A paper in the September 21 Journal of Neuroscience suggests that two prominent phenotypes considered relevant to Alzheimer’s disease are artifacts caused by overexpression of the amyloid precursor protein (APP). Scientists led by Takaomi Saido, RIKEN Brain Science Institute, Wako, Japan, found neither elevation of the CDK5 activator p25 nor reduction in sodium channels in a knock-in mouse model expressing human APP at endogenous levels. They surmise that these phenotypes reflect too much APP, not true AD-related pathogenic processes. Other researchers were skeptical. They said the authors were overreaching in their conclusions to the current set of experiments, but were optimistic about future insights from knock-in models.

To mimic Aβ pathology in the mouse brain, many research groups have explored dozens of models that overexpress mutant human variants of APP or presenilins (see Alzforum Research Models database). However, the field has long recognized that driving up gene expression beyond physiological levels can introduce artifacts, for example by disrupting other genes near the site of transgene insertion or by overwhelming protein homeostasis, triggering apoptosis (Kuang et al., 2006). 

Researchers have struggled to generate human APP knock-in (KI) models that would avoid such pitfalls. After more than a decade of trying, Saido’s lab finally succeeded, generating APPNL-F and APPNL-G-F models carrying Swedish/Iberian and Swedish/Arctic/Iberian mutations, respectively (see Apr 2014 webinar). The APPNL-F mice used in this study overproduce only Aβ, not APP, at an increased Aβ42:Aβ40 ratio, developing plaques and gliosis at six months, and cognitive impairments at 18. 

Overexpression Artifact?

APP23 mice produce an abundance of p25, but APP knock-in mice do not. [Courtesy of Saito, et al. The Journal of Neuroscience 2016.]

About 200 scientists currently work with Saido’s KI models, and they have begun to publish their findings. Early indications suggest that mushroom spines wane in synapses as these animals age and that the orphan G protein-coupled receptor GPR3 promotes amyloidogenic processing of APP (Oct 2015 newsHuang et al., 2015). In particular, some researchers have begun checking whether the KI mice recapitulate phenotypes previously published for overexpression models. However, many researchers contacted by Alzforum said their mouse colonies are not yet old enough to examine AD-related pathology or behavior. Still other groups are waiting to receive the mice. Some bemoaned a complicated and time-consuming process to obtain the animals, available only through the RIKEN BioResource Center. In multiple comments on Alzforum, Saido has encouraged the field to use his lab’s mice to validate phenotypes published in overexpression models.

For the current study, first author Takashi Saito and colleagues examined two phenotypes previously described in some overexpression models: downregulation of the sodium channel Nav1.1 in the interneurons prompting seizures; and calcium/calpain dysfunction leading to generation of p25, which activates CDK5 to hyperphosphorylate tau, ultimately triggering synaptic and memory deficits (Oakley et al., 2006; Seo et al., 2014). 

To test for the latter phenotype, Saito and colleagues compared p25 levels in homogenates of the neocortices and hippocampi from APPNL-F mice to similar preparations from APP23 mice, which overexpress APP. The knock-ins had wild-type levels of p25, while APP23 models had more than twice as much (see image). Even when Saito crossed APPNL-F mice with calpastatin-deficient animals (Cast KO), p25 levels in the offspring were normal (Takano et al., 2005). Calpastatin, the only endogenous inhibitor of calpain, normally suppresses p25 production. Saido concluded that even in Cast KO/APP knock-in mice, Aβ amyloidosis on its own fails to elevate intraneuronal calcium enough to activate calpain and trigger production of p25. Thus, Saido argues that p25 production in APP23 mice results from APP overexpression.

Other scientists were unconvinced. They noted that the conclusions hinge on one faint western blot (Figure 1) and said the experiment lacked appropriate controls, such as wild-type x Cast KO crosses. Li-Huei Tsai from Massachusetts Institute of Technology questioned if Saito and colleagues accurately measured p25. She noted that they previously reported a rise in p25 in wild-type calpastatin knockouts, and wondered why they did not see it in the APPNL-F x Calp-/- offspring (Sato et al., 2011). Saido responded that those previous results were from in vitro and ex vivo experiments that are less relevant. Tsai said data from other tissue and cell culture models indicate, as well, that Aβ generates excess p25 without APP overexpression (Zheng et al., 2005; Seo et al., 2014). 

That calpain may be overactive in AD brains fits with the idea that p25 contributes to pathology, noted Tsai (see Kurbatskaya et al., 2016). The calcium-dependent protease was therapeutically targeted for Alzheimer’s when the North Chicago-based pharmaceutical company AbbVie tested the inhibitor ABT-957 in a Phase 1 trial. AbbVie has halted development of the drug. A note on clinicaltrials.gov states the compound insufficiently engaged the target. An AbbVie representative declined to elaborate.

Saito and colleagues also used western analysis to quantify Nav1.1 sodium channels. Levels in wild-type, APPNL-F, APPNL-F x Cast KO, and in APP23 were all similar and higher than in J20 mice, which overexpress APP (Verret et al., 2012). These mice begin to lose hippocampal neurons by 12 weeks of age, which is not commonly seen in APP overexpression models. Saido and colleagues took their data to mean that Nav1.1 deficiency is unique to the J20 model, and therefore an artifact of transgene expression. However, APP23 mice also overexpress APP.

Again, others were not so sure. Tsai noted that because the reduction in sodium channel expression was previously reported to be specific to interneuron subpopulations, it would be hard to detect by western blot of crude lysates (see full comment below). Saido countered that Nav1.1-positive neurons are sufficiently abundant in the hippocampus and neocortex to measure changes in the sodium channel expression with the methodology used.

Ralph Nixon at the New York University School of Medicine hypothesized that APPNL-F mice develop too mild a phenotype to see a drop in sodium channels, or a rise in p25.

