Sometimes, the first sign of Alzheimer’s disease is subtle and may appear decades before a diagnosis—such as in the form of irregular behaviors caused by very early signs of brain dysfunction. Although scientists have tried to identify and measure these minimal behavioral changes in the past, it has not even been possible in mice models studying Alzheimer’s.
In a study posted online in Cell Reports, a team of scientists at Gladstone Institutes used a new video-based machine learning tool to focus on otherwise ‘hidden’ indicators of early disease in mice engineered to mimic aspects of Alzheimer’s. The work also illuminates a new strategy for identifying neurological disease earlier and monitoring how it develops over time.
Gladstone investigator Jorge Palop, Ph.D., senior author of the study, says: “We’ve shown the potential of machine learning to change the way we examine behaviors of brain function that suggest early abnormalities. It’s a valuable tool that opens the door to more completely understanding devastating brain disorders and where they start.”
To study video footage of mice exploring an open arena, the scientists used a machine learning platform called VAME — short for Variational Animal Motion Embedding. The open-source tool picked up subtle behavioral patterns captured on camera — changes that wouldn’t pop out by just eyeballing the mice.
Tracking the behavior:
Unlike traditional behavioral tests in mice that rely on preconceived tasks that the animals are forced to perform, VAME’s deep learning platform is very different. The team with VAME to simulate Alzheimer’s symptoms applied two different types of mice in the Gladstone study. Both models showed that the machine learning tool was able to properly identify a dramatically increased level of ‘disorganized behavior’ in the mice as they aged. For instance, the mice exhibited strange behaviors such as frequently shifting between activities they should not have been able to do, things that could be tied to memory and attention deficits.
“VAME analysis only requires smartphone-quality video and similar machine learning approaches could be used someday to study these spontaneous behaviors in humans as a way to diagnose neurological disease early,” Miller says. “The technology will be used, I hope, in the clinic and in the patients’ homes to evaluate them.”
“With this, we can give scientists and doctors a way to solve the very hard problem of diagnosing preclinical stages of disease,” he added.
Several years ago, VAME’s technology was still in its infancy, which is why Miller started experimenting with the technology. The platform was first developed by the team of Stefan Remy, MD, in Germany, and Miller and Palop worked with them. In a study published in Communications Biology, they helped to show VAME’s utility for neuroscience research.
The Gladstone team added an extra dimension to their new study by using VAME to see if a drug being tested as a potential Alzheimer’s intervention could stop the disorganized behavior in mice.
Prior research from Gladstone investigator Katerina Akassoglou, Ph.D., found that a blood clot’s protein, fibrin, is toxic to the brain when it enters it through damaged blood vessels — a pathway the scientists exploited in their experiment. Akassoglou’s lab has blocked the fibrin’s toxicity to prevent neurodegeneration and to protect from Alzheimer’s in animals.
The team then genetically blocked fibrin from inducing toxic inflammation in the brain to see if this therapeutic strategy could shield mice from Alzheimer-like behavior. The Alzheimer’s mice didn’t develop abnormal behaviors as much. But the fact that blocking fibrin’s inflammatory activity in the brain reduced almost all the spontaneous behavioral changes in Alzheimer’s mice ‘reinforces the idea that fibrin and the subsequent neuroinflammation are key drivers of the disease,’ Akassoglou adds, who is also an author of the study.
Aged AppNL-G-F mice exhibited mild spatial memory deficits and severe amyloidosis and gliosis. In the Morris water maze test, 22-month-old AppNL-G-F mice showed significant impairments in distance (p = 0.025) and latency (p = 0.017) compared to wild-type (WT) littermates, indicating learning and memory deficits. Spontaneous behavior analysis using VAME revealed significant age-dependent alterations, with 8 of 30 identified motifs showing abnormalities in 13-month-old AppNL-G-F mice. These mice displayed disrupted motif transitions and increased randomness in behavioral sequences. Blocking fibrinogen-induced neuroinflammation reduced behavioral alterations in Alzheimer’s models, validating fibrinogen as a therapeutic target and highlighting VAME’s precision in detecting subtle disease-associated changes.
“I do think that machine learning will eventually come to be a very useful clinical tool, and I think it can even provide an unbiased way of testing new treatments in the lab.” With other Gladstone teams that study neurological disease, Palop and Miller are now using the VAME technology to apply this technology for novel behavioral studies.
Miller says the tool and similar approaches should become more available for biologists and clinicians to speed up the process of developing powerful new medicines.
Reference: Miller SR, Luxem K, Lauderdale K, et al. Machine learning reveals prominent spontaneous behavioral changes and treatment efficacy in humanized and transgenic Alzheimer’s disease models. Cell Rep. 2024;43(11):114870. doi:10.1016/j.celrep.2024.114870
