Interactions of Medial and Lateral Prefrontal Cortex in Hierarchical Predictive Coding

Perceptual information processing in table tennis players: based on top-down hierarchical predictive coding

Long-term training induces neural plasticity in the visual cognitive processing cortex of table tennis athletes, who perform cognitive processing in a resource-conserving manner. However, further discussion is needed to determine whether the spatial processing advantage of table tennis players manifests in the early stage of sensory input or the late stage of processing. This study aims to explore the processing styles and neural activity characteristics of table tennis players during spatial cognitive processing. Spatial cognitive tasks were completed by 28 college students and 20 s-level table tennis players, and event-related potentials (ERP) data were recorded during the task. The behavioral results showed that the table tennis group performed better on the task than the college students group (control). The ERP results showed that the amplitude of the N1 component of the table tennis group was significantly lower than that of the control group. The amplitude of the P2 and P3 components of the table tennis group was higher than that of the control group. Table tennis players showed significant synergistic activity between electrodes in the β-band. The results of this study suggest that table tennis players significantly deploy attentional resources and cognitive control. Further, they employ stored motor experience to process spatial information in a hierarchical predictive coding manner.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11571-024-10171-4.

Mu and Delta Opioid Receptors Modulate Inhibition within the Prefrontal Cortex Through Dissociable Cellular and Molecular Mechanisms

Aberrant signaling within cortical inhibitory microcircuits has been identified as a common signature of neuropsychiatric disorders. Interneuron (IN) activity is precisely regulated by neuromodulatory systems that evoke widespread changes in synaptic transmission and principal cell output. Cortical interneurons express high levels of Mu and Delta opioid receptors (MOR and DOR), positioning opioid signaling as a critical regulator of inhibitory transmission. However, we lack a complete understanding of how MOR and DOR regulate prefrontal cortex (PFC) microcircuitry. Here, we combine whole-cell patch-clamp electrophysiology, optogenetics, and viral tools to provide an extensive characterization MOR and DOR regulation of inhibitory transmission. We show that DOR activation is more effective at suppressing spontaneous inhibitory transmission in the prelimbic PFC, while MOR causes a greater acute suppression of electrically-evoked GABA release. Cell type-specific optogenetics revealed that MOR and DOR differentially regulate inhibitory transmission from parvalbumin, somatostatin, cholecystokinin, and vasoactive intestinal peptide-expressing INs. Finally, we demonstrate that DOR regulates inhibitory transmission through pre- and postsynaptic modifications to IN physiology, whereas MOR function is predominantly observed in somato-dendritic or presynaptic compartments depending on cell type.

Monkey Prefrontal Cortex Learns to Minimize Sequence Prediction Error

In this study, we develop a novel recurrent neural network (RNN) model of pre-frontal cortex that predicts sensory inputs, actions, and outcomes at the next time step. Synaptic weights in the model are adjusted to minimize sequence prediction error, adapting a deep learning rule similar to those of large language models. The model, called Sequence Prediction Error Learning (SPEL), is a simple RNN that predicts world state at the next time step, but that differs from standard RNNs by using its own prediction errors from the previous state predictions as inputs to the hidden units of the network. We show that the time course of sequence prediction errors generated by the model closely matched the activity time courses of populations of neurons in macaque prefrontal cortex. Hidden units in the model responded to combinations of task variables and exhibited sensitivity to changing stimulus probability in ways that closely resembled monkey prefrontal neurons. Moreover, the model generated prolonged response times to infrequent, unexpected events as did monkeys. The results suggest that prefrontal cortex may generate internal models of the temporal structure of the world even during tasks that do not explicitly depend on temporal expectation, using a sequence prediction error minimization learning rule to do so. As such, the SPEL model provides a unified, general-purpose theoretical framework for modeling the lateral prefrontal cortex.

Non-overlapping sets of neurons encode behavioral response determinants across different tasks in the posterior medial prefrontal cortex

Higher mammals are able to simultaneously learn and perform a wide array of complex behaviors, which raises questions about how the neural representations of multiple tasks coexist within the same neural network. Do neurons play invariant roles across different tasks? Alternatively, do the same neurons play different roles in different tasks? To address these questions, we examined neuronal activity in the posterior medial prefrontal cortex of primates while they were performing two versions of arm-reaching tasks that required the selection of multiple behavioral tactics (i.e., the internal protocol of action selection), a critical requirement for the activation of this area. During the performance of these tasks, neurons in the pmPFC exhibited selective activity for the tactics, visuospatial information, action, or their combination. Surprisingly, in 82% of the tactics-selective neurons, the selective activity appeared in a particular task but not in both. Such task-specific neuronal representation appeared in 72% of the action-selective neurons. In addition, 95% of the neurons representing visuospatial information showed such activity exclusively in one task but not in both. Our findings indicate that the same neurons can play different roles across different tasks even though the tasks require common information, supporting the latter hypothesis.

Dynamic Causal Modelling of Hierarchical Planning

Hierarchical planning (HP) is a strategy that optimizes the planning by storing the steps towards the goal (lower-level planning) into subgoals (higher-level planning). In the framework of model-based reinforcement learning, HP requires the computation through the transition value between higher-level hierarchies. Previous study identified the dmPFC, PMC and SPL were involved in the computation process of HP respectively. However, it is still unclear about how these regions interaction with each other to support the computation in HP, which could deepen our understanding about the implementation of plan algorithm in hierarchical environment. To address this question, we conducted an fMRI experiment using a virtual subway navigation task. We identified the activity of the dmPFC, premotor cortex (PMC) and superior parietal lobe (SPL) with general linear model (GLM) in HP. Then, Dynamic Causal Modelling (DCM) was performed to quantify the influence of the higher- and lower-planning on the connectivity between the brain areas identified by the GLM. The strongest modulation effect of the higher-level planning was found on the dmPFC→right PMC connection. Furthermore, using Parametric Empirical Bayes (PEB), we found the modulation of higher-level planning on the dmPFC→right PMC and right PMC→SPL connections could explain the individual difference of the response time. We conclude that the dmPFC-related connectivity takes the response to the higher-level planning, while the PMC acts as the bridge between the higher-level planning to behavior outcome.

Learning at Variable Attentional Load Requires Cooperation of Working Memory, Meta-learning, and Attention-augmented Reinforcement Learning

Flexible learning of changing reward contingencies can be realized with different strategies. A fast learning strategy involves using working memory of recently rewarded objects to guide choices. A slower learning strategy uses prediction errors to gradually update value expectations to improve choices. How the fast and slow strategies work together in scenarios with real-world stimulus complexity is not well known. Here, we aim to disentangle their relative contributions in rhesus monkeys while they learned the relevance of object features at variable attentional load. We found that learning behavior across six monkeys is consistently best predicted with a model combining (i) fast working memory and (ii) slower reinforcement learning from differently weighted positive and negative prediction errors as well as (iii) selective suppression of nonchosen feature values and (iv) a meta-learning mechanism that enhances exploration rates based on a memory trace of recent errors. The optimal model parameter settings suggest that these mechanisms cooperate differently at low and high attentional loads. Whereas working memory was essential for efficient learning at lower attentional loads, enhanced weighting of negative prediction errors and meta-learning were essential for efficient learning at higher attentional loads. Together, these findings pinpoint a canonical set of learning mechanisms and suggest how they may cooperate when subjects flexibly adjust to environments with variable real-world attentional demands.