Alzheimer's disease and related disorders: neuropathology, genetics and clinical features

In 1907, the two key histopathological hallmarks of Alzheimer's disease (AD), β-amyloid plaques and neurofibrillary tangles (NFT), were for the first time described by Alois Alzheimer when he examined brain sections of his patient Auguste D. 1 Since then, the number of PubMed entries for ‘Alzheimer’ steadily increased and, in 2002 alone, added up to 5213 entries. The last 20 years of AD research, with the help of animal models, assisted in the elucidation of aspects of the pathophysiology of AD and the relationship of the two major lesions. Parallel to this increased insight into disease mechanisms, the number of AD patients is rising as the numbers of old people are increasing in many countries. Moreover, the incidence of AD rises from less than 2% for people under the age of 60 to about 30% in people older than 85.2

β-Amyloid plaques and Aβ processing

β-Amyloid plaques are one of the histopathological hallmarks of AD. The term amyloid has been introduced to describe a heterogeneous class of protein aggregates with a β-pleated sheet secondary structure, which confers affinity to the histochemical dye congo red. In AD, β-amyloid is deposited around meningeal and cerebral vessels, and in the gray matter as β-amyloid plaques. The major proteinaceous component is a 40–42 amino-acid polypeptide termed Aβ (Aβ40 and Aβ42), which is derived by proteolysis from the amyloid precursor protein (APP).3, 4

APP can be proteolytically cleaved by the membrane-associated α-secretase, which cleaves APP within the Aβ domain and secretes the amino-terminal portion of APP (APPs). This pathway is non-amyloidogenic, as APP cleavage precludes the formation of Aβ. Alternatively, cleavage may occur in the endosomal–lysosomal pathway: The β-secretase generates the aminoterminus of Aβ, a fragment called β-stub. This precursor is further processed by γ-secretase to generate the Aβ peptide. The cleavage site of γ-secretase is critical as it dictates the length of the peptide, with Aβ40 being the most common species, and Aβ42 the less common but more fibrillogenic and neurotoxic species.

β-Secretase activity has been attributed to a single protein, BACE.5 Reconstitution of γ-secretase activity was achieved in yeast lacking endogenous γ-secretase and shown to depend on the presence of four components: presenilin, nicastrin, APH-1 and PEN-2.6

Tau phosphorylation and neurofibrillary tangles

The second histopathological hallmark of AD are the neurofibrillary lesions that are found in cell bodies and apical dendrites as NFT, in distal dendrites as neuropil threads and in the abnormal neurites that are associated with some β-amyloid plaques (neuritic plaques). NFT are also abundant, in the absence of plaques, in additional neurodegenerative diseases such as Pick's disease (PiD), progressive supranuclear palsy (PSP), corticobasal degeneration (CBD), argyrophilic grain disease (AgD) and frontotemporal dementia (FTD), including the familial form of frontotemporal dementia with Parkinsonism linked to chromosome 17 (FTDP-17).7, 8 The neurofibrillary lesions contain abnormal filaments as their major proteinaceous component. In AD brain, these are mainly paired helical filaments (PHF).9, 10 The core protein of these filaments is tau, a microtubule-associated protein.11 Besides a major role in stabilizing microtubules, additional functions have been assigned to tau in signal transduction, organization of the actin cytoskeleton, intracellular vesicle transport and anchoring of phosphatases and kinases.12, 13, 14, 15, 16, 17, 18, 19, 20, 21 Tau is mainly a neuronal protein, but has also been found, although at low levels, in astrocytes and oligodendrocytes.22 In PSP and CBD, tau forms aggregates in both glial cell types.7

Tau is a phosphoprotein under physiological conditions.23 Under pathological conditions such as AD, tau is hyperphosphorylated, which means that it is phosphorylated to a higher degree at physiological sites, and at additional ‘pathological’ sites.24 Phosphorylation tends to dissociate tau from microtubules. Since this increases the soluble pool of tau, it might be an important first step in the assembly of tau filaments. In the diseased brain, tau also undergoes a conformational change, whereas soluble tau adopts a natively unfolded conformation.25, 26

Genetics of Alzheimer's disease and Frontotemporal dementia

In the complex etiology of AD, in addition to environmental conditions, genetic factors play a major role. Twin studies support the notion that 70–80% of the risk to develop AD is determined by genetic factors.27 Of the two dozen genes identified until today, only the apolipoprotein E (APOE) gene has been confirmed unanimously as a risk gene and found to be associated with sporadic AD (SAD) (reviewed in Rocchi et al28).

Epidemiological studies indicate that about 30% of AD patients have a family history of disease in which at least one first-degree relative is affected, and only 10% of these have a clear autosomal dominant inheritance.29 The histopathological hallmarks are indistinguishable when familial AD (FAD) is compared with SAD. The existence of early-onset familial forms of AD allowed the identification of causative genes and key proteins, the elucidation of pathogenic mechanisms, and the development of transgenic animal models. In early-onset familial FAD, mutations have been identified in three genes: in the APP gene itself, and in the presenilin 1 (PS1) and presenilin 2 (PS2) genes. These mutations account for less than 1% of the total number of AD cases.30

Mutations in the APP gene are estimated to account for up to 5% of FAD. The first missense mutation in FAD families was found in a British family in 1991.31 In mutation carriers, a valine was replaced by an isoleucine at codon 717 (V717I, the so-called ‘London mutation’). Since then, additional allelic variants have been reported. Immediately upstream of the Aβ domain, mutations at codons 670 and 671 were discovered in two Swedish FAD families, consisting in a double base pair substitution which results in lysine and methionine replaced by aspartic acid and leucine (K670D/M671L, ‘Swedish mutation’, APPsw).32 Both kinds of mutations have been expressed in transgenic mice.

Until today, a total of 20 pathogenic mutations has been identified in the APP gene. Of these, some cause cerebral amyloid angiopathy (CAA), a term used to describe deposition of amyloid in the walls of blood vessels of the central nervous system.33 CAA is an important cause of cerebral hemorrhage and may also result in ischemic lesions and dementia.

A close association between CAA and AD has been shown by several studies demonstrating that CAA is present in over 80% of AD cases. The high incidence of CAA in AD explains the relatively common occurrence of CAA-related hemorrhage in AD brains.34

In 1992, a second FAD locus was detected by linkage analysis on chromosome 14, and a few years later the PS1 gene (PSEN1) was identified by positional cloning.35, 36 PS2 (PSEN2) was identified by its homology to PS1 and mapped to chromosome 1.37 Most of the FAD cases are caused by mutations in PS1 and PS2. To date, more than 120 mutations have been identified in the PS1 gene, and only eight mutations in the PS2 gene.

In FAD, no mutations have been identified in the tau gene. However, in 1998, exonic and intronic mutations were identified in the tau gene that were linked to FTDP-17, a familial dementia related to AD.38, 39, 40 These findings established that dysfunction of tau in itself can cause neurodegeneration and lead to dementia. In addition to frontotemporal dementia, tau mutations can cause diseases as diverse as subcortical gliosis,41 cortico-basal degeneration (CBD)42 and pallido-ponto-nigral degeneration.43

Clinical features of AD, relationship of plaques and NFT

The clinical presentation of AD is dominated by early memory deficits, followed by gradual erosion of other cognitive functions such as judgment, verbal fluency or orientation. Although this sequential order may vary, memory impairment is normally the first and dominating feature. This correlates with the fact that the most severe neuropathological changes occur in the hippocampal formation, followed by the association cortices and subcortical structures, including the amygdala and the nucleus basalis of Meynert.44 NFT develop in specific predilection sites and spread in a predictable, non-random manner across the brain. This sequence of the tau pathology is subject to little inter-individual variation and provides a basis for distinguishing six stages of disease progression:45, 46 the transentorhinal stages I–II representing clinically silent cases; the limbic stages III–IV of incipient AD; and the neocortical stages V–VI of fully developed AD. By using the phospho-tau-specific antibody AT8, neuronal changes can be evaluated well before the actual formation of NFT.

