For two decades, researchers have relied upon mouse models that overexpress disease-related proteins to tease apart the connections between amyloid and Alzheimer’s disease. These models have moved the field forward, but a nagging sense that they capture only aspects of the human disease pathology persists. Now, more than a year after their debut, amyloid-precursor protein knock-in (APP-KI) mice, which express mutated human APP at physiological levels, are starting to yield interesting findings. Writing in the September 30 Journal of Neuroscience, Ilya Bezprozvanny and colleagues at the University of Texas Southwestern Medical Center report that the excessive Aβ42 peptide produced in these mice throws calcium regulation out of whack and limits the number of mushroom spines, which are thought to play a key role in memory. The results jibe with what the researchers previously reported in another knock-in mouse expressing mutated human presenilin, indicating that wonky calcium levels may be a common mechanism driving spine loss during AD. The results are among the first to come out of the APP-KI model, and could spell a win for knock-in models in general.

The development of APP knock-in mice was the culmination of a 12-year labor of love by co-author Takaomi Saido at RIKEN Brain Science Institute in Wako, Japan (see Apr 2014 Webinar). Surmounting sizable technical difficulties, the researchers generated strains that express physiological levels of human APP harboring the Swedish and Iberian mutations (APP NL-F), and the Swedish, Iberian, and Arctic mutations (APP NL-G-F) that cause familial AD. The animals develop amyloid plaques and eventually memory problems, with age at onset depending on the strain. Despite their normal levels of the APP protein, the mice express an elevated ratio of Aβ42 to Aβ40, as do most people with AD. Since publishing on the mouse strain last year, Saido said he has shared them with 120 labs around the world, which have been breeding the mice and are now carrying out their initial experiments.

Bezprozvanny, who studies the role of calcium homeostasis in Alzheimer’s disease, said he avoids using APP-overexpression models due to the potential caveats that come with pumping out excessive amounts of a given protein. APP and its peptide fragments have known, and maybe also unknown, functions that could confound findings in these models, he said. Furthermore, overexpression models tend to produce a lower Aβ42/Aβ40 ratio than do most people with AD, he said. “However, these APP-KI mice have more Aβ42, and the ratio looks a lot more like human disease,” Bezprozvanny said. Saido’s APP-KI mouse coaxed Bezprozvanny to take his first plunge into APP mouse model territory.

Previous work from Bezprozvanny’s lab relied on a different mouse to look at calcium homeostasis. The PS1 M146V knock-in expresses normal amounts of PS1 carrying the M146V mutation linked with familial AD. Bezprozvanny had previously reported that besides processing APP, presenilin 1 also acts as a relief valve for ER calcium, suggesting it was the elusive ER calcium leak channel (see Sep 2006 newsJun 2010 news). Last year, his group reported that PS1-M146V knock-in mice had elevated ER calcium (see Apr 2014 news). This heightened ER calcium turned down expression of stromal interaction molecule 2, a protein that dials up the influx of calcium into the cell Normally, STIM2 boosts the activity of calcium channels on the cell surface in response to flagging intracellular stores of the cation—a process called neuronal store operated calcium entry (nSOC, see image below). Defunct PS1 led to elevated ER calcium, which reduced STIM2 levels and thus the influx of calcium into the cell via nSOC. Ultimately, this reduced levels of phosphorylated calmodulin kinase II (pCamKII), which helps convert thin spines into the more stable mushroom variety.

OverSTIMulation? Excess Aβ42 in APPKI mice (right) ups ER calcium levels and triggers a chain of events that dismantles mushroom spines. [Image courtesy of Zhang et al., Journal of Neuroscience 2015.]

The scientists wondered if Aβ42 also affected the calcium-STIM2-nSOC pathway, and found that injecting Aβ42 oligomers into mice reduced mushroom spines through the STIM2 pathway (see Popugaeva et al., 2015). With the APP-KI mouse newly available, first author Hua Zhang and colleagues wanted to see if the same STIM2 calcium pathway faltered in mice due to aberrant Aβ processing. As they had done with the PS1 KI, they crossed APP-NL-F knock-in mice with animals expressing green fluorescent protein to allow for visualization of dendritic spines using confocal microscopy. 

