Recent Advances in the Modeling of Alzheimer’s Disease

1The deposition of amyloid β peptide (Aβ) in the brain is the major pathological hallmark of Alzheimer’s disease (AD), which is considered the most common type of dementia in the world. To date, disease-associated mutations in the presenilin 1 (PSEN1) and presenilin 2 (PSEN2) genes number more than 300, while more than 50 mutations have been reported in the amyloid precursor protein (APP) gene (Alzforum1). In response to these findings, many transgenic mouse models overexpressing mutant APP or APP/PSEN1 cDNAs have been developed (1st generation models), however they often suffer from experimental limitations resulting from the mislocalization of APP (Figure 1) and by the overproduction of APP fragments such as the C-terminal fragment of APP generated by β-secretase (CTF-β) and APP intracellular domain (AICD). Neither of these fragments appears to accumulate in AD brains, meaning that artificial endosomal abnormalities and transcriptional malfunctions, respectively, may be induced. Other overexpression artifacts include calpain activation, calpastatin deficiency-induced early lethality, and endoplasmic reticulum stresses. Furthermore, it was demonstrated that the random insertion of transgene(s) resulted in the destruction of unexpectedly large regions of endogenous gene loci in the host animal. We suggest that all transgenic mouse models being used in research in which APP or APP/PSEN1 are overexpressed should undergo whole genome sequencing (WGS) so that destroyed loci that possibly affect their phenotypes can be identified.

2To overcome these drawbacks, single App knock-in mice, i.e., AppNL–G–F/NLG–F knock-in (AppNL–G–F) and AppNL–F/NL–F knock-in (AppNL–F) lines, were generated that harbor the Swedish (KM670/671NL) and Beyreuther/Iberian (I716F) mutations with or without the Arctic (E693G) mutation (2nd generation models) (Figure 2). These mice, which exhibit typical Aβ pathology, neuroinflammation and memory impairment, are being used in more than 500 research laboratories world-wide. At present, the AppNL–G–F line is being used more frequently than the AppNL–F line given that it develops Aβ pathology approximately three times faster and can be used to analyze downstream events such as neuroinflammation, pericyte signaling, oxidative stress, tau propagation, and spatial memory impairment (;;; Table 1). Human Arctic mutation carriers are indistinguishable from other familial and sporadic AD patients in pathological and neurological terms except for low retention of 11C-labeled Pittsburgh compound B (PiB) in PET study, indicating that the mutant mice are relevant models for studying AD in general.

3In addition to AppNL–F and AppNL–G–F models, App knock-in mice devoid of the Swedish mutations (AppG–F mice) have been recently developed, in which the Swedish mutations (NL) were replaced by a wild-type sequence (KM) (Figure 3 and Table 2). The AppG–F mice are more suitable for preclinical studies of β-secretase inhibition given that the Swedish mutation affects the reactivity of APP to β-site amyloid precursor protein cleaving enzyme 1 (BACE1) and most AD patients do not carry Swedish mutations.

4Despite the advantages mentioned above, the AppNL–G–F line is not suitable for investigating the metabolism, clearance or deposition of Aβ because the Arctic mutation present in the middle of the Aβ sequence results in an Aβ that is resistant to proteolytic degradation and susceptible to aggregation. Moreover, this model is not suitable for use in preclinical immunotherapy studies due to its affinity for anti-Aβ antibodies, even in the presence of guanidine hydrochloride (GuHCl). The Arctic mutation may also directly or indirectly interfere with interactions between Aβ deposition and the apolipoprotein E genotype, although there is no experimental evidence for this. In contrast, the AppNL–F line accumulates wild-type human Aβ, but it may take up to 18 months for the pathology to become sufficiently evident for investigational studies to be carried out, which is too long for researchers to wait in a practical sense. Therefore, a new mouse model that accumulates wild-type human Aβ as quickly as the AppNL–G–F model, but did not depend on the presence of the Arctic mutation was desired.