Should researchers be concerned about years of data from overexpression models? Given that APPNL-F x Calp-/- mice only reproduced two of five phenotypes observed in APP23 x Calp-/- mice, i.e., amyloidosis and neuroinflammation, Saido and colleagues estimate that 60 percent of the phenotypes in APP-overexpressing mice generally could be artifacts (Saito et al., 2014). This would call into question findings from more than 3,000 papers, they wrote.

Other scientists doubt that such sweeping conclusions can be drawn just yet; after all, only two papers on the knock-ins have been published thus far on this particular topic. “It is quite a leap from reporting ‘only two of five phenotypes studied were reproduced’ to concluding that 60 percent of all phenotypes in thousands of publications are artifacts,” Mike Sasner at Jackson Laboratory in Bar Harbor, Maine, wrote to Alzforum.

Sasner said that APP knock-in models do avoid certain problems of transgenic animals, but that they have their own drawbacks. Because knock-ins express APP at near-endogenous levels, they have late-developing, mild phenotypes that make them impractical for many studies. However, he agreed that traditional APP overexpression models are not ideal for testing potential AD therapies.

Tsai pointed out that a duplication of the whole APP gene causes rare familial forms of AD, and that people with Down’s syndrome, who carry an extra copy of chromosome 21 that harbors the gene, also develop the disease. She stressed that APP overexpression is not necessarily an artificial way to model Alzheimer’s.

In addition, Nixon questioned the validity of putting multiple familial AD mutations into a single animal, as was done in both these knock-ins and some of the overexpression models. There is no indication the Swedish, Iberian, and Arctic mutations ever crop up in the same person. While Nixon agreed that overexpression causes problems, he questioned if the knock-ins are a better option. “The ultimate mouse model of Alzheimer’s disease would be the simplest and least artificial, while recapitulating the full spectrum of neuropathology,” he told Alzforum. “We still don’t have that.”

All in all, many commentators, including some who declined to be quoted, echoed concerns about protein overexpression, and agreed that the knock-in mice represent a major advance. Saido will co-chair a symposium on the animals at the Society for Neuroscience annual meeting November 12-16 in San Diego.—Gwyneth Dickey Zakaib 


  1. This paper from Takaomi Saido and colleagues states that 60 percent of the phenotypes observed in the traditional AD mouse models are merely “artifacts” based on the absence of these phenotypes in their APP knock-in mouse models carrying two or three distinct APP mutations (Saito et al., 2014). In one of these lines, they claimed that calpain activation, p25 production, and Nav1.1 reduction are all artifacts of APP and PS1 overexpression. Careful examination of the data, however, shows that this conclusion is not strongly supported by the presented, very limited data, which suffers from absence of controls and the lack of a comprehensive analysis needed to draw the sweeping conclusion that calpain activation is an “artifact” of transgene overexpression. The specific points that should be considered are highlighted below:

    1.     APP duplication is known to cause familial Alzheimer’s disease. Similarly, individuals with Down’s syndrome caused by trisomy of chromosome 21 (which contains the APP gene) also develop early onset Alzheimer’s disease. Moreover, there is evidence that individual neurons in sporadic AD patients show differences in APP gene copy number (Bushman et al., 2015). Therefore, given that APP duplication (and its overexpression) is a cause of Alzheimer’s disease, it is misleading to characterize the increased calpain activity as an “artifact” of APP overexpression.

    2.     The authors suggest that the differences they observe between their APP mutant knock-in model and other transgenic models are simply due to the presence or absence of APP overexpression. This is a very limited interpretation. It is well established that AD-related pathogenic phenotypes in mouse models not only relate to APP expression and total Aβ generation, but also to Aβ oligomer formation and localization, both neuroanatomic and subcellular. These centrally important factors have not been analyzed in this paper or the preceding report (Saito et al., 2014). 

    3.     Increased calpain activity and marked reduction of calpastatin (Cast) not only have been shown in mouse models of Alzheimer’s disease, but have also been widely reported in postmortem brain samples of Alzheimer’s disease patients (for instance, Saito et al., 1993; Rao et al., 2008; Laske et al., 2015Kurbatskaya et al., 2016). Thus, the lack of increased calpain activity in the mouse model used by Saido and colleagues in fact calls into question the physiological relevance of their model, and whether the APP23 and other transgenic models are closer approximations of the human disease.

    4.     Increased p25 generation in Cast-KO mice in an otherwise wild-type background has been shown previously (Sato et al., 2011). However, in Fig. 1 in this new paper, the Cast-KO/APPNL-F/NL-F mice show no increase in p25 signal, which should occur if, as expected, it reflects calpain-dependent p35 cleavage. This raises the question of specificity and whether the authors are in fact measuring p25. The authors discuss crosses with a Cast-KO mouse extensively in the paper, but they do not show any p25 data from Cast-KO mouse brain or APP23 crossed to Cast-KO. These blots should show elevated p25 signal and would serve as positive controls.

    5.     It is noteworthy that the knock-in APPNL-F/NL-F mice were examined only at 2 years of age. It is possible that calpain activity is downregulated by this late age. The authors also have not ruled out the possibility that one or more of the calpain gene(s) are disrupted in their model. These possibilities may account for why APPNL-F/NL-F x Cast-KO crosses did not increase p25 generation. Moreover, examination of the blot in Fig. 1 suggests that p35 may be elevated in APPNL-F/NL-F compared with the other genotypes, consistent with calpain downregulation. These control experiments should be conducted before making any strong conclusions. To rigorously assess p25 generation, a broader range of ages should also be carefully analyzed. A similar concern applies to Fig. 2, in which conclusions about sodium channel expression are based solely on a 2-year-old brain sample, and are generalized to suggest that a previous report of sodium channel downregulation in the J20 model was unique to that model. Furthermore, the present paper utilizes western blotting of whole brain homogenates, a highly insensitive approach for assessing sodium channel regulation in specific interneuron subpopulations. In particular, there is no in situ analysis by immunofluorescence, electrophysiology, or other methods.

