Primate anterior insular cortex represents economic decision variables proposed by prospect theory

Decision-making under uncertainty is a fundamental aspect of both human and animal behavior. A crucial element is risk attitude – the willingness to accept uncertainty for potentially better outcomes. Risk attitude isn't static; it's heavily influenced by context, varying with wealth levels and whether potential outcomes are gains or losses. Prospect theory, a prominent descriptive model of decision-making under risk, explains this context-dependence through two key concepts: (1) outcomes are evaluated relative to a reference point (current wealth); and (2) sensitivity to value changes differs between gains and losses (loss aversion). While prospect theory accurately describes behavior, its neuronal implementation remains unclear. Previous research implicated the anterior insular cortex (AIC) in risk-averse behavior and representing internal states. This study hypothesized that AIC neurons encode behaviorally relevant value information consistent with prospect theory, representing the reference point and reference-dependent value signals that are asymmetric in loss and gain contexts. These representations would then influence risk attitude.

Numerous human neuroimaging studies and lesion studies have identified brain areas active during risky decision-making, with the AIC being of particular interest. Human fMRI studies suggest AIC's role in representing internal states and risk-averse behavior. Lesion studies further support AIC's involvement in modulating risk attitude. Monkey recording studies have also demonstrated AIC's role in encoding reward expectation. This existing literature provided the foundation for the hypothesis that AIC neurons encode value information in a manner consistent with prospect theory's framework.

Two macaque monkeys were trained in a token-based gambling task. The task involved accumulating tokens to obtain a fluid reward. On each trial, monkeys chose between a gamble option (uncertain outcome) and a sure option (certain outcome), with trials presented in either a gain or loss context. The number of tokens to be gained or lost and the probabilities were varied. Single-neuron activity was recorded from the AIC during the choice period. The monkeys' risk attitude was assessed by analyzing their choices across different contexts (gain/loss) and wealth levels (starting token number). A prospect theory model was fitted to the behavioral data to quantify risk aversion, loss aversion, and probability weighting. Neuronal activity was analyzed using linear regression models to identify neurons encoding decision-related variables (expected value, start token number, risk). The relationship between neural activity and behavioral choice/risk attitude was assessed using receiver operating characteristic (ROC) analysis. This involved calculating the area under the curve (AUC) to measure a neuron's ability to distinguish between different choice or risk-attitude states.

The study revealed that monkeys exhibited context-dependent risk attitudes, mirroring human behavior. They were more likely to gamble in gain contexts and less likely in loss contexts. Furthermore, their choices were influenced by their current wealth (start token number), showing risk aversion with increasing wealth in gain contexts and risk-seeking behavior in loss contexts. The monkeys employed a relative value framework, consistent with prospect theory. The prospect theory model accurately captured the monkeys' choice behavior, outperforming an expected value model. AIC recordings identified neurons encoding various decision-related variables: (1) Value across gains and losses (general value signal); (2) Gain/loss category; (3) Behavioral salience (value encoding with inverse correlation in gain/loss contexts); (4) Value specifically for losses (Loss value neurons) or gains (Gain value neurons); (5) Start token number (parametric, categorical, or numerical encoding); and (6) Risk (parametric or categorical encoding). Loss-value neurons showed higher sensitivity to value changes in the loss context compared to Gain value neurons in the gain context. This difference in sensitivity was also modulated by wealth level, with sensitivity decreasing as wealth increased. Finally, a subset of AIC neurons (15%) showed significant correlations between their activity fluctuations and trial-by-trial variations in the monkeys' choices and risk attitude.

The findings strongly support the hypothesis that the primate AIC encodes key variables consistent with prospect theory. The AIC represents the reference point (current wealth), gain/loss context, and value in a relative, rather than absolute framework. The observed loss aversion in AIC neuronal responses mirrors behavioral patterns. The results demonstrate the AIC's crucial role in monitoring contextual information influencing risk attitudes. The observed tendency for monkeys to prefer gamble options, unlike most human studies, might be attributed to task-specific factors such as the small reward amounts and numerous trials. This discrepancy highlights the importance of considering species-specific differences, individual variability, and task design in future studies.

This study provides strong evidence for the AIC's role in encoding economic decision variables as predicted by prospect theory. AIC neurons represent wealth levels as a reference point, distinguish gains and losses asymmetrically, and show loss aversion. The study's limitations include the relatively small number of reward amounts and probabilities used. Future research could investigate the role of AIC in the broader neural circuitry of risk attitude, explore the potential causal role of AIC, and examine its interaction with emotional and autonomic processes.

The study's limitations include the relatively small number of reward amounts and probabilities tested, which might have limited the assessment of non-linearities in the utility and probability weighting functions. Furthermore, the study didn't determine if the observed neuronal signals are unique to AIC or if similar signals exist in other brain areas. Finally, causal relationships between AIC activity and risky decision-making weren't directly established.