3.1.1. Plaque Pathology in Mouse Models
As a mammalian model system, mice have the advantages of a short lifespan and rapid reproduction, which facilitates timely completion of experimental protocols. They are also comparatively easy to maintain and breed in a laboratory environment. Numerous tools, data, and standardised behavioural tests have been established for assessing phenotypes in mice. The development of embryonal stem cells and targeted mutagenesis has enabled the production of models that more accurately recapitulate the aetiology of human disease state. These factors combined has resulted in mice being the most common animal models of AD.
There have been a large number of mouse models constructed in various ways, far too many to include here. We have selected a representative group of models that were either notable because they were novel at the time or have been widely used in the field. Table 1 lists these selected mouse models.
Table 1. Selected key mouse models of AD and their major phenotypes.
Name |
Type of Modification |
FAD Mutations |
MAPT Mutations |
Plaques |
Tangles |
Neurodegeneration |
Reference |
PDAPP |
Transgenesis |
Indiana in APP |
|
X |
|
|
[21] |
Tg2576 |
Transgenesis |
Swedish in APP |
|
X |
|
|
[22] |
TgCRND8 |
Transgenesis |
Swedish and Indiana in APP |
|
X |
|
|
[23] |
PSAPP |
Transgenesis |
Swedish in APP, M146L in PSEN1 |
|
X |
|
|
[24] |
BRI-Aβ40 |
Transgenesis |
Aβ1–40 peptide |
|
|
|
|
[25] |
BRI-Aβ42 |
Transgenesis |
Aβ1–42 peptide |
|
X |
|
|
[25] |
5XFAD |
Transgenesis |
Swedish, Florida, London in APP. M146L and L286V in PSEN1 |
|
X |
|
X |
[26] |
JNPL3 |
Transgenesis |
|
P301L in MAPT |
|
X |
X |
[27] |
rTg4510 |
Transgenesis |
|
P301L in MAPT |
|
X |
X |
[28] |
3xTg |
Transgenesis |
Swedish in APP, M146L in PSEN1 |
P301L in MAPT |
X |
X |
X |
[29] |
TAPP |
Transgenesis |
Swedish in APP |
P301L in MAPT |
X |
X |
X |
[30] |
Plaques and tangles are the two main pathological hallmarks of AD, followed by neurodegeneration. In order to create models with high face validity, these phenotypes have been highly sought after. The first reported mouse models that developed plaque pathology were created via transgenesis (TG). Researchers introduced the human
APP gene (h
APP) containing mutations known to cause FAD. The first mouse model, the PDAPP line created in 1995, overexpressed the V717F Indiana mutation h
APP with the Platelet-Derived Growth Factor (PDGF) promoter via a minigene construct. Around 40 copies of the transgene were randomly inserted in this line at a single site, and all three major splice variants of h
APP (695, 751, and 770) were expressed. These mice developed both dense and diffuse plaque pathology by eight months of age in the entorhinal cortex, cingulate cortex, and hippocampus. By 18 months, the amyloid burden in these brain regions was thought to be greater than that seen in end stage human disease. This model also showed signs of synaptic loss, microgliosis, and astrocytosis, but no tau/tangle pathology or neurodegeneration
[21][31].
The next, and still commonly used mouse model, was the Tg2576 line, which overexpressed the K670M/N671L Swedish mutation in a transgene containing the 695 isoform of human h
APP transgene driven by the Prion Protein (PrP) promoter. Tg2576 mice develop plaques and memory deficits in a progressive manner. Similar to the PDAPP mice, they do not show the tangles or neurodegeneration
[31][32]. These mouse models developed memory deficits and synaptic loss preceding the accumulation of insoluble plaques, providing evidence for the hypothesis that it is the smaller soluble forms of Aβ that cause these symptoms
[33][34]. Several further mouse lines were also created by introducing the h
APP gene with various FAD causing mutations; most exhibited plaques and memory deficits in an age-dependant manner as well as some level of synaptotoxicity (reviewed in
[35]).