    6.     The authors state that calpain-mediated p25 generation is an “artifact” caused by overexpression of membrane proteins, including APP and PS1, but they never address that directly in the current study. There are multiple studies demonstrating that in the absence of APP overexpression, organotypic hippocampal slices, cultured primary neurons, and even primary microglia exposed to Aβ show increased p25 generation (for instance, Zheng et al., 2005; Ma et al., 2013; Seo et al., 2014). Moreover, PS1 mutations were shown to indisputably disrupt calcium homeostasis and activate calcium-dependent enzymes (Morgan et al., 2007; Müller et al., 2011), which can increase calpain activity and subsequent p25 generation.

    7.     Similarly, there is a large body of literature, covering many mouse models of neurological disorders, that shows p25 production is unrelated to overexpression of membrane proteins. It is unfortunate that the authors did not cite any of these studies. These include models of tauopathies (Rao et al., 2014), ischemia (Meyer et al., 2014), and Parkinson’s disease (Smith et al., 2006). Furthermore, the authors’ statement that “overexpression of a mutant tau protein, a cytosolic protein, exhibited no effect on p25 levels (data not shown)” directly contradicts previous literature, a discrepancy that was not discussed in their paper. Based on points 6 and 7, it is misleading to characterize the p25 generation as an “artifact” of membrane proteins overexpression. 


    . Single App knock-in mouse models of Alzheimer's disease. Nat Neurosci. 2014 May;17(5):661-3. Epub 2014 Apr 13 PubMed.

    . Genomic mosaicism with increased amyloid precursor protein (APP) gene copy number in single neurons from sporadic Alzheimer's disease brains. Elife. 2015 Feb 4;4 PubMed.

    . Widespread activation of calcium-activated neutral proteinase (calpain) in the brain in Alzheimer disease: a potential molecular basis for neuronal degeneration. Proc Natl Acad Sci U S A. 1993 Apr 1;90(7):2628-32. PubMed.

    . Marked calpastatin (CAST) depletion in Alzheimer's disease accelerates cytoskeleton disruption and neurodegeneration: neuroprotection by CAST overexpression. J Neurosci. 2008 Nov 19;28(47):12241-54. PubMed.

    . Increased cerebrospinal fluid calpain activity and microparticle levels in Alzheimer's disease. Alzheimers Dement. 2014 Sep 4; PubMed.

    . Upregulation of calpain activity precedes tau phosphorylation and loss of synaptic proteins in Alzheimer's disease brain. Acta Neuropathol Commun. 2016 Mar 31;4:34. PubMed.

    . Calpastatin, an endogenous calpain-inhibitor protein, regulates the cleavage of the Cdk5 activator p35 to p25. J Neurochem. 2011 May;117(3):504-15. PubMed.

    . A Cdk5 inhibitory peptide reduces tau hyperphosphorylation and apoptosis in neurons. EMBO J. 2005 Jan 12;24(1):209-20. PubMed.

    . Activity-dependent p25 generation regulates synaptic plasticity and Aβ-induced cognitive impairment. Cell. 2014 Apr 10;157(2):486-98. PubMed.

    . Defects of immune regulation in the presenilin-1 mutant knockin mouse. Neuromolecular Med. 2007;9(1):35-45. PubMed.

    . Constitutive cAMP response element binding protein (CREB) activation by Alzheimer's disease presenilin-driven inositol trisphosphate receptor (InsP3R) Ca2+ signaling. Proc Natl Acad Sci U S A. 2011 Aug 9;108(32):13293-8. PubMed.

    . Specific calpain inhibition by calpastatin prevents tauopathy and neurodegeneration and restores normal lifespan in tau P301L mice. J Neurosci. 2014 Jul 9;34(28):9222-34. PubMed.

    . Ischemic stroke injury is mediated by aberrant Cdk5. J Neurosci. 2014 Jun 11;34(24):8259-67. PubMed.

    . Calpain-regulated p35/cdk5 plays a central role in dopaminergic neuron death through modulation of the transcription factor myocyte enhancer factor 2. J Neurosci. 2006 Jan 11;26(2):440-7. PubMed.

  2. The Saido knock-in (KI) mice have advanced our knowledge about the biological activity of APP mutants in vivo, but, as with any experimental model, results must be carefully interpreted.

    I view currently available APP transgenic mice less as models of AD than as tools to assay the effects of APP and APP mutants on global network activity and local plaque-associated cytopathology. APP transgenic mice have taught us that two different APP cleavage products can impair global network activity: Aβ and CTF. They have also elucidated the effects of two different types of Aβ oligomers—type 1 and type 2 Aβ oligomers. Type 1 Aβ oligomers disrupt global network activity, in the absence of plaques, whereas type 2 Aβ oligomers induce local plaque-associated cytopathology.

    Although numerous papers have elucidated signaling pathways in APP transgenic mice, few studies delineate the APP fragment that is responsible for the molecular phenotypes. In some instances, calculating correlation coefficients between a given phenotype and either CTF or Aβ has helped to delineate the likely responsible fragment. In other cases, mice expressing equivalent levels of APP, with or without the Swedish mutation, have been used to control for the effects of APP overexpression. While this controls for ER stress, which is very important, it does not help distinguish between CTF- versus Aβ-induced effects, regardless of whether the lines are made by KI or conventional methods. If a particular phenotype is absent in plaque-bearing mice that overproduce CTF, then we can conclude that the phenotype is not due to CTF or type 2 Aβ oligomers, but could be caused by type 1 Aβ oligomers. It is unclear whether the Saido KI mice produce type 1 Aβ oligomers; they possess multiple APP mutations that promote Aβ fibrils to form, which may result in the production of type 2 at the expense of type 1 Aβ oligomers.   