A comparative study of the Aβ-associated pathology defined five phases. These differ markedly from the stages which define the spreading of NFT: The first phase is the neocortical phase 1, followed by the allocortical phase 2. In phase 3, the diencephalic nuclei, the striatum and the cholinergic nuclei of the basal forebrain develop Aβ deposits, and in phase 4 several brainstem nuclei become additionally involved. Finally, phase 5 is characterized by cerebellar Aβ deposition. These findings suggest that Aβ deposition expands anterogradely into regions that receive neuronal projections from regions already exhibiting Aβ.47 Stereotaxic injection of radioactively labelled Aβ into the brains of transgenic mice would be a possibility to experimentally address this hypothesis. Since different brain regions (such as, for example, the amygdala and the pontine nucleus) all receive input from one Aβ-containing region (the neocortex), but become involved in β-amyloidosis at different phases (the amygdala in phase 2 and the pontine nucleus in phase 5), regional susceptibility may play an additional role. This notion is supported by the finding of a region-specific induction of β-amyloid-mediated NFT formation in P301L tau transgenic mice48, 49 (see below).

The relative contribution of plaques and NFT to the clinical features of AD is a matter of debate that has been addressed in animal models. Initially, a correlation has been reported between NFT numbers and the severity of dementia.50, 51 Recently, however, Delacourte et al30, 52 reported a synergistic interaction between the APP-related and the tau-related pathology, despite a different spatiotemporal distribution of plaques and NFT. They also found that whenever Aβ aggregates were detected, a tau pathology was found, at least in the entorhinal cortex. The opposite was not true because cases were found with an advanced tau pathology, with no trace of Aβ aggregates.30 Bigenic animals which express both mutant APP and tau have been generated and should be used to address the interaction of the two lesions and their relative contribution to behavioral impairment.

Clinical features of frontotemporal dementia (FTD) and related tauopathies

In contrast to AD, which is characterized predominantly by memory loss, FTD is mainly initiated with behavioral impairment (e.g. disinhibition, loss of personal and social awareness), followed by affective symptoms, speech disorder and memory problems. Late Parkinsonism is common, and a positive family history is frequently observed.53, 54 Neuropsychological changes include short-term memory, attention and concentration.55, 56, 57, 58, 59 Neuroradiological examination reveals an often symmetrical atrophy of the frontal and temporal lobes.57, 59, 60 In many cases, additional degenerative changes are observed in subcortical brain regions, such as the substantia nigra, leading to Parkinsonian symptoms.61 Linkage analysis in families with an autosomal dominant inheritance of FTD revealed linkage to the same region on chromosome 17.62, 63 Subsequently, mutations were identified in the gene encoding tau (MAPT),38, 39, 40 and the term FTDP-17 is now widely used to describe familial forms of FTD.

Pick's disease (PiD) is an extreme form of frontotemporal dementia, but clinically it is difficult to differentiate. This rare disorder is characterized by progressive dementia and personality deterioration associated with verbal and behavioral stereotypes. High densities of tau-containing degenerative neuronal lesions, referred to as Pick bodies, are found in many brain areas.64, 65

Progressive supranuclear palsy (PSP) is characterized by early signs of vertical gaze paresis and progression to total external ophthalmoplegia. Additional features are dysarthria, dysphagia and Parkinsonian symptoms. Dementia is progressive and a dominant feature of the terminal stages of the disease.66, 67

Corticobasal degeneration (CBD) is a rare, sporadic and slowly progressive late-onset neurodegenerative disease that is clinically characterized by cognitive disturbances, cortical sensory loss, extrapyramidal motor dysfunction and unilateral rigidity.68 Higher mental functions are relatively preserved in most patients.69 In contrast to AD, where hyperphosphorylated tau forms filaments only in neurons, in PSP and CBD they also form abundantly in glial cells.70, 71

Animal models for AD and related disorders

To better understand the role of β-amyloid plaques and NFT in AD and related disorders, experimental animal models have been developed, which reproduce aspects of the neuropathological characteristics of these diseases. Their suitability largely depends on the purpose a model has to suit. If one wants to model aspects of the histopathology, one has to discriminate between the precise anatomical ‘reproduction’ of the pathology, and modeling at the cellular level. This is important when the animals (in particular transgenic mice) are employed in behavioral studies intended to correlate the histopathology with dementia. These animal models may either offer a general proof of principle or reproduce more specific aspects of the human disease. Animal models may be used to identify disease modifiers, components of pathocascades and susceptibility genes. They may be employed in drug screenings. Finally, insight gained from these models can be translated to human disease and assist in the development of treatment therapies.

APP transgenic mice: histopathology, behavior, combinatorial transgenics and therapy

Histopathology and behavior

After the very first APP transgenic animals had failed to show an extensive AD-like neuropathology, in 1995, Games and co-workers successfully expressed high levels of the disease-linked V717F mutant form of APP, under control of the platelet-derived growth factor (PDGF) mini-promoter. These PDAPP mice showed many of the pathological features of AD, including extensive deposition of extracellular amyloid plaques, astrocytosis and neuritic dystrophy.72 A histological analysis of the PDAPP mice revealed that plaques first formed in the hippocampus, followed by a spreading into cortical areas. Moreover, the β-amyloid load in the hippocampus increased as a function of age. The early Morris water maze testing showed memory impairment, but there was an absence of correlation between memory deficits and Aβ aggregation, as similar levels of impairment were found in both young plaque-free mice and older mice with abundant plaques. A new water-maze protocol was developed for mice in which the platform was moved to a new location, after the mice had learned to escape quickly and reliably onto the hidden platform at one location. Reassessment of the PDAPP mice using this modified version of the Morris water maze showed a memory impairment which was age-dependent and correlated with Aβ plaque load.73 The spatial learning impairment displayed by the PDAPP mice could be dissociated into age-related and age-independent components as the new task was sensitive to age in non-transgenic controls but, in PDAPP mice, the age-related decline was substantially greater. As AD is also a synaptic disorder, the authors speculated that due to an impaired synaptic transmission in the hippocampus and other brain areas, the PDAPP mice would become gradually less able to rapidly encode changes in their memory representation of an environment and/or to retrieve information selectively — a deficit in episodic-like memory.73, 74, 75

A second transgenic model by Hsiao et al76 expressed the APPSW mutation inserted into a hamster prion protein (PrP) cosmid vector. The resulting Tg2576 line developed numerous β-amyloid plaques that could be stained with the dye Congo red. Learning was assessed in the Y-maze, a spatial alternation task, and in the Morris water maze. At 9 months of age and older, impairment in learning and memory became apparent in transgenic mice. However, studies correlating individual performance in learning and memory tests with the concentration of Aβ had not been performed.

By expressing the Swedish double APP mutation under control of the mThy1.2 promoter, a research group at Novartis established the APP23 mouse model with a seven-fold overexpression of APP. Typical congophilic plaques appeared already at 6 months of age. In addition to inflammatory processes, the plaques were immunoreactive for hyperphosphorylated tau, reminiscent of an early tau pathology.77, 78 Stereologic analysis revealed a 14% decrease of the number of CA1 pyramidal neurons in the transgenic mice compared with controls, whereas the number of neocortical neurons was unaffected.79 A thinning of the pyramidal layer was evident, and the neurons appeared to be interrupted in the vicinity of plaques. In contrast, no neuronal loss has been reported in the PDAPP, the Tg2576 or APP/PS1 mice, demonstrating obvious differences between mice and men.80, 81, 82

