REVIEW ARTICLE
Animal models of amyloid-b-related pathologies in
Alzheimer’s disease
Ola Philipson
1
, Anna Lord
2
, Astrid Gumucio
1
, Paul O’Callaghan
1
, Lars Lannfelt
1
and Lars
N.G. Nilsson
1
1 Department of Public Health and Caring Sciences ⁄Molecular Geriatrics, Uppsala University, Sweden
2 BioArctic Neuroscience AB, Stockholm, Sweden
Introduction
Alzheimer’s disease (AD) accounts for 60–70% of
all dementia cases. Prevalence increases with age from
1% in the 60 to 64-year age group, to 24–33% in
those aged > 85 years. There is an insidious onset
with an initial loss of short-term memory, followed by
progressive impairment of multiple cognitive functions
that affect the activities of daily living. The AD diag-
nosis is based on a patient’s medical history, neurolog-
ical assessment and neuropsychiatric testing of
cognitive functions. Neuroimaging techniques and
biomarkers in cerebrospinal fluid (CSF) are invaluable
in differential diagnosis.
The neuropathological diagnosis takes into account
the regional distribution and frequency of histopatho-
logical hallmarks; specifically, extracellular neuritic pla-
ques and intracellular neurofibrillary tangles (NFTs) in
postmortem brain. Neuritic plaques mainly consist of
b-sheet-containing fibrils of amyloid-b(Ab) that are
surrounded by dystrophic neurites and reactive glial
cells. Diffuse Abdeposits are also present, but these
Keywords
Alzheimer’s disease; amyloid beta-protein;
amyloid beta-precursor protein; animal
model; apolipoprotein E; neuropathology;
presenilin-1; presenilin-2; tau proteins;
transgenic mice
Correspondence
L. Nilsson, Department of Public Health and
Caring Sciences, Molecular Geriatrics,
Uppsala University, Rudbeck Laboratory,
Dag Hammarskjo
¨lds va
¨g 20, SE-751 85
Uppsala, Sweden
Fax: +46 18 471 4808
Tel: +46 18 471 5039
E-mail: Lars.Nilsson@pubcare.uu.se
(Received 5 October 2009, revised 29
November 2009, accepted 30 December
2009)
doi:10.1111/j.1742-4658.2010.07564.x
In the early 1990s, breakthrough discoveries on the genetics of Alzheimer’s
disease led to the identification of missense mutations in the amyloid-b
precursor protein gene. Research findings quickly followed, giving insights
into molecular pathogenesis and possibilities for the development of new
types of animal models. The complete toolbox of transgenic techniques,
including pronuclear oocyte injection and homologous recombination, has
been applied in the Alzheimer’s disease field, to produce overexpressors,
knockouts, knockins and regulatable transgenics. Transgenic models have
dramatically advanced our understanding of pathogenic mechanisms and
allowed therapeutic approaches to be tested. Following a brief introduction
to Alzheimer’s disease, various nontransgenic and transgenic animal models
are described in terms of their values and limitations with respect to patho-
genic, therapeutic and functional understandings of the human disease.
Abbreviations
AD, Alzheimer’s disease; ApoE, apolipoprotein E; APP, amyloid-bprecursor protein; Ab, Amyloid-b; BACE-1, b-site APP cleaving enzyme-1;
CAA, cerebral amyloid angiopathy; CCR2, chemokine (C-C motif) receptor 2; CSF, cerebrospinal fluid; MWM, Morris water maze; NFTs,
neurofibrillary tangles; PDGF, platelet-derived growth factor; PS, presenilin; SMC, smooth muscle cells; wt, wild-type.
FEBS Journal 277 (2010) 1389–1409 ª2010 The Authors Journal compilation ª2010 FEBS 1389
lack b-sheet structure and are therefore by definition
not amyloid. Cerebral amyloid angiopathy (CAA)
results in the degeneration of vessel walls and hemor-
rhages. CAA is found in 80% of AD brains, but is
not a diagnostic criterion. NFTs are intracellular fila-
mentous lesions with amyloid properties. They contain
hyperphosphorylated and aggregated forms of tau, a
microtubule-associated protein that normally serves to
assemble and stabilize microtubules.