    What are the prospects for better mouse models of AD? I’m optimistic about the technology, but pessimistic about the sociological obstacles impeding the type of science that needs to be done to invent better models. It is exciting to think about combining advances in mouse genetics (e.g., diversity-outbred mice) with new technologies that permit engineering mammalian genomes with unprecedented speed and precision.

    However, the type of work needed to invent better models generates little enthusiasm in NIH study sections, in part because it is not narrowly defined hypothesis-testing. There is also a time constraint. Saido and colleagues spent 10 years creating and characterizing their mice; even with the new technologies, more than one five-year grant cycle will probably be required to produce a model that possesses the most desirable features of Alzheimer’s disease while lacking undesirable confounds. Finally, and personally most worrisome, the necessary human resources are vanishing. Investigators with the desired track record, breadth and depth of experience, and who have no need to carve out new research niches, are in or entering the last phase of their careers. In my own situation, I would like to leave the field with a better model than Tg2576. Last year my collaborators and I submitted two grant applications to the NIH proposing to apply modern technologies to invent better models of Alzheimer’s disease. Both applications were triaged.

  3. I appreciate the comments from Li-Huei Tsai and Karen Ashe. It is always good to see the same scenery from different angles and to discuss issues that matter. Here are some comments on the points raised.

    APP duplication in some forms of FAD and chromosome trisomy in Down’s syndrome do not justify the overexpression paradigm because the increase of APP in APP duplication and Down’s syndrome is only 1.5-fold, whereas APP transgenic mice overexpress APP approximately 10-fold or more. (See, for instance, Figure 1 of Games et al., 1995). 

    Who has ever proven the toxicity of Aβ oligomers in vivo in a relevant manner? I do not think anyone has. There are so many kinds of oligomers. Which of the oligomers are the major player(s)? How would you characterize detergent-sensitive oligomers that bind to cell membranes? Sixty percent to 70 percent of total Aβ in wild-type brain is already bound to the membrane. Besides, it is good to remember that tauopathy-accompanying neurodegeneration is the direct cause of dementia in AD and that this neurodegeneration arises more than 20 years after Aβ deposition starts. So, Aβ oligomers alone cannot cause clinical dementia.

    Hasegawa and colleagues have proven that p25 generation in the AD brain is a postmortem artifact (Taniguchi et al., 2001). Further, they experimentally showed that calpain starts to become activated approximately 30 minutes after death.

    Sato only used ex vivo and in vitro paradigms, so the results do not reflect in vivo phenomena (Sato et al., 2011). 

    Even wild-type mice live to less than three years, so two years old is very old for a mouse. The knock-in paradigm does not disrupt any gene. We agree that it is always good to have as many controls as possible, but the use of calpastatin-deficient APP knock-in mice alone makes a strong enough case. Calpastatin deficiency by itself does not activate calpain; rather, you need high intracellular calcium elevation. (Everyone in the calpain research community is aware of this fact.) Instead of trying to detect p25 at various ages or under different conditions, I suggest that scientists measure intraneuronal concentration of calcium in the mutant mice. (Calpain activation requires more calcium than you think.) Regarding Nav1.1 downregulation, because PV-positive interneurons account for the majority of Nav1.1 in the hippocampus and neocortex, western blotting is sufficient to quantify Nav1.1, as we had validated the antibody specificity using conditional knockout mice. Besides, Mucke and colleagues used western blotting to show downregulation of the sodium channel in their Cell paper (Verret et al., 2012). 

    Please note how difficult it is to activate calpain in vivo (Takano et al., 2005; Higuchi et al., 2005). In these studies, we were able to activate calpain only by injecting sub-lethal concentrations of kainic acid. In contrast, calpain is easy to activate in vitro. The discrepancy between the in vivo and in vitro paradigms, which is also often observed with apoptosis, should never be overlooked.

    AD is a disorder totally different from non-AD tauopathies, ischemia, and Parkinson’s disease.

    I have a counter-question: If p25 is so important in pathogenesis of diseases, including AD, why has no one ever crossed p25 Tg mice with conditional CDK5 knockout mice as a proof of concept? CDK5 deficiency should eliminate the pathological phenotypes of p25 Tg mice because p25 is supposed to exert its effect via activation of CDK5.

    Now is the time for us to ask why more than 400 medication candidates have failed in clinical trials over the past 20 years.


    . Alzheimer-type neuropathology in transgenic mice overexpressing V717F beta-amyloid precursor protein. Nature. 1995 Feb 9;373(6514):523-7. PubMed.

    . Calpain-mediated degradation of p35 to p25 in postmortem human and rat brains. FEBS Lett. 2001 Jan 26;489(1):46-50. PubMed.

    . Inhibitory interneuron deficit links altered network activity and cognitive dysfunction in Alzheimer model. Cell. 2012 Apr 27;149(3):708-21. PubMed.

    . Calpain mediates excitotoxic DNA fragmentation via mitochondrial pathways in adult brains: evidence from calpastatin mutant mice. J Biol Chem. 2005 Apr 22;280(16):16175-84. PubMed.

    . Distinct mechanistic roles of calpain and caspase activation in neurodegeneration as revealed in mice overexpressing their specific inhibitors. J Biol Chem. 2005 Apr 15;280(15):15229-37. PubMed.

  4. In response to Dr. Ashe's comment about being pessimistic about the obstacles to developing better models, I'm happy to say that things are improving. The NIA has just funded a collaboration between Indiana University, SAGE, and The Jackson Lab specifically to create and validate better models of late-onset AD.

    Dr. Masliah also mentioned funding of new animal models in his Alzforum Q&A earlier this week.