5To achieve this, the heterozygous Psen1P117L/WT mutant line (Psen1P117L) which, of the several Psen1 mutants, exhibits the largest increase in Aβ42/Aβ40 ratio in the brain was utilized. The Psen1P117L line was generated by base editing technology. The AppNL–F mice were crossed with Psen1P117L mice, despite it being unclear whether their pathogenic effects, both of which act on the γ-cleavage of CTF-β, would be additive or not in vivo (Figure 4). The pathological phenotypes of AppNL–F mice were markedly enhanced in a synergistic manner with the Psen1P117L mutation, with AppNL–F X Psen1P117L/WT mice showing a more aggressive cored plaque pathology and neuroinflammation than the AppNL–G–F mice (Figure 5). These double mutant mice (3rd generation model) will likely become highly relevant tools for examining the pathologic mechanisms upstream of Aβ deposition. Moreover, these mice can be highly useful for the preclinical screening of disease-modifying therapy candidates promoting Aβ degradation or disaggregation, without the added concern associated with artificial effects caused by the Arctic mutation. We expect the double homozygous line, AppNL–F X Psen1P117L/P117L, to exhibit an even more-aggressive pathology. In any case, the AppNL–F X Psen1P117L mice are superior to the AppNL–G–F mice or the AppNL–G–I mice for universal and unbiased drug screening particularly because the Aβ-degrading enzyme, neprilysin (NEP: Neutral endopeptidase), has become a therapeutic target. The AppNL–G–I mice are similarly designed as the AppNL–G–F mice, in which the Beyreuther/Iberian mutation was replaced by the Austrian mutation. The characteristics of the App knock-in mouse lines are listed in Table 3.

6There are several precautions to be aware of to make the best use of the App knock-in mice.

7A number of the App knock-in mouse users use incorrect nomenclature such as APP-NLF, APPNLF and APPNL–F instead of the AppNL–F mice, which accords with international rules of standard genomic nomenclature. Genetic names always need to be italicized.

8Approximately 20 lines of mutant mice, published or unpublished, can currently or in the very near future be provided to academic and not-for-profit researchers for non-commercial research in a timely fashion with minimum restrictions (Table 2;; ). It is also recommended that scientists use optimized protocols for isolating Aβ from animal brain as previously described (; Figure 6). The method allows the most sensitive quantification of both soluble and insoluble Aβ with the smallest protocol deviations.

9In most cases, the knock-in mice are used in a homozygous state to accelerate the generation of pathological and behavioral phenotypes. However, the number of recessive mutations increases over time if the breeding of mice is maintained in this way. It is necessary therefore to back-cross heterozygous mutant mice with the wild-type C57B6/J mice to remove these extraneous mutations, preferentially for 5–10 generations at an interval of 5–10 generations. Those groups dealing with poor reproductive output of mice due to their extremely inbred nature can contact RIKEN BioResource Research Center (email: animal.brc@riken.jp), a national mouse repository of Japan, to renew their strains.

10It was previously indicated that some anti-Aβ antibodies are inappropriate for biochemically and pathologically detecting the Arctic Aβ produced by AppNL–G–F mice (; Figure 7). Note that the AppNL–G–F is the most frequently used model because it recapitulates Aβ pathology and neuroinflammation much faster than other lines. This applies to all other transgenic and knock-in mice that carry intra-Aβ mutations, including the Arctic and Dutch mutations. In contrast, there is no restriction regarding use of antibodies for the AppNL–F X Psen1P117L mice.

11In our experience, the most sensitive and reproducible test involves contextual fear conditioning, although its irreversible nature can be problematic. In a more complex sense, the application of a multi-task paradigm such as IntelliCage (NewBehavior AG, Zurich, Switzerland) would be more informative. We recently showed that the mouse genotypes can be predicted from their behavioral parameters by machine learning. It should be emphasized that the App knock-in mice are models of preclinical AD because the App knock-in mice, like all the APP and APP/PS1 transgenic mice, do not recapitulate tau pathology alone. Consistently, we observe only mild cognitive decline in these mice. In contrast, the App knock-in mice crossbred with mutant MAPT knock-in mice exhibited accelerated tau pathology (Table 2).

12Most AD patients exhibit parenchymal and vascular Aβ deposition in the brains, and both pathologies seem to be driven by impaired Aβ clearance within the interstitial fluid and perivascular drainage pathways. injected fluorescent tracers into Tie2-GFP:NG2-DsRed double reporter mice, which express GFP in all cerebral blood vessels and DsRed in perivascular cells, and successfully observed glymphatic pathway; subarachnoid CSF influx into the brain parenchyma and bulk ISF solute clearance from the parenchyma within the perivascular spaces. Importantly, in AD model mice, glymphatic CSF influx is reduced and the clearance of Aβ is severely impaired. Impaired glymphatic pathway may contribute to the deposition of Aβ in the blood vessels of the brain, cerebral amyloid angiopathy (CAA). Although CAA is profoundly observed in most AD patients, limited model mice, such as those with Dutch/Iowa mutation, exhibit apparent CAA, thus making it difficult to determine how CAA contributes to the pathogenesis of sporadic AD. Notably, human vascular endothelial cells express significant level of APP770 and human plasma contains ∼100 ng/ml of sAPP770. Since peripheral blood cells other than platelet do not express APP, and platelets release sAPP770 upon their activation, it is considered that plasma sAPP770 is mostly derived from endothelial APP770. Because in rodents plasma sAPP is a markedly lower (∼100 pg/ml) than that of humans, it’s possible that low level of endothelial APP expression in mice could be one of the reasons that App knock-in mice exhibit mild CAA pathology. To overcome this, a mouse line that specifically expresses human APP770 in endothelial cells has just been generated (unpublished). In brief, floxed hAPP770NL mice under the CMV early enhancer/chicken β-actin promoter were first generated. These mice were then crossed with Tie2-Cre mice, in which the Tie2 promoter directs the expression of Cre recombinase in the endothelial cells to obtain double transgenic (Tg) mice.

