Neurobiology of Aging 24 (2003) 1063–1070
Amyloid deposition precedes tangle formation in a triple
transgenic model of Alzheimer’s disease
Salvatore Oddo, Antonella Caccamo, Masashi Kitazawa, Bertrand P. Tseng, Frank M. LaFerla∗
Department of Neurobiology and Behavior, University of California, 1109 Gillespie Neuroscience Research Facility, Irvine, CA 92697-4545, USA
Received 4 August 2003; accepted 14 August 2003
Abstract
Amyloid-␤ (A␤) containing plaques and tau-laden neurofibrillary tangles are the defining neuropathological features of Alzheimer’s
disease (AD). To better mimic this neuropathology, we generated a novel triple transgenic model of AD (3xTg-AD) harboring three mutant
genes: ␤-amyloid precursor protein (␤APPSwe ), presenilin-1 (PS1M146V ), and tauP301L . The 3xTg-AD mice progressively develop A␤ and
tau pathology, with a temporal- and regional-specific profile that closely mimics their development in the human AD brain. We find that A␤
deposits initiate in the cortex and progress to the hippocampus with aging, whereas tau pathology is first apparent in the hippocampus and
then progresses to the cortex. Despite equivalent overexpression of the human ␤APP and human tau transgenes, A␤ deposition develops
prior to the tangle pathology, consistent with the amyloid cascade hypothesis. As these 3xTg-AD mice phenocopy critical aspects of AD
neuropathology, this model will be useful in pre-clinical intervention trials, particularly because the efficacy of anti-AD compounds in
mitigating the neurodegenerative effects mediated by both signature lesions can be evaluated.
© 2003 Elsevier Inc. All rights reserved.
Keywords: Amyloid; A␤; ␤-Amyloid; Presenilin; Tau; Tangles; Transgenic
1. Introduction
Alzheimer’s disease (AD) is an age-dependent and irreversible neurodegenerative disorder that causes a progressive
deterioration of cognitive functions, including a profound
loss of memory. The hallmark neuropathological lesions of
AD include amyloid deposits, in the form of either diffuse
or neuritic plaques, and neurofibrillary tangles, which consist of hyperphosphorylated tau aggregates [9]. The occurrence of both of these signature lesions is necessary to neuropathologically confirm a diagnosis of AD.
AD can manifest either sporadically or be transmitted
in an autosomal dominant fashion. Three genetic loci have
been found to underlie autosomal dominant, early onset AD:
␤APP on chromosome 21, PS1 on chromosome 14, and PS2
on chromosome 1 [10]. Clinical mutations in each of these
genes alter the metabolism of APP processing, leading to
either increased levels of total A␤ or a selective augmentation of the longer more amyloidogenic A␤42 species [7]. It
is predominantly this genetic evidence that has provided the
strongest support for the amyloid cascade hypothesis, which
predicts that A␤ is the trigger for all cases of AD [3]. In con∗
Corresponding author. Tel.: +1-949-824-1232; fax: +1-949-824-7356.
E-mail address: [email protected] (F.M. LaFerla).
0197-4580/$ – see front matter © 2003 Elsevier Inc. All rights reserved.
doi:10.1016/j.neurobiolaging.2003.08.012
trast, mutations in the tau gene do not lead to AD, but rather
to another form of dementia called frontotemporal dementia with parkinsonism-17, which is marked by neurofibrillary pathology similar to that in AD, although without any
amyloid deposition.
Gene-targeted and transgenic mice have proven to be invaluable for studying the pathogenesis of AD, although no
transgenic model recapitulates its complete neuropathological spectrum [12]. For example, the overexpression of mutant isoforms of human ␤APP in transgenic mice leads to
amyloid deposition in the murine brain, but is insufficient
for triggering the full spectrum of AD neuropathology, including the development of neurofibrillary pathology. This
outcome is surprising because human genetic data indicate
that overproduction of ␤APP, such as in Down syndrome,
is sufficient to trigger the complete spectrum of AD neuropathology. Therefore, the concomitant manifestation of
both plaques and tangles in a mouse requires aggressive
biotechnical strategies, such as introducing multiple transgenes into a mouse or through alternative means such as
the microinjection of pathological proteins into the brains
of genetically modified mice.