  5. I agree with Drs. Ashe and Sasner that the field needs better models that more closely recapitulate AD phenotypes. I am optimistic that the next generation of AD models will be greatly facilitated by the recently developed genome-editing technologies. For instance, the CRISPR/Cas9 system markedly reduces the time and effort to engineer mutations into the genomes of mouse models and human-induced pluripotent stem cells (iPSCs). The iPSC system is very exciting as it permits the evaluation of single gene effects in otherwise identical genetic backgrounds when isogenic lines are produced (Paquet et al., 2016). The combination of better mouse and iPSC models will likely be insightful in understanding disease mechanisms, and provide new platforms for developing therapies.

    On Dr. Saido’s comments, I have the following points:

    1.     APP gene dosage in AD and DS. APP gene dosage does not reflect exact levels of APP proteins. In fact, in DS, APP expression levels exceed the 1.5X gene dosage (Netzer et al., 2010; Iulita et al., 2014). And while Games et al. showed 10X APP expression in their PDAPP mouse line, other transgenic models, including APP23 and 5XFAD, show lower overexpression of APP (Capetillo-Zarate et al., 2006; Py et al., 2014). Overall, 1.5X gene dosage, which is not physiological, constitutes overexpression.

    2.     The absence of p25 generation in the WB from Cast-KO/APPNL-F/NL-F is still questionable. In their previous report (Saito et al., 2014), the authors showed that Cast deficiency accelerates Aβ amyloidosis and neuroinflammation in this mouse model. These effects are most likely mediated through calpain activation.

    3.     It is interesting to learn that Dr. Saido does not consider that Aβ oligomers play a role in synaptotoxicity and dementia in AD. This certainly differs from other observations (Walsh et al., 2002; Xia et al., 2009; Um et al., 2012). 

    4.     My major concern is that there seems to be selective consideration of published literature in the manuscript’s interpretations. For instance, while an increased p25/p35 ratio in AD brains compared to unaffected controls has continued to be reported until now (for instance, Sadleir and Vassar, 2012; Kurbatskaya et al., 2016), these results are ignored. Similarly, my previous comments cited many reports showing p25 upregulation in neurodegenerative conditions, including incubation of brain tissues with Aβ or overexpression of mutant tau. None of these conditions are related to membrane protein overexpression. It is disappointing that these results do not factor into the interpretation of the data of Saito et al. 2016.

    5.     We have shown previously that inhibition of p25 generation attenuates Cdk5 hyperactivation and AD-like pathology in 5XFAD mice (Seo et al., 2014). While Dr. Saido suggests crossing p25Tg with conditional Cdk5 knockout mice as a proof of concept, it should be noted that Cdk5 is an essential kinase in the nervous system whose function is required for embryonic brain development and maintaining synaptic plasticity and cognitive function in the adult. Cdk5 null, either whole-body or brain-specific, is detrimental to the nervous system (Ohshima et al., 1996; Takahashi et al., 2010; Su et al., 2013). As Cdk5 is hyperactive in AD, alternatively, many groups have attempted to modulate Cdk5 activity using inhibitors, small molecules or peptides, rather than completely ablate Cdk5 activity. Beneficial effects of Cdk5 inhibition in the conditions of p25 overexpression have been reported in many studies. (For instance, Wen et al., 2008; Sundaram et al., 2013.) 

    In conclusion, the amount of data published in this particular manuscript seems rather limited to support sweeping conclusions. While the APP knock-in mouse model has some limitations that the community should critically evaluate, I do see merit in using this model to investigate phenotypes first described in APP/PS overexpression models. I look forward to much more data being generated with this model as more groups are aging and studying these mice. 


    . Selective vulnerability of different types of commissural neurons for amyloid beta-protein-induced neurodegeneration in APP23 mice correlates with dendritic tree morphology. Brain. 2006 Nov;129(Pt 11):2992-3005. PubMed.

    . Nerve growth factor metabolic dysfunction in Down's syndrome brains. Brain. 2014 Mar;137(Pt 3):860-72. Epub 2014 Feb 11 PubMed.

    . Upregulation of calpain activity precedes tau phosphorylation and loss of synaptic proteins in Alzheimer's disease brain. Acta Neuropathol Commun. 2016 Mar 31;4:34. PubMed.

    . Lowering beta-amyloid levels rescues learning and memory in a Down syndrome mouse model. PLoS One. 2010;5(6):e10943. PubMed.

    . Targeted disruption of the cyclin-dependent kinase 5 gene results in abnormal corticogenesis, neuronal pathology and perinatal death. Proc Natl Acad Sci U S A. 1996 Oct 1;93(20):11173-8. PubMed.

    . Efficient introduction of specific homozygous and heterozygous mutations using CRISPR/Cas9. Nature. 2016 May 5;533(7601):125-9. Epub 2016 Apr 27 PubMed.

    . Differential spatio-temporal regulation of MMPs in the 5xFAD mouse model of Alzheimer's disease: evidence for a pro-amyloidogenic role of MT1-MMP. Front Aging Neurosci. 2014;6:247. Epub 2014 Sep 18 PubMed.

    . Cdk5 Protein Inhibition and Aβ42 Increase BACE1 Protein Level in Primary Neurons by a Post-transcriptional Mechanism: IMPLICATIONS OF CDK5 AS A THERAPEUTIC TARGET FOR ALZHEIMER DISEASE. J Biol Chem. 2012 Mar 2;287(10):7224-35. PubMed.

    . Single App knock-in mouse models of Alzheimer's disease. Nat Neurosci. 2014 May;17(5):661-3. Epub 2014 Apr 13 PubMed.

    . Activity-dependent p25 generation regulates synaptic plasticity and Aβ-induced cognitive impairment. Cell. 2014 Apr 10;157(2):486-98. PubMed.

    . Forebrain-specific deletion of Cdk5 in pyramidal neurons results in mania-like behavior and cognitive impairment. Neurobiol Learn Mem. 2013 Oct;105:54-62. Epub 2013 Jul 11 PubMed.