Genetics and risk factors implicated in
Alzheimer’s disease pathogenesis
Familial forms of AD, with an autosomal dominant
mode of inheritance, account for < 2% of all AD
cases. Onset is most often before 65 years of age, and
the penetrance is nearly always complete. The purifica-
tion and partial sequencing of Abfrom amyloid depos-
its of AD brain in the 1980s [1], led to the cloning and
localization of the amyloid-bprecursor protein (APP)
gene on chromosome 21 [2]. The first identified AD
mutation was located in the APP gene [3], although
the majority of mutations were caused by genetic
lesions in the presenilin (PS) genes, PS1 and PS2. The
mutations either enhance the steady-state level of Ab,
like the Swedish APP mutation (K670N ⁄M671L) [4],
or selectively increase the level of Ab42 and ⁄or alter
the Ab42 ⁄Ab40-ratio, like the PS and London-type
APP mutations do [5]. Abis liberated following
cleavage of APP by b-site APP-cleaving enzyme-1
(BACE-1) and the c-secretase complex, in which
presenilin contributes to the catalytic activity (Fig. 1).
However, only a fraction of Abin postmortem
AD brain is full-length Ab1-42 or Ab1-40. N- and
C-terminally truncated variants are prevalent, and Ab
can undergo racemization, isomerization [6] and pyro-
glutamyl modification [7]. The biochemical processes
generating all these Abspecies and their significance
to AD pathogenesis are only partially understood.
Early-onset AD can also arise as a result of increased
APP gene dosage caused by APP gene duplication [8]
and Down’s syndrome with trisomy 21. Virtually all
Down’s syndrome patients aged 35–40 years develop
AD neuropathology, and most experience dementia by
60–70 years of age [9].
The major genetic risk factor for developing late-
onset AD is the apolipoprotein E (ApoE) e4 allele
[10,11]. One ApoE e4 allele increases the risk of AD by
two- to threefold, and two e4 alleles confer a 12-fold
increase in risk. In the brain, ApoE is primarily synthe-
sized by astrocytes and serves to regulate the transport
of cholesterol-containing lipoprotein particles. ApoE
binds to Aband becomes a component of amyloid in
AD senile plaques. The pathogenic mechanism of
ApoE likely relates to altered deposition and ⁄or clear-
ance of Abin the brain, although the details are still
not fully understood [12]. A large number of other dis-
ease-related loci and candidate genes have been pro-
posed, but not generally verified, indicating that these
genes have a modest impact on the pathogenesis. The
major risk factors for AD are age and a family history
of the disease. Low education or cognitive reserve
capacity, female gender, head trauma, hypertension,
cardiovascular disease and a high-cholesterol diet are
proposed risk factors for AD [13] (Fig. 2).
Nontransgenic animal models
Based on the cholinergic hypothesis, scopolamine-
induced amnesia, excitotoxic lesions of the basal
forebrain and aged primates have been used to assess
cognitive deficits. Current symptomatic drugs for AD
were successfully evaluated in these models, but their
etiological relevance is low [14]. Nontransgenic rodents
Fig. 1. Disease-causing APP mutations used in transgenic models. The Swedish mutation (1) favors b-secretase (b) cleavage, while the
Flemish mutation (2) partly disfavors cleavage of APP at the a-secretase (a) site. The Arctic, Dutch and Iowa mutations (3), which are located
in the Ab-domain, mainly increase aggregation. The London-type APP mutations (4) alter c-secretase (c) cleavage to increase Ab42 or the
Ab42 ⁄Ab40 ratio.
Animal models of Alzheimer’s disease O. Philipson et al.
1390 FEBS Journal 277 (2010) 1389–1409 ª2010 The Authors Journal compilation ª2010 FEBS
are poor natural animal models of AD, but intracere-
broventricular infusion of Ab[15] or lipopolysaccharide
in such animals has been used. The latter leads to
neuroinflammation with hippocampal neurodegenera-
tion and spatial memory deficits [16]. The models
require attention to methodological detail and are
difficult to standardize. Senescence-accelerated mice
were selectively bred from AKR ⁄J mice. In short-lived
SAM-P8, there is an age-related increase in diffuse Ab
deposits and cyclin-dependent kinase 5, cholinergic
deficits and increased blood–brain barrier permeability.
