Human cooperation and trust are fundamental, yet not all individuals are trustworthy. Learning to distinguish trustworthy from untrustworthy individuals is crucial for successful social interactions. Rodent studies suggest a crucial role for the basolateral amygdala (BLA) in this learning process. Previous research with humans showed that individuals with BLA damage struggle to adapt their trust behavior. However, functional reorganization after brain damage might limit the generalizability of these findings to neurotypical populations. Neuroimaging studies in neurotypical populations have yielded inconsistent results regarding amygdala involvement in trust, possibly due to difficulties in differentiating between the BLA and the central amygdala (CeA), which have opposing affective roles in trust behavior. This study aimed to overcome these limitations by using a high-resolution fMRI acquisition protocol optimized for imaging ventral brain areas, focusing on the time-resolved analysis of amygdala subnuclei, and investigating their interaction with other subcortical regions like the nucleus accumbens (NAC), substantia nigra/ventral tegmental area (SN/VTA), and bed nucleus of the stria terminalis (BST). The study investigated task-dependent BLA and CeA function across different task phases, learning success levels, and trust behavior and subjective trustworthiness ratings. The interplay between these brain areas was analyzed to build a more nuanced understanding of trust learning and decision-making.

Prior research using rodent models strongly implicates the BLA in social experience-based learning. A previous study by the authors demonstrated that human participants with selective bilateral BLA damage were impaired in adapting their trust behavior in a repeated trust game. However, the generalizability of these findings to neurotypical populations remains unclear due to potential functional reorganization after brain damage. Human neuroimaging studies on trust have yielded inconsistent results regarding amygdala involvement, potentially because they treated the amygdala as a homogeneous structure, ignoring its heterogeneous subnuclei, the BLA and CeA, which have distinct functional roles in affective processing, particularly in the context of trust.

Sixty-two healthy neurotypical volunteers (mean age 23.83 years; 31 females) participated in a repeated trust game within an fMRI scanner. The game involved interactions with two simulated players, one trustworthy and one untrustworthy, each consisting of 20 rounds. The task comprised four phases: preparation, investment, waiting, and outcome. Participants invested points, which were tripled and then (partially or fully) returned by the simulated players. Behavioral data included investment amounts and post-experiment subjective ratings of trustworthiness, attractiveness, likability, and reliability. High-resolution fMRI data were acquired using a multi-band MRI sequence to optimize ventral brain area imaging. Data preprocessing included realignment, coregistration, and smoothing. Statistical analysis utilized a general linear model (GLM) with regressors for each task phase and player. Time-series analyses were conducted on amygdala subregions (BLA and CeA) and other brain regions (NAC, SN/VTA, BST) to investigate phase-dependent activation patterns and their relationship to learning success and behavioral measures. ROI analyses were also performed on the mean values of each participant's histogram plots, and state-dependent activation in amygdala subregions was examined using time-series analyses.

Behavioral data showed that participants invested more in the trustworthy player (r = +0.39, p = 0.002). The sample was split into learners (those who adapted their investment behavior over time to favor the trustworthy player) and non-learners. Time-resolved fMRI analysis revealed that the BLA showed maximal activation during outcome evaluation, while the CeA showed maximal activation during the preparation phase. However, there were no overall differences in BLA and CeA activation between trustworthy and untrustworthy players across all phases. Crucially, learners exhibited stronger BLA and CeA activation for the untrustworthy player during the introduction phase compared to non-learners. The nucleus accumbens (NAC) showed increased activation for the trustworthy player during outcome evaluation, regardless of learning success. Similar patterns were observed in the substantia nigra/ventral tegmental area (SN/VTA) and bed nucleus of the stria terminalis (BST). In learners, BLA activation differences between trustworthy and untrustworthy players correlated with investment behavior and post-experiment subjective trustworthiness ratings. Functional connectivity analyses showed that the BLA was connected to sensory integration areas and lateral PFC, whereas the CeA was connected to the ventral striatum and medial PFC.

The findings confirm the BLA's role in learning to differentiate between trustworthy and untrustworthy individuals, extending previous research on amygdala-damaged participants to neurotypical individuals. The BLA's activity during outcome evaluation suggests its involvement in belief formation about others' trustworthiness. The CeA appears to be involved in the motivational aspects of trust behavior and affective value attached to players, evident from its activity during the preparation phase and correlation with subjective trustworthiness ratings. The NAC, SN/VTA, and BST seem to mediate reward and risk evaluation related to the outcomes of trust behavior. The lack of differences between learners and non-learners in these regions may indicate that outcome processing and motivation were comparable across groups. The observed absence of amygdala habituation suggests the BLA's continuous encoding and updating of social experience information. The interplay between BLA, CeA, and other subcortical regions suggests a complex neural network underpinning trust learning and decision-making.

This study demonstrates the critical roles of BLA and CeA in learning whom to trust. The BLA is crucial for discriminating between trustworthy and untrustworthy players based on experience and optimizing trust behavior, while the CeA is involved in the affective evaluation of players. Other regions like NAC, SN/VTA, and BST contribute to reward and risk processing. Future research could explore the interplay between these regions using computational modeling to gain deeper insights into human social cognition.

The study used simulated players, potentially limiting the ecological validity of the findings. The sample size, while relatively large for fMRI studies, might not fully capture the variability of trust-related behaviors and neural responses. Future studies could employ real-world interactions to increase ecological validity. Further research could benefit from exploring more diverse populations to enhance generalizability of findings.