We recently reported the derivation of a triple transgenic
model of AD [6]. The triple transgenic mice (3xTg-AD)
harbor three mutant transgenes: PS1M146V , APPSwe , and
1064
S. Oddo et al. / Neurobiology of Aging 24 (2003) 1063–1070
tauP301L . The 3xTg-AD mice develop an age-dependent and
progress neuropathology that includes plaque and tangle
pathology. Here we show that despite equivalent overexpression of human APP and tau, A␤ pathology precedes
typical indications of tau pathology such as conformational
or hyperphosphorylation changes in the tau protein; these
results are consistent with the amyloid cascade hypothesis
which predicts that A␤ deposition is the earliest pathological trigger of AD.
2. Material and methods
antibody. The blots were washed in Tween-TBS for 20 min
and incubated for 5 min with Super Signal (Pierce).
2.3. Immunohistochemistry
Paraformaldehyde-fixed, free-floating brain sections (cut
at 50 ␮m) were immunostained. The following antibodies
were used: anti-A␤42 (Biosource), anti-tau HT7, anti-tau
AT8, (Innogenetics), and anti-GFAP (Dako). Primary antibodies were applied at dilutions of 1:3000 for GFAP, 1:500
for AT8, and 1:200 for HT7 and A␤42. Sections were developed with diaminobenzidine (DAB) substrate using the
avidin-biotin horseradish peroxidase system (Vector Labs).
2.1. Generation of triple transgenic mice
2.4. SELDI
Human ␤APP cDNA harboring the Swedish double
mutation and human tauP301L cDNA were individually subcloned into the Thy1.2 expression cassette. Both fragments
were isolated from the cloning vector by digestion with
EcoRI and PvuI and purified by sucrose gradient fractionation. After overnight dialysis in injection buffer (10 mM
Tris, pH 7.5, 0.25 mM EDTA), the two constructs were microinjected into the pronuclei of single-cell embryos from
PS1M146V knockin mice at the Transgenic Mouse Facility at
the University of California, Irvine. Transgenic mice were
identified by Southern blot analysis of tail DNA isolated
from 10-day-old pups using established procedures [6,11].
Mice were maintained in either a hemizygous or homozygous genotype. Age- and gender-matched nontransgenic
and PS1 knockin mice were used as controls.
2.2. RNA and protein expression analysis
Brain tissue from 10-week-old transgenic mice was mechanically homogenized and total RNA was isolated via
the guanidinium isothiocyanate acid phenol method. RNA
(10 ␮g) was loaded onto a denaturing agarose/formaldehyde
gel and transferred to nitrocellulose. Northern blots were hybridized with a random-primed 32 P-labeled cDNA fragment
corresponding to either human ␤APP or human tau. Blots
were stripped and re-probed for ␤-actin mRNA to control for
RNA loading. For quantification of band intensities, phosphoimaging screens (Amersham Biosciences) were scanned
on the Storm system and densitometric analysis was carried
out with Imagequant software.
For immunoblot, brains from transgenic and control mice
were dounce homogenized in a solution of 2% SDS in H2 O
containing 0.7 mg/ml Pepstatin A supplemented with complete Mini protease inhibitor tablet (Roche 1836153), briefly
sonicated, and centrifuged at 4 ◦ C for 1 h at 100,000 × g.
Proteins were resolved by SDS/PAGE (10% Bis-Tris from
Invitrogen) under reducing conditions and transferred to nitrocellulose. The blot was incubated in a 5% solution of nonfat milk for 1 h at 20 ◦ C. After overnight incubation at 4 ◦ C
with primary antibody, the blots were washed in Tween-TBS
for 20 min and incubated for 1 h at 20 ◦ C with secondary
Eight spot SELDI chips (labeled A–H) with pre-activated
surfaces (PS20 Ciphergen) were incubated with antibody
6E10 (Signet) diluted in PBS (0.5 mg/ml) overnight at 4 ◦ C
in a humidity chamber. The pre-activated surfaces were deactivated by washing with 0.5 M ethanolamine pH 8.5 for
30 min at room temperature. The chip was washed with
10 ml PBS + 0.5% Triton X-100 for 15 min, and transferred
to 8-well bioprocessor, washed again with PBS + 0.5% Triton X-100 for 15 min, followed by a final wash in PBS for
5 min. Two hundred microliters of sample was placed in
each well and incubated overnight at 4 ◦ C with gentle vortexing. The wells were washed twice with PBS + 0.5% Triton X-100 for 5 min and three times with PBS for 5 min. The
spots were dried at room temperature. The energy absorbing
molecule CHCA was prepared fresh by dissolving it in 50%
acetonitrile + 0.5% TFA with vortexing for at least 5 min.