The phenotypes likely relate to oxidative stress and
mitochondrial dysfunctions [17]. Mice with segmental
trisomy of chromosome 16 have primarily been used to
dissect the genetic mechanisms of Down’s syndrome
phenotypes. Ts65Dn mice [18], the most frequently
used model, have interesting synaptic and cognitive
phenotypes with degeneration of cholinergic neurons
that depend on APP gene dosage.
Following observations in postmortem brain from
patients with coronary artery disease, rabbits fed a
cholesterol-enriched diet were used as an animal model
[19]. They are impared in classical eyelid conditioning
and show diffuse Abdeposits and vascular inflamma-
tion. Aged dogs (> 10 years) can show impaired atten-
tion, spatial disorientation and disturbed diurnal
rhythm. Cognitive dysfunction in old dogs is associated
with diffuse Abdeposits [20], neuritic dystrophy and gli-
osis, but few amyloid plaques and no NFTs. Ventricular
dilation, cortical and hippocampal atrophy, CAA with
degeneration of smooth muscle cells (SMC) and hemor-
rhages can all be found in aged canine brain. Interest in
nonhuman primate models has grown following the fail-
ure to predict meningoencephalitis as a side-effect of the
AN1792 vaccination trial from transgenic studies. The
efficacy and safety of an Abvaccine has been tested in
the Carribean vervet monkey [21]. Alternatives are aged
lemurs [22], cotton-top tamarins [23], rhesus monkeys
[24] or squirrel monkeys [25]. An aged chimpanzee with
complete AD neuropathology, including neuritic
plaques and paired helical filament-containing NFTs,
was recently reported [26].
Fig. 2. AD pathogenesis according to the
amyloid cascade hypothesis. This theory
suggests that altered metabolism of Ab,in
particular aggregation-prone Abspecies like
Ab42, initiates AD pathogenesis. Oligomeric
assemblies of Abtrigger aggregation of tau
and the formation of NFTs, but also inflam-
mation and oxidative stress, by rather
unclear mechanisms. These downstream
processes give rise to progressive neurode-
generation, which ultimately results in
dementia. The main pathogenic pathway of
AD is illustrated with red arrows, whereas
minor contributory pathways are shown
with thinner brown arrows. The experimen-
tal support for the hypothesis comes mainly
from studies of families in which AD is
inherited as a dominant trait due to muta-
tions in APP,PS1 or PS2. The evidence that
the theory applies to sporadic AD is less
solid, although risk factors such as age and
ApoE genotype both strongly impact on Ab
aggregation in transgenic models and post
mortem AD brain.
O. Philipson et al. Animal models of Alzheimer’s disease
FEBS Journal 277 (2010) 1389–1409 ª2010 The Authors Journal compilation ª2010 FEBS 1391
Transgenic animal models
Models devoid of any disease-causing APP
mutations
Animal models expressing wild-type (wt) human APP
are of interest because the great majority of sporadic
AD patients do not carry any disease causing APP
mutation. In early transgenic attempts, APP processing
was bypassed altogether and human Abwas directly
expressed under a promoter. Natural inclusions in the
brain were mistakenly identified as amyloid-like fibrils
in these mice [27]. Fusion proteins, in which the signal
peptide and C-terminal fragment (C99) of wild-type
APP were joined and expressed under the control of
the cytomegalus enhancer ⁄chick b-actin promoter,
were also generated. Ablevels in plasma from these
transgenic mice were in the nmrange, but Abdeposits
did not form in the brain. Instead, intracellular Ab
aggregates or amyloid deposists were found in the pan-
creas [28], intestine [29] or skeletal muscles [30]. The
level of plasma Abin C99-based models was similar to
Tg2576, an APP transgenic model with high peripheral
promoter activity.
In an alternative strategy, a yeast artificial chromo-
some, harboring the whole APPwt gene, was used to
maintain transcriptional regulation, alternative splicing
and normal APP processing. In these Py8.9 mice,
proper APP protein synthesis and alternative splicing
was demonstrated, but the brain was devoid of neuro-
pathology and the levels of Abwere low [31]. How-
ever, when wild-type human APP was expressed at
very high levels, under the Thy1 promoter, sparse
parenchymal and vascular amyloid deposits were
found in aged mice [32]. Thus a pathogenic APP muta-
tion is not a prerequisite for amyloid deposition.
