Dissociable neural mechanisms track evidence accumulation for selection of attention versus action

Abstract

Decision-making is typically studied as a sequential process from the selection of what to attend (e.g., between possible tasks, stimuli, or stimulus attributes) to which actions to take based on the attended information. However, people often process information across these various levels in parallel. Here we scan participants while they simultaneously weigh how much to attend to two dynamic stimulus attributes and what response to give. Regions of the prefrontal cortex track information about the stimulus attributes in dissociable ways, related to either the predicted reward (ventromedial prefrontal cortex) or the degree to which that attribute is being attended (dorsal anterior cingulate cortex, dACC). Within the dACC, adjacent regions track correlates of uncertainty at different levels of the decision, regarding what to attend versus how to respond. These findings bridge research on perceptual and value-based decision-making, demonstrating that people dynamically integrate information in parallel across different levels of decision-making.

Fig. 1

Behavioral paradigm. a Participants viewed random dot motion patterns and could indicate whether the dots were primarily moving up or down and/or whether they were majority red or blue. They responded with either a left or right button press. Responses were bivalent, denoting both a color and a motion direction, and participants were rewarded for each stimulus attribute they correctly discriminated on a given trial. b The coherence and correct response for motion and color dimensions were varied orthogonally across trials. Four participant-specific coherence levels were used for each attribute. c Participants performed three epochs (192 trials each) that varied in motion/color reward associations, rewarding both either equally (Epoch 1) or differently (Epochs 2 and 3). Reward contingencies were explicitly indicated to the participants at the start of each epoch. *Response mappings and Epochs 2 and 3 reward associations were counter-balanced across participants

Fig. 2

Behavioral sensitivity to attribute evidence and rewards. During Epochs 2 and 3, responses were highly sensitive to both the amount of evidence and the relative reward for the two attributes. a A psychometric curve shows that participants were much more likely to select a response the more evidence it provided for the high-reward attribute. b Regression coefficients for the influence of high- and low-reward coherence on choice. While high-reward attribute coherence exerted the strongest influence on responses, participants were still sensitive to the evidence supporting the low-reward attribute. c Consistent with psychometric patterns in a and b, RTs were also more sensitive to the (unsigned) coherence of the high-reward attribute relative to the low-reward attribute. See also Supplementary Fig. 1. Error bars reflect s.e.m

Fig. 3

vmPFC and dACC differentially encode the relative evidence for the high- and low-reward attributes. a vmPFC (yellow) and dACC (red) ROIs were defined a priori based on the relevant findings from research on integration of information from multi-attribute stimuli displayed on a normalized Montreal Neurological Institute (MNI) template^,. b vmPFC positively tracked the evidence each attribute provided for the chosen response (signed coherence), but it did not weigh the evidence for both attributes equally. Rather, responses to the two attributes were weighed in proportion to the reward expected for responding correctly to that attribute. For reference, the inset shows the reward amounts (in dollars) expected for each attribute. c dACC tracked how little evidence was available for these two attributes, weighing the evidence for the two attributes in proportion to the influence that attribute will have on the ultimate choice (inset from Fig. 2b), potentially reflecting the amount of attention placed on that attribute while forming a decision. Regression coefficients are plotted with their corresponding s.e.m

Fig. 4

dACC encoding of attribute coherence varies along rostrocaudal axis. a A nonparametric whole-brain analysis revealed that more caudal regions of dACC negatively tracked the coherence of the high-reward attribute (red) and more rostral regions positively tracked the coherence of the low-reward attribute (green). Activations reflect t-statistics (t > 3.35, p < 0.001), extent-thresholded to achieve a cluster-corrected family-wise error p < 0.05, and are displayed on the inflated CARET surface. b, c This rostocaudal pattern was confirmed with a set of independent ROIs drawn from an earlier study (shown in b), which proposed that these reflect a range of uncertainty/conflict levels, from low-level responses (e.g., motor actions) most caudally to more abstract responses (e.g., decisions and strategies) more rostrally. Coefficients and corresponding s.e.m. are plotted in c for regressions of BOLD activity on high and low attribute coherence, performed separately for each ROI. b Is republished with permission of the Society for Neuroscience, from Taren et al.. *p < 0.05, ***p < 0.005