The CHCA was then centrifuged at 13,200 rpm for 5 min.
The supernatant was diluted 1:5 in 50% acetonitrile + 0.5%
TFA and stored in the dark until use. After the spots dried,
1 ␮l of diluted CHCA was placed on each spot and allowed
to dry at room temperature. The chip was read and analyzed using Ciphergen’s protein chip software version 3.1
and protein chip reader (PBS II).
3. Results
3.1. Triple transgenic model
In an attempt to better phenocopy the major neuropathological features of AD, we generated a novel triple transgenic model, harboring three mutant human transgenes:
PS1M146V , ␤APPSwe , and tauP301L . Details about the generation of this model have been described elsewhere [6]. Briefly,
we co-microinjected two independent transgenes encoding
human ␤APPSwe and human tauP301L , both under control
of the mouse Thy1.2 regulatory element, into single-cell
embryos harvested from homozygous mutant PS1M146V
knockin mice (Fig. 1A). Both transgenes co-integrated at
the same locus as shown by Southern blotting, and analysis
S. Oddo et al. / Neurobiology of Aging 24 (2003) 1063–1070
1065
Fig. 1. Characterization of 3xTg-AD mouse model. (A) Schematic diagram of the ␤APPSwe and TauP301L transgenes. Both the human ␤APPSwe (695
isoform) and human tauP301L (4R/0N) were cloned into a modified Thy1.2 expression cassette. The entire mouse Thy1.2 genomic sequence is shown with
exons depicted as boxes and noncoding sequences as thin lines. Using the pronuclear microinjection technique, equimolar amounts of both transgenes
were co-injected into single-cell embryos harvested from PS1 knockin mice. Southern blot and analysis of transgene transmission frequency indicated that
both the human APP and tau transgenes integrated at the same locus, allowing for the easy production of triple transgenic offspring. (B) Representative
Southern blot of tail DNA obtained from F1 offspring from the B1 3xTg-AD line showing integration positive and negative mice. (C) Representative
Northern blot showing APP expression. Northern expression analysis shows that the steady-state levels of transgene-specific RNA products are double
in homozygous mice. The blot for tau is not shown. (D) Immunoblot comparing steady-state levels of human APP and tau proteins. Tau and APP
expression are doubled in the homozygous mice. For both APP (detected with antibody 22C11) and tau (detected with antibody tau5), the levels are
doubled in the homozygous mice. (E) Steady-state levels of the ␤APP and tau protein are approximately three–four-fold and six–eight-fold higher than
endogenous levels in hemizygous and homozygous mice, respectively. (F) SELDI analysis indicates that A␤40 and A␤42 are the two predominant A␤
species produced in the 3xTg-AD brain.
1066
S. Oddo et al. / Neurobiology of Aging 24 (2003) 1063–1070
of transmission frequency among the offspring confirmed
that both transgenes were co-inherited [6].
Compared to crossbreeding, this approach yielded several
major advantages. The integration of the ␤APP and tau transgenes at the same genetic locus renders it unlikely that either
transgene will independently assort in subsequent generations. Therefore, this tight linkage coupled to the ‘knockin’
of the PS1 mutation indicates that the 3xTg-AD mice breed
as readily as any single transgenic line, particularly because
these mice have also been bred to homozygosity. Thus, deriving a large colony is straightforward, cost-effective, and
does not require extensive genotyping of the progeny. Moreover, the easy propagation of this transgenic line facilitates
their crossing to other transgenic or gene-targeted mice to
assess the impact of other genotypes on the neuropathological or physiological phenotype. Lastly, another advantage to
this approach is that multiple transgenes are introduced into
an animal without altering or mixing the background genetic constitution. Thus, an important confounding variable
is avoided, which may be a crucial parameter for behavioral,
electrophysiological, and vaccine-based experiments.