    . Specific inhibition of p25/Cdk5 activity by the Cdk5 inhibitory peptide reduces neurodegeneration in vivo. J Neurosci. 2013 Jan 2;33(1):334-43. PubMed.

    . Conditional deletion of neuronal cyclin-dependent kinase 5 in developing forebrain results in microglial activation and neurodegeneration. Am J Pathol. 2010 Jan;176(1):320-9. PubMed.

    . Alzheimer amyloid-β oligomer bound to postsynaptic prion protein activates Fyn to impair neurons. Nat Neurosci. 2012 Sep;15(9):1227-35. Epub 2012 Jul 22 PubMed.

    . Naturally secreted oligomers of amyloid beta protein potently inhibit hippocampal long-term potentiation in vivo. Nature. 2002 Apr 4;416(6880):535-9. PubMed.

    . Transcriptional regulation of beta-secretase by p25/cdk5 leads to enhanced amyloidogenic processing. Neuron. 2008 Mar 13;57(5):680-90. PubMed.

    . A specific enzyme-linked immunosorbent assay for measuring beta-amyloid protein oligomers in human plasma and brain tissue of patients with Alzheimer disease. Arch Neurol. 2009 Feb;66(2):190-9. PubMed.

  6. I read with considerable interest the paper of Saido and colleagues describing the lack of calpain activation, p25 accumulation, and tau pathology in their mutant APP knock-in mouse line. Relevant to the discussion surrounding the validity of these findings, it is noteworthy that my colleagues and I first characterized a knock-in mutant mouse model exhibiting cerebral amyloid deposition, created by Andrew Reaume and Richard Scott while at Cephalon, starting with initial publications more than 15 years ago.

    The mouse lines we studied carried a KM670/671/NL knock-in mutation in the APP gene (with a humanized Aβ sequence), a P264L knock-in mutation in PS-1, or homozygous mutations in both genes, the latter referred to as the APP/PS-1 DKI mouse line. Despite neocortical amyloid deposition beginning around six months of age and progressing in anatomical spread and intensity for another 21 months, the DKI mice failed to develop any appreciable phenotype of the human disease beyond the plaques and their attendant gliosis and neuroinflammation (Flood et al., 2002; Malthankar-Phatak et al., 2012). There was no loss of neurons or synapses, no neuropil disruption beyond that occurring within core-containing plaques themselves, and no tau hyperphosphorylation. Using antibodies specific for calpain- or caspase-derived cleaved proteins, there was no evidence for activation of either family of cell death-associated protease. These results are consistent with the observations of Saido and colleagues using an entirely different mutant knock-in mouse line.

    Our understanding of the mechanisms responsible for the pathogenic progression of Alzheimer's disease remain hampered by the lack of a small animal experimental model that recapitulates not only features of the protein deposition pathology and the human genetics, but also of the profound, progressive, and stereotypical regional neurodegeneration and synapse loss. Given that neorcortical synaptic density remains the strongest inverse structural correlate for the magnitude of cognitive decline in the Alzheimer patient (Terry et al., 1991), this is a huge unmet experimental translational need.


    . FAD mutant PS-1 gene-targeted mice: increased A beta 42 and A beta deposition without APP overproduction. Neurobiol Aging. 2002 May-Jun;23(3):335-48. PubMed.

    . Amyloid deposition and advanced age fails to induce Alzheimer's type progression in a double knock-in mouse model. Aging Dis. 2012 Apr;3(2):141-55. PubMed.

    . Physical basis of cognitive alterations in Alzheimer's disease: synapse loss is the major correlate of cognitive impairment. Ann Neurol. 1991 Oct;30(4):572-80. PubMed.

  7. The research conducted by Dr. Saido and his colleagues is long overdue, relevant, and gets to the heart of the problem of using overexpressing APP models. There is now sufficient data to indicate the neurotropic properties of APP and Aβ and why strategies to remove Aβ are flawed (e.g. Atwood, 2010; Kumar et al., 2016). 

    More relevant models of AD would involve developing models of endocrine dyscrasia, since ultimately endocrine mechanisms are the main drivers of aging and late-onset AD (Bowen and Atwood, 2004Atwood and Bowen, 2015).

    See also:

    Atwood, C.S. (2010). Amyloid-β aggregation as a protective acute-phase response to injury/neurodegeneration: A barrier function for amyloid-β deposits. In: Functional Amyloid Aggregation. Eds. Stefania Rigacci and Monica Bucciantini.  Research Signpost, Kerala, India, pp 115-134.  ISBN: 978-81-308-0425-5


    . Alzheimer's disease: the potential therapeutic role of the natural antibiotic amyloid-β peptide. Neurodegener Dis Manag. 2016 Oct;6(5):345-8. Epub 2016 Sep 7 PubMed.

    . Living and dying for sex. A theory of aging based on the modulation of cell cycle signaling by reproductive hormones. Gerontology. 2004 Sep-Oct;50(5):265-90. PubMed.

    . A Unified Hypothesis of Early- and Late-Onset Alzheimer's Disease Pathogenesis. J Alzheimers Dis. 2015;47(1):33-47. PubMed.

  8. We would like to emphasize that the APP knock-in mice do not display any tau pathology or neurodegeneration. This suggests that Aβ pathology might account, at least in part, for cognitive dysfunction in AD via disturbance of neuronal activities, because we observed reduction of mushroom spines in the relatively early stage in these mice (Zhang et al., 2015). However, we believe that Aβ-induced memory failure alone is insufficient to explain all the symptoms of AD patients, because tauopathy-accompanying irreversible neurodegeneration has already taken place at the disease onset even in FAD-mutation carriers (Bateman et al., 2012). To be strictly correct, we should consider the APP knock-in mice “models of preclinical AD.”

