D2/D3 dopamine supports the precision of mental state inferences and self-relevance of joint social outcomes

Abstract

Striatal dopamine is important in paranoid attributions, although its computational role in social inference remains elusive. We employed a simple game-theoretic paradigm and computational model of intentional attributions to investigate the effects of dopamine D2/D3 antagonism on ongoing mental state inference following social outcomes. Haloperidol, compared with the placebo, enhanced the impact of partner behaviour on beliefs about the harmful intent of partners, and increased learning from recent encounters. These alterations caused substantial changes to model covariation and negative correlations between self-interest and harmful intent attributions. Our findings suggest that haloperidol improves belief flexibility about others and simultaneously reduces the self-relevance of social observations. Our results may reflect the role of D2/D3 dopamine in supporting self-relevant mentalising. Our data and model bridge theory between general and social accounts of value representation. We demonstrate initial evidence for the sensitivity of our model and short social paradigm to drug intervention and clinical dimensions, allowing distinctions between mechanisms that operate across traits and states.

Keywords: Computational models; Predictive markers.

Fig. 1. Experimental design and model space.

a, Participants were entered into a double-blind, placebo-controlled, within-subject experimental design. ECG, electrocardiogram. b, Participants engaged in a three-partner version of the sharing game (inset). Here, partners were assigned the role of Dictator and, on each trial, could either take £0.10 for themselves (unfair outcome), or take £0.05 and give the participant £0.05 (fair outcome). Participant reported two types of attributional intent concerning the motivations of the partner after each outcome. These included harmful intent attributions and self-interest attributions. Partner order was randomized, and partner change was signalled. c, Model space used to test whether dopamine manipulations were best explained by the full model (M1), a model that constrained policy updating to a single sensitivity parameter for each attribution (M2), or a model that constrained prior uncertainty to a single parameter (M3; Table 1). Although filled objects are free parameters. Grey shaded objects are probability distributions.

Fig. 2. Model comparison, recovery and generative performance.

a, Model responsibility across all three drug conditions. Greater model responsibility at the group and individual levels indicates the most likely generative model to explain the data. Ex. prob. = exceedance probability that a single model best defines group behaviour; freq = model frequency that each model is the best fitting model for participants. b, Model recovery. All recovery analyses used n = 28 synthetic participants—one for each real parameter set approximated from the data. The Hierarchical Bayesian Inference (HBI) algorithm correctly identified the correct model for most participants with trivial differences between model frequencies. c, Parameter recovery. Pearson correlation matrix of common parameters across all drug conditions for simulated (y-axis) and real (x-axis) data. All correlations were over 0.71 (P-values < 0.001). Crosses indicate non-significant associations. d, Parameter recovery. Individual Pearson correlations between common parameters across haloperidol and placebo conditions for simulated (y-axis) and real (x-axis) data. All correlations were over 0.71 (P-values < 0.001). Black lines indicate the linear model of perfect association (r = 1). e, Parameter recovery. Individual Pearson correlations between common parameters across all drug conditions for simulated (y-axis) and real (x-axis) data. Black lines indicate the linear model of perfect association (r = 1). f, Top panel: Pearson correlation (±s.e.m.) between simulated and real harmful intent (left) and self-interest (right) attributions across all Dictator policies (n = 28; P-values < 0.001). Bottom panel: simulated harmful intent (left) and self-interest (right) mean attributions (±s.e.m.) for each drug condition and Dictator policy.

Fig. 3. Influence of haloperidol on the winning model.

a, Bayesian t-test results (n = 28) assessing the difference and uncertainty (median ± 95% HDI) of the change in mean parameter estimates (∆μ; difference in mean) between placebo and haloperidol. Red distributions indicate that the 95% high-density interval (HDI) does not cross 0, suggesting reasonable certainty that the mean difference is not an artefact of statistical noise. The d values indicate the median effect size (Cohen’s d; Supplementary Fig. 4). The red box indicates the parameters where the effect size distributions were most robust, where the 95% HDI lay outside of the region of probable equivalence with the null hypothesis. b, Simulations (±s.e.m.) of the marginal effect of likelihood parameters on the precision (1/σ²; inverse variance) of harmful intent (red) and self-interest (black) attributions over all trials, controlling for Dictator style. Vertical lines are indicative of the median individual parameter estimates from both haloperidol and placebo groups. The blue arrow indicates the difference from placebo to haloperidol (see Supplementary Fig. 3 for trial-wise and within-Dictator precision changes). Simulations are consistent with the notion that w_HI increases flexibility within and between contexts, accentuating smooth learning. Note that there was no significant correlation between w₀, w_SI and w_HI in our parameter estimation from our real data (all P-values > 0.05; Supplementary Fig. 2), suggesting independent contributions from each to the attributional dynamics. c, Factor loading of each parameter on flexibility (factor 1) and learning (factor 2) dimensions. A loading filter of |0.4| was applied. Both of these factors can discriminate effectively between drug conditions. The w_SI term is not featured in this plot as it was not meaningfully loaded onto either factor. d, Factor scores (absolute value) for each individual participant (n = 28) for both haloperidol (red) and placebo (blue) conditions ordered from low to high. The panels on the right demonstrate the marginal loading across participants. e, Candyfloss plot factor scores for each individual participant. The grey lines indicate that the same participant was responsible for each connected point under placebo (blue) and haloperidol (red). f, Receiver operating characteristic describing the sensitivity and specificity of factors to differentiate drug conditions. Area under the curve = 0.91; sensitivity = 0.8; specificity = 0.78.

Fig. 4. Association of mental state attributions between drug condition.

a, In both real and simulated data (n = 28), haloperidol (red) versus placebo (blue) induced a trial-wise negative Pearson association (±s.e.m.) between harmful intent and self-interest, which decayed over time for both real (R = 0.52, P = 0.029) and simulated (R = 0.65, P = 0.0046) data. The right-most panel shows the marginal effect (box plots demonstrate minimum, interquartile range, median and maximum values) of trial-wise correlations between conditions. Using linear regression, we show that the difference between Pearson correlations between haloperidol and placebo was significant for both real (estimate = 2.26, SE = 0.33, P = 9.29 × 10⁻⁸) and simulated (estimate = 2.23, SE = 0.44, P = 1.84 × 10⁻⁵) data. *** = P < 0.001. b, There was a general negative Pearson association (±s.e.m.) between harmful intent and self-interest found under haloperidol for mean attributions across all 18 trials; this was not true for the placebo. c, Summary of main effects between drug conditions on self and other oriented intentional attributions following social outcomes. Both trial-wise and averaged associative analyses indicate that other-oriented attributions concerning self-interest of others (black), and self-oriented attributions concerning the harmful intent of others (red), are independent under the placebo (PLAC) but coupled under haloperidol (HALO). Under haloperidol this coupling is biased towards exaggeration of other-oriented attributions and diminishment of self-oriented attributions.

Fig. 5. Summary of experimental parameter changes from current and past works.

Experimentally observed effects on our model. The impact of haloperidol on model parameters is indicated by green arrows. Prior results from the impact of high trait paranoia^, are indicated by red arrows.