Neurobehavioral phenotyping of AD model mice using Motion Sequencing

Abstract Alzheimer's disease (AD) is caused by progressive changes in neural circuits that culminate in memory loss, confusion, difficulty completing tasks, withdrawal, mood changes and ultimately death. Alterations in body movement — such as slowed gait, and difficulty in avoiding obstacles — have been associated with AD, are predictive of AD, and often appear in the pre-clinical stage, before cognitive changes are apparent. These observations raise important questions about how AD targets the cognitive and motor systems that support movement and/or action selection; addressing these questions in turn requires a clear view of how AD pathophysiology influences behavior both early and late in disease, and particularly how these changes in behavior are distinguished from the many motor-related changes apparent during normal aging. However, to date behavioral analysis of movement in AD mouse models have not yet yielded a clear and consistent view of how AD affects the neural circuits responsible for selecting, composing, sequencing and implementing ongoing behaviors. At least in part this failure reflects the methods used to characterize behavior in mouse models, which depend upon a set of reductionist assays that capture limited aspects of a mouse's overall behavioral comportment within a given experiment; as a consequence we do not know whether different AD models share core movement phenotypes, nor do we understand whether or how AD targets the cortico-striatal circuits that create the coherent, moment-to-moment patterns of action used by mice to interact with the world. Our laboratory has recently developed a novel behavioral characterization technique, based upon 3D machine vision and unsupervised machine learning techniques, called Motion Sequencing (MoSeq). MoSeq automatically and without human supervision identifies the behavioral modules (“syllables” e.g., a left turn, the first half of a rear, etc.) out of which spontaneous and self-directed behavior is composed, as well as the statistical rules governing the sequencing of these syllables (“grammar”). We have previously demonstrated that the dorsolateral striatum (DLS) contains explicit neural correlates for both syllables and grammar, and that the DLS is causally required to assemble syllables into meaningful and adaptive sequences. Here we propose to use MoSeq to characterize behavioral phenotypes expressed by a variety of AD mouse models, to perform joint neural-behavioral recordings to probe circuit mechanisms that underlie these observed phenotypes and, finally, to develop MoSeq into a broadly-applicable platform for studying the movement-related signatures of cognition. Taken together, these experiments promise to revitalize the study of behavior and neuro-behavioral relationships in pre-clinical models of AD, and to reveal key mechanisms that tie together AD-related genetic lesions, neural circuit function, and ongoing naturalistic patterns of action.