Dissociable effects of surprise and model update in parietal and anterior cingulate cortex

Abstract

Brains use predictive models to facilitate the processing of expected stimuli or planned actions. Under a predictive model, surprising (low probability) stimuli or actions necessitate the immediate reallocation of processing resources, but they can also signal the need to update the underlying predictive model to reflect changes in the environment. Surprise and updating are often correlated in experimental paradigms but are, in fact, distinct constructs that can be formally defined as the Shannon information (IS) and Kullback-Leibler divergence (DKL) associated with an observation. In a saccadic planning task, we observed that distinct behaviors and brain regions are associated with surprise/IS and updating/DKL. Although surprise/IS was associated with behavioral reprogramming as indexed by slower reaction times, as well as with activity in the posterior parietal cortex [human lateral intraparietal area (LIP)], the anterior cingulate cortex (ACC) was specifically activated during updating of the predictive model (DKL). A second saccade-sensitive region in the inferior posterior parietal cortex (human 7a), which has connections to both LIP and ACC, was activated by surprise and modulated by updating. Pupillometry revealed a further dissociation between surprise and updating with an early positive effect of surprise and late negative effect of updating on pupil area. These results give a computational account of the roles of the ACC and two parietal saccade regions, LIP and 7a, by which their involvement in diverse tasks can be understood mechanistically. The dissociation of functional roles between regions within the reorienting/reprogramming network may also inform models of neurological phenomena, such as extinction and Balint syndrome, and neglect.

Keywords: Bayes; attention; eye movement; learning; prediction.

Fig. 1.

On each trial of the task, participants began by fixating a central cross. A target (colored dot) appeared on a circular perimeter. Its location was predictable because target locations were similar over runs of 10–20 trials (mean of 15 trials). Two types of unexpected target locations could be observed, as illustrated in A and B. (A) Transition between runs (update trial). Initially, red dots are observed in the upper right; subsequently, blue dots are observed in the lower left. (B) One-off trial. A target appears in an unexpected location (targets are expected in the upper right, but the one-off target appears in the lower left). However, this is a one-off trial, and future targets revert to the original distribution in the upper right. Targets on one-off trials are colored gray. (C) Plot of target locations (angle α from vertical) over 150 trials. Different colored targets are from different runs. One-off targets are shown in gray. (D) Distribution of target locations within a run is a combination of a circular Gaussian, shown in red, and a uniform distribution, shown in black, from which one-off trials are drawn.

Fig. 2.

RTs indicate that participants performed the task as instructed. (A) RTs (time for gaze to arrive at the target) on update and one-off trials, as well as on surrounding trials. The dashed line is the mean RT for all trials. (B) Breakdown of saccadic RTs into onset time (time to leave central fixation), arrival at target (onset of fixation at the target), and dwell time (duration of fixation at the target).

Fig. 3.

Effect of prior entropy, surprise, and updating on saccadic responses. Results of a GLM analysis on saccadic onset, arrival, and dwell times in which the regressors were entropy of the prior (the strength by which the stimulus location was predicted), I_S, and D_KL. The bar height is the mean effect size (beta value) for each parameter across participants, and error bars are the group SEM. a.u., arbitrary units.

Fig. 4.

Pupil dilation is affected by both surprise and updating. The data are plotted as a function of time in the trial (shown on the x axis; labels in the lower panel apply also to the upper panel). The shaded regions represent the period in which the warning stimulus (a brightening of the central fixation cross) and the target were present on the screen. (Lower) Average eye displacement as a function of time, which gives an indication of when participants were looking at the target. (Upper) Results of a GLM in which pupil dilation at each time point was modeled with the same regressors used in the analysis of RTs (Fig. 3) and fMRI data (Fig. 5): entropy of the prior, surprise (I_S), and updating (D_KL) on the trial. The main effect (mean pupil dilation at that time point across all trials) was also modeled. The results show that although surprise (I_S) was correlated with increased pupil diameter, updating (D_KL) was associated with a relative decrease in pupil diameter. The time periods in which these effects were significant [Z-score is >2.3 (i.e., P < 0.01 uncorrected, two-tailed)] are indicated by the starred bars above and below the plot. The effect of D_KL is significant in an interval 730–1,240 ms after target onset; for I_S, the interval is 540–908 ms.

Fig. 5.

Effect of updating in the ACC. (A) Results of whole-brain fMRI analysis. This region in the ACC and pre-SMA was the only area in which there was a significant effect of updating (contrast shows all voxels with a parametric effect of D_KL as defined in Methods), using cluster size-based multiple comparisons correction. The color scale is 2.3 < Z < 3; the peak Z-score is 3.1 at MNI coordinates (6, 10, 54). The ROI denoted by the yellow line is the ACC ROI, a region described as the rCMA zone in a diffusion-weighted parcellation of the cingulate cortex (32). This was the ROI used in the analyses shown in B and C. (B) Effect size for I_S and D_KL in the ACC ROI, where bar height is the mean effect across the group of participants and error bars are the SEM. (C) Raw activity in the ACC ROI plotted as a function of trial-in-run; plotting conventions are as in Fig. 2.

Fig. 6.

Effect of surprise in the posterior parietal cortex. (A) Results of whole-brain fMRI analysis. This region in the posterior parietal cortex was the only area in which there was a significant effect of surprise (contrast shows all voxels with a parametric effect of I_S as defined in Methods), using cluster size-based multiple comparisons correction. The color scale is 2.3 < Z < 3, and the peak Z-score is 4.8 at MNI coordinates (−18, −60, 58). The red circle is the IPS3 ROI (with coordinates ±26, −54, 60), and the yellow circle is the PGp ROI (with coordinates ±32, −64, 44) (33). (B) Effect size for the I_S and D_KL in each ROI; the bar height is the mean effect across the group of participants, and the error bars are the SEM. There is a significant ROI × condition interaction.

Fig. 7.

Parietal activity, but not ACC activity, predicts RT on both one-off and update trials. Bars show the effect size (beta value) ± SEM from a multiple regression in which activity in the three ROIs (ACC, PGp, and IPS3) was used to predict RT. The effects of PGp and IPS3 activity on RT are significant for both one-off trials and update trials, and the effect of ACC activity is not significant in either case. A breakdown of this effect into onset, arrival, and dwell times is shown in Fig. S8.