Transgenic amyloid precursor protein mouse models of amyloidosis. Incomplete models for Alzheimer's disease but effective predictors of anti‐amyloid therapies

Abstract

INTRODUCTION

Amyloid precursor protein (APP) transgenic mice are models of Alzheimer's disease (AD) amyloidosis, not all of AD. Diffuse, compacted, and vascular deposits in APP mice mimic those found in AD cases.

METHODS

Most interventional studies in APP mice start treatment early in the process of amyloid deposition, consistent with a prevention treatment regimen. Most clinical trials treat patients with established amyloid deposits in a therapeutic treatment regimen.

RESULTS

The first treatment to reduce amyloid and cognitive impairment in mice was immunotherapy. The APP mouse models not only predicted efficacy, but presaged the vascular leakage called ARIA. The recent immunotherapy clinical trials that removed amyloid and slowed cognitive decline confirms the utility of these early APP models when used in therapeutic designs.

DISCUSSION

New mouse models of AD pathologies will add to the research armamentarium, but the early models have accurately predicted responses to amyloid therapies in humans.

1.

The present definition of Alzheimer's disease (AD) is a dementia that is associated with amyloid plaques, neurofibrillary tangles, and neurodegeneration. ¹ Prior to 1980, the term “Alzheimer's disease” was restricted to dementia onset prior to the age of 65 (presenile dementia), and dementia cases with plaque and tangle pathology exceeding this age were termed “senile dementia of the Alzheimer‐type.” In the mid‐1980s the peptide sequence associated with the amyloid plaques (amyloid β peptide or Aβ ² , ³ ;) and the genes encoding the parent protein (called the amyloid precursor protein; APP ⁴ , ⁵ , ⁶ , ⁷ ;) were identified. Subsequently, genetic studies of families with autosomal dominant Alzheimer's disease identified mutations in both the APP protein and one of its processing enzymes (presenilin; PS) that caused these disorders. ⁸ Early attempts to generate transgenic mice with APP mutations had limited success. ⁹ The first mouse model with consistent deposits of amyloid was initially developed at a biotechnology company, Exemplar, in Worcester MA and subsequently licensed to Athena Neurosciences. ¹⁰ This mouse was referred to as the PDAPP mouse because it used the platelet‐derived growth factor‐β promoter and an APP V717F (Indiana) mutation associated with familial AD cases. Additional APP mouse lines using the KM670/671ML (Swedish) mutation soon followed (Tg2576, APP23, R1.40). ¹¹ , ¹² , ¹³ Mice with only PS mutations had little pathology ¹⁴ but when combined with mice expressing APP mutations (PS1+APP; APP/PS1; R1.40XG9) had a synergistic effect on production of Aβ peptides, especially the longer Aβ42 variant. ¹⁵ , ¹⁶ , ¹⁷ Mice with multiple APP mutations also became available (J20, TASD41; TgCRND8) ¹⁸ , ¹⁹ , ²⁰ including knock‐in mice (APP/PS1 gene targeted). ²¹ A slightly more recent commonly used mouse has five different mutations (5xFAD). ²² Another valuable model also included a tau mutation along with APP and PS transgenes (3xTg). ²³

Although each of these mouse lines have slightly different promoter constructs, isoforms of APP and pathological phenotypes, they shared a number of common features. ²⁴ , ²⁵ , ²⁶ In AD tissue, there are multiple types of Aβ deposits. The one believed most critical is the senile or neuritic plaque which is a dense aggregate (20 to 100 μm) of Aβ and a number of other proteins (including Apolipoprotein E), which are enveloped by a zone of dystrophic neurites, enlarged dendritic/axonal processes containing a variety of synaptic and filamentous proteins, sometimes including tau fibrils. ²⁷ These neuritic plaques have a core stained by β‐pleated sheet binding agents such as Congo red and thioflavin S. Also associated with these neuritic plaques are activated microglial cells in close apposition to the plaque with activated glial fibrillary acidic protein (GFAP) ‐positive astrocytes enveloping this complex proteo‐cellular mass. Most of the same proteins and cells associated with the neuritic plaques in AD cases are also found in the APP mouse models. ¹⁰ , ²⁸ A second form of Aβ deposit is commonly referred to as a “diffuse” deposit or plaque. These are labeled by anti‐Aβ antibodies but not markers of β‐sheet aggregation. They do not induce glial activation and are generally considered innocuous. A third form of Aβ deposit is along meningeal vessels and the arterioles descending into the parenchyma. This material conforms to the cylindrical shape of the vessels, and is stained by Congo red and thioflavin S, and often referred to as cerebral amyloid angiopathy (CAA). In AD, the major length variants of Aβ, Aβ40, and Aβ42, have different distributions in these structures. ²⁹ , ³⁰ The diffuse Aβ is most extensively labeled by antibodies selective for Aβ42 but less so for Aβ40. The neuritic plaques are stained by antibodies specific for both length variants, and the vascular amyloid is stained primarily by antibodies labeling Aβ40 but not Aβ42. Most mouse models share this same distribution of length variants. ²⁸

