Scientists agree that mouse models of Alzheimer’s only partially recapitulate the disease and, more ominously, that their strong overexpression of human AD genes renders them prone to artifacts. In 2014, the group of Takaomi Saido of RIKEN Brain Science Institute, Wako, Japan, generated three strains of mice with mutant human APP knocked into the endogenous locus, resulting in normal expression levels of the protein. Since then, Saido has lobbied the field to study his mice and today, nearly 200 laboratories around the world do. So what do scientists know by now? Do these mice better model Alzheimer’s disease? What are their limitations?
A symposium at the Society for Neuroscience annual meeting, held November 12-16 in San Diego, tackled these questions. Scientists who work with the mice presented their latest data, other groups who use them were in the audience, and an enthusiastic crowd made for lively debate during a panel discussion. Overall, researchers remained divided for the time being on whether the knock-ins model AD more faithfully than do overexpression mice, especially on aspects where data differ between the two. Some investigators consider the very mild behavioral phenotype of the knock-ins to be a limitation, while others complained that their pathology develops slowly, hence experiments take longer. However, most scientists agreed that the knock-ins represent a valuable alternative to current models and that their increasing use will settle these open questions.
Amyloid Unleashed. Three month-old APP knock-in mice (left) have no amyloid deposits (green) in the cerebral cortex. Without the amyloid protease kallikrein 7 (right), they do, and gliosis to boot (red). [Courtesy of Taisuke Tomita.]
Pros and Cons of Knock-In Mice
What’s wrong with overexpression models? While they have generated many insights into the disease, researchers worry that insertion of the transgenes disrupts endogenous genes, causing artifacts. Some models can only be maintained on specific genetic backgrounds, or produce unexpected results when crossed with other transgenics. Overexpression of APP may lead to ER stress and excessive production of toxic APP fragments other than Aβ, such as β-CTF.
Saido believes his knock-in APPNL-F mice, which carry the Swedish and Iberian mutations, and APPNL-G-F mice, which add the Arctic mutation, overcome these problems (see Apr 2014 webinar). The knock-ins live a normal lifespan but develop early amyloidosis and inflammation. Memory problems crop up at 18 months in APPNL-F and at six months in APPNL-G-F, which accumulates amyloid faster. A third model, APPNL, carries only the Swedish mutation and has mild, slowly developing pathology.
Several presenters discussed how the mice have performed in their hands. For example, Taisuke Tomita of the University of Tokyo had in vitro evidence that the serine protease kallikrein 7 (KLK7) degrades monomeric and fibrillar Aβ. Tomita generated KLK7 knockout mice, and found they accumulated 50 percent more soluble Aβ40 and Aβ42 than wild-types. Crossing them to APPNL-G-F knock-ins, he saw a sixfold increase in amyloid plaques, 20 percent more tau phosphorylation, and earlier astrogliosis than in the knock-in parents (see image above). The findings establish a role for KLK7 in controlling pathology, Tomita concluded. The knock-in mice provided a good model for this research because they develop robust plaques and neuroinflammation early in life, Tomita told Alzforum. In particular, amyloid deposition varied less among knock-in littermates than it does among overexpression littermates. Because the knock-ins can be maintained as homozygotes, crosses and genome editing are easier to perform, he added. He also found the mice easier to handle. They jump and try to bite their handlers less, and are generally more docile than overexpression models, he said.
Amantha Thathiah, now at the University of Pittsburgh, compared findings from APP/PS1 and APPDutch mice to APPNL and APPNL-F knock-ins. In collaboration with Bart De Strooper at KU Leuven, Belgium, Thathiah had previously reported that in all these models, lowering expression of the G-protein coupled receptor Gpr3, which interacts with γ-secretase, dampened Aβ production (see Oct 2015 news). At SfN, she noted that she saw a more dramatic drop in absolute soluble Aβ levels in the APP/PS1 mice than in knock-ins. This might be expected, since the overexpression model produces more Aβ. On the other hand, in the APP/PS1 mice, decreases in the Aβ42/Aβ40 ratio fell short of statistical significance, whereas in the knock-ins, which have a higher baseline Aβ42/Aβ40 ratio due to the presence of the Iberian mutation, that decrease was clear. “Each model mimics a slightly different aspect of the disease,” Thathiah wrote to Alzforum. “Although the APP knock-in mouse models have late-developing, mild phenotypes that make them less practical for some studies, they avoid problems associated with overexpression and would be useful for testing potential preclinical AD therapies.”
Cognitive Effects Subtle, but Present
At the SfN session, researchers complained most loudly about the lack of a robust behavioral phenotype in the knock-ins. Without behavioral data, papers are more difficult to publish, noted Ilya Bezprozvanny of the University of Texas Southwestern Medical Center, Dallas. Saido said the mice appear to model the early, preclinical stages of AD, when amyloid accumulates and inflammation burgeons, but only subtle cognitive problems are apparent. In people, this stage can last up to 20 years before notable neurodegeneration occurs. The APP knock-in mice, which live only two years, may never reach that later stage.
