Hippocampal contributions to dopamine prediction error signals

• Applicant: Dr. Kauê Machado Costa, Ph.D.
• Subject Area: Cognitive, Systems and Behavioural Neurobiology ; Molecular Biology and Physiology of Neurons and Glial Cells
• Term: from 2018 to 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 419309165

I have a strong interest in the behavioral function of midbrain dopamine (DA) neurons. The currently most influential theory states that DA neurons code for a cached-value reward prediction error (RPE) which acts as a teaching signal for reinforcement learning. However, this explains only a fraction of animal learning capabilities and there is evidence that DA support forms of learning that go beyond the purview of this model. It has recently been described that DA neurons also code prediction errors based on sensory identity (sensory prediction errors; SPEs). I will explore if the computing of different forms of prediction errors rely on separate circuit mechanisms by testing the contribution of distinct regions of the hippocampus (HPC) to phasic DA neurons activity.The HPC can be divided into the dorsal HPC (dHPC), which has been implicated in memory representation, and the ventral HPC (vHPC), which has been linked to the control of motivation. I hypothesize that each area contributes different information for RPE and SPE computation. These signals will be differentiated in rats performing a decision-making task in which reward size and identity are independently manipulated in contiguous trial blocks. Blocks in which reward number is changed allow for the recording of RPE signals, while those with identity changes allow for the recording of SPE signals. While animals perform this task, I will simultaneously record single-unit DA neuron activity and inactivate either the dHPC or vHPC using optogenetic methods.It is often thought that value and sensory information, which respectively inform RPE and SPE coding, are computationally discriminable. Following this hypothesis, we can expect that dHPC inhibition, given its role in the recall of specific sensory characteristics, will disrupt DA neuron coding of SPEs, while inhibition of the vHPC, which regulates reward valuation, will impair RPE coding. However, it is also possible that RPEs are a special instantiation of SPEs, i.e. the representation of changes in reward value reflects the presence or omission of reward-related sensory input. If this is true, then dHPC inactivation should impair DA neuron representation of both RPEs and SPEs, while vHPC inhibition should not affect either. A final possibility is that DA neurons code a multivariate prediction error, with value and sensory input being coadjuvant variables used to construct experience-specific predictions. Within this framework, dHPC inhibition might affect both SPE and RPE representations, as they would both be linked to the memory of a specific reward, while vHPC inactivation may selectively affect SPEs by changing the animal's motivational state. These experiments will inform us on the degree to which the coding of classical and heterodoxical teaching signals by DA neurons are interdependent at the circuit and computational levels, potentially uncovering fundamental learning mechanisms.

• DFG Programme: Research Fellowships
• International Connection: USA
• Host: Dr. Geoffrey Schoenbaum