In aged APP23 mice, a major deposition of Aβ was found in the cerebral vasculature that had striking similarities to CAA. β-Amyloid deposition occurred preferentially in arterioles and capillaries and within individual vessels. CAA was associated with local neuron loss, synaptic abnormalities, microglial activation, microhemorrhage and loss of vascular smooth muscle cells. Although several factors may contribute to CAA in humans, the neuronal origin of transgenic APP, high levels of Aβ in cerebrospinal fluid and regional localization of CAA in APP23 mice suggested transport and drainage pathways rather than local production or blood uptake of Aβ as a primary mechanism underlying cerebrovascular amyloid formation. APP23 mice on an APP knockout background developed a similar degree of both plaques and CAA, providing further evidence that a neuronal source of APP/Aβ is sufficient to induce cerebrovascular amyloid and associated neurodegeneration.83, 84 As AD is also characterized by a severe depletion of the cholinergic system, cholinergic alterations were investigated in the APP23 mice. In aged transgenic mice, modest decreases in cortical cholinergic enzyme activity were found, compared with wild-type (WT) mice. Total cholinergic fiber length was more severely affected, with up to 35% decreases in the neocortex of APP23 mice. However, as there was no loss of cholinergic basal forebrain neurons in APP23 mice, this suggested that the cholinergic deficit was locally induced by the deposition of Aβ and not caused by a loss of cholinergic basal forebrain neurons. Lesion studies in the APP23 mice suggested that the severe cholinergic deficit in AD is likely caused both by the loss of cholinergic basal forebrain neurons and locally by cerebral amyloidosis in the neocortex.85

In addition to the APP transgenic models described here, many more models (such as the TgCRND886 or J20 mice87) have been developed by both academic and industrial research groups. Aspects of Aβ toxicity have been addressed and therapies have been tested.

Combinatorial transgenics: role of presenilins, BACE, ApoE and TGF-β1

Following the identification of pathogenic mutations in the presenilin genes in AD, transgenic animal models have been generated to address the role of PS1 and PS2 in APP processing in vivo. Two approaches can be envisaged: a knockout of the endogenous presenilin genes and the expression of FAD mutant forms, either alone or in combination with mutant APP. PS1 knockout mice were not viable and developed a skeletal and CNS phenotype.88 This lethal phenotype could be rescued by expression of the FAD mutant PS1 A246E. The finding that Aβ levels were elevated supported a gain of toxic function as a result of the PS1 mutation.89 To study APP processing in the absence of PS1, neuronal cultures were derived from PS1-deficient mouse embryos. Cleavage of the extracellular domain of APP by α- and β-secretase was not affected, whereas cleavage of the transmembrane domain of APP by γ-secretase was prevented, causing a five-fold drop in the production of Aβ. These results pointed towards PS1 as a potential AD drug target.90 In contrast to PS1 knockout mice, PS2 knockouts were viable and developed only mild pulmonary fibrosis and hemorrhage with age.91

Although pathogenic mutations in APP and presenilins do not co-exist in human AD, it was tempting to cross APP and PS1 mutant mice and to assess whether mutant PS1 would cause elevated Aβ levels. Mice which co-expressed the PS1 mutant A264E together with APPSW contained higher levels of Aβ42 in brain than either APPSW single transgenic mice or mice which co-expressed APPSW and WT PS1.92 Not only were Aβ42 concentrations increased, but co-expression of APPSW and PS1 A264E accelerated the rate of Aβ deposition.93 In related in vivo studies, it was shown that mutant PS1 selectively influenced the processing of both endogenous mouse APP and WT human APP, and increased Aβ42 production.94, 95

When PS1 M146L mutant mice were crossed with the APP transgenic line Tg2576, this again caused an increase in Aβ levels and plaque counts. Furthermore, both single and double transgenic mice showed a reduced performance in the Y maze as early as 3 months of age, before substantial Aβ deposition was apparent.96 Aβ plaques developed by 6 months of age. Although the plaques increased in size and number to 9 months of age, no further changes in Y-maze performance were seen.97 These data suggest that it may be necessary to perform additional behavioral tests to identify subtle changes missed in the Y-maze.

One of the hallmark features of AD is the loss of synapses. Gene profiling of APP transgenic mice by microarray and quantitative RT-PCR revealed reduced expression of synaptic plasticity-related genes in areas with β-amyloid plaques.98 Interestingly, these genes are also down-regulated in AD cortical tissue.

When the β-secretase BACE was overexpressed in transgenic mice, this augmented the amyloidogenic processing of APP, as demonstrated by decreased levels of full-length APP and increased levels of the C-terminal fragments C99 and C89. In mice expressing BACE in addition to human WT APP or APPSW, the induction of APP processing, characterized by elevated C99, C89 and sAPPβ, resulted in increased brain steady-state levels of Aβ40 and Aβ42.99

To address the role of genetic risk factors in APP processing and memory, additional bigenic mice have been produced: Several risk factors have been proposed for SAD and CAA, but only the ɛ4 allele of apolipoprotein E (apoE4) has been confirmed in multiple independent studies (reviewed in Rocchi et al28). ApoE is a lipid transport protein which is produced and secreted predominantly by glial cells. Three common alleles exist in humans: apoE2, E3 and E4. The gene dose of apoE4 correlates with the age of onset in AD, although the underlying mechanisms are not known.100 Expression of mutant APP on an apoE knockout background did not affect the age of onset of β-amyloid plaque formation but reduced their numbers.101 Moreover, instead of fibrillar Aβ-containing mature plaques, only diffuse plaques formed. As neither APP levels nor Aβ processing were affected, this would suggest that apoE affects fibrillogenesis and/or clearance of Aβ.102 To assess the effects of human apoE isoforms on Aβ deposition in vivo, apoE3 and E4 were expressed under control of an astrocyte-specific promoter on a mouse apoE null background, and crossed with APP V717F mutant mice. By 9 months of age, APP V717F mice on an apoE null background developed Aβ deposits, but numbers were significantly lower than in APP V717F mice on a mouse apoE background. In contrast to mouse apoE, similar levels of human apoE3 and E4 markedly suppressed early Aβ deposition at 9 months of age in APP V717F transgenic mice. These findings suggest that human apoE isoforms suppress early deposition of Aβ by inhibiting aggregation and/or enhancing Aβ clearance.103 However, by 15 months of age, expression of apoE3 and apoE4 resulted in fibrillar Aβ deposits and neuritic plaques, and substantially (>10-fold) more fibrillar deposits were observed in apoE4-expressing mice. All fibrillar Aβ deposits in the mice were associated with a neuritic dystrophy. Together, these data demonstrated a critical and isoform-specific role for apoE in neuritic plaque formation.104 ApoE may influence the threshold level of Aβ required for conformational change. Moreover, transgenic mice revealed species-specific differences of apoE isoforms.

When apoE4 was expressed in neurons of transgenic mice, these mice developed motor problems accompanied by muscle loss, loss of body weight and premature death.105 Overexpression of human apoE4 in neurons resulted in tau hyperphosphorylation, and increased phosphorylation was correlated with apoE4 expression levels. This suggests a role for apoE in neuronal cytoskeletal stability and metabolism.105 ApoE undergoes proteolytic cleavage in AD brains and in cultured neuronal cells, resulting in the accumulation of carboxy-terminally truncated fragments of apoE that are neurotoxic.106 This fragmentation is caused by a chymotrypsin-like serine protease that cleaves apoE4 more efficiently than apoE3. Transgenic mice expressing the carboxy-terminally cleaved product apoE4(Δ272–299), at high levels in the brain died at 2–4 months of age.107 The cortex and hippocampus of these mice displayed AD-like neurodegenerative alterations, including roughly 10-fold higher levels of tau phosphorylated at the AT8 epitope S202/T205 and Gallyas-positive neurons that contained cytosolic straight filaments, although an immuno-electron microscopic analysis has not been performed. Transgenic mice with lower levels of apoE4(Δ272–299) survived longer, but showed impaired learning and memory at 6–7 months of age. Thus, truncated fragments of apoE4, which occur in AD brains, are sufficient to elicit AD-like neurodegeneration and behavioral deficits in vivo. Therefore, the apoE-cleaving enzyme represents a potential new AD drug target.107

Inflammatory mechanisms may also play an important role in the pathogenesis of AD. Therefore, the role for cytokines and inflammation has been addressed in transgenic animal models. For example, co-expression of transforming growth factor (TGF-β1) and mutant APP in transgenic mice accelerated the deposition of Aβ. Higher levels of TGF-β1 mRNA were found in post-mortem brain tissue of AD patients compared to controls, and these levels correlated strongly with Aβ deposition in damaged cerebral blood vessels of patients with CAA. These results indicate that overexpression of TGF-β1 may initiate or promote amyloidogenesis in AD and experimental models, and may, thus, be a risk factor for developing AD.108 Modest increases in TGF-β1 lowered the Aβ plaque load, and recombinant TGF-β1 stimulated Aβ clearance in microglial cell cultures.109

Together, the APP single transgenic mouse models demonstrated a role for Aβ in progressive plaque formation, synaptic loss and gliosis. Bigenic models confirmed a disease-enhancing role for the β-secretase BACE and the γ-secretase component PS1, various inflammatory molecules and the ɛ4 allele of apolipoprotein E. Together, these models are suited to test therapeutic strategies aimed at reducing Aβ levels and behavioral impairment associated with Aβ. However, although these animal models contributed enormously to an understanding of the pathogenesis of AD, they do not develop NFT nor do they exhibit massive cell loss. NFT formation as the second histopathological hallmark of AD has been mainly addressed in tau transgenic models.