Nonetheless, the knock-in mice do develop subtle behavioral defects. In a poster, Amira Latif-Hernandez, working with De Strooper and Rudi D’Hooge at KU Leuven, detailed synaptic and memory impairments in these animals. She tested APPNL mice and APPNL-G-F mice at three and six months of age. The APPNL-G-F mice spent more time in open arms of a maze than did the APPNL mice, particularly as they aged, indicating they were less anxious. Learning and memory seemed similar in both models, as they adapted to changes in the location of a hidden platform in the Morris water maze equally well. By six months, however, APPNL-G-F mice froze less often than APPNL mice in a location where they had previously received a shock, suggesting some hippocampal and amygdala deficits, Latif-Hernandez noted.
Corresponding differences cropped up in electrophysiology. Hippocampal slices from six-month-old APPNL-G-F mice showed weaker long-term potentiation than APPNL mouse slices, but similar basal transmission and long-term depression. In slices from the prefrontal cortex, both basal transmission and LTP faltered in the six-month-old APPNL-G-F animals. The mouse scanner revealed increased connectivity of the frontal network in three-month-old APPNL-G-F mice relative to APPNL mice, but no difference at six months. Resting-state functional MRI experiments in older mice are ongoing, Latif-Hernandez reported. “We conclude that an increased Aβ42/Aβ40 ratio in APPNL-G-F mice at six months drives synaptotoxicity, which in turn causes increased deficits in specific behavioral domains,” Latif-Hernandez wrote.
The Best Model? The Answer Is in the Works
How common will use of the knock-ins become? It depends on the lab. Some, such as De Strooper's, are switching over to knock-in models exclusively. Others switch over partially. For example, Bezprozvanny used the knock-ins to investigate disruptions in calcium signaling and plans to do future signaling studies in them, too, though for now he will stick to overexpression models for behavioral data (see Oct 2015 news). Bezprozvanny believes the knock-ins recapitulate Alzheimer’s pathology more faithfully than do overexpression models. He is currently repeating older experiments done with overexpression lines in the knock-ins, but told Alzforum he has not yet found discrepant data.
Conflicting data has turned up in other experiments, however. Saito recently reported two findings from overexpression models that failed to reproduce in the knock-ins: a drop in Nav1.1 sodium channels, and a rise in the CDK5 activator p25 (see Sep 2016 news). At SfN, researchers debated the meaning of the findings, but did not reach consensus. Some argued that the knock-in mice themselves might produce artifacts. Robert Vassar of Northwestern University, Chicago, said that combining several familial mutations is artificial and could cause effects not seen in familial or sporadic AD. The Arctic mutation lies in the Aβ sequence and is known to alter how the peptide aggregates. Lennart Mucke of the Gladstone Institute of Neurological Disease, San Francisco, suggested that overexpression models may represent some aspects of Alzheimer’s disease more faithfully than do the knock-ins. For example, high levels of APP may accumulate at synapses in Alzheimer’s patients, as they do in Tg2576 mice, he said (see Tiwari et al., 2016).
In the end, researchers agreed that whether the new models are better than overexpression transgenics will become clear as more data on the knock-ins come online. “The jury is still out,” David Brody of Washington University, St. Louis, wrote to Alzforum. Mucke summed up the mood when he quoted the statistician George Box: “All models are wrong, but some are useful.” To derive the fullest picture, researchers need to compare several models, with the ultimate validation being human data, Mucke argued.
De Strooper concurred that to be considered truly robust, findings should be consistent across multiple approaches. Overexpression may remain useful for some studies, De Strooper said, especially behavioral ones, but he believes the proper control for those would be mice that overexpress wild-type human protein. Vassar agreed that knock-ins are an additional tool, writing to Alzforum, “The APP knock-in mice are an advance, but all animal models have limitations. We must keep ourselves constantly aware of these limitations and include rigorous controls in our experiments so that we can correctly interpret our results.”
Some researchers believe that adding normal levels of human tau to the mice might improve the model; for more on that, see next SfN story.—Madolyn Bowman Rogers
Research Models Citations
- G-Protein Receptor Knockout Rescues Several Models of Alzheimer’s
- In APP Knock-Ins, Calcium Chaos Dismantles Mushroom Spines
- Do APP Knock-ins Call Overexpression Models of AD into Question?
- Next-Generation Mouse Models: Tau Knock-ins and Human Chimeras
- Tiwari SS, Mizuno K, Ghosh A, Aziz W, Troakes C, Daoud J, Golash V, Noble W, Hortobágyi T, Giese KP. 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.