Figure 1. Homogeneity and heterogeneity of chosen value correlates.

(A) At decision time, chosen value correlates appeared homogenous across regions in their expression. The coefficient of partial determination (CPD) for chosen value averaged across populations of DLPFC (n = 310), OFC (n = 214) and ACC (n = 333) neurons (lines denote mean ± SE for each region). CPD was calculated by regressing chosen value onto firing rate during the choice period of a cost-benefit decision making task (see Materials and methods). Chosen value correlates were not significantly different between any brain region (permutation tests; DLPFC v OFC, no cluster survived thresholding, DLPFC v ACC, p=0.2706, OFC v ACC, no cluster survived thresholding; see Materials and methods). Dashed lines mark the null hypothesis level for CPD in each cortical area (see Materials and methods). (B) Population averages when chosen value was regressed onto firing rate during reward delivery. OFC showed stronger chosen value correlates following reward onset than ACC and DLPFC (permutation tests; OFC v DLPFC, p=0.0010, OFC v ACC, p=0.0028; see Materials and methods). (C and D) Within each region, chosen value correlates were heterogeneous across neurons. Chosen value correlates of the individual neurons contributing to the population averages in A and B respectively. Within each matrix: each row is a neuron (sorted by maximum CPD within the corresponding epoch and area), each column is a 10 ms time bin. Hence, neurons are sorted in a different order in C and D. Chosen value coding at reward delivery was weaker than at choice. Figure 1—figure supplement 1 shows the fraction of neurons with reliable coding of chosen value at choice and at the outcome. Figure 1—figure supplement 2 shows that OFC codes chosen value, as opposed to chosen benefit alone, at the time of reward delivery.

DOI: http://dx.doi.org/10.7554/eLife.18937.002

Figure 1—figure supplement 1. Fraction of neurons with reliable coding of chosen value at choice (A) and at outcome (B).

The regression coefficient for each neuron at each time point was assessed for significance to p<0.05 level. The proportion of significant neurons within each cortical area was smoothed across five 10 ms time points. The fraction of significant neurons exceeds the null hypothesis (dashed magenta line) for all time points after choice onset.

DOI: http://dx.doi.org/10.7554/eLife.18937.003

Figure 1—figure supplement 2. Orbitofrontal cortex codes chosen value, as opposed to chosen benefit alone, at the time of reward delivery.

The regression analysis from Figure 1 was repeated, with the chosen value predictor split into chosen benefit and chosen cost components for each trial. Both chosen benefit and chosen cost coding are present at the time of reward delivery. (A) and (B) plot the population average coefficient of partial determination for the chosen benefit and cost regressors timelocked to reward onset. Dashed lines mark the null hypothesis level for CPD in each cortical area (see Materials and methods). (C) and (D), as in Figure 1—figure supplement 1, show the fraction of neurons coding chosen benefit (C) or cost (D) within each cortical area. The dashed magenta lines indicate the null hypothesis fraction of significant coding neurons.

DOI: http://dx.doi.org/10.7554/eLife.18937.004