Another intriguing aspect of Aβ distribution is the regional pattern of deposition. In AD, amyloid deposition is largely in cerebral cortex association areas and hippocampus. Little amyloid staining is found in brainstem or cerebellum. ³¹ The same is true for the APP mouse models. This even extends to the striatum where the only form of Aβ deposition in AD is the diffuse form. ³⁰ , ³² The same holds true for APP mice. ²⁸

Although there are differences among the multiple mouse models of amyloidosis, there is a core of features preserved in most of them and these overlap with the same features found in AD brain post mortem (see Drummond and Wisniewski ²⁴ for a detailed review of these common pathologies in the different mouse lines). While there have been arguments that some APP mice may be “better” than others, it is likely that any intervention which would succeed in AD should be effective in almost all of them.

RESEARCH IN CONTEXT

1. Systematic review: The authors review the early literature regarding the use of mouse models of amyloidosis using traditional methods including PubMed. They further integrate these early findings with the more recent human clinical trial literature.

2. Interpretation: The conclusion is reached that the mouse models of amyloidosis were, in fact, predictive of both successes and failures of amyloid targeting drugs in human clinical trials when results are interpreted in the context of therapeutic study designs compared to prevention study designs.

3. Future directions: The models are incomplete models of Alzheimer's disease in that they lack the development of tauopathy, brain atrophy, and neurodegenerations found in late stages of Alzheimer's disease. Additional models that incorporate these other pathological influences on cognition should produce more complete models for screening drugs other than those targeting amyloid.

The first intervention to modify the establishment of amyloidosis in a mouse model was immunotherapy using a vaccine against the amyloid peptide. ³³ These studies treated mice from 1.5 to 13 months and from 11 to 18 months and found vaccinated mice accumulated less Aβ. This benefit of vaccination was replicated by other groups ³⁴ , ³⁵ , ³⁶ and included demonstration that the cognitive impairments associated with the amyloid phenotype in these mice were prevented by active immunization. All of this work was performed in a prevention treatment regimen, as were the wealth of additional studies confirming the effectiveness of anti‐amyloid immunotherapy in different mouse models with different vaccine variants and later monoclonal antibodies (reviewed in ³⁷ ).

However, unlike most other interventions targeting amyloid in clinical trials, the immunotherapy approach was also tested in therapeutic treatment regimens, initiating treatments in older mice with established amyloidosis. Pfeifer et al. ³⁸ treated APP23 mice at 21 months of age for 5 months with an anti‐Aβ monoclonal antibody. They reported significant reductions of 20%–30% for both detergent insoluble (formic acid soluble) Aβ and immunohistochemical Aβ, which they described as primarily clearance of diffuse Aβ deposits. They further detected microhemorrhages in the antibody‐treated, aged mice. Wilcock et al., ³⁹ examined Tg2576 mice with monoclonal antibody treatments from 18  to 21 months, and a second cohort with treatments starting at 21 months and ending at 27 months. Within 3 months, the antibody reduced neuritic Congophilic plaques by 90%. This was accompanied by a three‐fold elevation in the vascular amyloid deposits, and a time dependent increase in microhemorrhages. However, the treatment still led to reversal of the memory impairments despite the vascular leakage. In both models the amyloid had been accumulating for at least half of the mouse's lifespan up to the time of treatment. In a post mortem study of cases from the AN1792 trial of amyloid vaccination, Boche and Nicoll observed similar changes with reduced parenchymal, increased vascular amyloid deposits and microhemorrhage associated with cases autopsied at various intervals after vaccination. ⁴⁰ In a more recent study, 23 to 24 months old PDAPP mice were treated for 3 months with a murine version of donanemab which lowered Aβ42 levels by 40%–50% by ELISA but had no effect on histological Aβ deposition. ⁴¹ There was no increase in microhemorrhages in this study. The authors also commented that the bulk of deposits in the PDAPP mice were diffuse. Another study supporting the use of immunotherapy in a therapeutic treatment regimen was Aβ vaccination for 10 months in 23 year old vervet monkeys, which showed 65% reductions in insoluble Aβ42 and reduced histological Aβ immunostaining. ⁴² Thus, unlike many other agents designed to prevent amyloid, immunotherapy had ample preclinical evidence supporting its use in a therapeutic treatment regimen. Perhaps most surprising is that the mice also predicted the adverse events common in the human studies, now referred to as amyloid related imaging abnormality (ARIA). ⁴³ A recent study using the PDAPP mouse model found that vessels with CAA in antibody treated mice were associated with activation of perivascular macrophages, infiltration of plasma proteins, and macrophages plus development of hemosiderin indicative of microhemorrhage. ⁴⁴ Another recent study determined that manipulating Aβ clearance through the lymphatic system could also modify the rate of amyloid deposition. ⁴⁵ Thus, it is reasonable to expect that the APP mouse models could be further used to test adjunctive treatments to diminish the risk of ARIA in people treated with anti‐amyloid immunotherapy. They may also be useful in developing treatments for individuals where CAA predominates.

