pathomechanisms, disease modifiers and differentially expressed genes and proteins. This has been followed by validation in tissues from patients with AD and FTD. In addition, these models have also been used to develop novel therapeutic strategies (55).

Figure 1.

Interaction model illustrating the potential of human wild‐type and mutant tau transgenic animal models. AD = Alzheimer’s disease; CBD = corticobasal degeneration; FTDP‐17 = frontotemporal dementia and Parkinsonism linked to chromosome 17; PiD = Pick’s disease; PSP = progressive supranuclear palsy.

Wild‐type tau transgenic mice.  The first transgenic models used expression of wild‐type four‐repeat (49) and three‐repeat (13) human tau. These mice reproduced aspects of human pathology, such as the somatodendritic localization and hyperphosphorylation of tau. Subsequently, stronger promoters were used to drive expression and this resulted in more pronounced phenotypes (71, 120, 138). Non‐filamentous tau aggregates formed, which could be stained with phosphorylation‐ and conformation‐dependent anti‐tau antibodies [reviewed in (48, 53)]. However, despite an age‐dependent decrease in tau solubility, NFTs did not form, unless the mice reached a very old age (72).

Signs of Wallerian degeneration, including axonal breakdown and segmentation of myelin into ellipsoids (so‐called “digestion chambers”), were observed in these mice. Furthermore, neurogenic muscle atrophy, with groups of small angular muscle fibres, was present in the hindleg musculature of transgenic, but not control, mice (120). A prominent characteristic of the transgenic mice was the presence of large numbers of axonal spheroids in brainstem and spinal cord, displaying accumulations of neurofilaments, mitochondria and vesicles (71, 120, 138). Axonal spheroids are a characteristic of most cases of amyotrophic lateral sclerosis (64, 103, 123). They have also been reported in cases of FTD of unknown etiology (155). In one transgenic line, microglial cells with phagocytosed myelin debris and myelin ovoids were present (138). These findings demonstrated that overexpression of human tau can lead to a central and peripheral axonopathy that results in nerve cell dysfunction and amyotrophy. Interestingly, similar axonal swellings have been described in Aβ plaque‐forming transgenic mice (140).

Axonopathy is also found in AD brain, as evidenced by choline acetyl transferase (ChAT)‐immunoreactive swellings (140). Interestingly, this axonopathy is dependent on the Braak stage (used to stage the neurofibrillary pathology of AD), as stages I–III revealed many ChAT‐reactive swellings, whereas these were only half as frequent at stages IV–VI. In contrast, phosphotau immunoreactivity, although present at stages I–III, became much more significant at stages IV–VI. It was concluded that axonal swellings occur before detectable Aβ deposits and pathological tau phosphorylation, and may lead to the formation of dystrophic neurites (140). How spheroids form and further implications of the axonal transport failure hypothesis of neurodegeneration are discussed elsewhere (54).

Mutant tau transgenic mice.  Following the identification of pathogenic mutations in Tau in FTDP‐17, several groups reported on the formation of NFTs in neurons (1, 50, 88, 91, 141, 144, 145) and glial cells (36, 52, 62, 63, 109, 125, 129) in mice transgenic for mutant human tau, with a tetracycline‐regulated transactivator system being used in two models (52, 125). The first published NFT‐forming model expressed human P301L tau under the control of the murine PrP promoter (88). These mice developed motor disturbances by 10 months of age, which were more pronounced than those observed in wild‐type tau transgenic lines (71, 120, 138). NFTs were present in brain and spinal cord and the number of motor neurons in spinal cord was reduced by 50% (88). The P301L mice also showed abnormal tau filaments in astrocytes and oligodendrocytes, reflecting regulation of transgene expression by the PrP promoter (93). Using a different human tau isoform, mutation P301L was expressed under the control of the mThy1.2 promoter, which also led to the formation of NFTs, as identified by Gallyas silver staining and thioflavin S fluorescence (50) (Figure 2). In humans, the P301S mutation causes an earlier onset of clinical signs of FTD than mutation P301L (16). The shortest human four‐repeat tau isoform with the P301S mutation was expressed in mice using the mThy1.2 promoter (1). In brain and spinal cord, the levels of human tau were twofold higher than those of mouse tau. Human tau was expressed in neocortex and hippocampus, but levels were highest in brainstem and spinal cord. A twofold reduction in the number of spinal cord motor neurons was reported, indicating that motor neurons were either especially vulnerable to tau aggregation, or that expression of tau was particularly high in these cells. When antibodies specific for activated MAP kinase, JNK and p38 were used, tau‐positive neurons in P301S tau transgenic mice, but not in mice expressing wild‐type tau, were labeled (1). Interestingly, mice with a tau phenotype due to reduced protein phosphatase 2A (PP2A) activity also showed activation of these signaling pathways, suggesting that tau phosphorylation may have been brought about, at least in part, by activation of MAP kinase and JNK (78, 79).