Some of these mouse lines were subsequently crossed to produce mouse lines with multiple
APP transgenes; the result was usually a similar phenotype that appeared at an earlier age, which shows that these mutations have cumulative phenotypic effects. One example is the TgCRND8 line, engineered with a single transgene to contain the h
APP isoform 695 with both the Swedish and Indiana mutations under the control of the Prp promoter. These mice develop plaque pathology by three months of age, with earlier signs of cognitive impairment relative to the models with a transgene carrying a single AD mutation. The brain concentration of Aβ
1–42 in this compound model at six months was equivalent to the original PDAPP mouse line at 16 months. This compound model also showed an increase in Aβ
1–42 to Aβ
1–40 ratio (now considered to be an important indication of amyloidogenesis)
[23]. However, these mice still did not exhibit the other major neuropathological hallmarks of AD such as the tangles and neurodegeneration.
Some of the APP overexpression mouse lines were later crossed with mice carrying a human
PSEN1 transgene (h
PSEN1) with various mutations responsible for FAD. Interestingly, mice overexpressing h
PSEN1 mutations do not develop plaques or other symptoms, but do exhibit an increased ratio of the more amyloidogenic Aβ
1–42 relative to Aβ
1–40 in the brain
[36][37][38]. Crossing transgenic mice that overexpressed
APP with
PSEN1 transgenic mice greatly increased amyloid pathology. An example is the crossing of Tg2576 mice (
APP Swedish mutation) with both the PS-1 line (
PSEN1 M146L mutation)
[24][38] and the
PSEN1 A246E line
[36][39]. Taken together, this animal model work helped confirm that APP metabolism, and in particular, the production of the Aβ
1–42 peptide, was affected by mutations in
APP and
PSEN1, and that these mutations are likely acting on a single pathway. This work also provided supporting evidence for the hypothesis that the Aβ
1–42 fragment is more toxic than Aβ
1–40.
Attempts to confirm the role of individual Aβ peptides led to the creation of transgenic mouse lines that selectively expressed either the Aβ
1–40 or Aβ
1–42 amyloid fragment in the absence of the h
APP transgene (BRI-Aβ40 and BRI-Aβ42)
[25]. These models showed that high expression of Aβ
1–40 caused no overt plaque pathology, but even low expression levels of Aβ
1–42 was sufficient to cause plaque formation in both parenchymal brain tissue and blood vessels (cerebral amyloid angiopathy).
Attempts to capture a more complete AD phenotype led to crossing transgenic mice or creating constructs to overexpress multiple transgenes and mutations within these genes. Cell loss and neurodegeneration was ultimately achieved in the 5XFAD mouse model that expressed three
APP (Swedish K670M/N671L, Florida I716V, and London V717I) and two
PSEN1 (M146L and L286V) mutations under the murine Thy-1 promoter
[26][40]. The severe phenotype again supported the hypothesis that FAD mutations have an additive effect. However tangles, which are the other main hallmark of AD, were absent in these mice.
3.1.2. Replicating AD Tau Pathology
Interestingly, unlike other mammalian species (see below), wild type mice do not develop tangles as they age
[41]. Mutations in the human
MAPT gene (microtubule associated protein tau), which codes for the human TAU (hTAU) protein, cause frontotemporal dementia (FTD), but not AD
[42]. However tangle pathology, neurodegeneration, and memory loss were seen in transgenic mice models expressing human
MAPT (h
MAPT) with FTD causing mutations. The first mouse model with this phenotype was the JNPL3 line, which expressed the 4R0N isoform of h
MAPT with the P301L mutation
[27]. Subsequently, a hTAU expression tetracycline repressible mouse line (rTg4510) demonstrated that the smaller soluble forms of oligomeric TAU caused memory loss and neurodegeneration
[43][44]. Many overexpression h
MAPT transgenic lines have been produced and some have been crossed with transgenic mouse lines overexpressing FAD mutations in
APP and/or
PSEN1. The resulting lines demonstrated that the mechanisms leading to amyloid and TAU pathology interact. The 3xTg mice (Swedish mutation in
APP, M146V in
PSEN1, and P301L in
MAPT) develop plaques before tangles
[44][45], as observed in AD patients. A line developed by crossing the aforementioned
APP mutant mice Tg2576 with the
MAPT JNPL3 mice (called the TAPP line) altered the spatial distribution of tangles in the brain relative to original TAU expressing strain, with TAPP mice exhibiting tangles in the subiculum, hippocampus, and isocortex that were not present in JNPL3 mice. TAPP mice also had greatly increased numbers of tangles in the olfactory cortex, entorhinal cortex, and amygdala. This suggests that Aβ fibril deposition can alter the amount and distribution of insoluble TAU as tangles
[30].