Tau transgenic mouse models: histopathology and behavior

Histopathology of tau transgenic and Pin1 knockout mice

The first tau transgenic models were established in 1995 in mice by expression of the longest human brain tau isoform under control of the hThy1 promoter. 110 At that time, pathogenic mutations in tau had not yet been identified. Despite the lack of an NFT pathology, these mice modeled aspects of human AD, such as somatodendritic localization and hyperphosphorylation of tau and, therefore, represented an early pre-NFT AD-like phenotype. The subsequent use of stronger promoters to drive transgene expression caused a more pronounced phenotype.111, 112, 113 The tau aggregates in these mice could be stained with many phosphorylation- and conformation-dependent anti-tau antibodies (reviewed in Gotz7). In addition, due to a high expression level of the transgene in motor neurons, large numbers of pathologically enlarged axons with neurofilament- and tau-immunoreactive spheroids were present, especially in the spinal cord. Axonal spheroids are a neuropathological characteristic of most cases of amyotrophic lateral sclerosis (ALS), where they are believed to impair slow axonal transport.114, 115, 116 Spheroids have also been reported in cases of frontotemporal dementia with unknown etiology.117 Signs of Wallerian degeneration and neurogenic muscle atrophy were observed. When motor function was tested, the transgenic mice showed signs of muscle weakness. Tau protein extracted from transgenic brain and spinal cord became increasingly insoluble as the mice became older. Despite the decreased solubility of tau, NFT did not form, unless the mice reached a very old age.118 Taken together, these findings demonstrated that overexpression of human tau can lead to a central and peripheral axonopathy, resulting in nerve cell dysfunction and amyotrophy.

With the identification in 1998 of pathogenic mutations in tau in FTDP-17, several groups achieved NFT formation both in neurons119, 120, 121, 122, 123 and in glial cells of transgenic mice.124, 125 When a human tau isoform lacking the two amino-terminal inserts was expressed together with the P301L mutation by using the murine PrP promoter,119 in a line with high expression levels, 90% of the mice developed motor and behavioral disturbances by 10 months of age. These were more pronounced than reported in the previously published WT tau transgenic mouse models.111, 112, 113 The P301L mice showed a delayed righting response, and eventually became unable to go into the upright position.119 In hang tests, they fell after grasping the rope briefly, whereas WT human tau transgenic mice with similar transgenic tau expression levels held without falling. Within 2 weeks of phenotype onset, the transgenic mice could not ambulate. Weakness spread to all limbs and dystonic posturing developed, and within 3–4 weeks of initial signs, mice became moribund. This motor disturbance precluded the mice from being tested in the Morris water maze. In these mice, NFT were identified in brain and spinal cord, and motor neurons were reduced two-fold in spinal cord.119 The same mutation was expressed in a second model using the longest human tau isoform. In addition, the mThy1.2 promoter was chosen instead of the PrP promoter, which may account for different expression patterns in these mice.120 As above, NFT were identified by Gallyas silver stainings and thioflavin S-fluorescent microscopy. Activated astrocytes were detected in those cortical brain areas that contained numerous tau-positive neurons. To identify cells that underwent apoptosis, TUNEL stainings were used. Many TUNEL-positive neurons were found in the somatosensory cortex that contained many tau-positive neurons, whereas only very few TUNEL-positive neurons were present in WT mice. AD is not a motor neuron disease, and as P301L tau levels were low in motor neurons of the spinal cord, these mThy1.2-driven mice are ideally suited, in contrast to the PrP-driven P301L mice, to be tested in the Morris water maze.

The P301S mutation causes an early onset of clinical signs of FTDP-17.42 P301S mutant tau was expressed under control of the mThy1.2 promoter.123 In brains and spinal cords from transgenic mice, human tau levels were two-fold higher than levels of total mouse tau. Perchloric-acid soluble tau was phosphorylated at many phospho-epitopes of tau, with the exception of the AT100 phospho-epitope S214. This site has been shown, together with S422, to be linked to NFT formation in P301L mice.48 In the P301S mice, sarkosyl-insoluble tau was strongly immunoreactive with all antibodies including AT100. This indicates that immunoreactivity for phospho-S214 closely mirrors the presence of filaments, suggesting that phosphorylation of this site occurs in the course of, or after, filament assembly. Tau expression was found in the neocortex and hippocampus, but levels were strongest in brain stem and spinal cord. A two-fold reduction in the number of motor neurons in the anterior horn of the spinal cord was reported, indicating that motor neurons are either particularly vulnerable to tau aggregates or that levels of tau were particularly high in these cell types. Staining with apoptotic markers was nearly absent, in contrast to P301L transgenic mice.120 Antibodies specific for activated MAP kinase, phospho-JNK and phospho-p38 labeled tau-positive neurons. Neurons of WT tau transgenic mice with a tau pathology without NFT did not show altered activities of members of these kinases.123 For comparison, mice with a tau phenotype due to impaired phosphatase function also showed activation of the JNK and MAP signaling pathways.126 Together, these data suggest a role for the MAP kinase family in the human tau pathology.

To develop treatment strategies for pathological tau phosphorylation, it is important to identify sites which are linked to tau aggregation and filament formation, and to identify specific phospho site-directed kinases and phosphatases.127 Candidate kinases (such as GSK-3β and activated forms of cdk5) and phosphatases (such as PP2A) have been expressed in transgenic mice, either alone or in combination with tau. Their expression was shown to modulate tau phosphorylation.128, 129, 130, 131, 132, 133 However, their specific role in the pathogenesis of AD is still unclear.

Glial tau inclusions are uncommon in AD, but are significant in tauopathies such as PSP, CBD and PiD.134 To address the glial tau pathology, G272 V mutant tau was expressed by combining a PrP-driven expression system with an autoregulatory transactivator loop that resulted in high expression in a subset of neurons and oligodendrocytes. Electron microscopy established filament formation associated with hyperphosphorylation of tau. Thioflavin S-positive fibrillary inclusions were identified in oligodendrocytes and motor neurons in spinal cord.124 The clinical phenotype of these mice was subtle. However, when a total of three isoforms of human WT tau was overexpressed in neurons and glia using the mouse Tα1 α-tubulin promoter, this caused a glial pathology resembling the astrocytic plaques found in CBD and the coiled bodies in CBD and PSP.125 In contrast, no neuronal inclusions were found. An age-related progression of tau insolubility was observed, which was associated with increases of phosphorylated tau. 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 above the age of 24 months, this finding would suggest that the accumulation of tau leads to cell death even before the formation of abnormal filamentous aggregates. Behavioral studies have not been performed. Whether and to which extent the glial pathology contributes to the clinical features of diseases, such as PSP and CBD, could be best addressed by behavioral studies in these transgenic mice with a pronounced tau-related glial pathology.