One frequent criticism of the mouse amyloidosis models is the overwhelming number of interventions which modify the mouse amyloid phenotype, ⁴⁶ and apparent failure of these interventions in clinical trials. ⁴⁷ , ⁴⁸ One frequently overlooked nuance in making the comparisons of preclinical and clinical investigations is the stage of disease when the intervention is applied. The vast majority of preclinical studies targeting amyloid in mice have initiated treatment at an age prior to or early in the establishment of amyloidosis in a prevention study design (Figure 1). The reason for this is the practical need to maximize the probability of success in order to publish the results. When treatments are started early, both interventions that block amyloidosis, and interventions that remove established deposits or reduce their effects would be successful. However, if the intervention is delayed until the amyloidosis is well established (a therapeutic study design) only drugs that remove pre‐existing deposits or reduce their down‐stream effects (symptoms) would be successful. As one example, drugs reducing the production of Aβ may be successful in a prevention study but may fail in a therapeutic design because the already established amyloid pathology may still interfere with synaptic plasticity. Some of the first unsuccessful treatment trials directed at amyloid, including tramiprosate, tarenflurbil, and gamma‐secretase inhibitors, were criticized for basing preclinical support upon prevention studies rather than therapeutic approaches, or having little preclinical testing at all. ⁴⁹ , ⁵⁰ , ⁵¹ , ⁵² Aware of this criticism, beta secretase inhibitor (BACE‐I) trials developed considerable amounts of preclinical data. Prevention studies initiating BACE‐I treatments prior to amyloid deposition found dramatic reductions of amyloid plaques, ⁵³ , ⁵⁴ , ⁵⁵ , ⁵⁶ , ⁵⁷ but rarely resulted in prevention of cognitive decline. Preclinical therapeutic studies in older APP mice carefully included a baseline group collected at the start of the BACE‐I and vehicle treatments. Almost uniformly, these studies found some slowing of amyloid accumulation in the BACE‐I treated animals relative to vehicle treated mice, but none found a reduction below the amyloid content of the baseline group. ⁵⁸ , ⁵⁹ , ⁶⁰ , ⁶¹ Surprisingly, none of these therapeutic studies examined whether BACE‐I treatment reversed cognitive deficits in these mouse models. Thus, the relatively extensive preclinical literature predicted that BACE‐I would block initiation of plaques, and slow growth of existing plaques, but would not remove pre‐existing deposits, and provided no support for modifying cognitive changes in people with symptomatic disease (unlike the immunotherapy literature). These predictions are largely consistent with the outcomes in the clinical trials with small reductions in Aβ‐PET signals and no significant benefits on cognition (reviewed in ⁶² ). Of potential interest are preclinical therapeutic studies combining anti‐Aβ immunotherapy with BACE‐I. These studies found that the combination lowered amyloid more effectively than either intervention alone, and the amyloid dropped below the level of the baseline group, indicating removal of existing deposits. ⁶³ , ⁶⁴

FIGURE 1.

Comparison of prevention study design with therapeutic study design and human clinical trials. In preclinical prevention designs, interventions are administered either prior to the onset of amyloid deposition or in the early phases. Mice are often tested behaviorally at a time point shortly after memory deficits are apparent. Tissues are collected shortly after behavioral testing. In a therapeutic treatment design, mice are matured to an age when amyloid deposits are well established before treatment starts. Mice are then tested behaviorally and tissues collected. In a clinical trial, both amyloid and tau have been present for years, and participants are tested for memory loss and presence of pathology before interventions are administered. After 1 to 2 years participants are re‐tested and pathology and biomarkers measured in placebo and treatment groups. No tissue is collected until later at autopsy. Very few drugs have been tested in mouse therapeutic trials before human clinical trials. Created with Biorender.