    The absence of tauopathy and neurodegeneration in these mice, which live less than three years in general, is unsurprising because it takes more than two decades for Aβ amyloidosis to induce tauopathy and neurodegeneration in humans (Bateman et al., 2012). We can instead utilize the preclinical AD models, i.e. APP knock-in mice, for a number of purposes:

    1. identification of biomarker(s) for preclinical AD,

    2. identification of molecule(s) that evoke tauopathy in an Aβ pathology-dependent manner,

    3. application to preclinical studies of preventive medicine(s) and

    4. generation of a “perfect” AD model(s) by crossbreeding them with appropriate mutant mice.

    Biomarkers, which we could use for both diagnosis and prognosis of preclinical AD, will shorten the time necessary for drug development. The AppNL-F and AppNL-G-F mice are the only available single knock-in models that develop Aβ pathology and memory deficits. The presence of multiple mutations in the APP gene, while indeed not observed in human patients, should not impede the above purposes because, at least to our knowledge, they do not interact with each other. AppNL-F mice are probably more suitable for analyzing the mechanisms that affect Aβ deposition, AppNL-G-F for mechanisms that affect the downstream cascade.


    . Clinical and biomarker changes in dominantly inherited Alzheimer's disease. N Engl J Med. 2012 Aug 30;367(9):795-804. PubMed.

  9. To What Extent Do APP Knock-ins Call Overexpression Models of AD into Question?

    There are many reasons to question the claim that all phenotypes of hAPP transgenic mouse models that are not found in APP-KI mice are artifacts. Most importantly, this conclusion relies on mouse-to-mouse comparisons to determine what an “artifact” is, instead of considering the human condition as the gold standard. An alternative interpretation to consider is that APP-KI mice fail to mimic some aspects of the human condition that hAPP transgenic mice replicate more faithfully. It is worth emphasizing that overexpression of wild-type hAPP causes early onset familial AD in human carriers with APP duplications and most likely also in Down’s syndrome. Interestingly, recent findings suggest that CYFIP2-dependent local synthesis of hAPP at synapses may be increased not only in Tg2576 mice, but also in humans with sporadic AD (Tiwari et al., 2016), suggesting that overexpression of hAPP could be even more relevant to the human condition than is widely presumed.

    Multiple molecular alterations identified in hAPP transgenic mice are also found in people with AD, including altered levels of Nav1.1, calbindin, collagen VI, EphB2, klotho, activated group IVA phospholipase A2, and adenosine receptor A2A (Cheng et al., 2009; Chin et al., 2007; Cissé et al., 2011; Dubal et al., 2015; Meilandt et al., 2008; Orr et al., 2015; Palop et al., 2003; Sanchez-Mejia et al., 2008; Suberbielle et al., 2015; Verret et al., 2012). If some of these or other alterations that occur in hAPP transgenic mice and in AD patients were not detectable in APP-KI mice, as suggested by Saito et al., it is unclear to us why the failure of APP-KI mice to reproduce these aspects of the human condition should be interpreted as a strength of this model and a flaw of hAPP transgenic mice.

    We would like to comment in particular on Nav1.1 depletions in this context. We identified Nav1.1 reductions in the parietal cortex of AD patients and of hAPP-J20 mice (Verret et al., 2012). In contrast, Saito et al. did not find such reductions in APP-KI mice when they analyzed lysates of whole mouse brains, an approach that can obscure Nav1.1 depletions in the parietal cortex by dilution. Furthermore, their Nav1.1 experiment was probably underpowered (four mice per group to make 10 potential comparisons) and no statistical analyses were performed to assess the potential decrease in Nav1.1 levels in APPNL-F/NL-F x Cast KO mice that is suggested by the data shown in Figure 2. In comparison, we used 18-22 mice per group to make six comparisons and performed rigorous statistical analyses. Lastly, the authors concluded that Nav1.1 alterations are specific to hAPP-J20 mice, disregarding papers that have documented similar alterations in other mouse models of AD. For example, hypofunction of Nav1.1 has been demonstrated also in Tg2576 mice (Corbett et al., 2013) and in BACE1 transgenic mice (Kim et al., 2007). 


    . Collagen VI protects neurons against Abeta toxicity. Nat Neurosci. 2009 Feb;12(2):119-21. PubMed.

    . Reelin depletion in the entorhinal cortex of human amyloid precursor protein transgenic mice and humans with Alzheimer's disease. J Neurosci. 2007 Mar 14;27(11):2727-33. PubMed.

    . Reversing EphB2 depletion rescues cognitive functions in Alzheimer model. Nature. 2011 Jan 6;469(7328):47-52. PubMed.

    . Sodium channel cleavage is associated with aberrant neuronal activity and cognitive deficits in a mouse model of Alzheimer's disease. J Neurosci. 2013 Apr 17;33(16):7020-6. PubMed.

    . Life extension factor klotho prevents mortality and enhances cognition in hAPP transgenic mice. J Neurosci. 2015 Feb 11;35(6):2358-71. PubMed.

    . BACE1 regulates voltage-gated sodium channels and neuronal activity. Nat Cell Biol. 2007 Jul;9(7):755-64. PubMed.

    . Enkephalin elevations contribute to neuronal and behavioral impairments in a transgenic mouse model of Alzheimer's disease. J Neurosci. 2008 May 7;28(19):5007-17. PubMed.

    . Astrocytic adenosine receptor A2A and Gs-coupled signaling regulate memory. Nat Neurosci. 2015 Mar;18(3):423-34. Epub 2015 Jan 26 PubMed.

    . Neuronal depletion of calcium-dependent proteins in the dentate gyrus is tightly linked to Alzheimer's disease-related cognitive deficits. Proc Natl Acad Sci U S A. 2003 Aug 5;100(16):9572-7. Epub 2003 Jul 24 PubMed.

    . Phospholipase A2 reduction ameliorates cognitive deficits in a mouse model of Alzheimer's disease. Nat Neurosci. 2008 Nov;11(11):1311-8. PubMed.

