The formation of flexible memory is a key ability that allows tracking the value of a choice outcome for adaptive decision in dynamic environments. How do brain circuits compute such values? This question is central to understanding how multiple behavioral disorders develop due to the awry computation of values associated with choices, e.g. in drug addiction. In both mammals and insects, associative circuits allow the constant update of memory upon repetitive experience. Their function relies on modulatory teaching signals delivered by dopaminergic inputs, which themselves receive recurrent connections from the output of the circuit. Many questions remain regarding the mechanisms for the computation of learned value, notably about the role of recurrent connections within the circuits to continue or arrest the gating of synaptic plasticity and the molecular pathways following dopamine signals that set a specific level of plasticity. I study these questions in the associative circuit within the larval brain of Drosophila melanogaster.
My research considers two main mechanisms:
On one hand, the role of recurrent connections in adjusting dopamine-gated plasticity. The hypotheses I focus on are based on reinforcement learning theories as well as on knowledge from the larval brain connectome (entirely recontructed and published in Winding et al., Science 2023). Using advanced neurogenetics approach, I transiently manipulate dopamine reinforcing neurons as well as the neurons projecting feedback onto dopamine neurons, while testing the effect of this manipulation on fine-tuning learned behaviors and on dopamine responses to unpredicted or predicted stimuli.
On the other hand, the role of different dopamine receptors and second messengers to implement synaptic plasticity over the course of training. Here, I base my hypotheses on previous findings on the involvement of distinct types of dopamine receptors expressed in the same memory cells in setting associative memories of different kinds. I plan to use RNA interference techniques to knock down specific receptors in specific neurons and set up in vivo imaging of 2nd messengers.
Drosophila larva uniquely permits a combination of sophisticated, multi-scale, approaches. With online tracking, specific genetic targeting of neurons (to e.g. activate, silence, or image neuronal activity), computational modelling of networks, and the use of a detailed connectome as a road map, we can study the way recurrent networks implement reinforcement learning to an unprecedented level of precision.