As tau in AD is subject to conformational changes, the role of the prolyl isomerase Pin1 has been addressed, which specifically acts on the phosphoThr231-Pro motif of tau and induces a conformational change, thereby restoring tau function and promoting its dephosphorylation through the conformational specificity of phosphatases, such as PP2A.135, 136 Interestingly, reducing PP2A activity in transgenic mice also increases tau phosphorylation.126, 130 Pin1 can be copurified with tau filaments and a depletion of soluble Pin1 was observed in AD brains.135, 136 In the only tau transgenic animal models (P301S) where Pin1 expression has been investigated, it was shown to be present in only a minority of tau-positive neurons.123 As depletion of Pin1 induces mitotic arrest and apoptotic cell death, sequestration of Pin1 into tau filaments may contribute to neuronal death.135 Following the observation of an inverse correlation between Pin1 expression and predicted neuronal vulnerability in AD brain, Pin1 knockout mice were generated.137 These developed progressive age-dependent motor and behavioral deficits, as do tau transgenic mice.111, 112, 113, 119 Moreover, Gallyas-positive NFT were observed and tau filaments were isolated from Pin1 knockout brain extracts, demonstrating that murine tau is capable of forming filaments in vivo and not only human tau, as has been previously thought (discussed below).137

Behavioral studies in tau transgenic mice

The PrP promoter-driven P301L tau transgenic mice strongly overexpress mutant tau in several neuronal cell-types, including motor neurons. Therefore, they develop a progressive motor phenotype not commonly observed in AD.119 The V337M tau mutant mice express mutant tau only in the hippocampus. They show an 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.138 R406W tau mutant mice express tau at highest levels in the hippocampus and, to a lesser extent, in other cortical and subcortical brain areas. However, in the amygdala, only a few cells strongly expressed mutant tau, even in 16–23-months-old animals.122 These mice showed a slight decrease in locomotor activity during the first 5 min of the open-field test. Whereas overall evaluation revealed no significant difference, they showed a significant impairment in the contextual and cued fear conditioning test. The mThy1.2 promoter-driven P301L mice accumulate tau in many brain areas, but NFT develop mainly in the amygdala.48 Therefore, these mice were assessed in several amygdala-dependent tasks.139 The amygdala is involved in mediating effects of emotion and stress on learning and memory, as determined in fear conditioning and conditioned taste aversion (CTA) tests.140, 141 The P301L mice showed an increased exploratory behavior but normal anxiety levels and no impairment in fear conditioning.139 Similarly, aged APP mutant mice showed no deficits in fear conditioning when compared to corresponding WT mice although a reduction in function might have been expected as they developed β-amyloid plaques in the hippocampus and the amygdala.142 The general lack of massive neurodegeneration in all animal models of AD may explain the almost normal performance of both P301L and APP mice in fear conditioning, which is different from the impaired fear conditioning in AD.143 In the P301L mice, fear conditioning is probably unaffected due to the absence of tau aggregates in the central and lateral nucleus of the amygdala.

CTA is a well-established learning and memory paradigm in which subjects learn to associate a novel taste with nausea and, as a consequence, avoid consumption of this specific taste at the next presentation. Acquisition and consolidation of CTA memory was not significantly affected by the P301L transgene. However, transgenic mice extinguished the CTA more rapidly than did WT mice.139 For comparison, AD patients exhibit an impairment at all levels of gustatory information processing.144, 145 Rapid extinction of CTA memory in the P301L mice may be due to the presence of tau aggregates in the basolateral nucleus of the amygdala, which has been shown to be essential for the extinction of CTA memory, whereas acquisition is dependent on an intact central nucleus, where no tau aggregates were found. Taken together, the behavioral analysis of tau transgenic mice showed that tau aggregation, in the absence of NFT formation, is sufficient to cause behavioral deficits. Not unexpectedly, the form of mutant tau and the type of promoter used for its expression determine the pattern of tau pathology in transgenic mice, and can result in very different behavioral phenotypes. However, the studies revealed a relationship between the presence of tau aggregates in specific brain areas and the impairment of memory functions which are controlled by these respective brain areas.

Interaction of β-amyloid and tau

To assess the relationship of plaques and NFT in AD, animal models, in which selected aspects of this relationship can be addressed are useful. Studies in monkeys have shown that synthetic Aβ42 fibrils induced tau-containing NFT in aged, but not young adult rhesus brains. Aβ42 toxicity in vivo was also highly species-specific: Toxicity was greater in aged rhesus monkeys than in aged marmoset monkeys, and not significant in aged rats.146 In mice, the expression of P301L mutant tau appeared to accelerate tau filament formation compared to WT tau transgenic mice.119, 120 P301L mutant mice are therefore suitable models to determine whether β-amyloid affects the tau pathology in these mice.

When synthetic Aβ42 fibrils were stereotaxically injected into the somatosensory cortex and the hippocampus of P301L and WT human tau transgenic mice and non-transgenic control littermates, five-fold increases of NFT were found 18 days following the injections in the amygdala of P301L transgenic, but not WT tau transgenic or control mice.48 The phospho-tau-specific antibody AT8 revealed neuronal changes well before the actual formation of NFT. The latter were, in contrast, tightly correlated with the pathological phosphorylation of tau at the phospho-epitopes S422 and AT100 (T212/S214). The finding that β-amyloid was not capable of inducing NFT formation in non-NFT-forming WT tau transgenic mice may reflect species differences between men and mice. It may also imply that β-amyloid cannot induce NFT formation de novo. However, in a human tissue culture system, Aβ was capable of inducing tau filament formation, in the absence of pathogenic tau mutations.127

An alternative experimental in vivo approach was chosen by Lewis et al49 who crossed Aβ-producing Tg2576 mice with P301L tau mutant mice. Compared to P301L single transgenic mice, double transgenic mice at 9–11 months of age showed a more than seven-fold increase in NFT numbers in the olfactory bulb, the entorhinal cortex and amygdala, whereas plaque formation was unaffected by the presence of the tau lesions. Levels of total soluble endogenous and transgenic tau protein and tau mRNA were similar in single and double transgenic brains. In situ hybridization analysis showed no differences in tau transgene expression, indicating that there was no global or region-specific increase in tau expression that could explain the enhanced NFT pathology in the double mutant mice. The region-specific induction of β-amyloid-mediated NFT formation in P301L tau transgenic mice mirrors the regional vulnerability observed in AD brains.48, 49

To address the relationship of plaques and NFT yet another approach was chosen: PS1 M146I knockin mice were co-injected with constructs harboring mutant APPSW and P301L tau transgenes under control of the mThy1.2 promoter.147 Compared to crossbreeding, this approach had major advantages such as speed, a reduced requirement for genotyping, and the possibility of introducing multiple transgenes without altering or mixing the genetic background. In the triple transgenic mice, transgenic APP and tau were both expressed at comparable levels. Levels of tau were much higher than in the previously published P301L transgenic models,119, 120 but, in contrast to these, by immunohistochemistry no tau alterations were apparent by 6 months of age, whereas extracellular Aβ deposits were visible. This delayed tau pathology is difficult to explain. However, the temporal and regional pattern of plaque and NFT formation closely mimicked that observed in AD. Aβ deposition began in cortical regions and occurred only later in the hippocampus, whereas NFT formation initiated in limbic brain structures and then progressed to cortical regions. A third hallmark of AD, synaptic dysfunction, was modeled in these mice. Together, these experiments demonstrate pathological interactions between Aβ and tau that lead to increased NFT formation. Besides their major advantages for understanding the pathophysiology of NFT formation, these models are useful to develop therapies designed to reduce Aβ-mediated NFT formation.

The stereotaxic injection experiments had identified distinct phospho-epitopes of tau that are linked with NFT formation. In taking this finding a step further, in vitro studies were performed which used tau-transfected human SH-SY5Y neuroblastoma cells to demonstrate that β-amyloid can induce tau filament formation, and that both tau filament formation and the β-amyloid-mediated reduction in tau solubility can be abrogated by the mutagenesis of the tau phospho-epitope S422.127

Active and passive immunization paradigms are currently being applied by several laboratories to determine whether these would reverse the β-amyloid-mediated induction in NFT formation in APP/tau double transgenic mice. Obviously, aspects of selective vulnerability are more difficult to address, as the distribution of the lesions varies from AD, due to species, promoter and transgene differences.