Figure 2.

Phosphotau staining of the hippocampus of a P301L mutant human tau transgenic mouse.

PSP, CBD and some cases of FTDP‐17 are characterized by a substantial glial tau pathology, which has been reproduced in animal models. The first model expressed G272V mutant tau by combining a PrP‐driven expression system with an autoregulatory transactivator loop that resulted in high tau expression in a subset of neurons and oligodendrocytes. Electron microscopy established that tau filament formation was associated with hyperphosphorylation of tau. Thioflavin S‐positive fibrillary inclusions were identified in oligodendrocytes and motor neurons of the spinal cord (52), but the clinical phenotype was only subtle. In contrast, when human wild‐type tau was overexpressed in neurons and glial cells under the control of the mouse Tα1 α‐tubulin promoter, a glial pathology was found that resembled the astrocytic plaques of CBD and the coiled bodies of PSP and CBD (62). In contrast, no neuronal inclusions were found. The solubility of tau decreased with age, while its phosphorylation increased in parallel. A significant age‐related loss of neurons was only found in 18‐month‐old mice, whereas oligodendrocytes were already lost at 6 months of age. As Gallyas‐positive tau filaments in oligodendrocytes were only observed after 24 months, this does suggest that the accumulation of tau led to cell death well before the formation of abnormal filamentous aggregates. Subsequently, using the 2′,3′‐cyclic nucleotide 3′‐phosphodiesterase promoter, transgenic mice were generated that expressed human P301L tau exclusively in oligodendrocytes (63). Interestingly, structural disruption of myelin and axons preceded the emergence of thioflavin‐S reactive tau inclusions in oligodendrocytes, as did impaired axonal transport, which occurred even earlier, preceding motor deficits (63). The question of whether and to what extent glial pathology contributes to the clinical features of PSP and CBD may best be addressed by behavioral studies using mice with a pronounced glial tau pathology.

To reproduce the Aβ plaque and NFT pathologies in a single model without the need for crossings, triple transgenic mice carrying the PS1 M146V, APP K670N/M671L (Swedish) and P301L tau transgenes were developed. In these mice, synaptic dysfunction, including deficits in long‐term potentiation, were reported to precede Aβ plaque and NFT formation (109). In order to allow a better side‐by‐side comparison of the effects of wild‐type and P301L mutant human tau expression, an approach was chosen whereby two strains were generated, both expressing the longest human tau isoform at similar levels, with one bearing the P301L mutation and the other being wild‐type (145). The two strains developed very different phenotypes. Mice expressing wild‐type human tau developed early motor impairement at around 6–8 weeks of age and an axonopathy in the absence of tau aggregates. However, their survival rates were normal. In contrast, the P301L mutant mice started to develop NFTs at 6 months of age, in the absence of an axonal phenotype, but despite having only minor motor problems, all the mice died before 13 months of age. It was concluded that excessive binding of wild‐type human tau to microtubules, as opposed to reduced binding of P301L tau, may have been responsible for the development of axonopathy and tauopathy, respectively. In addition, a conformational change in P301L mutant tau may have been a determinant in triggering tauopathy. A third line was a knock‐in replacing the mouse gene with a single copy of the wild‐type human tau expression construct used above. These mice had normal survival rates without any obvious pathology and with only minor motor problems later in life (145). When compared with results obtained in the other tau models, it can be concluded that the range of phenotypes encountered was likely due to the use of different promoters for transgene expression, the integration sites of the transgenes, the expression levels of human tau and/or the mouse strains used, without necessarily shedding any light on the underlying mechanisms (53). At the molecular level, much remains to be done to understand how tau impairs neuronal and glial function and causes nerve cell death.

To determine whether NFTs are central to the neurotoxic cascade in AD or represent a protective response, P301L mice were generated in which transgene overexpression could be reduced (125). Mice expressing doxycycline‐repressible human P301L mutant tau developed progressive age‐related NFTs, remarkable neuronal cell loss (70% in the CA1 region) and behavioral impairment. Following a reduction of transgenic tau from 13‐ to 2.5‐fold overexpression, memory function recovered and the number of neurons stabilized, but NFTs continued to accumulate. These findings have shown that dysfunction of tau impairs memory, when mutant tau is massively overexpressed. They further imply that NFTs per se (as entities of fibrillar accumulation that are visible at the light microscopic level) are not sufficient to cause cognitive decline or neuronal death in this model (125). In agreement with this, cognitive impairment in our P301L tau transgenic mice occurred in the absence of NFT formation (112, 114). NFTs are found in only a small percentage of neurons in the animal models published so far and are outnumbered by dysfunctional neurons containing tau aggregates, but lacking NFTs. Therefore, it is not surprising that functional impairment in these mice did not correlate with NFTs, but rather with the larger number of cells containing pre‐NFT aggregates.