    . DNA repair factor BRCA1 depletion occurs in Alzheimer brains and impairs cognitive function in mice. Nat Commun. 2015 Nov 30;6:8897. PubMed.

    . Alzheimer-related decrease in CYFIP2 links amyloid production to tau hyperphosphorylation and memory loss. Brain. 2016 Oct;139(Pt 10):2751-2765. Epub 2016 Aug 14 PubMed.

    . Inhibitory interneuron deficit links altered network activity and cognitive dysfunction in Alzheimer model. Cell. 2012 Apr 27;149(3):708-21. PubMed.

  10. I wondered if Lennart might comment. He has had our single APP knock-in mice since 2012, two years before we published our Nat Neuroscience paper. “Collaboration and competition together” is our scientific policy because we believe that this is the best way to accelerate research worldwide. Lennart’s group may have examined the Nav1.1 levels in the parietal cortex of the single APP knock-in mice, but they have not informed us of the results.

    It is also important that there was no Nav1.1 reduction even in our postmortem mouse brains. I suggest that the research community be cautious about pursuing Nav1.1 as a therapeutic target.

    One must be very careful interpreting the human data because they are almost always taken from postmortem tissues in which calcium-dependent enzymes, such as phosphorylases, phospholipases, and proteases artificially alter intra- and extracellular environments. Lee and colleagues showed that the large portion of phosphorylated tau from human brain undergoes dephosphorylation postmortem (Matsuo et al., 1994). One must therefore step back when making comments based on postmortem tissues. Ideally, the best way would be to use biopsied tissues rather than autopsied tissues, although this is extremely difficult.

    The pathological phenotypes of APP knock-in mice are not mild at all and include amyloidosis, astrocytosis, and microgliosis. Approximately 40 percent of the data obtained using APP- or APP/PS-overexpressing models may be relevant. Non-overexpression paradigms are better than overexpression paradigms because of the following drawbacks to the latter:

    • Destruction of endogenous gene loci
    • Unphysiological interaction of overexpressed APP with cellular elements
    • Unphysiological interaction of overproduced non-Aβ APP fragments with cellular elements
    • Non-specific ER stress
    • Accumulation of Aβ species different from those in AD brains
    • Atypical region specificity of Aβ pathology
    • Disturbance of transcription factor dynamics
    • Inconsistent effects of secretase inhibitors
    • Phenotype-biased screening of mouse lines
    • More artificial phenotype(s) generated by crossbreeding with other mutant mice

    Finally, please join us for the SfN symposium entitled “Second Generation AD Mouse Models for Reproducible Preclinical Studies,” which will take place in the morning of Sunday, Novemberv13th. You will see a lot of new findings and hear some interesting discussions.


    . Biopsy-derived adult human brain tau is phosphorylated at many of the same sites as Alzheimer's disease paired helical filament tau. Neuron. 1994 Oct;13(4):989-1002. PubMed.

Make a Comment

To make a comment you must login or register.


Research Models Citations

  1. APP NL-F Knock-in
  2. APP NL-G-F Knock-in
  3. APP23
  4. J20 (PDGF-APPSw,Ind)

Webinar Citations

  1. Good-Bye Overexpression, Hello APP Knock-in. A Better Model?

Paper Citations

  1. . ER stress triggers apoptosis induced by Nogo-B/ASY overexpression. Exp Cell Res. 2006 Jul 1;312(11):1983-8. Epub 2006 May 9 PubMed.
  2. . Neuronal Store-Operated Calcium Entry and Mushroom Spine Loss in Amyloid Precursor Protein Knock-In Mouse Model of Alzheimer's Disease. J Neurosci. 2015 Sep 30;35(39):13275-86. PubMed.
  3. . Loss of GPR3 reduces the amyloid plaque burden and improves memory in Alzheimer's disease mouse models. Sci Transl Med. 2015 Oct 14;7(309):309ra164. PubMed.
  4. . Intraneuronal beta-amyloid aggregates, neurodegeneration, and neuron loss in transgenic mice with five familial Alzheimer's disease mutations: potential factors in amyloid plaque formation. J Neurosci. 2006 Oct 4;26(40):10129-40. PubMed.
  5. . Activity-dependent p25 generation regulates synaptic plasticity and Aβ-induced cognitive impairment. Cell. 2014 Apr 10;157(2):486-98. PubMed.
  6. . Calpain mediates excitotoxic DNA fragmentation via mitochondrial pathways in adult brains: evidence from calpastatin mutant mice. J Biol Chem. 2005 Apr 22;280(16):16175-84. PubMed.
  7. . Calpastatin, an endogenous calpain-inhibitor protein, regulates the cleavage of the Cdk5 activator p35 to p25. J Neurochem. 2011 May;117(3):504-15. PubMed.
  8. . A Cdk5 inhibitory peptide reduces tau hyperphosphorylation and apoptosis in neurons. EMBO J. 2005 Jan 12;24(1):209-20. PubMed.
  9. . Upregulation of calpain activity precedes tau phosphorylation and loss of synaptic proteins in Alzheimer's disease brain. Acta Neuropathol Commun. 2016 Mar 31;4:34. PubMed.
  10. . Inhibitory interneuron deficit links altered network activity and cognitive dysfunction in Alzheimer model. Cell. 2012 Apr 27;149(3):708-21. PubMed.
  11. . Single App knock-in mouse models of Alzheimer's disease. Nat Neurosci. 2014 May;17(5):661-3. Epub 2014 Apr 13 PubMed.

Other Citations

  1. Alzforum Research Models

External Citations

  1. RIKEN BioResource Center
  2. Phase 1 trial
  3. symposium

Further Reading

Primary Papers

  1. . Calpain Activation in Alzheimer's Model Mice Is an Artifact of APP and Presenilin Overexpression. J Neurosci. 2016 Sep 21;36(38):9933-6. PubMed.