Revealing sudden transitions from goal-directed to habitual behavior during learning in mice

The transition from goal-directed to habitual decision-making during learning is thought to be gradual, yet permanent. Current approaches for distinguishing between the two decision processes require discrete ‘test’ sessions that preclude assessment of the nature and timing of the transition. Here, we devised a naturalistic devaluation approach to assess the underlying decision mode en passant, without discrete ‘test’ sessions. We hypothesized that by shifting a need for plain water, to a preference, we would favor action rate variability in animals during the goal-directed phase, unmasking transitions towards habitual behavior, expected to be characterized by a reduction in action rate variability. Mice received ad libitum non-palatable water in the home cage but received small plain water droplets for correct actions during an auditory stimulus recognition task. These mice exhibited naturalistic fluctuations in the cue-driven response rate during discriminative instrumental training. Then, abruptly and overnight, this variability ceased, and mice transitioned to a high and stable response rate despite no measurable changes in weight or general motivation. An HMM-GLM model revealed that animal’s choices pre-transition rapidly switched between two states (Go-biased/NoGo-Biased) while choices post-transition were dominated by one persistent Go-biased state. Pre-transition performance was sensitive to outcome devaluation (a hallmark of goal-directed behavior) but insensitive afterwards (a feature of habitual behavior). The sudden transition was accompanied by signatures of automaticity in lick microstructure and pupillary dynamics. Surprisingly, some animals reverted to goal-directed mode after several sessions in habit mode suggesting that transitions to habitual decision-making are not permanent. Lesioning the dorso-lateral striatum (area essential for habit formation) impaired the appearance of a transition. Thus, this naturalistic devaluation paradigm provides a powerful en passant approach to study habit formation and shows that transitions to habitual decision-making are strikingly abrupt, but also reversible, suggesting that control of decision-making is more agile than previously thought.