Compared with wild-type mice expressing GFP, at three months of age the APP-KI-GFP mice had a slightly lower proportion of mushroom spines, and higher fractions of thin and stubby spines. These differences were exacerbated by 6 months, when the fraction of mushroom spines in WT-GFP mice increased to 35 percent, while it stayed at 22 percent in APP-KI-GFP mice.

The researchers next sought to determine whether Aβ42 played a role in mushroom spine loss. Hippocampal neuron cultures generated from APP-KI mice contained elevated levels of Aβ42, and the neurons amassed fewer mushroom spines than cultures from wild-type mice did. When the researchers bathed WT neurons in culture medium from APP-KI neurons, their mushroom spines shrank to levels seen in the APP-KI neurons. Conversely, when the APP-KI culture medium was regularly refreshed with medium from WT neurons, mushroom spines sprouted to WT levels. Together, these results hinted that extracellular Aβ was responsible for the loss of mushroom spines.

Through a series of experiments similar to those conducted in PS1 knock-in mice, the researchers determined that APP-KI mice had abnormally high levels of ER calcium in their neurons. This reduced expression of STIM2, which prevented nSOC-mediated calcium influx and slashed pCamKII levels by more than half. Ergo, fewer mushroom spines.

Unlike PS1-KI mice, in which a malfunction in PS1’s calcium leak function triggered ER calcium overload, the cause of this imbalance in the APP-KI mice was unclear. The researchers ultimately pinned it on overstimulation of the metabotropic glutamate receptor mGluR5. Excessive Aβ42 kicked ER calcium levels into overdrive through this receptor. Blocking mGluR5 normalized ER calcium stores, STIM2 expression, and nSOC calcium influx. Overexpression of STIM2 in APP-KI cultured neurons circumvented the ER calcium overload altogether—rescuing nSOC calcium influx as well as mushroom spines. The same was true when researchers delivered excess STIM2 to the brain: 6-month old APP-KI mice that had been injected with STIM2 viral vectors at 2 months of age had normal proportions of mushroom spines.  

The results indicate that whether starting with a mutation in PS1 or in APP, mechanisms that lead to mushroom spine loss converge on calcium dysregulation, Bezprozvanny said. Given that previous work from his lab also found reduced levels of STIM2 in brains of sporadic AD patients, the pathway could represent a general mechanism in AD, and thus a broad therapeutic target, he said.

Grace Stutzmann of Rosalind Franklin University in Chicago agreed. “Importantly, the proposed pathway incorporates many of the key suspected features of AD (e.g., Aβ42, ER calcium dyshomeostasis, glutamate signaling) and thus provides a means to accommodate seemingly disparate proposals describing synaptic loss into a linear sequence of events culminating in impaired memory function,” she wrote to Alzforum.

Gregory Cole of the University of California, Los Angeles commented that the APP-KI mouse served as a useful tool to confirm the presence of the STIM2/mushroom spine pathway without overexpressing APP. “This provides an elegant in vivo system to confirm in vitro evidence from electrophysiological studies that simply added Aβ42 to neurons, and implicated mGluR5 as an Aβ oligomer target relevant to synaptic deficits,” he wrote to Alzforum.

However, several commentators pointed out that unlike people with AD, the APP-KI mice never develop full-blown neurodegenerative disease. “What these mice lack are tau inclusions and extensive neurodegeneration,” commented Michel Goedert of MRC Laboratory of Molecular Biology in Cambridge, England. The link between tau aggregation, cognitive impairment, and neurodegeneration is a well-established hallmark of AD, he added.

Saido was quick to make the same point. However, he added that the mice do decline on some cognitive tests by 18 months of age, and on other tests by 12 months. He speculated that loss of mushroom spines, hence impaired synaptic function, could contribute to this decline, even though there is a lag time of six months between the onset of the mushroom spine and the cognitive phenotype. Bezprozvanny told Alzforum that his lab is currently exploring whether problems in the nSOC-STIM2 pathway underlie this later cognitive decline in APP-KI mice.