Transgenic mouse models: suitability, limitations and discrepancies

The two histopathological hallmarks of AD, amyloid plaques and NFT, have been reproduced in several animal models. Inflammatory aspects have been addressed, and the role of presenilins in APP processing and the importance of phosphorylation in tau aggregation has been shown. However, selective and massive neuronal loss, a hallmark of AD and related tauopathies, has not been recapitulated in any of the tau transgenic models published so far, with the exception of motor neuron loss in spinal cord.119, 123 In the APP transgenic models, a 14% nerve cell loss has been reported in one single model, where it was confined to the CA1 region of the hippocampus.79 Somehow, it seems to be difficult to achieve massive cell loss in the mouse, either due to the short lifespan of mice, or because mouse neurons may be less vulnerable. Implanting human neurons into the mouse brain followed by a viral delivery of AD-related transgenes may help to solve this enigma.

The exact spatiotemporal distribution of plaques and NFT seen in AD patients has not been reproduced in any of the currently available mouse models, as the expression pattern of any transgene is dependent on promoter choice, copy number and integration site of the transgene. In addition, it is virtually impossible to compare different transgenic lines (even when the same transgene construct is used by two laboratories), as the genetic background and the integration site significantly influence the histopathology, mortality, learning and memory performance, with heterogeneous backgrounds exhibiting superior learning and memory performance. With the exception of one model (see below), the APP and tau models all express the transgene under control of an exogenous promoter. This raises some concerns as, for example, the endogenous tau gene carries axonal targeting signals and sequences governing mRNA stability in the 3′ untranslated region (UTR) that are not present on the transgenic constructs.148 Also, as in classical transgenic approaches, only one of the six known human brain tau isoforms has been expressed, it is not possible, for example, to model the preferential accumulation of three-repeat tau in PiD vs that of four-repeat tau in PSP and CBD. Transgenic technology suggests two basic approaches to address this question, either a knock-in approach by homologous recombination of an FTDP-17-linked tau mutation into the mouse genomic tau locus, or a transgenic model in which the entire human WT or FTDP-17 mutant tau gene is expressed. The latter approach was pursued and the entire WT human tau gene was overexpressed at levels that were two- to three-fold higher than those in the classical human tau transgenic mice113 but, in the absence of a tau mutation, these mice did not develop any obvious histopathological or neurological phenotype.149 When these mice were crossed with knock-in mice in which a cDNA for the enhanced green fluorescent protein EGFP had been inserted into exon 1 of tau, this resulted in mice which expressed all the six human tau isoforms but not endogenous mouse tau.150 These mice developed tau filaments, and pathological tau accumulated in cell bodies and dendrites of neurons in a spatiotemporally relevant distribution. As in AD, the pathology was predominantly found in the neocortex and hippocampus, it was minimal in the brain stem and spinal cord and there was no evidence of gross motor or behavioral disturbances. When the two strains are compared, these data would suggest that endogenous mouse tau is inhibitory to tau filament formation. In PrP-driven P301L tau transgenic mice, murine endogenous tau has, if not inhibitory, at least been shown to be excluded from the tau filaments.119 Similarly, in transgenic mouse models for Parkinson's disease, insoluble inclusions were reported, which were strongly immunoreactive for human but not mouse α-synuclein.151, 152 These findings are difficult to reconcile with the identification in Pin1 knockout mice of filaments composed of endogenous mouse tau.137 In APPSW/PS1M146L mutant mice, rare PHF-like structures clearly distinguishable from neurofilaments and microtubules have been reported; whether these were composed of tau has not been demonstrated.153 Obviously, a detailed analysis at the molecular level of the differences between murine and human tau and the different tau isoforms is required to clarify these discrepancies.

A correlation of cognitive performance with the tau and β-amyloid pathology has been demonstrated in behavioral studies. In addition, the immunization studies convincingly demonstrated that β-amyloid plaques can be removed and behavioral deficits reverted; however, the precise mechanism and relationship between β-amyloid removal and memory function is not fully understood, partly because the toxic species of β-amyloid (protofibril vs fibril vs soluble monomer) has not been entirely defined.154

In addition to a pharmacological or genetic interference in animal models, it is likely that the techniques of functional genomics will contribute to a better understanding of the pathogenesis of AD. Sophisticated proteomic and transcriptomic technologies in a microarray format will identify components of pathocascades, but also lead to the isolation of genes whose transcription is turned on or off in response to the pathology. For example, in the APP/PS1 transgenic mouse model, gene expression profiling by microarrays and quantitative RT-PCR revealed reduced expression of several genes essential for long-term potentiation (LTP) and memory formation. In cortical AD tissue, the same memory-associated genes were downregulated, suggesting that reduced expression of genes associated with memory consolidation is linked to memory loss in both the transgenic animal model and the human disease.98 The same techniques may be exploited by dissecting brain areas, with and without a pathology, to identify genes which confer selective vulnerability to AD.155 Despite their limitations in modeling all aspects of the human pathology, transgenic mice have been (and continue to be) extremely valuable tools in AD research. However, the production of transgenic mice is a time-consuming, laborious and inefficient task (Figure 1). This is true already for single transgenic mice, but even more pronounced when one wants to crossbreed two or three transgenic lines. The lower species discussed in the next paragraph are more accessible to genetic manipulations. They are easy and fast to breed and their housing is considerably cheaper. Whereas ethical concerns limit the numbers of mice employed in experiments, experimentation with lower species is either less restricted or not restricted at all.

Figure 1
figure 1

Experimental species: Suitability and limitations. Mice, fish, worms and flies offer distinct species-related advantages to model the pathophysiology of AD and related disorders. The inherent experimental limitations and aspects of the human pathology which have been successfully modeled in the four species are listed.

Full size image

Lessons learned from models other than transgenic mice

Species as diverse as the fruit fly Drosophila, the nematode Caenorhabditis elegans, and the sea lamprey have been employed to provide additional insight into the pathogenesis of AD, as each system offers its unique advantages (Figure 1). However, these lower species have a brain anatomy much different from humans, making a direct comparison difficult.

Studies in the fly

Studies in Drosophila addressed both the physiological and pathological role of APP. They showed a physiological role for APP as a receptor for vesicular kinesin 1, a motor protein involved in anterograde vesicle trafficking.156 Ectopic overexpression of WT and mutant APP caused increased cell death in the larval brain. This toxicity seems to be dependent both on the Aβ sequence and the carboxy-terminal tail of APP that putatively binds kinesin 1. Although the presence of the β-amyloid peptide was not reported in this study, a previous study showed that the Aβ peptide could be generated from a modified fragment of human APP in Drosophila both in vitro and in vivo.157

When WT and FTDP-17 mutant forms of human tau were expressed in Drosophila, these showed 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.158 It is worthwhile mentioning that, in AD, around 15% of the neuronal loss cannot be explained by NFT.159, 160 This implies that non-NFT-related mechanisms of neurodegeneration may also compromise vulnerable subsets of neurons. A quantitative analysis of NFT in human brain revealed that a substantial number of pyramidal cells may persist either unaffected or in a transitional stage of NFT formation in both neocortical layers. This suggests that a considerable number of neurons with intracellular NFT exists in the neocortex until late in the course of AD. Whereas it is not possible to assess whether such transitional neurons are fully functional, these affected neurons might respond positively to therapeutic strategies aimed at protecting the cells that are prone to neurofibrillary degeneration in AD.161

Overexpression of WT tau in combination with phosphorylation by the Drosophila GSK-3 homolog and wingless pathway component Shaggy lead to a neurofibrillary pathology with tau filaments that had periodicities of 45 nm, whereas PHF in human brain have periodicities of 75–80 nm.162 This implies that factors other than tau itself define the ultrastructural characteristics of the tau filaments.