3.1.3. Construct Validity of Transgenic Mouse Models of AD
Although mice expressing a transgene with a single FAD mutation display some symptoms of the disease, it is evident from the literature that three or more AD and FTD associated mutations are required to replicate the majority of the human pathology. In contrast, multiple mutations have not been reported in humans with AD, and in nearly all cases of FAD, only a single mutation is required to develop the entire phenotype.
There are good reasons for the requirement of a compound approach to create equivalent AD pathology. Unlike human h
APP, the proteolytic cleavage products of murine
App (m
App) do not naturally form plaques. This is due to three amino acid substitutions in the amyloid beta sequence compared to human (
Figure 1), which reduces the ability of murine Aβ peptides to aggregate
[45]. In addition, murine β-secretase enzymes typically cleave m
App to form Aβ
11-x, even though it cleaves h
APP to form Aβ
1-x [46][47]. Deposition of cleavage products from m
App is only apparent in models with high expression levels of m
App and only after an extended period. This is one of the reasons why h
APP is typically used instead
[48].
Figure 1. A comparison of the human and mouse Aβ peptide sequence, showing the three amino acid substitutions responsible for the functional difference between the two.
The ratio of h
APP isoforms differs within brain regions and also in other organs. The ratio also changes during the course of development and ageing
[49][50]. The two longer isoforms of h
APP, 751 and 770, are more prevalent in the AD brain relative to healthy controls
[51]. Overexpressing the h
APP 751 isoform also causes more obvious amyloid pathology in mice than overexpressing the short (APP695) isoform
[52]. The pathology generated in a mouse model therefore depends on which of the three isoforms of h
APP is overexpressed, or whether the full h
APP gene sequence is used.
Several different promoters have been used to drive overexpression of h
APP in mouse models of AD including the promoters for PDGF-B (platelet derived growth factor B-chain) and the PrP (prion protein gene motifs). Different promoters drive different levels and spatial patterns of expression including outside the brain. For example, the PDGF-B and Thy-1 (thymocyte differentiation antigen 1) promoters are neuron-specific
[53][54], while the PrP promoter has less specificity, also driving expression in glial cells and other non-brain tissue
[55]. The Thy-1 promoter included in the construct to make the APP23 model (Swedish mutation in
APP) is active only after birth, preventing potential developmental effects
[56]. Various Tet-controlled lines have been created that allow for more control over the timing and location of transgene expression, but have the added complication of requiring an extra transgene
[57][58][59]. All of these promoters are selected for ease of use or particular benefits, but because none of them are the endogenous promoter, the natural expression pattern of
APP is not replicated in any of the models.
3.1.4. Murine APP Knock in Models
In an attempt to overcome the limitations of
APP TG models, a small number of knock in (KI)
App models have been created with targeted gene editing. Inserting selected mutations in the endogenous genes should mean expression is quantitatively, spatially, and temporally appropriate. Mouse
App was ‘humanised’ in these models by converting the codons for the three amino acids that differ between human and mice in the Aβ coding portion of m
App. This allows murine BACE1 to cleave mAPP at the human equivalent position
[60][61][62][63]. These mice did not develop overt phenotypes such as memory deficits, synaptic loss, and/or plaque pathology. These phenotypes only became evident after the insertion of multiple
APP mutations (combinations of Swedish, London, Dutch, Iberian, and Artic)
[64], and usually only after breeding to homozygosity in concert with homozygous FAD
PSEN1 mutations
[63][65].
