An agent independent axis for executed and modeled choice in medial prefrontal cortex

Abstract

Adaptive success in social animals depends on an ability to infer the likely actions of others. Little is known about the neural computations that underlie this capacity. Here, we show that the brain models the values and choices of others even when these values are currently irrelevant. These modeled choices use the same computations that underlie our own choices, but are resolved in a distinct neighboring medial prefrontal brain region. Crucially, however, when subjects choose on behalf of a partner instead of themselves, these regions exchange their functional roles. Hence, regions that represented values of the subject's executed choices now represent the values of choices executed on behalf of the partner, and those that previously modeled the partner now model the subject. These data tie together neural computations underlying self-referential and social inference, and in so doing establish a new functional axis characterizing the medial wall of prefrontal cortex.

Figure 1

Experimental Design and Behavior (A) Trial timeline for the fMRI task of delegated intertemporal choice. In blocks of 40 trials, the frame of reference is changed between choosing for self and for partner. (B) Illustrative valuations from two example trials. Here, the subject exemplifies a relatively low discounter, as shown by the small influence of delay on value, while the partner is a relatively higher discounter. Note that there are four values to consider in the task: preferred and nonpreferred values for the subject and the partner. By computing value differences in the two frames of reference, we could dissociate both the different discount rates and the different choices of the two individuals. In this example, the subject’s value difference would be 15 in both trials, but the partner’s would be 5 and 10. (C) Correlation matrix between predicted value-related BOLD responses for partners with different temporal discount rates (k). Predicted response is assumed to align with the difference between chosen and unchosen values of each player. Reward and delays in the choice set were optimized to minimize overall predicted correlations (see Supplemental Information). Participants were prescreened to measure their discount rates and then paired to minimize correlations. Green dots represent the pairs of participants. Indeed the self and other value differences were broadly decorrelated in the experimental data (mean r = −0.11). (D) Average percent choice of the high-value long-delay option for high discounters (light) and low discounters (dark). Shown are choices in the prescreening questionnaire, during training on their partner’s choice preferences, and when choosing for self and for other in the fMRI experiment. Error bars show the standard error of the mean. Note that high discounters were paired with low discounters and vice versa. The behavioral flipping indicates that subjects learnt their partner’s preferences. See Figure S1.

Figure 2

Gradient of Functional Organization in the Rostromedial Prefrontal Cortex (A) Regions of prefrontal cortex responding to the average contrast of value difference over all trials, i.e., across both self and other values (p < 0.01). Yellow dots are equally spaced along the activity profile and serve as the spatial markers for the analysis in (C). (B) Within the region of overall value sensitivity (shown in A), some regions respond more to the value difference that will be acted on in the current trial (executed in red), and some to the currently irrelevant value difference (modeled in blue). (C) Left: A spatial gradient analysis of functional contrast against position along the ventral-dorsal axis of medial prefrontal cortex (see colored dots in A). Left: In each subject, data from five anatomical locations are mapped onto a line and the spatial regression slope is computed. Right: Across subjects there is a strong gradient, with executed value effects expressed in more ventral and modeled value effects in more dorsal zones. No such dorsoventral gradient exists for the contrast of self-versus-other. The difference between the two gradients (indicating a difference of a difference) was significant (p < 0.00001). Error bars show the standard error of the mean. Right: Results of an equivalent spatial gradient analysis in the temporoparietal cortex. Sagittal sections through TPC are shown in Figure S2E.

Figure 3

Neural Correlates of Value Difference during Delegated Intertemporal Choice (A) Activity while subjects choose for themselves. Green: Activity associated with subject’s chosen minus unchosen value difference. Yellow: Activity associated with partner’s preferred minus nonpreferred value difference. (B) Activity while the subject chooses for their partner. Green: Activity associated with subject’s preferred minus nonpreferred value difference. Yellow: Activity associated with the partner’s values in the frame of reference of the choices made by the subject on behalf of the partner (chosen value minus unchosen value). (C) Voxels showing a conjunction of the contrasts shown in green in (A) and yellow in (B) (executed value difference signals) are shown in red. Voxels showing a conjunction of contrasts shown in yellow in (A) and green in (B) (modeled value difference signals) are shown in blue. Individual maps are thresholded at p < 0.05 before the conjunction analysis. (D) Formal test that brain regions exchange agents between choice conditions in medial prefrontal and temporoparietal cortices. In each case, data are extracted from clusters defined by the opposite condition, ensuring no selection bias is present (see text and Supplemental Experimental Procedures). Bars show average effects of value (chosen/best – unchosen/worst) across subjects. Error bars show the standard error of the mean. All p values are two tailed, and refer to interaction effects. Significant effects of each bar against 0 (p < 0.05 two tailed) are marked with stars. See Table S1 and Figure S3.