To explore the possibility that the functions of APP and tau may somehow converge on the same cellular process, the Drosophila APP homologue APPL was overexpressed in Drosophila neurons along with tau. Flies with panneural coexpression failed to expand wings and to harden the cuticle upon eclosion, suggestive of neuroendocrine dysfunction. Transport of axonal cargo was disrupted, as shown by the increased retention of synaptic proteins in axons and the scarcity of neuropeptide-containing vesicles in the distal processes of peptidergic neurons.163 Together, these data show that APP may have a function in axonal transport and that axonal transport may be impaired in AD.

Reproducing the cellular biochemistry of neurodegenerative diseases is likely to be the most fruitful feature of Drosophila models. Moreover, their low cost, small size and short lifespan make Drosophila an attractive organism for the use in drug screenings. Modifier screenings can be used to identify new genes that are involved in the pathogenesis of AD. In contrast, Drosophila models might be less useful to address the behavioral abnormalities of AD. Similarly, the extensive anatomical difference between the fly and vertebrates will probably limit the contribution of the Drosophila model to questions about how the progressive cognitive deterioration of AD patients is caused by the loss of specific neuronal populations.

Studies in the fish

The sea lamprey, Petromyzon marinus, is a parasitic fish whose central nervous system is characterized by a set of six giant neurons in the hindbrain, the so-called anterior bulbar cells, or ABCs, which are readily accessible for manipulation. This is a unique advantage of this experimental model, as ABCs have been characterized morphologically and studied on a single-cell level in great detail.164 The ABCs resemble most large vertebrate neurons by having extensively branched, tapered dendrites which receive large numbers of synaptic inputs.165 Self-replicating mRNAs derived from Semliki Forest Virus have been microinjected to chronically overexpress tau. As in AD, a stereotyped sequence of degenerative changes was induced throughout time, with the earliest and most severe changes occurring in the distal-most dendrites. Moreover, this sequence of degenerative changes was spatiotemporally correlated with the appearance of AD-related phospho-epitopes of tau.166 A lipid-soluble, low-molecular-weight proprietary compound was shown to retard the progressive degeneration of human tau expressing ABCs.167

Studies in the worm

The major signaling pathways of the soil nematode C. elegans and vertebrates are remarkably conserved, making it an attractive animal model for research into neurodegeneration. C. elegans was the first multicellular organism of which the genome has been completely sequenced. Somehow surprisingly, 65% of the known human disease genes turned out to have a worm counterpart.168 As C. elegans can be easily grown in microtiter plates, screens to identify gene knockouts can be automated and large-scale set-ups have been devised which allow the temporal gene inactivation of hundreds to thousands of genes in parallel by RNA interference.169 For non-biased screens for suppressors and enhancers of neurodegeneration, C. elegans is, together with Drosophila, particularly well suited.170 For example, two genes have been isolated in C. elegans, aph-1 and pen-2, that interact genetically with presenilins and play a critical role in γ-secretase activity.170 Furthermore, genes have been analyzed whose human counterparts carry mutations leading to AD and other neurodegenerative diseases. Whereas mutations in the human PS1 gene are associated with early-onset AD, the mutants of the respective C. elegans gene sel-12 show neurological defects only in a few neurons, but a highly penetrant egg-laying defect.171, 172 Despite this obvious phenotypic discrepancy between invertebrate and mammalian dysfunction, a remarkable degree of functional conservation was demonstrated in several studies by rescuing the worm defects through transgenic expression of the respective human WT genes.171, 172 Expression of both WT and mutant tau in C. elegans lead to behavioral, synaptic and pathological abnormalities, although mutant tau caused an earlier and more severe phenotype.173 An important difference between C. elegans and the mouse models described above is that behavioral impairment, such as uncoordinated movement, was noted before insoluble tau accumulated, suggesting that while aggregation may contribute to eventual neuronal loss it is not required to cause neuronal dysfunction. It may well be that the assessment of hind limb function as done for some mouse strains as a read-out of neuronal dysfunction is not sensitive enough to detect early phenotypic changes due to non-aggregated tau.111, 112, 113, 119

Together, these studies reveal that the lower species offer distinct advantages compared with mouse models. To develop treatment therapies for AD, their distinct potential needs to be exploited. In a translational approach, once modifier genes have been identified in C. elegans and Drosophila they can be subsequently co-expressed in mice together with APP and tau.

Therapeutic strategies

The main efforts in the development of treatment strategies for AD focus on the prevention of Aβ production, Aβ aggregation or downstream neurotoxic events (Figure 2). The observation that Aβ may not only affect the tau pathology, but also the Lewy body pathology in Parkinson's disease and related disorders, suggests that potential AD drugs might benefit a broader spectrum of neurodegenerative disorders than previously anticipated.152, 174

Figure 2
figure 2

Experimental therapies. The amyloid cascade hypothesis of AD suggests several possibilities of therapeutic intervention.

Full size image

The two proteolytic activities involved in the production of Aβ, β- and γ-secretase, are attractive drug targets of the pharmaceutical industry.175 A major problem in drug development arises from the finding that APP is not the only substrate of these secretases. However, oral administration of a γ-secretase inhibitor to APP V717F transgenic mice reduced brain levels of Aβ in a dose-dependent manner, without apparent toxicity due to impaired Notch signaling.176

Other approaches aim to block the assembly of Aβ into oligomers, protofibrils or fibrils. Following the demonstration that a five-residue peptide, designed as β-sheet breaker, inhibited Aβ fibrillogenesis, disassembled preformed fibrils in vitro and prevented neuronal death induced by fibrils in cell culture,177 a modified peptide with improved pharmacological properties was developed which showed a high rate of penetration across the blood–brain barrier, and the ability to induce a reduction in β-amyloid deposition in APP V717I singly and APP V717I/PS1 A246E doubly transgenic mice.178 Multiple weekly injections of the breaker seemed to be required, but, in principle this approach should be applicable to any disease with abnormal protein aggregates.

As zinc and copper can form complexes with Aβ, another strategy involves lowering of zinc levels. Oral administration of the antibiotic clioquinol, which chelates zinc and copper, led to a 50% reduction in brain Aβ deposition in Tg2576 mice.179

Interestingly, β-amyloid deposition was markedly reduced when the Tg2576 mice were crossed with synaptic ZnT3 zinc transporter knockout mice.180 Instead of preventing Aβ aggregation, another approach involves clearing enzymes of Aβ, such as insulin-degrading enzyme (IDE) and neprilysin. A lentiviral approach demonstrated a role for neprilysin in reducing Aβ deposits in transgenic mice,181 and in IDE knockout mice, a 50% decrease in Aβ degradation was found in both brain membrane fractions and primary neuronal cultures.182

A further approach aims to affect cholesterol levels and the lipid metabolism. Since apoE, a risk gene for SAD, is a regulator of lipid metabolism, it is reasonable to assume that lipids such as cholesterol are involved in the pathogenesis of AD. Recent epidemiological and biochemical studies have strengthened this assumption by demonstrating an association between cholesterol and AD. Elevated cholesterol levels increased Aβ in cellular and most animals models of AD, and drugs that inhibit cholesterol synthesis lowered Aβ in these models.183

Furthermore, non-steroidal anti-inflammatory drugs (NSAIDs), such as ibuprofen, are currently being tested.184 Also, antioxidant treatment and the use of inhibitors of proteases (such as calpain) have been envisaged as both APP and tau are calpain substrates.185

Finally, modulating tau phosphorylation and truncation is a reasonable therapeutic strategy which may not only benefit AD, but also the many tauopathies with little or no β-amyloid deposition (Figure 2). Several candidate kinases (such as cdk5 or GSK3) have been targeted in vivo, a major problem, however, is that these kinases have many different functions in both neuronal and non-neuronal cells, making drug development not an easy task.186, 187, 188

Immunization trials in APP transgenic mice

As the PDAPP mice were generally available, and despite the fact that immunizations had so far been mainly applied to viral and bacterial diseases, Schenk and co-workers, rapidly followed by other research groups, applied both passive and active immunization strategies to Aβ-plaque-forming mice (Figure 2).