The necessity of including multiple mutations to induce human equivalent disease confounds the use of these models, but they have helped differentiate between phenotypes due to the TG process, and those that represent the disease in a mouse. Consistent phenotypes observed in TG and KI models include plaque formation, changes to glial cells and astrocytes, and lowered rates of hippocampal neurogenesis, although some artifacts such as transgene calpain activation have been noted
[66][67][68]. The presence of cognitive impairment appears to vary more between KI models than TG models. The KI models with cognitive impairment have plaque pathology prior to memory impairment, unlike the commonly used TG mice models
[62][69]. Memory impairment following plaque formation is the order of events seen in patients
[70], so KI models do appear to more faithfully replicate symptom clusters. Despite this, the higher variability of phenotypes in KI models, along with their milder symptom profile, means that transgenic models are still widely used.
3.1.6. Construct Validity of MAPT Mouse Models
Compared to hTAU with six isoforms (named 4R2N, 4R1N, 4R0N, 3R2N, 3R1N, and 3R0N)
[79][80][81] murine TAU (mTAU) only has three of the human equivalent isoforms (4R0N, 4R1N, 4R2N). There is also variability in TAU protein conservation. Some regions of mTAU tau are very similar to hTAU, while other regions differ greatly. There are species-specific differences in the presence of different isoforms during development, and spatially across the brain
[81][82]. In TG models, the presence of endogenous mouse Mapt (m
Mapt) can alter the splicing ratios of introduced h
MAPT [83][84].
As stated above, unlike in humans, tangles do not form naturally with age in the mouse brain. Indeed, it appears that replacing the m
Mapt gene with the human equivalent, and in some lines with a FTD mutation, is necessary to create a TAU dysfunction phenotype in mice
[85]. The inclusion of FTD mutations to ensure a tangle phenotype in murine models is a major issue for construct validity. There are probably better models of frontotemporal dementia and other tauopathies than AD, even though they have provided insights about TAU toxicity
[29][30]. Unexpected non-disease associated deficits have been found in some models, for example, the commonly used JNPL3 line (P103L mutation in
MAPT) has motor impairments and develops eye irritations
[27][86]. Further the Tau P301S line develops severe paraparesis at 5–6 months
[87]. However severe motor impairment is not usually observed in AD until late in the disease course
[88].
3.1.7. Predictive Validity of Murine Models
Almost no mouse model of AD has shown predictive validity in human clinical trials to date, despite many therapeutic agents ‘curing’ a mouse of AD symptoms (for reviews, see
[85][86]). Those that have been successful were based on the cholinergic system or NMDA receptors and only provide temporary symptomatic relief. While symptomatic relief is important, the predicted increase in the prevalence of AD means that finding a method to prevent or cure the disease is now becoming an urgent priority.
In addition to drug failures, there is the issue of differences in drug metabolism between species; something well tolerated in mice may not be so in humans
[89][90]. Many clinical trials have failed to make it to later stages due to adverse side effects, which were not present in mice. For example, immunisation of mice with Aβ
1–42 (named AN1792 in the clinical trial) was able to lower the volume of plaque material in the brain and preserve cognitive function. Unfortunately, this approach failed to show benefits in clinical trials and 6% of the immunised patients developed meningoencephalitis
[91][92]. The adverse effects were thought to be due to a T-cell response in humans against the large Aβ
1–42 fragment. Subsequent immunisation trials with smaller epitopes that were beneficial in mice including the drugs Bapineuzumab
[93] and Solanezumab
[94] showed a similar lack of efficacy and/or adverse side effects
[95][96][97].
To date, well over 200 compounds have failed to affect the disease course
[10], and this appears to have led to some controversial decisions. Recently, the drug aducanumab (sold as Aduhelm) was approved by the FDA through an accelerated approval pathway, on the condition that follow-up trials are performed to determine efficacy. This drug showed mixed results in clinical trials, with a benefit seen at the highest dose, but only in one of the two trials. Given that 35% of patients developed brain swelling (cerebral adema) and 19% brain bleeds (intracerebral haemorrhage), there are serious safety considerations
[98]. It is clear that models of AD with higher predictive validity are desperately needed.