13App knock-in mouse models were previously produced by who used multiple pathogenic mutations. These mice carry the Swedish (K670N/M671L), Dutch (E693Q), and London (V717I) mutations with the humanized Aβ sequence. The Dutch mutation results in an intensive CAA pathology in humans, thereby causing brain hemorrhage and early mortality. This mutation is therefore not specifically responsible for causing FAD. These mice did not develop prominent Aβ deposits over their lifespan, but when they were crossbred with Psen1M146V knock-in mice, an age-dependent deposition of Aβ was seen in the resultant double knock-in mice. The deposition of Aβ was detected not only in the parenchyma of the cerebral cortex but also in the cerebral vasculature, similar to that seen in CAA in humans. Double knock-in mice that did not have the Dutch mutation exhibited virtually no vascular pathology. In this way, if the authors had used the Beyreuther/Iberian or Austrian mutation instead of the London mutation in the mouse App gene then they probably would not have had to introduce the Psen1 knock-in mice. Knock-in mice harboring the Dutch mutation can still serve as relevant models for CAA; however they may not be appropriate for examining the effect of immunotherapy on CAA because the Dutch mutation is present in the middle of the Aβ sequence.

14NEP and insulin-degrading enzyme (IDE) are considered the two major catabolic enzymes that degrade Aβ. NEP is capable of degrading both soluble and insoluble Aβ, but it is not clearly shown whether endogenous IDE could degrade insoluble Aβ in the mouse brains, rather IDE appears to be involved in metabolism of AICD. Unfortunately, their roles in Aβ metabolism in vivo have never been compared in an impartial and side-by-side manner. Once double mutants crossbred single App knock-in mice with NEP (Mme) KO mice and with IDE (Ide) KO mice were analyzed in detail for their biochemical properties and Aβ pathology properties, it would be clear their distinct roles in APP metabolism and the AD pathogenesis.

15Further to the above, a deficiency of NEP had no significant impact on the levels of various neuropeptides as well as enkephalins that are well known to be in vitro substrates for NEP in the cerebral cortex and hippocampal formation of mice, presumably because NEP is mainly expressed in secretory vesicles and on the presynaptic membranes of excitatory neurons, while most if not all neuropeptides are secreted from inhibitory neurons. This argues against the concern that NEP up-regulation for the treatment of preclinical AD would reduce the levels of these neuropeptides. These findings also indicate that NEP relatively selectively degrades Aβ in the brain. Whereas familial AD (FAD) is unambiguously caused by an increased anabolism of Aβ, and of Aβ42 and Aβ43 in particular, the anabolism of Aβ appears unaffected prior to its deposition in the brain that subsequently leads to the onset of sporadic AD (SAD). These observations suggest that NEP-sensitive amyloidogenic Aβ likely plays a primary pathogenic role in the etiology of SAD. Our findings are consistent with the aging-dependent decline of NEP expression in the human brain and with recent genome-wide association studies (GWAS) indicating that variants of the gene encoding NEP (MME) are associated with the risk of SAD development. Taken together, our results imply that the aging-associated decrease in NEP expression is a primary cause of SAD and could thus be a target for the treatment of preclinical AD once other factors such as apolipoprotein E genotypes have also been considered.

16Since NEP is a major Aβ-degrading enzyme and it is downregulated upon aging, its decreased levels in the brain will most likely lead to increased Aβ levels. NEP is regulated by the neuropeptide somatostatin. Somatostatin, which binds to somatostatin receptors, is also decreased upon aging and in AD possibly due to loss of somatostatin-positive interneurons. Somatostatin, which was first identified to regulate secretion of growth hormone from pituitary, has been traditionally abbreviated as SRIF (somatotropin release-inhibiting factor), so we will keep to this nomenclature in this review. SST1, SST2, SST3, SST4, and SST5 are used to express somatostatin receptor subtypes 1–5. Interestingly, mutations in SRIF are linked to AD. By using a combination of in vitro and in vivo approaches to identify the subtype specificity of the five SSTs expressed in the brain and considered to be involved in the regulation of NEP. We would like to emphasize that it is necessary to use a co-culture system of primary neurons from the cortex, hippocampus, and striatum for in vitro experiments.

