Researchers rely on transgenic mice to model Alzheimer's disease, but these models are far from perfect. Some can only be maintained on specific genetic backgrounds, some die young or have surprising phenotypes when crossed with other mouse lines, and in others amyloidosis drifts later in successive generations, suggesting gene expression changes from parents to offspring. Enter a new line of mice that promises to circumvent many of these problems. Researchers led by Takaomi Saido have engineered APP knock-ins. These animals recapitulate much of the pathology seen in presymptomatic AD, without overexpressing APP or interrupting other mouse genes. How do these knock-ins compare?

Saido led the discussion on April 22 with Karen Hsiao Ashe, Ronald DeMattosDavid Holtzman, Mathias Jucker, and Mike Sasner


Check out these knock-ins and other mouse models in our new Research Models database.



Alzforum thanks Nature Neuroscience for making Saido's paper freely available to Alzforum readers. 


View the full Webinar below:

See slide presentations below; click on blue circles to switch between Fagan, Saido, and Ashe slide presentations.



Over the last two decades, researchers have generated dozens of different mouse models of Alzheimer's disease. The most commonly used lines carry mutant forms of human Aβ precursor protein (APP), either alone or in combination with human presenilin, tau, or other genes. Used for both basic research and drug discovery, these animals have provided a wealth of information. However, there have been niggling doubts from the beginning that some of their phenotypes may have little to do with the human disease, and negative clinical trials later reinforced critical debate about the validity of overexpression models.

The concern arises because the human transgenes are randomly inserted into the mouse DNA, where they may disrupt essential genes or regulatory elements. For example, the Tg2576 mice only survive on a mixed B6/SJL mouse genetic background; if the transgene is maintained on a pure B6 background, the animals die prematurely, suggesting the human APP gene may be interrupting mouse DNA sequences. Transgenic mice tend to lack cell-specific alternative splicing of the APP transcript, and some strains die young of unknown causes.

Moreover, the transgenes are expressed at much higher levels than normal. Many mice with APP transgenes not only produce more Aβ than wild-type mice, but also make more full-length APP and APP fragments, including the soluble ectodomains (sAPPβ),  C-terminal fragments, and the APP intracellular domain (AICD), all of which have poorly characterized functions. APP itself interferes with axonal transport because it interacts with microtubule motors. Researchers recognize that targeted replacement of the mouse genes with human versions is a cleaner approach, but such “knock-ins” have proven a tough nut to crack.

After a 12-year effort, Takaomi Saido's group at RIKEN Brain Science Institute, Wako, Japan, has made APP knock-in mice. Humanizing the Aβ sequence and introducing various mutations was time-consuming. In addition, the researchers had removed a long intron between exons 16 and 17 to make manipulations easier, but discovered that expression plummeted without this DNA. The intron probably contains important regulatory elements, said Saido. 

As described in the April 13 Nature Neuroscience, first author Takashi Saito and colleagues generated three strains of knock-in mice. One, called NL, contains the Swedish mutation KM670/671NL, which enhances β-secretase cleavage of APP. A second, NL-F, carries the Swedish plus the Beyreuther/Iberian mutation I716F, which favors γ-secretase cleavage of C-terminal fragments at the 42 position and increases the Aβ42:Aβ40 ratio. The third, NL-G-F, carries the Swedish, Iberian, and the Arctic mutation E693G, which accelerates Aβ aggregation. All these mice carry humanized Aβ sequences in the mouse APP locus and are driven by normal mouse promoters. The NL-F and NL-G-F models are curated on the Alzforum Research Models database.

How do these mice compare to stalwarts such as APP23 or Tg2576? Both these lines express the Swedish mutation. Unlike APP23, which overexpresses the precursor protein sevenfold, NL and NL-F mice produce normal amounts of APP and AICD. Like APP23, they generate more β-CTFs than do wild-type mice, in keeping with the known stoking of β-secretase cleavage by the Swedish mutation. NL-F mice produced more Aβ42 than either NL or APP23 mice, leading to an accumulation of soluble and insoluble Aβ42 as the mice aged. Plaques appeared in the cortex and hippocampus at around 6 months in NL-F mice, as opposed to 12 months in the APP23 animals, and steadily grew as the knock-ins aged to 24 months. In plaques, Aβ1-42 appeared first, followed by Aβ3(pE)-42, the N-terminal truncated, pyroglutamylated form. This finding hints that full-length Aβ seeds plaques, a point of debate in the field.