Instead it seems to depend upon producing sufficient
Ablevels in the brain to ensure fibrillization. To
explore the pathogenic impact of individual Abspe-
cies, a fusion protein, BRI–wt-Ab42, was designed
from which Abwas released by furin-like enzymes on
the cell surface. BRI is a transmembrane protein that
is involved in amyloid deposition in British familial
dementia. The fusion design permitted the synthesis of
high Ablevels in the brain in a manner similar to APP
transgenic mice, but in the absence of APP overexpres-
sion. Transgenic mice expressing BRI–wt-Ab42 devel-
oped extensive vascular and parenchymal amyloid
pathology, accompanied by dystrophic neurites and
astrogliosis. In contrast to many APP transgenic mod-
els, amyloid deposition in BRI–wt-Ab42 mice began in
the cerebellum, where both furin and the transgene
were highly expressed. This illustrates how the anatom-
ical location of AD neuropathology can be manipu-
lated simply by enhancing the regional dosage of
amyloidogenic proteins and enzymes regulating their
metabolism. In contrast to BRI–wt-Ab42, no neuropa-
thology was found in aged BRI–wt-Ab40 transgenic
mice although they had higher Ablevels when they
were young. Thus the identity of Abdetermines if
neuropathology will develop [33].
Models with the London-type APP mutation
The London mutation (V717I) was the first genetic
lesion to be discovered in a family with AD [3], and
shortly thereafter the Indiana mutation (V717F) was
found in an American pedigree [34]. Patients with the
Indiana mutation develop short-term learning impair-
ments in the fifth decade, followed by progressive
cognitive impairment and dementia with typical AD
neuropathology. An abundance of NFTs and senile
plaques was observed at autopsy, as well as mild CAA
[35]. Games et al. described the neuropathology of
PDAPP mice, the first transgenic AD model [36]. A
mini-gene encompassing a human APP cDNA with the
Indiana mutation interposed with introns had been
designed. Alternative splicing and synthesis of all three
isoforms APP
695
, APP
751
and APP
770
, with strong and
selective neuronal expression was enabled by the plate-
let derived growth factor (PDGF)bpromoter. Impor-
tantly, young PDAPP mice produced high Ab42 levels
in the brain, particularly in the hippocampus. The ani-
mals preferentially accumulated Ab42 peptides and
developed senile plaques, but also a substantial num-
ber of diffuse Abdeposits at 9–10 months of age [37].
Plaque formation began in the cingulate cortex and
was accompanied by phospho-tau immunoreactive
dystrophic neurites, synaptic loss and gliosis in the
adjacent tissue, but not by overt neuronal loss [38,39].
Ultrastructural analyses revealed neurons in close
proximity to senile plaques and amyloid fibrils. The
latter had a diameter of 9–11 nm and were surrounded
by neuronal membranes and vesicles [40]. Young
PDAPP mice showed deficits in spatial learning and
memory, which worsened with increasing age and Ab
burden, although their performance in a novel
object-recognition task was unimpaired [41]. By
contrast, others found age-dependent deficits in object
recognition and place learning impairments that were
independent of age [42]. These discrepancies could be
because of differences in experimental procedures or
unintentional genetic drift of mouse colonies. PDAPP
mice are typically bred on a mixed genetic background
(Swiss Webster, DBA ⁄2 and C57Bl ⁄6). Hippocampal
volume and corpus callosum length is reduced in
Animal models of Alzheimer’s disease O. Philipson et al.
1392 FEBS Journal 277 (2010) 1389–1409 ª2010 The Authors Journal compilation ª2010 FEBS
PDAPP mice and this depends on APP gene dosage,
but it is unrelated to age-dependent Abaccumulation
[43]. Certainly this abnormality could impact on the
behavior of PDAPP mice, but the molecular mecha-
nism and its relevance to macroscopic atrophy in AD
(if any) is still unclear.