In the first study by Schenk et al,189 the PDAPP mice were immunized with Aβ42, either before the onset of an AD-like neuropathology or at an older age, when Aβ deposition was well established. Immunization of young animals essentially prevented the development of β-amyloid plaques, neuritic dystrophy and astrogliosis, while treatment of older animals reduced the extent and progression of the AD-like neuropathology, suggesting that immunization with Aβ may be effective in both preventing and treating AD.189

Similarly, peripheral administration of antibodies directed against Aβ was sufficient to reduce β-amyloid plaques. The passively administered antibodies, despite their modest serum levels, were able to enter the central nervous system, decorate plaques and induce clearance of pre-existing β-amyloid. When examined in an ex vivo assay with sections of PDAPP or AD brain tissue, β-amyloid-specific antibodies were shown to trigger microglial cells to clear plaques through Fc receptor-mediated phagocytosis and subsequent peptide degradation.190 By immunizing with F(ab′)2 fragments, which lack the Fc region of the antibody, non-Fc-mediated mechanisms were also shown to be involved in Aβ clearance.191 Together, these results indicated that antibodies can cross the blood–brain barrier to act directly in the central nervous system, although the amount of antibodies that entered the brain was very low.190

Subsequently, other research groups confirmed the efficacy of the Aβ immunization by various routes, including intranasal administration.192, 193 Further studies revealed that immunization also reduced age-dependent learning deficits. Vaccination with Aβ protected APP/PS1 double transgenic mice from age-related memory deficits that normally occur in these mice: At an age when untreated transgenic mice showed memory deficits on the radial-arm water maze test of working memory, the Aβ-vaccinated transgenic mice showed a cognitive performance that was superior to that of the control transgenic mice and, ultimately, performed as well as non-transgenic mice. The Aβ-vaccinated mice had also a partial reduction in β-amyloid burden at the end of the study.194

In another study, Aβ immunization of the murine APPSW/V717F model TgCRND8 reduced the number of dense-cored plaques and cognitive dysfunction as assessed in the reference memory version of the Morris water maze task, without altering total levels of Aβ in the brain. This may imply that either an approximately 50% reduction in plaques is sufficient to affect cognition, or that vaccination may modulate the activity and abundance of a small subpopulation of especially toxic Aβ species.86 In the mice, complete removal of Aβ deposits does not seem to be necessary for memory improvement, as even a single injection of an anti-Aβ antibody could improve both working and reference spatial memory in PDAPP mice.195 The same holds true for passive immunization, as three anti-Aβ antibody injections over a period of 3 weeks caused significant memory improvement of Tg2576 mice, with no effect on the pools of saline-soluble, SDS-soluble or formic acid-soluble Aβ.196 The initial assumption from the amyloid cascade hypothesis, in which fibrillar amyloid plaques are the main culprits, may not apply to the APP mouse models. In particular, the results from the vaccination trials seem to indicate that memory deficits can be reversed without significant depletion from the fibrillar Aβ pool. In summary, these studies suggest that Aβ formation contributes to memory loss in AD and that immunization with either the Aβ peptide or an Aβ-specific antibody reduces learning and memory impairment in mice.197 This approach has been used as a potential strategy to prevent and/or treat AD in humans. Studies in WT mice of two different genetic backgrounds showed differences in the time course of antibody generation and in antibody titers. This suggests that the high degree of genetic variability in humans needs to be taken into consideration when human vaccination trials are designed.198

Therapeutic vaccination trials in humans

Currently, there is no cure for AD available. To ameliorate the cholinergic deficits in AD patients, the first cholinergic drugs were developed and approved in 1993 for the treatment of AD in the USA, followed by the NMDA antagonist memantine, that was approved for the first time in Europe in 2002. As transgenic animal models provided evidence that both active and passive Aβ immunization can reduce cognitive dysfunction in APP transgenic mouse models without any side effects, an immunization trial was initiated in humans. First, good safety and tolerability data were obtained from different species, including mice, rabbits, guinea-pigs and monkeys. Then, phase I clinical trials were initiated with the AN-1792 vaccine, a pre-aggregated synthetic Aβ42 preparation, and the adjuvant QS-21. First, a single-dose phase I study with 24 patients and, subsequently, a multiple-dose phase I study with 70 patients were reported with good safety and tolerability data.199 Additionally, it could be shown that approximately 25% of the patients produced Aβ-specific antibodies. In October 2001, a phase IIa clinical trial was started in a multicenter placebo-controlled double-blind design and patients were randomized to receive AN1792/QS-21 or placebo (in a ratio of 4:1) by intramuscular injection.200 It was planned to obtain pilot efficacy data together with more extensive safety and tolerability data. However, the dosing was prematurely halted in January 2002 when four patients who had received the AN-1792/QS-21 vaccine developed signs of subacute meningoencephalitis.201, 202 Until then, 372 patients had received injections; four received one injection; 338 received two and 30 patients three injections. Of these, 298 patients received the AN1792/QS-21 vaccine. Over the entire period, 18 patients (6% of the AN1792/QS-21-treated group) developed signs of meningoencephalitis. Upon unblinding, these patients turned out to belong to the verum (AN1792/QS-21-treated) group. The latency from the final AN1792/QS-21 injection to the development of the first symptoms varied between 5 and 71 days; only for two patients, longer intervals have been reported (156 and 168 days). The severity of the clinical picture ranged from mild to severe disabilities, and 12 of 18 patients recovered completely.200 In a cohort of 30 patients from one of the centers, additional data were obtained. First, it could be demonstrated that 24 of the 30 patients developed specific antibodies against β-amyloid which did not cross-react with human full-length APP. These antisera stained β-amyloid plaques on brain slices of APP transgenic mice and post-mortem brain sections from patients with AD. Nine of the 24 patients gave a weak staining with pre-immune serum.203 In the 24 patients, the specific immune reaction against Aβ was still stable after 1 year. Moreover, this immune reaction showed a significant correlation with slowed cognitive decline. These beneficial clinical effects could also be demonstrated in two of three patients who had experienced immunization-related aseptic meningoencephalitis.204 Together, these results indicate that vaccination against Aβ might be an interesting, causal treatment for AD, provided that a safe treatment modality can be introduced.

The first analysis of a brain from a patient who had died after immunization with AN-1792 revealed unusual features, despite diagnostic neuropathological features of AD, including extensive areas of neocortex with very few Aβ plaques, compared with unimmunized cases of AD (n=7). However, those areas of the cortex that were devoid of Aβ plaques contained densities of NFT, neuropil threads and cerebral amyloid angiopathy (CAA) similar to unimmunized AD brains, but lacked plaque-associated dystrophic neurites and astrocyte clusters.205 For comparison, in the APP23 animal model, it was shown that passive immunization resulted in a significant reduction of mainly diffuse amyloid; however, it also induced an increase in cerebral microhemorrhages associated with β-amyloid-laden vessels, suggesting a possible link to the neuroinflammatory complications of the Aβ immunization in the human trial.206 In contrast to the APP transgenic mouse models which had been used to establish the active and passive immunization protocol in mice,86, 189, 190, 194 the APP23 mice had a pronounced vascular amyloid pathology (CAA). Taking into account that CAA is a common feature of AD, this could be a reason why immunization-related complications were encountered in the human trial. The safety of the Aβ immunization may depend on the degree of CAA in human AD patients. Whether immunization would prevent NFT pathology and synaptic dysfunction is still open and requires further studies. Furthermore, it should be kept in mind that although AD is the most prevalent human neurodegenerative disorder, there are many additional tauopathies where NFT formation occurs in the absence of Aβ asking for other types of strategies.

Outlook

Taken together, a wealth of information has been gained with animal models of human neurodegenerative diseases, both with respect to the cell biology of neurodegeneration and also in the light of recent advances of the immunotherapy in human patients. It can be hoped that continued research in this fast-moving field will provide further insight, leading to a better understanding of the pathogenic mechanisms in AD. As pathocascades are likely to be shared between a range of diseases, these findings may also contribute to other fields of research, such as Parkinson's disease.152 Ultimately, these efforts will assist in the development of a safe treatment of AD.