Transgenic mice were readily identified by Southern blot
(Fig. 1B). Hemizygous (PS1M146V/M146V , h␤APPSwe +/0,
hTau +/0) and homozygous (PS1M146V/M146V , h␤APP +/+,
hTau +/+) transgenic mice were maintained. As expected,
steady-state levels of both transgene-specific mRNA and
protein products were doubled in homozygous as compared
to hemizygous 3xTg-AD mice (Fig. 1C and D). Notably, the
steady-state levels of both human APP and tau products are
comparable in mice of both genotypes (Fig. 1E).
Using sandwich ELISA, we previously showed that the
3xTg-AD mice produced increasing amounts of A␤ with
age, particularly the larger more amyloidogenic 42-amino
acid isoform [6]. To obtain a more thorough qualitative assessment, we used SELDI mass spectrometry. As shown in
Fig. 1F, both A␤40 and A␤42 are evident in total brain extracts of 3xTg-AD mice.
3.2. Aβ deposition precedes tau pathology
Because of the approach used to generate the 3xTg-AD
mice, both the tau and APP transgenes are expressed to
comparable levels in mice of the same genetic background
(Fig. 1C–E). We, therefore, chronicled the sequence of
neuropathology in hemizygous and homozygous 3xTg-AD
mice. As the steady-state levels of the human APP and tau
transgenes are roughly twice as high in the homozygous versus hemizygous mice, the neuropathology is more severely
manifested in these animals, occurring with an earlier onset
than hemizygous mice (Fig. 2). Moreover, we found that
the development of the plaque and tangle formation was
graded, with A␤ deposition emerging well in advance of tau
pathology. This was the case for both the hemizygous and
homozygous mice. It appears that A␤ pathology precedes
tau pathology by several months in this model, consistent
with the amyloid cascade hypothesis [3].
In 3xTg-AD mice of both genotypes, intraneuronal A␤
immunoreactivity is one of the earliest neuropathological
changes, which is first detectable in cortical brain regions
(cf. Figs. 2C and D). This is soon followed by the emergence of extracellular A␤ deposits (Fig. 2G). As with the
intraneuronal staining, the largest cluster of extracellular A␤
deposits are found in the frontal cortex and occur predominantly in layers 4–5 (Fig. 2C, E, and G). By 15 months,
extracellular A␤ deposits are apparent in posterior cortical
regions such as the occipital and parietal cortices, suggesting that there is an age-related, regional dependence to the
A␤ deposits in the 3xTg-AD mice. Also by this time point,
numerous extracellular A␤ deposits are observed in both
limbic and cortical regions of the hemizygous mice (Fig. 2I
and J). Many of these deposits were also thioflavin S positive (data not shown).
Whereas A␤ deposition is present by 6 months of age in
hippocampus of homozygous mice, the only tau immunoreactivity that was apparent at this time point was with antibody HT7, which recognizes human-specific tau. Antibodies such as MC1, which recognizes conformational changes
in tau, an early requirement for tangle formation failed to
yield any immunostaining in 6-month-old mice; it was not
until the 3xTg-AD mice reached about 12 months of age that
MC1-positive immunostaining became apparent [6]. Likewise, staining with a phospho-specific tau antibody, such as
AT8, which recognizes tau phosphorylated at serine 202 and
threonine 205, is also negative in younger mice (Fig. 3E–H).
Like the MC1 staining, it is not until the mice are approximately a year old before AT8 immunoreactivity is readily
observed (data not shown). In contrast to the A␤ staining,
we observed that in the majority of the 3xTg-AD mice, the
most extensive tau immunostaining was first apparent in the
CA1 region of the hippocampus and then progressively affected neurons in the cerebral cortex of older mice. The intraneuronal tau aggregates were also evident with histological stains including Gallayas and even by hematoxylin and
eosin and thioflavin S [6].
Both the hemizygous and homozygous 3xTg-AD mice
develop a progressive and age-dependent A␤ and tau pathology. Our analysis indicates that A␤ pathology precedes tau
pathology. Because both transgenes are expressed to comparable levels, these data provide in vivo experimental support
for the amyloid cascade hypothesis.
3.3. Astroglial response in 3xTg-AD mice
Astrogliosis is a prominent feature of AD neuropathology [1,3]. We double labeled brain sections from hemizygous 3xTg-AD mice with antibodies to A␤ and the
astrocytes-specific marker GFAP. We found that some reactive astrocytes colocalize with extracellular A␤ deposits
(Fig. 4A). To obtain a more quantitative assessment, we
measured GFAP levels in the cortex and hippocampus of
hemizygous 3xTg-AD and NonTg mice by western blotting.
