Machine learning behavioral analysis reveals cervical instability as an early biomarker of Amyotrophic Lateral Sclerosis

Abstract

Early detection of neuromuscular disorders is a major clinical challenge, with most diagnoses only occurring after considerable motor neuron degeneration has already taken place. The central problem for early diagnosis of neuromuscular diseases is the subtlety of early symptoms and where to look for them. Without defined behavioral markers, the earliest stages of disease go undetected, delaying intervention and limiting neuroprotective therapeutic evaluation. Here, we present a machine learning (ML) based framework that identifies subtle postural alterations in freely behaving animals. Using longitudinal pose data from SOD1G93A mice, a widely used Amyotrophic Lateral Sclerosis (ALS) mouse model, we focused on postural states in idle periods, behavioral states usually overlooked in disease monitoring. Our analyses revealed consistent deviations in posture and a feature analysis pinpointed cervical instability during adolescence as a key distinguishing feature. We validated these findings through two independent behavioral assays engaging cervical musculature: rearing and wet-dog shakes, both of which showed significant impairments in male SOD1G93A mice as early as 3 weeks of age, many weeks earlier than conventional muscle function assays. This approach establishes an unbiased, non-invasive, scalable strategy for detecting early-stage neuromuscular dysfunction, and provides a foundation both for clinical behavioral biomarker development in ALS and related disorders and will enable evaluation of early neuroprotective interventions.