Aberrant neural computation of social controllability in nicotine-dependent humans

The study investigates how nicotine dependence affects social controllability—the capacity to influence others during interpersonal exchanges. Because smoking is often embedded in social contexts, deficits in social control may contribute to maladaptive behaviors. Prior computational psychiatry work links addiction to altered reinforcement learning, preference for immediate rewards, and impairments in model-based control and forward planning, particularly in complex environments. The authors hypothesized that, compared to non-smokers, smokers would show reduced ability to exert social control, reflected computationally by underestimation of how current actions influence future social outcomes, and neurally by reduced ventromedial prefrontal cortex (vmPFC) encoding of forward-projected values. They also examined midbrain responses to social norm prediction errors as a secondary neural mechanism.

Prior research has shown: (1) addiction-related alterations in reward prediction error signaling within mesolimbic circuits; (2) stronger temporal discounting in substance users, suggesting a bias toward immediate rewards; (3) deficits in model-based planning and forward thinking in addiction, exacerbated by complex environments; and (4) a central role of the vmPFC in representing cognitive maps and tracking downstream consequences of choices in social tasks. Additionally, midbrain structures (VTA/SN) contribute to social norm updating and reward learning. Despite these insights, empirical evidence on neural and computational mechanisms of social decision-making in nicotine dependence has been limited, motivating a direct test of social controllability in smokers versus non-smokers.

Design: Two samples were studied. (1) In-person fMRI sample (n=42): smokers (n=17) and non-smokers (n=25). (2) Independent online replication sample (n=219): smokers (n=72) and non-smokers (n=147). Both studies were preregistered. Participants: In-person smokers smoked >10 cigarettes/day for ≥1 year; exclusions included major medical/neurological/psychiatric conditions, MRI contraindications, and other substance dependence (except nicotine/alcohol for smokers). Non-smokers had no substance dependence. Online smokers reported ≥1 cigarette/week, with exclusions for medical/psychiatric diagnoses; non-smokers reported zero tobacco use and no cravings. Task: Modified ultimatum game with two conditions. In the Controllable condition, participants’ accept/reject choices stochastically altered the next offer by 0, 1, or 2 dollars (accept → decrease; reject → increase; each with 1/3 probability). In the Uncontrollable condition, offers were random (mean $5), unaffected by current choices. Participants interacted with members of two labeled teams without explicit controllability information, learned contingencies through experience, and afterward rated perceived controllability (0–100%). Initial offer was $5, offers to the participant were always ≤$9. Smokers played 30 trials/condition; non-smokers’ analyses were truncated to the first 30 trials to match. For modeling, early (first 5) and late (last 5) trials for smokers were excluded, and equivalent middle trials for non-smokers were used. Computational modeling: Compared forward-thinking (FT) models with planning horizons of 1–4 steps against a 0-step norm-learning model and a model-free RL model. FT models computed total projected value by combining current utility and mentally simulated future utilities under an estimated influence parameter (δ; −$2 to $2) indicating how much one’s action shifts subsequent offers. Internal norm expectations were updated trial-by-trial via a Rescorla–Wagner rule, yielding norm prediction errors (nPEs). Action selection followed a softmax function (inverse temperature β). Model comparison used Deviance Information Criterion and parameter recoverability; the 2-step FT model was selected for analyses. fMRI acquisition and analysis: Philips 3T scanner; structural multi-echo MP-RAGE (1 mm isotropic); EPI: TR=2000 ms, TE=25 ms, voxel 3.4×3.4×4.0 mm, 37 slices. Preprocessing in SPM12: slice timing correction, co-registration, normalization (2 mm isotropic), smoothing (8 mm FWHM). Two first-level GLMs modeled: (1) forward-projected choice values at choice submission; (2) norm prediction errors; both with standard task regressors and six motion covariates. Group-level ANOVAs compared smokers vs non-smokers (PFDR<0.05, k>50). ROI analyses used independent coordinates: vmPFC [−2, 50, −2] (8-mm sphere) for FT value; midbrain/SN-VTA [−4, −26, −11] (8-mm sphere) for nPEs. Statistics: In-person comparisons used two-sample t-tests. Online sample used non-parametric bootstrapping/permutation (10,000 iterations) due to non-normality and unbalanced groups. Additional GLMs assessed potential contributions of mood, impulsivity, and risk aversion to behavioral/model parameters. Data and code were publicly shared.

