Aberrant neural computation of social controllability in nicotine-dependent humans

Social controllability, the ability to influence social interactions, is vital for effective decision-making. Impaired social controllability may contribute to maladaptive behaviors like smoking, a highly social activity. While research exists on nicotine addiction's relation to cue reactivity and impulsive control, the mechanisms underlying social cognitive deficits remain unclear. Reinforcement learning models have explored how drugs alter decision-making neural computations, particularly reward prediction errors. Economic models like temporal discounting highlight substance-dependent individuals' preference for immediate rewards, potentially reflecting interactions between cognitive control, time perception, and risk preference. Recent computational models link addiction to model-based control and forward planning dysfunctions, potentially amplified by complex environments. However, empirical evidence supporting these computational frameworks in social decision-making, particularly in nicotine addiction, is limited. This study hypothesized that smokers would exhibit reduced social control ability due to impaired neural computations of social forward planning signals, specifically focusing on the ventromedial prefrontal cortex (vmPFC).

Existing research demonstrates the importance of the vmPFC in tracking the downstream effects of choices and exploiting controllability in simulated social environments. The vmPFC encodes cognitive maps crucial for model-based planning. Studies have used reinforcement learning algorithms to model how drugs affect decision-making, including reward prediction error encoding by the mesolimbic circuit and temporal discounting. Research also suggests that model-based planning deficits are further amplified by complex environments. However, evidence supporting these computational frameworks in substance-dependent individuals within the context of social decision-making is scarce. This study aims to bridge this gap by directly examining the neural computations of social controllability in smokers versus non-smokers.

Two independent samples were studied: an in-person fMRI sample (17 smokers, 25 non-smokers) and an online replication sample (72 smokers, 147 non-smokers). Participants performed a modified ultimatum game where their choices probabilistically influenced future offers from simulated partners. A computational model incorporating forward thinking (FT) computations at varying depths (1-4 steps) was used to analyze choices. Model comparison determined the best-fitting model, which was used to extract key parameters, including 'estimated influence,' reflecting the perceived impact of actions on future outcomes. fMRI data were analyzed using SPM12, with GLMs examining neural correlates of forward thinking value and norm prediction errors. ROIs in the vmPFC and midbrain were defined based on prior research. Model-agnostic behavioral measures (offer sizes, rejection rates, perceived controllability) were also compared between groups. The online study included additional measures of risk aversion, mood (Beck Depression Inventory-II), and impulsivity (Barratt Impulsivity Scale). Statistical analysis involved two-sample t-tests for the fMRI sample and non-parametric bootstrapping for the online sample due to non-normal distribution and unequal sample sizes.

Smokers failed to exploit the controllability of their interactions, receiving significantly lower offers than non-smokers in both samples. Although overall rejection rates weren't significantly different, smokers showed lower rejection rates for medium-sized offers, suggesting a lack of strategic behavior to increase future offers. Smokers also reported a lower sense of control, although this difference wasn't consistently statistically significant across both samples. Computational modeling revealed that while smokers used a similar 2-step FT model as non-smokers, they significantly underestimated the influence of their choices on future offers (lower 'estimated influence' parameter). This finding replicated across both samples. Neurally, smokers exhibited reduced vmPFC activation related to forward thinking values (significantly lower in the fMRI sample and a trend in the online sample). They also showed reduced midbrain tracking of norm prediction errors (significantly lower in the fMRI sample). The reduced activity of the vmPFC was negatively correlated to projected total values in smokers, deviating from the pattern seen in non-smokers, suggesting the vmPFC's role in both value representation and future-oriented planning is impaired.

The findings suggest that underestimation of future consequences is a key feature in nicotine-dependent individuals, hindering their ability to exert control in social situations. This provides a neurocomputational account for observed social cognitive deficits. The results contrast with expectations of simply a reduced planning horizon, suggesting that the issue lies not in planning depth but rather in inaccurate estimation of influence. This aligns with and provides a computational explanation for increased temporal discounting in SUD, and suggests that accurate prediction of how current actions impact future outcomes could be a key element of interventions for SUD. Reduced vmPFC activity aligns with previous findings linking it to reduced preference for delayed rewards and impaired valuation processes. Aberrant midbrain activity regarding norm prediction errors extends existing work on altered reward prediction error encoding in addiction to the context of dynamic social interactions.

This study demonstrates that nicotine dependence is associated with an underestimation of the future consequences of actions in social settings, leading to impaired controllability. This is supported by computational and neural findings of reduced vmPFC activity in tracking projected values and impaired midbrain processing of norm prediction errors. Future research should investigate the role of craving and beliefs about the humanness of the other players in these processes, and address limitations such as sample size and gender representation.

The study has limitations, including a relatively small sample size in the fMRI study and low representation of females, potentially affecting generalizability. Planned analyses on the impact of craving were not conducted due to difficulties in enforcing overnight abstinence. Future studies should address these limitations by using larger samples, diverse representation and improved methodology to control for craving levels. While simulated players were used, future studies could explicitly measure beliefs about the 'humanness' of interaction partners to better understand their influence.