Neural encoding of perceived patch value during competitive and hazardous virtual foraging

Foraging decisions are crucial for survival and are influenced by various factors including competitor density, food availability, and predation risk. In safe, competitive environments, animals often avoid competition to maximize individual gain. Conversely, in hazardous environments, risk-dilution strategies, such as aggregating with conspecifics to reduce individual predation risk, become advantageous despite potential competition for resources. This study examines whether human foraging decisions align with these patterns and aims to identify the underlying neural mechanisms involved. Prior research on foraging has largely focused on virtual two-patch tasks without predation threats, highlighting the roles of the mid-cingulate cortex (MCC) and ventromedial prefrontal cortex (vmPFC) in action selection and value encoding. However, the influence of ecological threats on foraging decisions and their neural correlates remains under-investigated. This study addresses this gap by employing a virtual foraging task that incorporates both competition and predation risk, allowing for a nuanced examination of human foraging strategies and their neural underpinnings. We hypothesized that humans would adapt their strategies based on a computed 'perceived patch value' (PPV), a measure that integrates reward, threat, and competition levels, and that this PPV would be neurally encoded in brain regions associated with action selection and value.

A substantial body of research explores foraging decisions across various species. Studies have consistently shown that competitor density and food quantity significantly impact patch selection, feeding behavior, and foraging duration. In predation-free, competitive environments, competition avoidance is a common adaptive strategy to maximize individual gains. Conversely, under predation risk, risk-dilution strategies, based on the principle of safety in numbers, become prevalent. Hamilton's selfish herd theory provides a framework for understanding how prey animals aggregate to reduce their individual risk of predation. However, these strategies often entail efficiency costs, creating a trade-off between maximizing resource acquisition and minimizing predation risk. Optimal foraging theory suggests that this trade-off is resolved by considering the overall fitness or perceived value of a patch, which depends on reward, threat, and social context. Although research has examined the neurobiological mechanisms of foraging decisions focusing on economic risk, reward, and uncertainty, the influence of ecological threats remains largely unexplored. Previous studies using virtual foraging paradigms consistently involve regions such as the MCC and vmPFC in action selection and value encoding, but they lack the integration of ecological threats.

Twenty-two participants (6 female; mean age 31) underwent fMRI scanning over two days while performing a virtual two-patch foraging task. The task involved choosing between patches with varying competitor densities under safe (no predator) and hazardous (predator present) conditions. Each session included four blocks, each with a unique cycle of competitor states in the two patches, allowing participants to learn the repeating sequences. Trials involved three phases: (1) a preview of the competitor state in each patch, (2) a decision phase where participants chose a patch, and (3) a foraging phase where participants competed to collect rewards in the chosen patch. In the hazardous condition, a virtual predator could appear and capture a participant (resulting in an aversive electric shock), introducing predation risk. The electric shock intensity was individually calibrated for each participant. The experimental design controlled for decision effort, energy costs, and reward density across safe and hazardous conditions. Three trial types were used: short duration, immediate decision (SI); short duration, later decision (SL); long duration, immediate decision (LI). Data analysis focused on the immediate decision trials (SI and LI). A computational model was developed to estimate individual participants' perceived patch value (PPV), integrating reward, shock cost, and probability of being captured. Multivariate representational similarity analysis (RSA) was used to identify brain regions encoding PPV, along with univariate analyses. fMRI data were preprocessed using fMRIprep, and first-level models included regressors for each trial, along with regressors of no interest. The RSA analysis compared neural similarity patterns across trials with the similarity of task conditions (competitor numbers, PPV, threat levels) using linear regression. Univariate analyses assessed how overall activity levels varied with key decision variables.

Behavioral data revealed that participants adopted a competition-avoidant strategy (selecting patches with fewer competitors) in safe conditions (89% of decisions) and a risk-dilution strategy (selecting patches with more competitors) in hazardous conditions (32% of decisions). The difference in the number of competitors chosen between the safe and threat conditions was highly significant (paired t-test, p < 0.0005). This difference was further validated by analyzing the mean difference scores for safe and threat conditions (paired t-test, p < 0.0005). Choosing patches with more competitors in threat conditions significantly increased the probability of avoiding capture, exhibiting the risk-dilution effect. In safe conditions, avoiding competition resulted in greater reward acquisition. The computational model accurately predicted participants' choices (69.89% accuracy), demonstrating that decisions were driven by PPV. The model's predictive probabilities were well-calibrated with true choice probabilities. Model comparison indicated that the initial model (considering PPV) provided a better fit than variants incorporating learning or choice stickiness. RSA identified a distributed network of regions encoding PPV, prominently including the MCC, PCC, mPFC, and OFC. The vmPFC and MCC encoded PPV for both the current and alternative patches, with the current patch's value also encoded in the premotor cortex, hippocampus, and anterior insular cortex. Crucially, these regions encoded PPV independently of social density. Univariate analyses corroborated these findings and also revealed activity in the visual cortex linked to the number of competitors. No regions showed specific representation for the difference in the number of competitors or the PPV difference between patches.

The findings demonstrate that humans adaptively adjust their foraging strategies based on a computed PPV, reflecting competition avoidance in safe situations and risk dilution under threat. The identification of the MCC and vmPFC as key neural substrates encoding PPV suggests that these regions integrate multidimensional information relevant to survival. The MCC's role may be in coordinating emotional responses and motor actions according to learned values, especially under imminent threat. The vmPFC's involvement aligns with its established role in valuation. The lack of strong amygdala involvement in encoding PPV is notable, possibly suggesting a less direct role in this specific foraging decision-making process. Although the task design allowed for model-free learning, the model fitting showed that the consideration of PPV was a better explanation of participant decision-making than a simple learning model. Future studies using more complex experimental designs, coupled with detailed behavioral measures (e.g., reaction times, decision confidence), could provide further insights into the precise nature of PPV computation and its interplay with other cognitive processes.

This research provides substantial evidence for the role of perceived patch value (PPV) in human foraging decisions, highlighting the adaptive nature of strategies employed in different environments. The study's key contribution is the identification of a distributed neural network, particularly the vmPFC and MCC, underlying PPV computation. This work advances our understanding of human decision-making in ecologically relevant contexts. Future research could explore the role of other factors in determining PPV, investigate the neural mechanisms underlying PPV computation in more detail, and examine how individual differences in risk aversion influence foraging decisions. Furthermore, examining scenarios with more unpredictable resource availability and incorporating a more complex predator model could extend these findings.

Several limitations warrant consideration. The virtual environment may not perfectly reflect real-world foraging challenges. The use of a fixed cycle of competitor states may have allowed participants to learn optimal strategies, potentially reducing the ecological validity. The lack of direct measures of decision confidence or subjective value may have confounded PPV estimation. The study was conducted in a virtual environment, and the findings might not directly translate to real-world situations. Furthermore, the relatively simplistic nature of the predator model may not capture the full complexity of predation risk in natural environments.