Two time points were selected: 2 months, in which there is
S. Oddo et al. / Neurobiology of Aging 24 (2003) 1063–1070
Fig. 2. A␤ deposition initiates in the cortex and progresses to the hippocampus. Coronal sections from 6- and 15-month-old hemizygous and homozygous mice were evaluated with an A␤42-specific
antibody. Intracellular A␤ accumulation, the earliest pathological manifestations observed in these mice, is markedly more extensive in the homozygous mice as compared to hemizygous mice in both
the cortex and hippocampus. Note also that extracellular A␤ deposits are evident by 6 months of age in the cortex of homozygous mice, but do not appear in the hippocampus until the mice are older.
This was also the case for hemizygous mice. By 15 months of age, A␤ deposition was readily apparent in mice of both genotypes. Original magnifications, 10×.
1067
1068
S. Oddo et al. / Neurobiology of Aging 24 (2003) 1063–1070
Fig. 3. Tau pathology initiates in the hippocampus and progresses to the cortex. Coronal sections from 6- and 15-month-old hemizygous and homozygous mice were evaluated with a human-specific tau
antibody (HT7) and antibody AT8, which detects serine 202 and threonine 205 phosphorylated residues. Although human tau immunoreactivity is apparent in 6-month-old mice, it does not appear to be
hyperphosphorylated until the mice are older. By 15 months, extensive AT8 immunoreactivity is evident in the hippocampus as well as in cortical structures. Original magnifications, 10×.
S. Oddo et al. / Neurobiology of Aging 24 (2003) 1063–1070
1069
Fig. 4. Steady-state levels of GFAP increase with age in the 3xTg-AD mice. (A) Reactive astrocytes colocalize with some extracellular A␤ deposits.
Double labeling immunohistochemistry in which A␤ deposits are labeled with antibody 6E10 and stained with True Blue, whereas astrocytes are reacted
with an anti-GFAP antibody and stained with DAB. Steady-state levels of the astrocyte marker GFAP was measured in brain extracts prepared from the
cortex (B) and hippocampus (C) from a 15-month-old hemizygous and NonTg mice. Blots were normalized to ␤-actin.
no obvious neuropathology, and 18 months in which both
A␤ and tau pathology is quite advanced. At the younger
time point, steady-state levels of GFAP were comparable
between hemizygous 3xTg-AD and NonTg mice (Fig. 4B
and C). By 18 months of age, GFAP levels were markedly
elevated in the transgenic mice, particularly in the cortex.
4. Discussion
In this study, we characterized the regional and temporal
profile of A␤ and tau aggregates in the 3xTg-AD hemizy-
gous and homozygous mice. We find that the anatomical
and temporal pattern of both signature lesions develop
analogously to that observed in the AD brain [5]. For A␤
deposits, we observed both diffuse and fibrillar A␤ aggregates, each of which was reactive with an anti-A␤42-specific
antibody. Cortical areas were initially more affected than
subcortical regions such as the hippocampus. As the mice
age, however, amyloid deposits also developed in limbic
areas such as the hippocampus. The subiculum and CA1
subfield of the hippocampus were particularly affected, although we also noted that in older mice, the CA3 subfield
and dentate gyrus contained extensive A␤ deposits. In con-
1070
S. Oddo et al. / Neurobiology of Aging 24 (2003) 1063–1070
trast, tau aggregates were first apparent in the hippocampus,
and become markedly more widespread in cortical regions
as the mice age.
We also find that A␤ pathology precedes tau pathology
in this model. Whereas extracellular amyloid deposits manifested by 6 months of age, we did not observe conformational changes in tau as evidenced by MC1 immunostaining or immunoreactivity with phospho-specific tau markers
until the mice were about 1 year of age. There was a hierarchal pattern to the tau staining with MC1 immunoreactivity emerging first, followed by phospho-specific markers
such as AT8 and AT100, and then lastly PHF [6]. Because
the APP and tau transgenes were expressed to comparable
levels, we believe this observation provides strong experimental support for the amyloid cascade hypothesis. This hypothesis posits that A␤ accumulation, which may occur as a
consequence of overproduction, mismetabolism, or failures
in clearance, is the initiating trigger that underlies all forms
of AD [3].