Behavioral studies have also been performed with the V337M tau mutant mice, which expressed transgenic tau only in the hippocampus. These mice showed increased locomotor activity and memory deficits in the elevated plus maze, increased spontaneous locomotion in the open‐field, but no significant impairment in the Morris water maze (142). R406W tau mutant mice expressed highest tau levels in the hippocampus and, to a lesser extent, other cortical and subcortical brain areas (144). A slight decrease in locomotor activity was shown in these mice during the first 5 minutes of the open‐field test and, in addition, a significant impairment in the contextual and cued fear conditioning test was present.

Another approach is to take advantage of virus‐based expression to address the role of tau in disease. In one study Sindbis virus was used to express human tau in hippocampal slices. Tau acquired a pathological conformation, as evidenced by Alz50/MC1‐reactivity. It became hyperphosphorylated and insoluble, resulting in region‐specific neurodegeneration. Live imaging demonstrated that tau‐dependent degeneration was associated with the development of a “ballooned cell” phenotype. Surprisingly, spine densities and morphologies were not altered, suggesting that synaptic integrity was remarkably stable in the presence of tau‐dependent degeneration (131).

Crosses onto a tau null background.  The first tau knock‐out model was established in 1994 by insertion of a neomycin resistance cassette into exon 1 of Tau (60). In large‐calibre axons, an increase in MAP1A appeared to compensate for the loss of tau. In some small‐calibre axons, however, microtubule stability was reduced and microtubule organization was significantly changed (60). Mice lacking tau developed muscle weakness, as determined by the wire‐hanging test, hyperactivity in a novel environment and impairment of contextual fear conditioning. They also had a tendency to fall off the rod more easily in the narrow beam test (70). In cases of sporadic and familial forms of FTD without tau pathology, reduced levels of tau protein, but not mRNA, have been reported (6, 156). It suggests that the loss of tau protein itself may lead to some of the neurological characteristics observed in FTD and that a loss of physiological function of tau in the presence of FTDP‐17 mutations may contribute to disease in mutation carriers. A reduction in tau levels may be a cause of mental retardation in humans, as suggested by the recent reports of heterozygous microdeletions on chromosome 17q21.31 (80, 95, 132, 133, 149).

A second tau knock‐out strain was established by inserting a neomycin resistance cassette into exon 1 of Tau (26). In contrast to the above strain, primary hippocampal cultures from these mice showed a significant delay in maturation, as judged by axonal extension. The technique of selectively enhancing axonal growth on laminin substrates failed to restore normal neuronal maturation of the tau knock‐out neurons. By crossing tau transgenic with knock‐out mice, tau‐deficient neurons were reconstituted with human tau, thus restoring a normal pattern of axonal growth and neuronal maturation. The ability of human tau protein to rescue tau‐deficient mouse neurons confirms that tau expression can affect the rate of neurite extension (26). The third tau model is a knock‐in strain generated by targeting a cDNA encoding enhanced green fluorescent protein (EGFP) to exon 1 of Tau (147). These mice did not show any abnormalities in either central or peripheral nervous system.

In separate work, the entire wild‐type human tau gene was expressed at levels that were twofold to threefold higher than those observed in mice transgenic for a single isoform of wild‐type human tau (32). Surprisingly, these mice did not develop any obvious histopathological or neurological phenotype. However, when they were crossed with the EGFP knock‐in mice (147), they developed tau filaments and pathological tau accumulated in cell bodies and dendrites of neurons in a spatiotemporally relevant fashion (2). This may suggest that endogenous mouse tau can inhibit filament formation of human tau. In P301L and P301S tau transgenic mice, murine tau has been shown to be excluded from the filaments (1, 88). Similarly, in transgenic mouse models of Parkinson’s disease, insoluble inclusions were reported which were immunoreactive for human, but not mouse, α‐synuclein (84, 96).

Invertebrate models.  Although the mouse has been the major species used for modeling AD and FTD, other organisms, such as the fruit fly Drosophila melanogaster, the nematode Caenorhabditis elegans and the sea lamprey Petromyzon marinus have also been used [reviewed in (53, 87)] (Figure 1).

Drosophila offers the advantage of low cost, small size and short lifespan, which makes it an attractive organism for drug and modifier screens. When wild‐type and FTDP‐17 mutant forms of human tau were expressed in Drosophila, flies exhibited key features of the human disorder, including adult‐onset, progressive neurodegeneration, early death, enhanced toxicity of mutant tau, accumulation of abnormal tau and anatomical selectivity. However, neurodegeneration occurred without NFT formation (152), which is not entirely contradictary to findings in human AD brain, where around 15% of the neuronal loss cannot be explained by NFTs (39, 45). Interestingly, when expression of wild‐type human tau was combined with that of the Drosophila GSK‐3 homolog “Shaggy”, neurofibrillary pathology with tau filaments was observed (73). In a related study, expression of wild‐type human tau caused impaired axonal transport with vesicle aggregation and associated loss of locomotor function, in the absence of neuronal loss (102). Interestingly, co‐expression of a constitutively active form of GSK‐3β enhanced both axonal transport and locomotor phenotype and two GSK‐3 inhibitors, lithium and AR‐A014418, reversed these changes. Both studies have suggested that GSK‐3 may be an attractive drug target.

