The role of mPFC and MTL neurons in human choice under goal-conflict

Humans frequently face decisions involving conflicting motivations, a classic example of approach-avoidance conflict. These conflicts, central to anxiety, involve weighing potential rewards against potential punishments. While animal studies implicate the striatum, medial prefrontal cortex (mPFC), and medial temporal lobe (MTL) in processing outcome values and guiding behavior, their precise roles in human approach-avoidance conflict remain unclear. The mPFC and MTL are known to process outcome values and valence, with the hippocampus and amygdala involved in forming contextual and emotional associations respectively. However, existing research has not definitively differentiated the neural substrates involved in learning outcome valence from those mediating conflict resolution itself. Animal studies, using paradigms like the elevated plus maze, point to the amygdala, hippocampus, and mPFC's involvement in avoidance behavior. Some suggest the ventral hippocampus plays a role in decision-making during conflict, while others emphasize the hippocampus and amygdala's role in learning from outcomes. Studies on depression and schizophrenia show disrupted prediction-outcome associations in the amygdala-hippocampal complex, highlighting the MTL's significance. This research aims to clarify the distinct contributions of the mPFC and MTL in human approach-avoidance conflict using intracranial recordings during a goal-conflict task.

Existing research indicates a significant role for the mPFC and MTL in processing reward and punishment information, particularly within the context of reinforcement learning and approach-avoidance conflicts. Animal studies have extensively explored the neural mechanisms underlying reinforcement learning, highlighting the striatum's role in signaling prediction errors and the mPFC and MTL's involvement in processing outcome values and valence. The hippocampus and amygdala are crucial for forming contextual and emotional associations guiding future behavior. However, the literature lacks a clear differentiation between the neural substrates that encode outcome valence and those mediating approach-avoidance conflict resolution. While studies utilizing tasks like the elevated plus maze in animals implicate the amygdala, hippocampus, and mPFC in avoidance behavior, it remains unclear how these regions interact to influence behavior during goal conflicts in humans. Furthermore, research into psychopathology shows the importance of these brain regions in processing rewards and punishments, for instance in depression and schizophrenia, but this research mostly focuses on situations that lack the inherent conflict of our paradigm. Therefore, there is a significant gap in our understanding of how these brain areas contribute to adaptive behavior when both reward and punishment are possible outcomes of a single decision.

Fourteen patients with intractable epilepsy, implanted with depth electrodes for clinical reasons, participated. Electrophysiological data were recorded from the mPFC (dmPFC and cingulate cortex) and MTL (amygdala and hippocampus) while patients played a computer game (PRIMO) designed to induce approach-avoidance conflict. In the game, players controlled an avatar to catch falling coins (reward) while avoiding falling balls (punishment). A ‘controlled’ condition required active participation in achieving or avoiding outcomes, while an ‘uncontrolled’ condition presented outcomes independently of player actions. High goal conflict (HGC) trials featured multiple balls between the avatar and the coin, while low goal conflict (LGC) trials had fewer obstacles. Single- and multi-unit activity was recorded and analyzed to assess neuronal response selectivity to outcome valence (reward vs. punishment), controllability (controlled vs. uncontrolled), and the relationship between neural activity and subsequent behavioral choice. Data analysis involved spike sorting, peri-stimulus time histogram (PSTH) generation, bootstrapping for determining neuronal responsiveness, and statistical tests (e.g., McNemar's exact test, ANOVA, GLMM) to assess neuronal response probability, selectivity, and brain-behavior interactions. The study carefully accounted for potential confounds such as movement artifacts.

The study identified 310 neurons across mPFC and MTL regions. Under controlled conditions, mPFC neurons showed significantly greater responsiveness to punishments than rewards. This negative outcome bias was not observed in the MTL, which exhibited similar responses to both rewards and punishments. Critically, MTL neuronal firing following punishment, but not reward, significantly correlated with a lower probability of subsequent approach behavior in high-conflict trials. This effect was not observed in mPFC neurons alone. Interestingly, mPFC responses to punishment consistently preceded similar MTL responses. Further analysis revealed a significant interaction between mPFC and MTL activity following controlled punishment, predicting subsequent avoidance behavior. The temporal precedence of mPFC activity suggests a potential mechanism where mPFC initially processes the negative valence of punishments, subsequently influencing MTL encoding of this information and shaping future behavioral choices. The effect was specific to the controlled condition suggesting a role for agency. Analysis showed that the hippocampus, within the MTL, was the primary driver of this effect on subsequent choices. The study also controlled for movement, ensuring the observed effects weren't solely due to motor planning or related artifacts.

The findings challenge existing models by demonstrating a sequential interaction between mPFC and MTL during approach-avoidance conflict. The mPFC's sensitivity to negative outcomes, especially under controlled conditions, aligns with its role in processing negative emotions and potentially reflects the importance of a sense of agency. The MTL's role in modifying subsequent approach behavior, despite lacking valence selectivity, highlights its significance in learning from outcomes and updating behavioral tendencies. The hippocampus, in particular, appears to be crucial for this learning process. These results integrate findings from reinforcement learning and anxiety-related research, suggesting a cooperative interplay between mPFC valuation of negative outcomes and MTL-mediated learning which guides subsequent adaptive choices. The observed interaction between mPFC and MTL neuronal activity underscores the complexity of approach-avoidance conflict resolution and the dynamic interplay between these brain regions.

This study provides compelling evidence for a sequential and interactive role of mPFC and MTL in human approach-avoidance conflict. The mPFC's negative outcome sensitivity, particularly under controlled conditions, and the MTL's influence on subsequent behavior after punishment suggest a mechanism where mPFC valuation interacts with MTL learning to guide future choices. Further research should explore the computational mechanisms underlying this interaction and investigate how these findings relate to psychiatric disorders characterized by impaired reward processing and maladaptive avoidance behaviors. Investigating the role of agency and the specific contributions of different MTL substructures would further refine our understanding.

The study used a sample of epilepsy patients, potentially limiting the generalizability of the findings to the broader population. The high approach probability in the task limited the analysis of avoidance trials. The ecological nature of the task made it difficult to analyze neural activity during the anticipation phase due to movement artifacts. Finally, the aggregation of neurons from different MTL substructures could have masked subtler region-specific effects. Future studies should address these limitations using larger, healthier samples and tasks designed to more evenly distribute approach and avoidance trials.