Microglia and activated astrocytes accumulated around the plaques, reflecting the neuroinflammation that is also seen in the brains of people with AD. Reductions in the proteins synaptophysin and PSD95 in the cortex and hippocampus suggested the knock-in mice were losing synapses. Beginning at 18 months, NL-F mice had trouble learning the location of a treat in a Y maze. Age-matched NL mice, on the other hand, navigated the Y maze with aplomb, suggesting that the increases in Aβ42, rather than β-CTF, caused the learning and memory deficits.

In short, the NL-F mice display many of the pathologies seen in AD. Saido believes these mice can better answer questions about the pathogenesis of AD. "Conventional transgenic mice overproduce so many other APP fragments that their phenotypes could be misleading, and they make it difficult to study the underlying mechanisms driving pathology," Saido told Alzforum. 

Which reported phenotypes did the knock-in mice not confirm? In other words, what parts of this extensive literature might be artifacts due to APP overexpression? The answer is far from complete. It will require further analysis of the knock-in mice and an in-depth review of the literature. That said, drawing this full comparison is important to weed the literature of reported phenotypes that could mislead scientists who use the models for mechanistic studies or drug discovery. To get the process started, Saido’s paper offers a prior phenotype from his own lab that did not hold up. He had reported previously that deficiency of the enzyme calpastatin exacerbated pathology in APP23 mice, causing the mice to die young with hyperphosphorylation of tau and somatodendritic atrophy (see Higuchi et al., 2012). While this seemed plausible at the time, crossing NL-F mice with calpastatin knockouts reproduced none of these phenotypes. 

"There have been nearly 3,000 research papers using APP transgenic mice. Many describe crosses with other gene knockouts or knock-ins, such as ApoE and tau, but we are unsure if the observed phenotypes are relevant in terms of human AD pathogenesis," said Saido. He suggested much of the earlier work should be reexamined.

Because NL-F mice take 18 months to show signs of memory impairment, Saito and colleagues generated the NL-G-F mice to try to speed things up. The addition of the Arctic mutation had the desired effect, accelerating phenotypes by a factor of three. These mice initially deposit Aβ in the brain by two months, as opposed to six months of age for the NL-F mice, and falter in the Y maze at six months.

In another example of how these mice perhaps mimic human disease more faithfully than prior models, NL-G-F mice also develop subcortical amyloidosis. This recapitulates pathology seen in human carriers of the Arctic mutation, yet transgenic mice overexpressing APP with the Arctic mutation lay down no plaques in subcortical brain regions. 

Saido believes the new knock-in mice simulate preclinical AD, that is, aggressive Aβ pathology with mild memory impairment but prior to overt neurofibrillary pathology. They could be used to screen for preventive medicines or biomarkers of presymptomatic disease, Saido said. 

Saido offers to freely distribute these animals to academic labs and not-for-profit institutions.—Tom Fagan


Q: The memory deficits occur after plaque deposition in your models. Do you think these findings challenge the current Aβ model, which suggests soluble Aβ oligomers precede plaques and are responsible for memory deficits?

Saido: I think the memory impairment that we observed corresponds to the very mild impairment seen in preclinical AD patients because there is no tauopathy or neurodegeneration not only in our mice but also in the conventional APP Tg mice. Unless we can reconstitute all the major AD pathologies without using mutant tau, we cannot really tell whether Aβ oligomers play a central role in AD pathogenesis. Besides, if Aβ oligomers are important, membrane-associated ones must be the most toxic, but extraction of such oligomers would require the use of detergent(s), which might destruct the oligomeric structure. So far, we have only detected a monomer and a dimer in our models.

Q: Did you check the APP processing in endosome in these mice?

Saido: No, not yet. We are currently examining the mice lacking Atg-7 based on our previous observation (Nilsson et al., 2013). This paper is open-access.

Q: How about autophagy markers in these APP knock-in mice?

Saido: Per Nilsson in my lab is working on this issue.

Q: Do you see any increased inflammatory responses prior to plaque pathology?  Also, do you see any activated microglia or reactive astrocytes that phagocytose Aβ?

Saido: I believe the inflammatory responses arise only after plaque formation. You made a good point about phagocytosis. We will collaborate with astrocyte and microglia specialists to address these questions.

Q: Do brain interstitial fluid or cerebrospinal fluid measurements indicate that any of the downstream changes of neurodegeneration—for example tau increase—are happening in the APP knock-in mice as they age to or past two years?

Saido: We have collected CSF but have not yet analyzed tau or other biomarker candidates.

Q: Is there any tau pathology in these mice?

Saido: We see some signs in the dentate gyrus and CA3 sector of hippocampus from very old (24 months or more) NL-F mice, but they appear to be stochastic and thus could be artifacts. I believe we need to humanize murine tau to investigate this.

Q: Did you look at phospho-tau levels in later stages of these mice?