Van Leuven et al. generated several transgenic mod-
els, including mice with only the London mutation,
APP-London. As expected, a markedly increased level
of Ab42 was found in young mice and predominantly
Ab42-immunoreactive diffuse and neuritic plaques in
aged animals, compared with models harboring the
Swedish mutation. Impaired long-term potentiation in
hippocampal slices and deficits in spatial learning and
memory in the Morris water maze (MWM) were
reported. The mice were on a FVB ⁄N background and
displayed neophobic behavior. They were sensitive to
glutamate antagonists and died prematurely. These
phenotypes were noted in young mice prior to the
onset of plaque formation, and could possibly be
caused by the combination of APP overexpression and
FVB ⁄N genetic background [44]. In older mice
(> 15 months), CAA was found in the arteries and
pial arterioles in association with disruption of external
elastic lamina and the formation of aneurysm. Further-
more, the ratio of Ab42 ⁄Ab40 levels in leptomeninges
was eight times lower than in neocortical tissue
extracts. Ab42 could still have initiated deposition of
Abin vessels, because some focal lesions were only
Ab42-immunoreactive [45]. By contrast, brains from
patients with the London mutation contained mainly
Ab-immunoreactive plaques and cytoskeletal pathol-
ogy, but only modest or little CAA [46]. Thus, the
CAA phenotype in the transgenic mice might not have
been caused by the London mutation. Instead it may
be the result of strong APP expression, advanced age,
strain background (FVB ⁄N) and ⁄or differences in APP
processing between species.
Models with the Swedish APP mutation
The Swedish mutation (KM670 ⁄671NL) is located just
outside the N-terminus of the Abdomain in APP. It
was identified in 1992 [47] and shown to increase Ab
levels by six- to eightfold [4]. These discoveries created
intense interest in APP processing and paved the way
for the development of more sophisticated ELISAs to
selectively measure Ab40 and Ab42 [48]. Later, the
Swedish mutation became essential in the identification
and characterization of BACE-1 [49]. The clinical and
neuropathological features associated with the Swedish
mutation are those of typical AD [47,50]. Tg2576
mice, the most frequently used APP transgenic model,
harbor the Swedish mutation and display both AD-like
Abneuropathology and cognitive deficits [51]. The
Swedish mutation redirects APP processing to secre-
tory vesicles en route to the cell surface in cell culture
[52], whereas APPwt is largely processed in recycling
endosomes [53]. This difference may be largely irrele-
vant in the brain, because Absynthesis along the
endolysosomal pathway is clearly important in Tg2576
[54]. A substantial amount of CAA is often found in
transgenic mice with the Swedish mutation, which is
likely to be because of the high rate of synthesis and
accumulation of Ab1-40.
In Tg2576, more than fivefold overexpression of the
human APP
695
isoform with the Swedish mutation is
generated by the prion promoter. APP cDNA was
cloned into a 40 kb genomic fragment (cosSHaPrP)
[55] from the hamster prion protein gene. A significant
proportion of Tg2576 mice die at a young age, and the
severity of this phenotype depends upon the genetic
background. It has been found that colonies are best
maintained by mating heterozygous C57BL ⁄6 males
with B6SJLF1 females. At around 11 months of
age, Tg2576 mice show extracellular Abdeposits which
are largely soluble in SDS and mainly contain Ab40
(75%). CSF levels of Ab42, but not Ab40, decrease
with age and amyloid deposition [56], and pathogenesis
is accelerated in female Tg2576 mice [57]. Both these
observations fit well with biochemical and epidemio-
logical findings in AD [13].
Borchelt et al. used the prion promoter to generate
line C3-3 [58]. A chimeric cDNA clone encoding
murine APP
695
was used, in which the region in and
around the murine Abdomain was replaced with the
human Absequence and the Swedish mutation. The
MoPrP.Xho vector was much smaller than the cos-
SHaPrP, and selectively directed twofold overexpres-
sion of APP to the brain [58,59]. In an even more
refined strategy, the murine Absequence was human-
ized and the Swedish mutation introduced with gene
targeting. In this knockin model, APP
NLh ⁄NLh
, only
five single amino acids were altered in the entire mur-
ine genome. Consequently, APP protein synthesis
remained unchanged in terms of its spatial and tempo-
ral expression pattern and mRNA localization. The
Swedish mutation led to markedly enhanced b-secre-
tase activity and a ninefold increase in Ablevel,
compared with normal aged human brain [60]. By
22 months of age, APP
NLh ⁄NLh
mice had not devel-
oped Abneuropathology [61], but young mice were
elegantly used to estimate the turnover of Ab, APP
and APP fragments in vivo [62]. In another genomic
approach, a 650 kb yeast artificial chromosome vector
harboring the whole human APP gene locus with the
O. Philipson et al. Animal models of Alzheimer’s disease
FEBS Journal 277 (2010) 1389–1409 ª2010 The Authors Journal compilation ª2010 FEBS 1393