Non-stationary Markovian dynamics shape swim bias fluctuations in zebrafish larvae

Abstract

Animals are capable of internally modulating behavior as a function of internal state, environmental sensory conditions, and context. Classical models often assume that sensorimotor decisions are driven primarily by external stimuli, with individual biases treated as either constant or evolving monotonically over time. Here, we show that 5-day-old zebrafish larvae display slowly changing directional swim biases even in stable, homogeneous environments, with fluctuations unfolding over many hours. Computational modeling suggests that these biases arise from a non-stationary Markovian process, with two largely independent internal input streams modulating the tendency to repeat swim directions across consecutive swim bouts. These slow fluctuations are present across sensory conditions, including different light intensities and global motion cues, although the switching statistics are modulated by these conditions. Our findings reveal an intrinsic, context-sensitive source of behavioral variability and provide a framework for further study of computational principles that generate spontaneous and adaptive behavior.