Directed expression of wild‐type human and Drosophila tau in adult mushroom body neurons, which are centres of olfactory learning and memory, strongly impaired associative olfactory learning and memory, but olfactory conditioning‐relevant osmotactic and mechanosensory responses remained intact (99). For comparison, studies in P301L tau transgenic mice revealed an impairment of conditioned taste aversion memory, a specific type of olfactory learning in mice (14, 112, 128). In another Drosophila model (102), wild‐type human tau was expressed in larval motor neurons. This caused significant morphological and functional disruption of the neuromuscular junction. Mitochondrial numbers were reduced in presynaptic nerve terminals. Collapse of the mitochondrial membrane potential in wild‐type larvae phenocopied the effects of tau overexpression on synaptic transmission. The authors suggested that overexpression of tau caused synaptic dysfunction, which may have resulted from a reduction in the number of functional mitochondria (19). Mitochondrial dysfunction has also been reported in human P301L tau transgenic mice, based on a proteomics screen (24).

Drosophila models have been used to address the mitosis failure hypothesis of AD. It is based on the concept that the mechanisms underlying synaptic plasticity are largely based on external morphoregulatory cues and internal signaling pathways. Non‐neuronal cells use these mechanisms to sense their relationship to the local neighborhood and to control proliferation and differentiation. This puts nerve cells at risk of erroneously converting signals derived from plastic synaptic changes into positional cues that will activate the cell cycle. Cell cycle activation potentially links synaptic plasticity to cell death, as shown in mice expressing human tau on a mouse tau null background (4). Both apoptotic and non‐apoptotic modes of neurodegeneration have been described in these mice. Data from wild‐type and mutant tau‐expressing Drosophila support a role for cell cycle activation downstream of tau phosphorylation and upstream of apoptosis. It was shown that ectopic cell‐cycle activation leads to apoptosis of postmitotic neurons in vivo. TOR (target of rapamycin) kinase activity was increased and required for neurodegeneration (77). Apoptosis was confirmed by TUNEL staining, visualization of fragmented nuclei and membrane blebbing, but non‐apoptotic forms of cell death may also have been present. Furthermore, evidence for a role of cell cycle‐regulated genes in neurodegeneration has been obtained in mice transgenic for wild‐type human tau (3) and in tau‐expressing human neuroblastoma cells (Hoerndli et al, unpublished). However, in human P301S tau transgenic mice, none of the tested activators and co‐activators of the cell cycle was overexpressed and levels of two inhibitors of cell cycle progression, p21Cip1 and p27Kip1, were increased in brain and spinal cord (28). As these inhibitors accumulated in the cytoplasm of nerve cells, the majority of which contained tau aggregates, this implies that reactivation of the cell cycle was not involved in tau hyperphosphorylation or filament formation. Taking all experimental models into consideration, the role played by cell cycle re‐entry and apoptosis in tauopathies remains controversial.

In contrast to Drosophila, the sea lamprey Petromyzon marinus is not routinely used as an experimental organism. It is a parasitic, eel‐like fish, with a set of six giant neurons in the hindbrain; these so‐called anterior bulbar cells, or ABCs, are readily accessible for manipulation (56). They resemble most large vertebrate neurons, in that they have extensively branched, tapered dendrites, which receive a large number of synaptic inputs (57). Self‐replicating mRNAs derived from Semliki Forest Virus were microinjected into ABCs to overexpress tau. It was possible to reproduce the type of degeneration observed in AD, with the earliest and most severe changes occurring in distal dendrites. Degeneration correlated with the appearance of AD‐related phosphoepitopes of tau (58). A lipid‐soluble, low‐molecular weight proprietary compound, was shown to retard the progressive degeneration of human tau‐expressing ABCs (59).

