Dementia Insights: What Do Animal Models of Alzheimer’s Disease Tell Us?

Alzheimer’s disease (AD) is characterized by amyloid β (Aβ) plaques, neurofibrillary tangles (NFT) made of hyperphosphorylated tau, and chronic neuroinflammation. Exactly how each contributes to AD-related neurodegeneration is only partially understood. Much of what we know comes from studying each pathology in animal models that provide an opportunity for detailed cellular and molecular investigation and serve as test beds for novel therapeutics in AD.

There are 2 forms that comprise 95% of all AD occurrence, sporadic AD and late-onset AD (LOAD), the latter of which has onset after age 65. Familial (fAD) occurs before age 65 and comprises the other 5% of cases. Most animal models for AD are mouse genetic models using mutations identified from people with early-onset fAD. Among these are mutations in genes coding for amyloid precursor protein (APP), presenilin 1 (PS1), and presenilin 2 (PS2). Alternative genetic models include mutations that induce changes in tau. Other genetic models have been developed for mutations associated with increased risk of LOAD (eg, apolipoprotein E4 alleles and a missense mutation in TREM2). In addition to genetic models, a number of risk factors that increase the risk of LOAD (eg, age, diabetes, and obesity) are being investigated in rodent models to determine mechanisms that may underlie the increased risk.

Rodents are the most widely used animal for study of AD because of their biological similarity to humans, short life-cycles, and existing tools to induce mutations. Additionally, researchers can have total control of the environment for rodents, increasing the ability to limit extraneous variables in experiments. The investigation of AD mechanisms in these models range from very precise mutations to study specific pathological features to combined mutations to generate more comprehensive models of AD pathology.

In this article, we briefly highlight some animal models and recent approaches being used to further our knowledge of AD pathophysiology. Major categories of nonclinical AD models employed in AD-based investigations are presented, including some of the most recent preclinical models and translational approaches to mirror LOAD in the clinical population.

Amyloid Models

Discovery of APP, PS1, and PS2 mutations provided an opportunity to test the Hardy and Higgins amyloid-cascade hypothesis that Aβ deposition is the pathophysiologic cause of AD, because PS1 and PS2 cleave APP into Aβ isoforms. Inducing fAD mutations in mouse models allows study of the amyloidosis process and effects in the evolution of AD.

The first successful mouse model (Figure) to present with Aβ aggregation was the PDAPP transgenic mouse, developed using a human fAD APP mutation, V717F, with platelet derived growth factor-B (PDGFB) as a promoter. The PDAPP mice develop extracellular Aβ at age 6 months, exhibit apparent plaques at age 9 months, and have increasing plaque burden through age 22 months.¹ The PDAPP model allowed evaluation of novel amyloid-reducing agents for treatment of AD.^2,3 There are some limitations of PDAPP mice, including a disproportionate elevation in Aβ42 and limited vascular amyloid deposition, unlike that seen in LOAD.

Another mouse model,Tg 2576, was generated with the Swedish mutation, K670N and M671L in APP, driven by the hamster prion protein (PrP) promoter. Deposition of Aβ in Tg2576 mice also begins at age 6 months and Aβ plaques develop around age 12 months. In contrast to PDAPP mice, Aβ40 and Aβ42 increase nearly proportionally and Tg2576 mice present with larger plaques and vascular amyloid deposition.¹ Structural inconsistencies between Tg2576 mice and human AD amyloidosis, however, limit use of this model for investigation of AD plaque formation and elimination.⁴

The APP23 mouse model was created with the same Swedish mutation as the Tg2576 mice, but with an APP isoform and Thy-1 as a promoter. Because of lower breeding efficiency, APP23 mice are typically maintained as heterozygotes. At about age 6 months, APP23 mice develop Aβ plaques similar to those seen in human AD but with a denser core and smaller halo. A prominent hallmark of APP23 mice is the considerable amount of vascular amyloid deposition and subsequent cerebral amyloid angiopathy (CAA) these mice develop.⁵ A strength of this model is the significant neuronal loss (14%-28%) in the hippocampus at age 14 to 18 months that also mimics what is seen in humans with AD.⁵