17Using a battery of Sst double knockout (dKO) mice, we found that NEP is regulated by SST1 and SST4 in a redundant manner. Sst1 and Sst4 dKO mice exhibit a specific decrease of presynaptic NEP in the Lacunosum molecular layer. Moreover, a genetic deficiency of Sst1 and Sst4 in the App knock-in mice aggravated the Aβ pathology in the hippocampus. As a first proof of concept towards an Aβ-lowering strategy involving NEP, a treatment with an agonist selective for SST1 and SST4 could ameliorates the Aβ pathology and improves cognitive outcomes in the App knock-in AD mouse model as schematized in Figure 8. These results indicate that a combination of SST1 and SST4 homodimers or the SST1 and SST4 heterodimer may become a target for pharmaceutical intervention to treat preclinical AD. Of note, the expression of SRIF in human brain declines with aging and in AD and may causally contribute to AD pathogenesis via reduction of NEP activity/expression.

18Although SRIF is known to regulate Aβ catabolism by enhancing NEP-catalyzed proteolytic degradation, the mechanism by which SRIF actually regulates NEP activity is yet to be fully elucidated. Proteomic analyses enabled α-endosulfine (ENSA), an endogenous ligand of the ATP-sensitive potassium (KATP) channel, to be identified as a negative regulator of NEP downstream of SRIF signaling. The expression of ENSA is significantly increased in AD mouse models and in patients with AD. In addition, NEP directly contributes to the degradation of ENSA, suggesting a substrate-dependent feedback loop regulating NEP activity.

19It was also discovered the specific KATP channel subtype [sulfonylurea receptor subunit 1 (SUR1) and inwardly rectifying K+ channel 6.2 (Kir6.2)] that modulates NEP activity, resulting in altered Aβ levels in the brain. Pharmacological intervention targeting this particular KATP channel by diazoxide attenuated Aβ deposition, with impaired memory function rescued via the NEP activation in our AD mouse model. These findings provide a mechanism explaining the molecular link between KATP channels and NEP activation. They also provide new insights into how ENSA and the KATP channel could profile as a new therapeutic target for lowering Aβ and thus provide an alternative strategy to prevent AD. Figure 9 summarizes the NEP activation mechanism that involves SRIF receptor subtypes, ENSA and KATP channel involvement.

20To date, most if not all, mouse models of tauopathy have been unable to recapitulate the tau pathology without overexpressing mutant human tau protein. As a novel in vivo platform for studying human tauopathy, human MAPT knock-in mice have been developed in which the entire Mapt gene including all exons and introns are humanized. In each strain, the MAPT and Mapt genes encoded human and murine tau proteins, respectively. This was done by crossing MAPT knock-in mice with single App knock-in mice in order to study the role of the Aβ-tau axis in the etiology of AD. The double knock-in mice exhibited a more pronounced tau phosphorylation status than single MAPT knock-in mice but lacked evidence of tau pathology and neurodegeneration (in a manner similar to that of single App knock-in mice) even after waiting until the mice were 24 months old.

21In both the absence and presence of Aβ amyloidosis, the tau humanization has been found to significantly accelerate the propagation of AD brain-derived pathological tau (Figure 10; ). Tau accumulation was intensified in the latter case and closely associated with dystrophic neurites, consistently showing that Aβ amyloidosis affects tau pathology. These results indicated that pathological human tau interacted better with human tau than with murine tau, and suggest the presence of a species-defined preference between the pathogenic proteins. The MAPT knock-in mice also facilitate the investigation of behavioral properties and of human tau characteristics in living animal models. In addition, mutant MAPT knock-in mice carrying various pathogenic mutations have been generated (Table 2). These mice exhibit aging-dependent tau aggregation and cognitive impairment in a manner accelerated by Aβ pathology and are being provided to the research community upon request.

22Pathological tau causes synaptic dysfunction and loss of synapses. One of promising molecules that mediates tau pathology-induced neurotoxicity is N-methyl-D-aspartate receptor (NMDAR). For example, tau accumulation disturbs synaptic plasticity through JAK2/STAT1-mediated suppression of NMDAR. Phosphorylation of tau at Tyr18, which is mediated by the tyrosine kinase Fyn, enhances NMDAR-dependent excitotoxicity. Several reports indicated that glutamate-induced excitotoxicity was prevented by downmodulation of tau. We also previously identified a NMDAR-related molecule as a tau binding protein which is involved in tau pathology- induced neurodegeneration.