The nematode C. elegans has also been used as a model organism in tau research. It can be grown in microtiter dishes, screens to identify gene knock‐outs can be automated and large‐scale set‐ups have been devised that allow inactivation of thousands of genes in parallel through RNA interference (10). Moreover, small single‐stranded regulatory RNAs, called microRNAs, capable of interfering with intracellular mRNAs that contain either complete or partial complementarity, were first identified in C. elegans (92) and have potential for AD research. C. elegans and Drosophila are particularly well suited for non‐biased screens for suppressors and enhancers of neurodegeneration (37). Expression of wild‐type and mutant human tau in C. elegans led to similar behavioral, synaptic and pathological abnormalities, although expression of mutant tau caused an earlier and more severe phenotype. Moreover, an uncoordinated movement (Unc) phenotype was noted prior to the accumulation of insoluble tau, suggesting that, while aggregation may contribute to neuronal loss, it is not required for neuronal dysfunction (81). Expression in mechanosensory neurons revealed differences for wild‐type and mutant forms of tau, in that worms expressing wild‐type tau showed a smaller reduction in response to touch than those expressing mutant tau. As neurons started to degenerate, they accumulated tau in nerve cell bodies and processes, in the absence of apoptosis (100).

POTENTIAL OF ANIMAL MODELS

Dissection of pathophysiology.  Transgenic mice have been used to address the question of cross‐talk between tau inclusions and amyloid plaques. The major plaque component is the 39–43 amino acid long Aβ (mainly Aβ₄₀ and Aβ₄₂), which is derived from APP through proteolytic cleavage (40, 97). In cerebral cortex of aged rhesus monkeys, injection of plaque‐equivalent concentrations of fibrillar, but not soluble, Aβ resulted in neuronal loss, tau phosphorylation and microglial proliferation. As the same preparations were not toxic to younger monkeys, this suggested that Aβ toxicity may be age‐dependent. Similar effects were not observed in rodents following the injection of fibrillar Aβ (38). These results in primates have suggested that Aβ neurotoxicity may be more pronounced in the aging brain. Therefore, longevity could contribute to the susceptibility to AD by rendering the brain more vulnerable to Aβ neurotoxicity.

To determine whether Aβ can affect tau pathology, preparations of synthetic fibrillar Aβ₄₂ were injected into the somatosensory cortex and the hippocampal CA1 region of human wild‐type and P301L tau transgenic mice and non‐transgenic controls. This caused a fivefold increase in the number of NFTs in the amygdala of P301L tau mice, but not in wild‐type tau transgenic mice or control mice (51). In contrast, when the non‐fibrillogenic reverse peptide Aβ_42–1 was injected, the number of NFTs was not altered. A different approach was adopted by Lewis et al, who crossed their P301L tau transgenic line with Aβ plaque‐forming APP mutant mice (89). They found that double transgenic mice had a more than sevenfold increase in NFTs in selected brain areas, as compared with P301L tau transgenic mice. When triple transgenic mice were injected intracerebrally with anti‐Aβ antibodies or a γ‐secretase inhibitor to prevent Aβ formation, this resulted in the disappearance of somatodendritic tau staining in young, but not old, mice, again demonstrating an interaction between Aβ and tau (110).

An interaction between Aβ and tau has also been demonstrated in cell lines (35, 113, 115, 151) and through the functional validation of proteomics findings (24). Transcriptomics and proteomics, collectively known as “Functional Genomics”, are being used increasingly to dissect the pathogenesis of neurodegenerative diseases (23, 68) (Figure 1). Using proteomics, mitochondrial proteins, antioxidant enzymes and synaptic proteins have been found to be differentially expressed in the brain of human P301L tau transgenic mice. Interestingly, the reduction in the levels of mitochondrial complex V was confirmed in human carriers with the P301L mutation. Functional analysis demonstrated mitochondrial dysfunction in P301L tau mice, together with reduced NADH‐ubiquinone oxidoreductase activity and, with age, impaired mitochondrial respiration and ATP synthesis. Mitochondrial dysfunction was associated with higher levels of reactive oxygen species in aged transgenic mice. The increased tau pathology in aged homozygous P301L tau mice revealed modified lipid peroxidation and an up‐regulation of antioxidant enzymes in response to oxidative stress. Therefore, it may be that changes in the respiratory chain linked to the appearance of oxidative damage lead to neuronal degeneration in AD brain.

To investigate whether brain cells from P301L tau mice are more susceptible to Aβ than controls, the mitochondrial membrane potential of isolated cortical brain cells was measured in the presence and absence of Aβ (24). Previous experiments using PC12 cells had shown that addition of Aβ leads to a significant decrease in mitochondrial membrane potential (76). Interestingly, it was found that in nerve cells from P301L tau mice the basal mitochondrial membrane potential was normal. A secondary insult with Aβ₄₂ resulted in a greater reduction in the membrane potential of mitochondria from P301L tau mice than from controls. Importantly, this effect was region‐specific and therefore probably dependent on the presence of P301L mutant tau, as cells from the cerebrum (high P301L tau) were more susceptible than cells from the cerebellum (low P301L tau). These data have suggested a synergistic action of Aβ and tau on mitochondrial function. Moreover, it can be concluded that the tau pathology involves a mitochondrial and oxidative stress disorder distinct from that caused by Aβ (24).