Cognitive deficits are observed in all these mouse models, demonstrated through common working and recognition memory tests (eg, novel object recognition [NOR], Morris water maze [MWM], and Y-maze).¹ Although all of these mouse models also show dystrophic neurites and an activated immune system, neuron loss does not occur, and even in the AP23 mice that have hyperphosphorylated tau around Aβ plaques, NFT are not observed.⁵

Mouse models with mutations in PS1 and PS2 consistent with fAD have also been created. The PS1 mouse uses the PS1 mutation, M146L. The PS1 mice are unique in having calcium dysregulation that enhanced apoptotic cascades (programmed cell death processes), allowing researchers to analyze the role of calcium and apoptosis in neurodegeneration. The PS1 mice also present with selective increases in the amount of Aβ42, with no change in Aβ40 levels but do not develop Aβ plaques and or evidence of cognitive deficits. This has led to breeding mice with both APP and PS mutations.

Breeding mice to have both APP and PS mutations results in a more aggressive form of disease in the APP / PS1 mouse model generated by crossing the Tg2576 and PS1 mice. The APP / PS1 mice develop Aβ deposits at about age 3 months, earlier than mouse models with single mutations. The APP / PS1 model has a large amount of neuron loss (35% in hippocampus)¹ and learning impairments in both working and spatial memory as measured with Y-maze and MWM at age 4 to 5 months,⁶ making this strain an excellent model to investigate disease pathogenesis. Similarly, the 5xFAD mouse model was created by combining 5 fAD mutations. By age 1.5 months, intraneuronal Aβ42 is present. By age 2 months, high amounts of Aβ42 are found extracellularly and Aβ deposits form. This makes the 5xFAD model important in understanding the effect of Aβ42, however, the ratio of Aβ42 to Aβ40 is greater than that observed in human AD, creating a concern for higher Aβ42 toxicity than occurs in humans. The 5xFAD model is unique in that the plaque formation occurs at age 4 to 5 months, prior to behavioral changes. Both the APP / PS1 and 5xFAD mouse models also exhibit inflammatory responses with activation of astrocytes and microglia, dystrophic neurites, and neuron loss—important features of AD—but still lack NFT, a key component of the disease, NFT.⁷

Although some variation is inherent in these models, the Aβ mouse models can be used to study AD pathogenesis, particularly amyloidosis. The disease cannot be completely understood, however, without also modeling other core pathological features, specifically NFT and neuroinflammation.

Tau Models

The aggregation of intracellular microtubule-associated protein tau (MAPT) causes tauopathy in several neurodegenerative diseases, including AD. There are 6 isoforms of MAPT on chromosome 17, and the primary function of tau includes stabilizing microtubule-associated proteins essential for axon structure and transport.⁸ Transgenic mouse models have been developed in attempts to isolate and investigate tauopathy, in particular tau hyperphosphorylation, which occurs in AD.

Substantial data establish that NFT are composed of hyperphosphorylated tau,⁸ and phosphorylation at specific sites on tau may be an underlying mechanism for attachment anddetachment of tau from microtubules; this provides targets for drug development. The first transgenic mouse model of a tauopathy (Alz17) manifested hyperphosphorylation through altering the largest tau isoform (2N4R; 441 amino acids). The 2N4R isoform is the most favorable substrate for hyperphosphorylation by rodent kinases, however, these mice did not develop true NFT’s. The 8c mouse model overexpresses all 6 tau isoforms, but NFT are still not seen.

A groundbreaking discovery of frontotemporal dementia with parkinsonism linked to chromosome 17 (FTDP-17) supports the hypothesis that mutated tau can cause neurodegeneration by altering both extrinsic and intrinsic MAPT mutations.^9-12 A group of autosomal-dominant diseases that have mutations in MAPT, FTDP-17 leads to neuron loss and gliosis. This discovery initiated a second-generation of transgenic mice strains that mimic human tauopathy more closely. The P301L or JNPL3 mice have a missense mutation associated with FTDP-17,^10,13 and progressively express tau deficits similar to those seen in humans but do not have behavior or motor abnormalities. Data from this mouse model also suggest that tau overexpression is more detrimental than tau aggregation, a topic that is still controversial. The JNPL3 line remains a widely-utilized mouse model to study tauopathies.

