Throughout our life we learn to perform a variety of different actions to impact the world around us, and to obtain the outcomes we want. Many of these actions are learned by associating specific stimuli to particular responses, thereby building sensorimotor repertoires, where the occurrence of the stimulus leads to the execution of the action. But many other actions are self-paced, and learned/shaped based on their consequences. Such actions are initiated and performed in a spontaneous manner, and if some desirable outcome (a reward, a stimulus, even another action) occurs as a consequence, we learn to do them again, and to shape their execution to achieve that goal. This process of learning from consequence could be viewed conceptually as a motorsensory process, which is impaired in conditions affecting self-paced action generation, such as Parkinson’s disease, and action control, such obsessive-compulsive disorder.
The efforts of our laboratory have been dedicated to understanding the circuits and the mechanisms underlying the generation of novel self-paced actions, and the shaping of these actions into complex action repertoires based on the consequences of their execution. Understanding this process, through trial and feedback, requires mechanistic insight into how self-paced actions are initiated, how they can be selected/initiated again, how feedback can shape their execution and organization, and what drives the actions to be performed in a particular situation.
To achieve this understanding, we carry out experiments that address four distinct but related questions through four subgoals:
1. Action initiation/generation: What mechanisms underlie the self-paced initiation of actions and the self-paced generation of diverse actions in a particular context?
2. Action shaping and automatization: What mechanisms underlie the changes in how actions are executed through iterative processes of repetition of action and consequence (trial and feedback)? This implies understanding how particular parameters of action execution can be shaped through feedback to more accurately or more rapidly reach a particular goal. Because most action repertoires we learn to execute are sequences of individual elements, understanding how actions are shaped implies understanding how actions are organized into complex and precise sequences.
3. Action selection: What are the neural bases underlying the choice of which actions to perform or not perform in a given context (and when to perform them), based on the previous experience of the consequences of each action in that particular context? This requires understanding how credit is assigned to a particular action for a given consequence, and how that action is selected again in lieu of others.
4. Action control: What mechanisms determine why actions are performed in particular states, i.e. are actions performed to obtain a particular goal (intentional, goal-directed or model-based), or are they triggered by a particular context or state (feedforward stimulus-response or habits, model-free)?
A growing body of evidence from our laboratory and others suggests that cortico-basal ganglia circuits are involved in action generation and selection, in skill learning, and in learning goal-directed actions and habits. There is also a large body of evidence that the neurotransmitter dopamine plays a key role in regulating basal ganglia function to control action generation and reward learning. Therefore, we investigate the cortico-basal ganglia mechanisms underlying the processes described in the different subgoals above, by using an across-level approach, from behavior, to circuits, to molecules. We mostly use mice in these studies because they combine the power of genetics in a mammalian species with relatively conserved cortico-basal ganglia loops, that can generate and propagate oscillatory activity, with the possibility of accurately quantifying simple behaviors such as action initiation (using electromyographic [EMG] recordings or inertial sensors), as well as the possibility of learning complex skills and goal-directed planning tasks. Our results have demonstrated key mechanisms by which dopamine and basal ganglia modulate action generation and action learning, and invite a reassessment of the classic model of dopamine and basal ganglia function.