What of the dozens of other labs working with these mice? Several researchers responded to a query from Alzforum with news that their work is still in its earliest stages, and thus too preliminary to share. However, a new paper published by Bart De Strooper and Amantha Thathiah at KU Leuven in Belgium featured the mice. The researchers reported that loss of the G-protein coupled receptor GPR3 keeps amyloid production in check by reducing Aβ production. Thathiah had initially reported this finding using APP/PS1 mice (see Feb 2009 news). 

“We plan to predominantly use APP-KI strains in our future experiments,” Thathiah told Alzforum. “At the moment, they are the best AD model we have.”

Saido expects more to come from APP-KI in the future. While the mice do not develop profound neurodegeneration in their lifetime, their moderate pace of disease progression may be more akin to the human situation, he said. “Neurons are very tough cells, which could be why it takes many years to develop disease,” he added. “That means we have a large window for prevention. That’s the hopeful part.”—Jessica Shugart


  1. The APP-KI mouse model is valuable in that it avoids the impact of chronic high overexpression of APP that signals through sAPPα and a multitude of APP binding proteins, most obviously those that interact with the C terminal cytoplasmic domain when it is intact as well as when it is released as a γ-secretase fragment. This multitude of confounding signal transduction events generated by APP overexpression was deliberately circumvented in the APP-KI model, which uses multiple mutations rather than APP overexpression to achieve pathogenic levels of Aβ42. Because the knock-in retains the natural regulation, it can elevate APP expression in feedback loops that may upregulate APP focally in response to inflammation or other events, but it avoids the consequences of tonic abnormal APP overexpression. PS1 mutations, the other commonly used tool to increase pathological Aβ42, also has consequences for many substrates that will not be seen in other forms of FAD or sporadic AD. Since PS1 mutations are sufficient to cause changes in calcium signaling, the use of PS1 mutants to study events in spines downstream from aberrant calcium regulation has been problematic. Hence, Zhang and colleagues have very reasonably employed the APP KI model to confirm the presence of their STIM2 neuronal calcium entry-mediated, mushroom spine-loss pathway in a model with Aβ42 elevation in the absence of mutant PS1 or APP overexpression.

    This provides an elegant in vivo system to confirm in vitro evidence from electrophysiological studies that simply added Aβ42 to neurons and implicated mGluR5 as an Aβ oligomer target relevant to synaptic deficits. Since Aβ42 does many things to glia and other targets and precipitates tauopathy, the clinical impact of simply targeting mGluR5 is open to question. That said, elevated calcium in vivo in other AD models has been beautifully demonstrated by Backsai and colleagues and it is hard to believe that it wouldn't be a problem for normal neuronal function.  Since the elevated calcium in neurons is manifested in a kind of subclinical excitotoxicity and there is some evidence for a benefit from Levetiracetam/Keppra, this paper points to one reasonable approach to try to go upstream and prevent the chronic aberrant elevation of neuronal calcium influx.

    Seeing the advantages of the APP-KI model, we got permission to use Saido's APP-KI mice several years ago. However, due to difficulty of NIA funding in recent years, we have not had resources to develop a colony.

  2. In recent years, the AD field has been undergoing a bit of a paradigm shift in attempts to understand mechanisms of memory loss, with an increased focus on synaptic structure and function. This study by Zhang et al., and their earlier Neuron paper (Sun et al., 2014), provide a much-needed, step-by-step walk through the mechanisms by which the memory-supporting mushroom spines are lost in AD brains. Importantly, the proposed pathway incorporates many of the key suspects of AD pathology (e.g., Aβ42, ER calcium dyshomeostasis, and glutamate signaling) and thus provides a means to accommodate seemingly disparate theories of synaptic loss into a linear sequence of events culminating in impaired memory function. I look forward to the next study!   