In hindsight, it is not surprising that drugs designed to slow Aβ42 production (β and γ‐ secretase inhibitors) were not effective when applied in a therapeutic treatment study design. They can block amyloid accumulation, but not remove that which already exists. Immunotherapy was demonstrated to remove established amyloid deposits in multiple preclinical amyloidosis models, predicting the positive outcomes seen in the recent second generation antibody trials. ⁶⁵ , ⁶⁶ , ⁶⁷ Moreover, the mouse amyloidosis models presaged the increased microhemorrhage and increased vascular amyloid deposition detected post mortem in cases from the active immunization AN1792 study ⁶⁸ and later by the MRI studies demonstrating ARIA in cases treated with monoclonal antibodies. ⁴³

In summary, the first generation of mouse amyloidosis models have several advantages over other screening methods for developing anti‐amyloid medications. They result in a variety of amyloid deposits that are similar to those found in humans both histologically and chemically plus they populate the same brain regions. Generally, drugs that either reduce amyloid production or remove pre‐existing amyloid in mice have similar effects in humans. Moreover, when tested in therapeutic experimental designs, the models are predictive of outcomes in clinical trials in symptomatic cases of AD. The disadvantages are that they model only a portion of the core pathologies of AD, and generally lack the tauopathy that appears required for neurodegeneration and brain atrophy. Most of the models also include overexpression of Aβ, which might mask potentially deleterious effects of excessive Aβ reduction (for example by BACE inhibitors). By referring to them as Alzheimer's mice, this misleads some to conclude they should predict all phases of the disease, which they do not.

Increasingly it is realized that there are myriad causes of cognitive decline with age, and even summing all the pathologies found post mortem, at most half the variance in rate of decline can be accounted for by pathology. ⁶⁹ Much of the criticism of the mouse models of amyloidosis derives from the misclassification of them as models of the entire disease, plus our lack of understanding what role amyloidosis plays in the development of age‐related cognitive decline. The recent clinical studies with immunotherapy, two decades past the initial observations in mice, are giving us our first glimpses into what amyloid's role might be and how effective reducing existing amyloid will be in the disease process.

There are certainly enormous opportunities to improve these animal models. One example is that typically mouse models are evaluated with terminal biomarkers at the conclusion of the study (except for behavior), while in humans all biomarkers are collected during the treatment and, often, also at baseline before the study starts. Opportunities for greater homology exist. One study, using a transgenic rat model (McGill‐Thy1‐APP rat), studied an Aβ oligomer binding fusion protein using pre and post treatment PET scans, CSF measures of Aβ42/40 and neurofilament light, volumetric MRI, and resting state fMRI followed by post mortem histochemistry. ⁷⁰ This parallels similar measures used in human trials and should translate more directly to clinical trial outcomes. Behavioral tests in humans and mouse models are also quite distinct, although Hort developed spatial navigation memory tests in humans, analogous to the Morris maze. ⁷¹ It is unlikely that mice will ever be performing tests of verbal learning or clock drawing, but we can attempt to find some nonverbal tasks that could be performed by both mice and humans such as delayed match to sample and other touchscreen types of tests, ⁷² with the consideration that caloric restriction, a requirement for most of these tests, has been demonstrated to modify amyloidosis directly. ⁷³ Additionally, the models can incorporate more human genes, such as immune proteins and other variants that confer AD risk (e.g., TREM2; apolipoprotein E4). Such models are currently under development in a number of different laboratories around the world including the NIH‐funded MODEL‐AD consortium.

Clearly, there will be a host of medications to slow the pace of memory loss with age, and other models of cognitive aging and other disease pathologies will be needed to test these agents. However, the existing models of amyloidosis have served us well, and predicted those agents that can be used to edify the role of amyloid in AD. They may be wrong for a complete understanding of AD pathological processes, but they proved very useful for understanding amyloidosis and justifying the first disease modifying therapies.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. Author disclosures are available in the supporting information. DGM is a member of the Scientific Advisory Board of Synaps Dx and MindImmune. He also receives research support from Hesperos and Bright MindsBiosciences.

Supporting information

Supporting Information

Morgan DG, Lamb BT. Transgenic amyloid precursor protein mouse models of amyloidosis. Incomplete models for Alzheimer's disease but effective predictors of anti‐amyloid therapies. Alzheimer's Dement. 2024;20:1459–1464. 10.1002/alz.13566