In a subsequent study, both pathologies were combined by injecting Aβ₄₂ into the brain of P301L tau mutant mice or by incubating neuroblastoma cells expressing human P301L tau with Aβ₄₂ (25). By using proteomics, a significant fraction of the proteins altered in each system was found to be involved in responses to stress or to play a role in metabolic control. Model‐specific effects were also identified, such as differences in cell signaling proteins in the cellular model and cytoskeletal and synaptic proteins in the case of the animal model. By western blotting and immunohistochemistry, 72% of tested candidates were also altered in AD brain, with a major emphasis on proteins involved in the stress‐related unfolded protein response. These processes may therefore be important initiators of the Aβ₄₂‐mediated pathogenic cascade in AD (25). In a separate study, proteomics was used in a line of mice transgenic for wild‐type human tau (146).

In addition to proteomics, transcriptomics has been applied to tau transgenic animal models. Using UniGem mouse microarray chips, inflammatory mediators and inhibitors of apoptosis were down‐regulated in human P301L tau transgenic brains (65). A second P301L tau transgenic line was analyzed using Affymetrix microarrays (21). A single up‐regulated gene, glyoxalase I, was identified. This enzyme plays a crucial role in the detoxification of dicarbonyl compounds by reducing the formation of AGEs. Levels of glyoxalase I mRNA and protein were significantly elevated in the brains of human P301L tau mice. Moreover, a glyoxalase I‐specific antiserum revealed many intensely stained, flame‐shaped neurons in AD brain, compared with control brains, suggesting a previously unidentified role for glyoxalase I (21). The finding that anti‐AGE antibodies reacted predominantly with intracellular NFTs, as opposed to extracellular NFTs, has suggested that AGE modifications may be involved in the early stages of disease and may not just be secondary consequences of tau deposition (15, 126).

Chaperones have been proposed to play a role in tau pathology (31, 134, 154), as supported by proteomics findings (25). The E3 ligase CHIP interacts with both Hsp90 and Hsc70 and targets the associated unfolded proteins to the proteasome (104). Hsc70 and CHIP have both been demonstrated to mediate tau degradation by the proteasome and Hsp90 has been found to inhibit tau aggregation (31, 134). To determine a role for CHIP in suppressing NFT formation in vivo, a quantitative analysis of CHIP levels was performed in human and mouse brains (124). Increased levels of CHIP and Hsp70 were found in AD compared with controls. In AD brain, the amount of CHIP was inversely proportional to that of sarkosyl‐insoluble tau. In human P301L tau transgenic mice, CHIP was widely distributed, but it was only weakly expressed in spinal cord, the region with the most tau inclusions and neuronal loss. In cerebellum from transgenic mice, CHIP levels were significantly higher than in controls. Human tau was more highly expressed in this region than in spinal cord, but only moderate levels of sarkosyl‐insoluble tau were detected. To further determine its role in tau aggregation, CHIP was deleted in mice (29). This led to the accumulation of non‐aggregated, ubiquitin‐negative and hyperphosphorylated tau. When P301L mutant tau was expressed on the CHIP null background, neither pre‐NFTs nor NFTs developed, despite an increased accumulation of phosphorylated tau. These findings, in conjunction with RNAi studies in C. elegans, have led to the suggestion that polyubiquitination of tau by CHIP may facilitate the formation of NFTs (29). In line with this, recent mass spectrometry studies of paired helical filament (PHF) tau (22) have suggested that ubiquitination of PHF‐tau may be an earlier pathological event than previously thought (8).

In search of modifiers.  Tau protein is hyperphosphorylated in human neurodegenerative diseases. Consequently, much work has gone into the mapping of phosphorylation sites and the identification of candidate protein kinases and phosphatases. PP2A is the major tau phosphatase in brain (94). Its activity has been reported to be reduced in AD brain and dephosphorylation of tau can be blocked in cells by okadaic acid, an inhibitor of PP1 and PP2A (5, 46, 117). When rat brain slices were incubated with okadaic acid, tau became phosphorylated at several sites (111). Moreover, transgenic mice with neuronal expression of a dominant‐negative mutant form of the catalytic subunit of PP2A developed a pre‐NFT phenotype with hyperphosphorylation of murine tau (78). Subsequent analysis of these mice revealed that chronic inhibition of PP2A activity resulted in the activation of ERK and JNK, in the absence of apoptosis (79).

Tau phosphorylation has also been studied using transgenic expression of GSK‐3β and cyclin‐dependent kinase 5 (cdk5)/p25. Phosphorylation of serine 9 inactivates GSK‐3β and mice expressing the constitutively active S9AGSK‐3β displayed increased kinase activity, in the absence of neurofibrillary pathology. When they were crossed with mice transgenic for wild‐type human tau, axonopathy and motor deficits markedly improved (139). These findings have suggested that GSK‐3β may be protective. However, other studies using different models have reached the opposite conclusion.