The question of whether tau expression or tau aggregation causes the toxicity seen in AD is unanswered. A study using the P301L mouse model genetically suppressed tau expression and allowed progression of tau aggregates.¹⁴ Neuron and cognitive deficits were rescued despite the accumulation of NFTs, suggesting that NFT accumulation is not highly correlated with cognitive and behavioral deficits in this model. Additional studies confirm that tau expression is more detrimental than NFT’s aggregation.¹⁵ Tau mouse models have advanced our understanding of tau biology, as well as our understanding of the role of tau (expression or NFT) in AD pathology.

Recent focus has shifted to studying the tau prion hypothesis that misfolded tau propagates between neurons via cell-to-cell transfer. The P301L tau transgenic model has been used with a specific promoter to induce tau expression in entorhinal hippocampus, suggesting that human tau progressively spreads in a manner similar to prion related disorders.^16,17

Most work using tau mutant models has focused on understanding tau biology and how tau phosphorylation may influence AD onset and progression. Additionally, attempts to arrest or alter tau overexpression or NFT formation with phosphorylation modulators, immunotherapies, therapeutics to dissociate NFT’s, and microtubule stabilizing compounds have emerged and are being evaluated in tau transgenic models.

Models of Neuroinflammation

Neuroinflammation is characterized by persistent microglial activation that results in increased proinflammatory cytokines, aberrant phagocytosis, and microglial aging and death. Early studies described neuroinflammation as a secondary effect of Aβ plaque accumulation and tau hyperphosphorylation. Recent data, however, demonstrate that neuroinflammation exacerbates both Aβ production and tau pathology, suggesting a more central role for neuroinflammation in AD.¹⁸ The discovery that a missense mutation in a microglial receptor (TREM2) increased risk for AD, led to further investigation of microglia and neuroinflammation because microglia are the resident immune cells in the brain.

Most studies of inflammation in AD are conducted in the transgenic rodent models of Aβ or tau pathologies discussed, but a number of inflammation-specific models without associated AD pathology are also being used to better understand mechanisms and potential therapies in AD. Several rodent inflammation LOAD models develop neuroinflammation early in disease pathogenesis and are not genetically manipulated by mutations related to Aβ, tau, or fAD.¹⁹ In immune-challenged, neurotoxin-induced, and transgenic models of neuroinflammation, streptozotocin, polyI:C, and p25 models best represent the chronologic progression seen in LOAD. These models are most fitting for studying how inflammation contributes to AD and potential treatment with anti-inflammatory agents.

Much effort has been directed at investigating microglia. A recent study on prostaglandin E2 type 4 (EP4) receptor-associated protein (EPRAP), a cytoplasmic signaling partner of EP4 present on microglia, demonstrated a role for EP4 / EPRAP in neuroinflammation. Transgenic mice generated by crossing EPRAP^-/- with J20^+/- (APP-based mouse model) had suppressed microglial activation and reduced anxiety-like behavior, but no change in production and accumulation of Aβ.²⁰

Alternative approaches have focused on microglia phenotypes, including the novel dysfunctional microglia type, disease-associated microglia (DAM),²¹ commonly found near or surrounding Aβ plaques in postmortem brain tissue from people with AD and a transgenic mouse model. The APP / PS1 mouse model has been used to study receptor-interacting serine / threonine protein kinase (RIPK1), which is highly expressed in microglia of people with AD and has been shown to be involved inflammation and necroptosis.²² Pharmacologic and genetic inhibition of RIPK1 both reduced Aβ plaque accumulation, levels of inflammatory cytokines, and memory deficits. More importantly, the inhibition stimulated microglial degradation of Aβ in vitro. Further assessment revealed that RIPK1 mediates the induction of cystatin F, leading to endosomal/lysosomal pathway impairment. Together, these findings suggest that RIPK1 promotes microglial dysfunction causing reduced phagocytic activity, increased inflammation, and enhanced progression of AD. Examination of Aβ plaque-associated microglia in postmortem brain tissue from 5XFAD mice, APP23 mice, and people with fAD or LOAD shows high expression of interleukin 1 β and upregulation of phagocytic genes (eg, CD11c, dectin-1, and TREM2) in the plaque-associated interleukin α^+/- microglia after LPS injection.²² In human tissues, however, higher expressions of phagocytic markers and AD-related genes (eg. APOE, AXL, TYROBP, and HLA-DRA) are observed in fAD compared with LOAD. This study demonstrates that elevated levels of phagocytic genes can increase chronic inflammation, microglial dysfunction, and cell death that may enhance the progression of AD.