Saido: We see increased phospho-tau immunoreactivities in dystrophic neurites around Aβ plaques.

Q: Between different groups who have developed different strains, could the field come up with a consensus list or table of which previously described phenotypes are still considered valid, and which, like the calpastatin example, are most likely artifacts of APP overproduction?

Saido: I believe so. As David Holtzman mentioned during the Webinar, the involvement of ApoE4 in Aβ deposition is likely solid. Our mice will resolve the controversy over the role of the environmental enrichment in prevention of Aβ pathology and memory impairment.

Q: Given that so many technical issues are associated with the Morris water maze, maybe we shouldn't use such variable tests in the first place. Are we trying to get too much out of APP mice?

Saido: I agree that assessing memory impairment in mouse models by behavior experiments is difficult, especially in a manner that can distinguish between neurological conditions of preclinical AD, MCI, and AD in humans. Even changes in long-term potentiation (LTP), which must reflect more basic plastic processes, could be artificial because the electrophysiological stimuli used to induce LTP are higher than physiological ones. For these reasons, I prefer to define the disease conditions by pathological features, i.e., Aβ amyloidosis, tauopathy, and neurodegeneration. It would be even better if we could observe shrinkage of the brain and enlargement of ventricles by MRI.

Q: In the final animal model, i.e., NL-G-F, four major mutations are accumulated to induce the APP and cognitive decline in six months instead of 18 months. Would this complicated animal model design be helpful to test drugs?

Saido: The mice would not be suitable for testing anti-Aβ immunotherapy because antibodies to wild-type Aβ1-40/42 would not bind to the Arctic Aβ with the same avidity. Drugs targeted to mechanisms downstream of Aβ amyloidosis, such as neuroinflammation, can be tested using the NL-G-F mice.

Q: At a time when amyloid targeting therapies for AD are failing, would the APP-KI model be a good one to test drug candidates?

Saido: I believe that some of the amyloid-targeting therapies will become effective preventive medications if they are used for preclinical AD patients (it would be like using statin for hypercholesterolemia to prevent stroke), because Aβ amyloidosis starts about two decades before the disease onset. For the preventive medicines to succeed, we need to identify plasma biomarker(s). I think the new mice will help us to re-examine amyloid-targeting therapies and establish plasma biomarker-based presymptomatic diagnosis. 

Q: Do you have any idea why the APP-NL-F mice deposit so much more aggressively than Schroder's earlier APPsw/Lon knock-in line?  I understand that the earlier line used the London 717 rather than the Beyreuther 716 mutation, but is that the only difference? 

Saido: Although I do not have much information about the genomic structure of APPsw/Lon knock-in line (this can be important because intron 16 affects the expression of APP), I think the difference is caused by the strength of the Beyreuther/Iberian mutation, which increases the ratio of Aβ42/Aβ40 approximately 30-fold.

Q: Aβ42 is a misfolded protein of the beta type. CJD amyloid is a misfolded protein of the alpha type. Is there any new insight into the earliest possible precursors for β42 amyloids (plural) versus alpha misfolded amyloids in CJD?

Saido: Although I am not an expert on CJD amyloidosis, I think the major difference lies in the presence of the GPI anchor of prion protein, which induces more membrane surface-associated interactions.

Q: Is there any plan to make the new APP KI mice commercially available?

Saido: The mice will be available to nonprofit research organizations with a material transfer agreement and to for-profit organizations with a license. RIKEN BRC will soon take care of the procedures.

Q: I was wondering about the tangential comment addressing the difficulties of maintaining the Tg2576 mutation on a pure C57BL/6 background. Could you tell us more? Is there a good paper I could read to learn more about that? I'm quite interested in the impact of strain background in these mouse models.

Saido: Tg2576 mice survive only on the mixed background of B6 and SJL, and will die when back-crossed to B6 background. I believe this notion is well-recognized in the research field. This is also what I heard directly from Karen Ashe when I visited her in Minnesota to give a seminar in 2012. I do not know if there is any literature describing this phenomenon.


No Available Comments

Make a Comment

To make a comment you must login or register.


Mutations Citations

  1. APP K670_M671delinsNL (Swedish)
  2. APP I716F (Iberian)
  3. APP E693G (Arctic)

Research Models Citations

  1. APP23

Paper Citations

  1. . Mechanistic involvement of the calpain-calpastatin system in Alzheimer neuropathology. FASEB J. 2012 Mar;26(3):1204-17. PubMed.
  2. . Aβ secretion and plaque formation depend on autophagy. Cell Rep. 2013 Oct 17;5(1):61-9. PubMed.

Other Citations

  1. NL-F

External Citations

  2. paper 

Further Reading

No Available Further Reading

Primary Papers

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