One approach has been to use a tetracycline/doxycycline‐driven transactivation system. Transgenic mice overexpressing doxycycline‐regulatable GSK‐3β (Tet/GSK‐3β mice) were crossed with mice expressing tau with three FTDP‐17 mutations (VLW mice), giving rise to prefibrillar tau aggregates. They developed thioflavin S‐positive tau aggregates and filaments. Moreover, the atrophy of the dentate gyrus of the hippocampus, which was present in Tet/GSK‐3β mice, developed much faster in the Tet/GSK‐3β/VLW mice (33). To explore whether the phenotype resulting from increased GSK‐3 activity could revert following restoration of normal GSK‐3 levels, transgene expression was shut down in symptomatic mice. This led to normal phosphotau levels, diminished neuronal cell death and suppression of the cognitive deficits (34).

In another study, P301L tau transgenic mice were given lithium chloride, a reasonably specific GSK‐3 inhibitor. This resulted in significant inhibition of GSK‐3 activity, accompanied by a reduction in tau phosphorylation and a decrease in the levels of aggregated, sarkosyl‐insoluble tau. Administration of a separate GSK‐3 inhibitor also resulted in a reduction in sarkosyl‐insoluble tau, supporting the notion that lithium chloride exerted its effects through the inhibition of GSK‐3. Levels of aggregated tau correlated with the degree of axonal degeneration, which was ameliorated temporarily when lithium chloride was administered during early stages of NFT formation (107). In C. elegans, co‐expression of GSK‐3 and wild‐type tau was reported to bring about a deterioration in the touch response which was greater than that present in worms transgenic for tau alone. Co‐expression of the chaperone Hsp70 led to some improvement in the touch response (100).

A second protein kinase investigated through transgenic expression in mice is cdk5 and its activator p25. When human P301L tau transgenic mice were crossed with p25‐overexpressing mice, a fivefold increase in the number of NFTs was observed, alongside increased phosphorylation of tau at several sites (106). A similar phenotype has also been obtained when P301L tau transgenic mice were crossed with mice expressing a dominant‐negative mutant form of PP2A (Deters et al, unpublished). In a separate study, however, transgenic mouse lines that co‐expressed cdk5 and its activator p35, as well as wild‐type human tau protein, failed to exhibit increased phosphorylation of tau in nerve cells (148).

Cdk5, GSK‐3, and PP2A have been reported to co‐immunoprecipitate, suggesting a functional association (118). Overactivation of cdk5 has been shown to inhibit GSK‐3 activity in young mice. This effect was lost in older mice, resulting in increased GSK‐3 activity and phosphorylation of tau. The authors concluded that GSK‐3 is a key mediator of tau hyperphosphorylation, with cdk5 acting as modulator (118). More recently, a potential role for tyrosine phosphorylation of tau has emerged and a correlation between tyrosine phosphorylation and tau aggregation has been demonstrated (150). Disease‐related modifications of tau, that is, isoforms and FTD mutations, affected the interaction between tau and fyn, a non‐receptor tyrosine kinase of the Src family (12).

Parkin is a ubiquitin ligase, loss‐of‐function mutations of which cause familial forms of Parkinson’s disease (84). Some of these patients exhibit significant tau pathology (47). When parkin null mice were crossed with the VLW human tau transgenic mice, a number of abnormalities were observed that were not present in the single transgenic lines (98). They included progressive gait abnormalities, loss of dopamine and dopamine markers in striatum, nuclear tau‐immunoreactive deposits in motor neurons, abnormal expression of glial markers and enhanced levels of pro‐apoptotic proteins. In addition, the mice had reduced levels of CHIP‐Hsc70 and pSer9GSK‐3β. These findings suggest that interactions between tau and parkin could be relevant for the pathogenesis of at least some tauopathies.

Drosophila has been used in modifier screens (Figure 1). In a model tauopathy, protein kinases and phosphatases comprised the major classes of modifiers and several candidate kinases (GSK‐3 and cdk5 among them) enhanced tau toxicity in vivo (18, 135). Overall, increased tau phosphorylation correlated with an increase in toxicity. One exception was MARK/PAR‐1, which behaved as a genetic suppressor (135). MARK is a kinase that phosphorylates the KXGS motifs in the microtubule‐binding region of tau (7). In contrast to these findings, in a separate study, MARK/PAR‐1 enhanced tau toxicity by triggering an ordered phosphorylation process. Through loss‐of‐function and overexpression studies, it was shown that PAR‐1 plays a central role in regulating tau phosphorylation and toxicity in Drosophila, but without promoting NFT formation (105). In a third modifier screen, puromycin‐sensitive aminopeptidase (PSA) was identified and shown to protect against tau‐induced neurodegeneration in vivo, with loss‐of‐function of PSA exacerbating neurodegeneration (75).

