An experimental and mathemetical framework with internal model update capturing a visuo-motor adaptation process in Zebrafish larvae

The ability to maintain accurate motor performance in a constantly changing world is a fundamental feature of the nervous system, a process known as motor adaptation. This thesis investigates the long-term dynamics of this process using larval zebrafish as model organism. Through a behavioral analysis based on the optomotor response (OMR), this work characterizes the animal's response to a sustained sensorimotor perturbation. In response to a reduction in visual reafference gain, I found a bimodal response, with larvae exhibiting an immediate and robust increase in motor energy at the beginning, followed by a slow, progressive decay to the baseline. A temporal decomposition of the swimming bouts, revealed a clear dissociation between two distinct control mechanisms operating to coordinate the single response: a fast, reactive component, in the late phase of the bout, consistent with an online feedback controller; a slow, predictive component, in the early phase of the bout, associated with a gradual recalibration of the motor program, a hallmark of the updating of a feedforward internal model. A dual-pathway mathematical model for the motor adaptation process was developed relying on a single, common prediction error signal sufficient to drive both the immediate reactive correction and the slow, integrative update of the internal model. Fitting the model to the kinematic data from individual fish, I demonstrate its ability to accurately reproduce the observed multi-timescale dynamics and to infer the evolution of characteristic latent variables such as the Internal Model (IM) and Prediction Error (PE) states. I leveraged this model to investigate alterations in motor learning in zebrafish lines carrying mutations in the high-confidence Autism Spectrum Disorder (ASD) risk genes fmr1 and nlgn2b. The analysis revealed that these mutations do not cause a uniform deficit but instead induce distinct, atypical learning strategies: nlgn2b mutants display impaired predictive learning and an over-reliance on reactive control, whereas fmr1 mutants exhibit a faster and potentially more volatile learning system. In conclusion, this thesis provides a formal, algorithmic-level description of motor adaptation, bridging behavioral observations and the underlying computational principles.