23To elucidate key molecules underlying tau accumulation-induced neurodegeneration, a comprehensive screening of tau-interacting proteins (tau interactome) was constructed. Tau-binding proteins were isolated by immunoprecipitation-LC-MS/MS (IP-MS) using a Flag-tag antibody and wild-type tau Tg (wtau-Tg) mice, which is expressing human tau tagged with a flag epitope. Considering that tau is a microtubule-binding protein, we validated the methods used to generate the tau interactome by identifying the tubulin beta-4A chain as one of the tau-binding proteins.

24Of the many proteins identified in the tau interactome, we focused on carboxy-terminal PDZ ligand of neuronal nitric oxide synthase (CAPON), which is an adaptor protein of neuronal nitric oxide synthase (nNOS). CAPON acts as an enzyme for the production of nitric oxide (NO) and is involved in NMDAR-mediated excitotoxicity. It is thought to recruit substrates to nNOS and facilitate their NO-mediated modification through protein-protein interactions. The presence of CAPON polymorphisms associated with schizophrenia and other psychiatric disorders has been reported in several studies. Moreover, CAPON was shown to positively regulate spine density and to regulate neuronal cell death downstream of the NMDAR. These findings suggest that CAPON contributes to neurotransmission and neuronal excitotoxicity. In addition, one report showed that CAPON is upregulated in CA1 pyramidal cells in the AD brain, implying that CAPON may play an important role in the pathogenesis of AD. The mechanism(s) underlying these effects nevertheless remain(s) unknown.

25To further elucidate the effects of CAPON on AD pathology, we introduced CAPON cDNA into the brains of AppNL–G–F and AppNL–G–F X MAPT (hTau) double-KI mice using a newly developed adeno-associated virus (AAV)-mediated approach. We analyzed the effects of human tau protein as it is known that the hTau-KI mouse expresses an endogenous level of WT human tau. These experiments revealed that CAPON expression facilitates hippocampal atrophy in conjunction with neuronal cell death, and that a deficiency of CAPON in the P301S-Tau-Tg tauopathy mouse model suppressed tau pathology and neurodegeneration (Figure 11). From our results, an intervention in the interaction between CAPON-tau or CAPON-nNOS could be a new approach for the treatment of AD.

26Common marmosets (marmosets, Callithrix jacchus) are small non-human primates that belong to the New World Primates (Figure 12; ). They have been increasingly utilized in neuroscience because of advantages that were observed over other research primates. Marmosets possess physiological functions, brain structures and complex cognitive/social behaviors similar to those of humans; they communicate mainly via visual and auditory measures. In association with AD research, the amino acid sequence of Aβ in marmosets is identical to that of humans, with aged wild-type marmosets starting to accumulate Aβ from 7 years of age or even earlier. In addition, adolescent marmosets exhibit tau hyperphosphorylation, but not neurofibrillary tangle formation, in the brain that increases with aging. Their life spans in captivity are as long as 10–15 years, making them suitable for age-related research. Their immune systems and metabolic functions resemble those of humans and thus may affect the pathogenic processes related to AD. Because sleep disorder is an early clinical symptom of AD, it is noteworthy that marmosets share with humans the sleep phases composed of rapid eye movement (REM) and non-REM cycles. Among various non-human primate species, the marmoset seems most applicable to genetic manipulation, i.e., generation of designed mutants, for which their high reproductive efficacy is advantageous. Furthermore, fecundity characteristics of marmosets, such as a short period of sexual maturity, multiple births, and short gestation interval, are suitable for producing genetically modified disease models.

27Majority of FAD-causing mutations reside in the PSEN1 gene. Typically, deletion mutations in exon 9 or point mutations at the 3′ splice site (acceptor site) of exon 9 in the PSEN1 gene cause dominantly inherited FAD. The point mutations instigate exon 9 elimination and S290C modification in the corresponding mRNA at the junction sites of exons 8 and 10 via the conversion of alternative splicing. Thus, generation of a marmoset model of AD is set out in which exon 9 of the PSEN1 gene product is deleted using gene-editing technologies to produce AD marmoset models. Since TALEN exhibited high genome-editing efficacy, generates few off-target effects, and produces little mosaicism, the TALEN would be a suitable tool for producing exon 9 deletion in the PSEN1 gene. Although it is a non-peer review data, the exon 9 deletion in the PSEN1 gene that is an AD causing mutation has been successfully introduced into non-human primates by TALEN.

28All authors listed have made a substantial, direct, and intellectual contribution to the work, and approved it for publication.

29The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

30All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.