Several microglial receptors have significant functions in neuroinflammation (eg, TREM2, CX3CR1, and GABA_B receptors). The TREM2 receptor modulates inflammatory responses, facilitating recovery and phagocytosis of damaged neurons, clearance of plaques, and activation of DAM.^21,24 Mutations in TREM2 confer variable degrees of increased risk for developing LOAD. Complete deletion and haplo-insufficiency of TREM2, however, have different effects in AD models vs cell culture.²⁵ It has been proposed that to play a neuroprotective role, microglia must surround plaques to create physical barriers limiting fibril outgrowth and spreading of plaque-associated toxicity.²⁶

The receptor for CX3CL1 (fractalkine, FKN), CX3CR1, is solely expressed on macrophage populations. Studies on FKN signaling have produced contradictory results. In a tau model, CX3CR1 knockout increases tau phosphorylation and aggregation, enhances microglial activation, and worsens learning deficits. In amyloidogenic models, however, CX3CR1 knockout mitigates neuron loss and activation of microglia early in the progression of AD without affecting amyloid production and reduces Aβ deposition in both APP / PS1 and R1.40 models.¹⁸

As AD pathology progresses, these changes appear to be lost with some reports of worsened AD pathology at older ages in mouse AD models. Because CX3CR1 is expressed nearly more than 1000-fold higher in microglia than in peripheral myeloid cells or other CNS cell types, examination of fractalkine signaling in AD pathophysiology is warranted.

GABA plays a central role in learning, memory and immune signal regulation.²⁷ GABAergic signaling is linked to neuroinflammatory responses through GABA_B receptors present on microglia. Activation of GABA_A and GABA_B receptors produces anti-inflammatory responses. When GABA is secreted by astrocytes into the extracellular fluid, inflammatory responses of activated microglia and astrocytes are inhibited and proinflammatory signaling in activated microglia is suppressed. Further, GABA_B receptor activation in microglia and astrocytes reduces proinflammatory cytokines. It is hypothesized that the large amount of GABA secreted by astrocytes is directed at regulating microglial responses. If GABAergic homeostasis is jeopardized, the loss of microglial regulatory mechanisms is expected.^18,27

Conclusion

Several rodent model studies of AD highlight the diversity of mechanisms and targets being evaluated in transgenic AD animals. Each model has advanced our understanding of the underlying biology of core features of AD. Despite tremendous advances, we still lack an ideal model for LOAD. This presents difficulties for developing therapeutics for AD. Most novel therapeutic interventions carried out in clinical trials were initially tested in animal model systems. To advance to clinical investigations, there must be demonstrated benefits in at least 1 nonclinical model. Unfortunately, none of the candidate therapies that showed promise in nonclinical systems has led to success in human AD clinical trials.

There are several possible explanations for the lack of success in clinical trials for any agent that showed promise in animal models of AD. The most common argument is that many animal models are based on genetic mutations seen in fAD, which differs from LOAD. Additional arguments suggest the degree of pathology (severity and time course) seen in many animal models lends itself well to understanding the underlying biology but does not appropriately mimic what is seen in clinical populations with LOAD.

Based on concerns regarding how well existing animal models translate to humans with AD, there has been more effort over the last 5 years to generate and test new mouse models of AD. This work includes generating mice with the CRISPR / CAS9 approach that results in viable genetic models very quickly. Some of the most notable efforts include a collaborative effort between the NIH and Jackson Laboratories for the MODEL AD Center. This program involves identification of mutations in LOAD populations and rapidly translating them into novel mouse models. Mutations of TREM2, a novel ApoE4, and others are used to create models being studied by deep phenotyping including behavioral, imaging, and biochemical studies. The models are then made available to other investigators for further work. An additional effort includes the search for new models that do not rely solely on genetic mutations. Because there are numerous risk factors for AD, including diabetes, hyperglycemia, inflammation, and compromised vasculature, several groups have focused on these risk factors to investigate how they may link to the core pathologies of AD.

Use of animal models has contributed to understanding AD biology and plays an important role in characterizing candidate interventions. As the diversity of mouse models increase, and our understanding of LOAD in clinical populations increases the models will continue to provide a necessary means for understanding and treating AD.