Misclassification in memory modification in App^NL-G-F knock-in mouse model of Alzheimer’s disease

Alzheimer’s disease (AD) is a progressive neurodegenerative disease, which is the leading cause of dementia. In its early stages, amyloid-ꞵ (Aβ) plaque accumulation in the brain may occur for several years without overt cognitive symptoms. While advances in neuroimaging and biochemical tests have improved early diagnosis, a mechanistic gap remains between neuropathologies and cognitive symptoms. Investigating the behavioral changes in early-stage AD from the perspective of an internal model would bridge this gap by explaining which computations are impaired by pathological traits leading to cognitive symptoms. However, how internal states of the model are altered in AD progression and contribute to cognitive symptoms remains unclear. Understanding such processes may help identify AD-related changes before the onset of dementia and allow the development of more precise assessment methods to support early intervention.

Previously, Gershman et al. (2017)1 proposed the latent cause model, which provides a normative account of memory modification phenomena in Pavlovian fear conditioning by describing how new observations are classified into old memories or a new one. Given that memory impairment is the predominant early symptom in AD, estimating such internal states could provide insights into the cognitive alterations that may underlie memory-related symptoms. We hypothesized that the internal state of this model is altered from the early disease stage, where Aβ plaques have already accumulated. We used App^NL-G-F knock-in mice2 as a model of early-stage AD and estimated their internal states using the latent cause model. App^NL-G-F knock-in mice, along with age-matched controls, underwent memory modification learning, which consisted of fear conditioning, extinction, and reinstatement.

App^NL-G-F mice successfully formed fear associative memory and its extinction memory, but exhibited a lower extent of fear memory reinstatement after being given an unsignaled shock. Computational modeling suggested that their internal states were biased toward overgeneralization when facing similar stimuli with different outcomes; overdifferentiation when facing dissimilar stimuli with the same outcomes. The altered internal states of App^NL-G-F mice illustrated misclassification during the memory modification.

Investigating internal cognitive states provides a new approach for understanding how AD pathology alters behavior. This would advance precision medicine by supporting early detection and personalized cognitive assessment. This may benefit individuals at risk for AD and medical professionals. However, relationships among the internal, pathological, and behavioral states should be carefully interpreted from various perspectives. In particular, further research is needed to link the internal model with the specific neural mechanisms affected by Aβ plaque accumulation.

This research was published online in the international scientific journal eLife on November 4, 2025.