  3. Postsynaptic dendritic spines are believed to play an important role in learning and memory. They are classified into mushroom, thin, and stubby spines, based on morphology. Of these, mushroom spines are believed to be stable “memory spines.” Persistent activation of Ca2+/calmodulin-dependent protein kinase II is needed for the long-term stability of mushroom spines. It requires neuronal store-operated Ca2+ entry (nSOC), which is gated by stromal interaction molecule 2 (STIM2). This study shows that hippocampal mushroom spines were reduced in an amyloid precursor protein knock-in model of Aβ deposition (the APP-NL-F line). This line has the advantage that APP is not overexpressed, yet Aβ42 production is increased and Aβ deposits form. The loss of mushroom spines was caused by dysregulated Ca2+ homoeostasis, consequent to overstimulation of mGluR5 receptors by Aβ42. Similar findings had previously been obtained in other mouse models.

    However, the behavioral changes were always subtle. Do these animals develop Alzheimer’s disease? Would a human being with this pathology and these symptoms be diagnosed with Alzheimer’s disease? Probably not. What these mice lack are tau inclusions and extensive neurodegeneration. The association between cognitive impairment and tau deposits is well established; it has been shown that tau aggregation is a key mediator of neurodegeneration. It will be interesting to see if the APP NL-F line will help us to understand the largely enigmatic connection between the assemblies of Aβ and tau.

  4. The idea of upregulating STIM2 to reduce calcium and thereby treat AD is interesting, but problematical with regard to how to do it practically. Targeting the downstream protease activation may achieve similar outcomes in a more translatable way. Several years ago, Arancio and colleagues showed that E64, which inhibits cysteine proteases including calpains, improved memory and long-term potentiation deficits in APPswe-overexpressing mice. In hippocampal cells, Halpain and colleagues showed that CA074Me, which inhibits the cysteine proteases cathepsin B and L, prevented NMDA induced dendritic loss. It would be interesting to see if E64 and CA-074Me treatment also improves mushroom budding in the APP KI mice.


    . Inhibition of calpains improves memory and synaptic transmission in a mouse model of Alzheimer disease. J Clin Invest. 2008 Aug;118(8):2796-807. PubMed.

    . Cathepsin B-like proteolysis and MARCKS degradation in sub-lethal NMDA-induced collapse of dendritic spines. Neuropharmacology. 2004 Oct;47(5):706-13. PubMed.

  5. While the APP-KI model offers an invaluable lens for elucidating insights on how amyloid influences neurons and disease cascades downstream of amyloid, an important objective is the identification of candidate therapeutics that modify disease progression during preclinical stages, possibly before significant amyloid deposits are observable—an endeavor that would be facilitated by characterizing changes to neurons, most likely associated with aging, that influence malformed protein quantities in aging brains.

    A challenge that continues to complicate drug development for targeting the preclinical progression of disease is integrating epidemiological evidence into AD-modeling paradigms (to model the possible influence of aberrations in metabolism and decreased mitochondrial capacity associated with aging, for example); incorporating the cellular processes associated with aging that may influence malformed proteins into animal models (to better observe how an aging cell influences amyloid/tau, rather than how amyloid/tau influences a cell); and then evaluating candidate therapeutics that abrogate processes that influence levels of malformed proteins utilizing animal models that capture the influences of aging on disease.

Make a Comment

To make a comment you must login or register.


Webinar Citations

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

Research Models Citations

  1. APP NL-F Knock-in
  2. APP NL-G-F Knock-in
  3. PSEN1(M146V) Knock-In
  4. APPPS1

News Citations

  1. Presenilins Open Escape Hatch for ER Calcium
  2. Perplexing Presenilins: New Evidence for Calcium Leak Channels
  3. Calcium Sensor STIM2 Maintains Synapses, Ebbs in Alzheimer’s
  4. Big Haul? A G Protein-coupled Receptor Regulates Aβ Production

Paper Citations

  1. . STIM2 protects hippocampal mushroom spines from amyloid synaptotoxicity. Mol Neurodegener. 2015 Aug 15;10:37. PubMed.

Further Reading


  1. . Presenilins function in ER calcium leak and Alzheimer's disease pathogenesis. Cell Calcium. 2011 Sep;50(3):303-9. PubMed.

Primary Papers

  1. . 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.