Behavior (in-person fMRI sample):

• Smokers failed to increase offers over trials compared to non-smokers. Mean offers: smokers $4.50 ± 2.14 vs non-smokers $5.98 ± 1.95; t(40)=2.31, p=0.0131, Cohen’s d=−0.72.
• Overall rejection rates did not differ (smokers 43.23% ± 23.75 vs non-smokers 50.26% ± 14.79; p>0.05; d=−0.35), but smokers rejected fewer medium offers ($4–$6): 46.72% ± 33.53 vs 66.93% ± 33.20; t(40)=2.27, p=0.0144; d=−0.61. No differences for low or high offers.
• Perceived controllability was lower in smokers (52.40% ± 20.76) vs non-smokers (65.91% ± 22.39), but not significant: t(37)=−1.93, p=0.062; d=−0.63.

Computational modeling (in-person):

• FT models outperformed 0-step and model-free RL models; 2-step FT model chosen (good recoverability). Predictive accuracy: non-smokers 86.21%; smokers 86.47%.
• Estimated influence (δ) was lower in smokers: 0.352 ± 1.544 vs non-smokers 1.396 ± 0.654; t(40)=−3.018, p=0.002; large effect (d≈0.89). No group differences in other parameters (β, α, initial norm, adaptation rate) at conventional thresholds.

Replication (online sample):

• Smokers received lower offers: $5.53 ± 1.85 vs $6.06 ± 1.68; bootstrapping p=0.0266; d=−0.30. Trajectories rose slightly for both groups but remained lower for smokers.
• Overall rejection rates similar (smokers 51.57% ± 12.36 vs non-smokers 53.97% ± 9.85; p=0.077; d=−0.21), but smokers again rejected fewer medium offers (57.59% ± 29.34 vs 66.40% ± 27.24; p=0.0175; d=−0.31). Risk aversion did not account for this difference (GLMs).
• Perceived controllability lower in smokers: 52.68% ± 34.46 vs 61.32% ± 34.63; p=0.0442; d=−0.25.
• Estimated influence (δ) reduced in smokers: 1.119 ± 1.016 vs 1.351 ± 0.833; p=0.0447; d=0.25. No order effects; effects robust to mood and impulsivity covariates.

Neuroimaging (in-person):

• vmPFC encoding of forward-projected choice value was reduced in smokers. ROI parameter estimates: non-smokers 0.347 ± 1.05 vs smokers −0.749 ± 2.00; t(40)=−2.31, p=0.013; d=0.69. Whole-brain confirmed non-smokers > smokers in vmPFC (PFDR<0.05, k>50).
• Midbrain (SN/VTA) encoding of norm prediction errors was reduced/inverted in smokers. ROI parameter estimates: non-smokers 0.302 ± 1.10 vs smokers −0.306 ± 1.04; t(40)=−1.80, p=0.040; d=0.57. Whole-brain confirmed non-smokers > smokers in midbrain for nPEs (PFDR<0.05, k>50).

Findings support the hypothesis that smokers exhibit impaired social controllability. Behaviorally, smokers failed to strategically reject medium offers and consequently obtained lower offers over time. Computationally, both groups used a 2-step forward-thinking strategy, but smokers underestimated the causal impact of their actions on future offers (lower estimated influence δ), aligning with their diminished perceived controllability. Neurally, smokers showed aberrant vmPFC encoding of forward-projected choice values and reduced midbrain tracking of norm prediction errors, indicating disruptions in circuits supporting model-based, future-oriented valuation and social learning. These results provide a neurocomputational explanation for smokers’ difficulties in exploiting controllability in social environments and suggest links to broader addiction phenomena such as steep temporal discounting and deficits in flexible adaptation. The work extends addiction research beyond non-social contexts by demonstrating specific alterations in social forward planning and norm updating.

The study identifies underestimation of one’s social influence as a core feature of nicotine dependence, linked to reduced vmPFC encoding of forward-projected values and midbrain norm prediction error signaling. These neurocomputational deficits explain smokers’ impaired ability to exploit social controllability. Replication in a larger online sample strengthens generalizability. The findings suggest potential intervention targets: enhancing accurate mental simulation of how current actions affect future outcomes may improve decision-making in addiction. Future work should examine sex differences, the effects of abstinence/craving on social controllability, belief about partner humanness, and extend paradigms to include negative outcomes to test generalization to avoidance and punishment contexts.

• In-person fMRI sample size was modest with low female representation, limiting power and generalizability and precluding robust sex-difference analyses.
• Planned abstinence manipulation failed (smokers did not remain abstinent), preventing tests of craving/withdrawal effects on social controllability.
• Beliefs about the “humanness” of partners were not measured, which may influence social decisions.
• The temporal discounting factor was fixed in the model for technical reasons, potentially limiting inference about discounting–influence interactions.
• Task variants differed in trial counts between groups; analyses mitigated this by matching trial windows but residual effects cannot be fully excluded.