Distributed cortico-striatal computations underlying prediction learning in Pavlovian Markov processes

• Applicants: Professor Dr. Wolfgang Kelsch; Dr. Jonathan Reinwald
• Subject Area: Experimental and Theoretical Network Neuroscience ; Cognitive, Systems and Behavioural Neurobiology
• Term: since 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 544257695

In many natural scenarios, temporal or spatial sequences of cues can be used to find a good or predict an outcome. The formation of such higher-order predictions is critical to learning associations along these chains of cues and, thereby, making accurate predictions about the world. Markov chains, where a defined set of probabilities regulates the sequential transition between states, provide a formal description of such real-world processes. In the brain, predictions are assumed to be produced by the functional interaction of distributed cortical and mesolimbic regions. We aim here to increase the understanding of how distributed cortical and mesolimbic circuits interact to form predictions in higher-order Pavlovian conditioning. Such inter-regional interactions have been explored only in a few cases during prediction coding and Pavlovian learning of cue sequences. In preliminary work to this proposal, we established a Pavlovian Markov chain in mice to study the neuronal mechanisms underlying the learning of predictions using behavioral fMRI and multisite recordings. We propose to examine first the natural interactions within mesolimbic regions that evolve during learning of value predictions and error signals. These interactions will then be tested causally by subpopulation-selective optogenetic silencing. We will particularly determine here how different components of prediction coding expressed in ventral striatal sub-circuits contribute to dopamine neuron signaling. These ventral striatal sub-circuits, in turn, are assumed to produce predictions with the contribution of higher cortices. Their exact intra-cortical interaction and their specific influence on ventral striatum prediction coding are however not entirely understood. To address this, we will perform simultaneous recordings from multiple cortical regions recently identified by us with fMRI uring the Pavlovian Markov chain, and examine their natural interaction patterns. Based on the reconstruction of these interregional interactions, we will causally test by projection- and cue-selective silencing, what information cortical regions feed into ventral striatal sub-circuits. Importantly, the probability of transitioning from one cue to the next is independent of previous transitions in such first-order Markov chain processes. In many real-life scenarios, however, the probability of future outcomes depends on the conditional occurrence of multiple cues. Higher-order Markov chains model this latter scenario. We will leverage different types of Markov chains to delineate how far the prediction coding processes generalize. Together, this work shall contribute to a better understanding of fundamental neuronal coding mechanisms of higher-order association learning producing complex predictions. This shall inform, through mechanistic insight, about processes that, when dysfunctional, are thought to give rise to psychosis and addictive behaviors.

• DFG Programme: Research Grants