In C. elegans, neuronal expression of human tau resulted in the accumulation of insoluble phosphorylated tau, neurodegeneration and uncoordinated movements. In a genome‐wide RNA interference screen, 16 757 genes were tested and 75 were found to enhance the transgene‐induced behavioral phenotype (82). Forty‐six of these genes had sequence similarities to known human genes and fell into a number of broad classes, including kinases, phosphatases, chaperones and proteases. They comprised some of the proteins discussed above, such as MARK, GSK‐3β, PP2A and CHIP. The other 29 modifiers only had sequence similarities with other nematode genes.

In conclusion, the tau transgenic animal studies carried out to date support a role for serine/threonine kinases (in particular ERK, JNK, p38, GSK3‐β, cdk5/p25 and MARK) and phosphatases (PP2A), as well as the chaperone system in diseases with filamentous tau deposits.

Treatment strategies.  The search for novel therapeutic approaches to AD (Figure 1) has been stimulated by studies in Aβ plaque‐forming transgenic mice, in which active and passive immunization strategies have led to a reduction or the prevention of Aβ deposition in brain (9, 127, 136). Further work revealed that immunization can also reduce cognitive deficits in mouse models with Aβ accumulation (30, 74, 101) and, likewise, in AD patients (66, 67). Oddo et al demonstrated that a single intracerebral injection of anti‐Aβ antibodies into the hippocampus of triple transgenic mice significantly reduced extracellular and intracellular accumulation of Aβ and resulted in the clearance of early tau pathology (110). When the selective M1 muscarinic receptor agonist AF267B was assessed in this model, it was found to rescue the cognitive deficits in a spatial task, but not the deficits in contextual fear conditioning (17). These effects predicted the neuropathological findings, as Aβ and tau pathologies were reduced in cerebral cortex and hippocampus, but not in the amygdala. The reduction in tau pathology may have been mediated through a decrease in GSK‐3β activity. This study has suggested that selective M1 agonists may be efficacious for the treatment of AD.

To determine if the formation of NFTs could be reduced or even prevented using Aβ‐based vaccination, P301L tau transgenic mice were immunized with pre‐aggregated Aβ₄₂, until high serum anti‐Aβ₄₂ titers were reached (83). NFT formation was then induced through bilateral intrahippocampal injection of Aβ₄₂ aggregates (51). To determine whether the previously reported Aβ₄₂‐mediated increase in NFTs could be blocked by active immunization, the animals were analyzed 28 days after the injection. The intracerebral injection of Aβ₄₂ caused a significant increase in NFTs, but this was not affected by immunization. It was concluded that active immunization was not sufficient to prevent the effects of Aβ₄₂ on tau aggregation in this model system. Further studies are needed to determine whether modifications of the protocol used could affect the Aβ₄₂‐mediated induction of NFT formation (83).

Modulation of tau phosphorylation and cleavage is another therapeutic strategy. It could benefit patients with AD and those with diseases characterized only by tau deposits. Several candidate tau kinases have been targeted in vivo. A major problem is that these kinases have many other functions besides phosphorylating tau (61, 90, 116). A recent study has evaluated this concept further by using an orally bioavailable and blood‐brain barrier‐penetrating analog of the relatively non‐specific protein kinase inhibitor K252a (85). This compound prevented motor deficits in the P301L tau transgenic mouse line JNPL3 (88) and reduced the levels of soluble aggregated, hyperphosphorylated tau. However, NFT numbers were not reduced, suggesting that the main cytotoxic effects of tau were not exerted by NFTs, but by as yet ill‐defined aggregates, named tau*, in analogy with Aβ*56 (85). This is in line with studies where a reduction in the expression of transgenic human P301L tau led to a recovery of memory and the stabilization of neuronal numbers, despite continued accumulation of NFTs (125). The protein kinase inhibitor rapamycin has been reported to reduce toxicity in Drosophila expressing wild‐type or mutant forms of tau, probably by reducing the amount of insoluble tau (11). Rapamycin induces autophagy through inhibition of TOR (108). It had a similar effect in a Drosophila model of Huntington’s disease (121, 122). It remains to be seen whether the beneficial effects of rapamycin may have had something to do with the TOR‐dependent cell cycle activation that has been described in Drosophila tauopathy models.

OUTLOOK

The identification of pathogenic mutations in AD and FTD has helped to understand disease pathogenesis and has paved the way for the development of animal models. These have been invaluable for dissecting pathogenic mechanisms, for identifying disease modifiers and for developing novel therapeutic strategies (Figure 1). Although AD is the most common dementia (48, 53, 86), tau‐containing lesions are also present in several other neurodegenerative diseases, where they form in the absence of extracellular plaques (130). It follows that Aβ‐directed therapies are likely to be ineffective in these diseases, unless they target common pathogenic mechanisms or structural characteristics shared between tau and Aβ (55).