Although A␤ deposits manifest prior to any overt tau
pathology in the 3xTg-AD mice, it is likely that A␤ aggregates influence the development of tau pathology. The same
appears to be true in other mouse models [1,4]. Although
A␤ and tau pathology initiate in different brain regions
in the 3xTg-AD mice (i.e. cortex for A␤ and hippocampus for tau), A␤ aggregates (particularly the intracellular
species) may alter cellular metabolism, leading to conformational and phosphorylation-specific changes in tau. Notably,
A␤42-specific epitopes are found to be associated with intracellular neurofibrillary tangles in the AD brain [2,8].
Despite the development of both plaques and tangles in
the 3xTg-AD mice, we observed that synaptic dysfunction
manifests prior the accumulation of these hallmark pathological lesions [6]. This suggests that synaptic dysfunction
is an early event in the pathogenesis of AD. Notably, in our
model, we correlated the deficits in synaptic plasticity with
APP overexpression (mostly intracellular A␤ accumulation)
and not with tau aggregates [6]. Nevertheless, it is probable
that altered tau will be found to also affect synaptic plasticity in older transgenic mice.
The development of both A␤ and tau pathology in the
3xTg-AD mouse model is significant as it should enable a
more accurate evaluation of potential therapeutic interventions (such as A␤ immunizations) in an animal model that
more closely mimics the neuropathology of AD. In addition,
this model will be useful for determining if modulation of
either the A␤ or tau pathology impacts the development of
the other.
Acknowledgments
This work was supported by a grant from the Alzheimer’s
Association and by the National Institutes of Health Grants
AG17968 and AG0212982.
References
[1] Gotz J, Chen F, van Dorpe J, Nitsch RM. Formation of neurofibrillary
tangles in P3011 tau transgenic mice induced by Abeta 42 fibrils.
Science 2001;293:1491–5.
[2] Grundke-Iqbal I, Iqbal K, George L, Tung YC, Kim KS, Wisniewski
HM. Amyloid protein and neurofibrillary tangles coexist in the same
neuron in Alzheimer disease. Proc Natl Acad Sci USA 1989;86:2853–
7.
[3] Hardy J, Selkoe DJ. The amyloid hypothesis of Alzheimer’s
disease: progress and problems on the road to therapeutics. Science
2002;297:353–6.
[4] Lewis J, Dickson DW, Lin WL, Chisholm L, Corral A, Jones G, et al.
Enhanced neurofibrillary degeneration in transgenic mice expressing
mutant tau and APP. Science 2001;293:1487–91.
[5] Mesulam MM. Neuroplasticity failure in Alzheimer’s disease:
bridging the gap between plaques and tangles. Neuron 1999;24:521–
9.
[6] Oddo S, Caccamo A, Shepherd JD, Murphy MP, Golde TE, Kayed
R, et al. Triple-transgenic model of Alzheimer’s disease with plaques
and tangles: intracellular A␤ and synaptic dysfunction. Neuron
2003;39:409–21.
[7] Scheuner D, Eckman C, Jensen M, Song X, Citron M, Suzuki N, et
al. Secreted amyloid beta-protein similar to that in the senile plaques
of Alzheimer’s disease is increased in vivo by the presenilin 1 and 2
and APP mutations linked to familial Alzheimer’s disease. Nat Med
1996;2:864–70.
[8] Schwab
C, McGeer PL. Abeta42-carboxyterminal-like immunoreactivity is
associated with intracellular neurofibrillary tangles and pick bodies.
Exp Neurol 2000;161:527–34.
[9] Selkoe DJ. Alzheimer’s disease: genes, proteins, and therapy. Physiol
Rev 2001;81:741–66.
[10] Selkoe DJ, Podlisny MB. Deciphering the genetic basis of
Alzheimer’s disease. Annu Rev Genom Hum Genet 2002.
[11] Sugarman MC, Yamasaki TR, Oddo S, Echegoyen JC, Murphy MP,
Golde TE, et al. Inclusion body myositis-like phenotype induced by
transgenic overexpression of beta APP in skeletal muscle. Proc Natl
Acad Sci USA 2002;99(9):6334–9.
[12] Wong PC, Cai H, Borchelt DR, Price DL. Genetically engineered
mouse models of neurodegenerative diseases. Nat Neurosci
2002;5:633–9.