Post-decision processing in primate prefrontal cortex influences subsequent choices on an auditory decision-making task
- University of Pennsylvania, United States
- Iwate University, Japan
Figures
Figure 1
Task and stereotactic location of vlPFC.
(a) Each monkey decided whether a temporal sequence of tone bursts was predominantly ‘low frequency’ or ‘high frequency’ and responded with a rightward or leftward movement, respectively, of the joystick. The monkey could report its choice any time after stimulus onset. (b) Schematics of the auditory stimulus (+100% and 0% coherence stimuli). The auditory stimulus consisted of a sequence of tone bursts (50 ms duration; 10 ms inter-burst interval). Stimulus coherence refers to the percentage of high-frequency bursts (up to +100%) or low-frequency bursts (down to −100%). (c) vlPFC (pink square) is ventral to the posterior aspect of the principal sulcus (PC) and anterior to the arcuate sulcus (AS; Romanski and Goldman-Rakic, 2002). The dotted box indicates the circumference of the recording chamber. Arrows indicate the anterior (A)-posterior (P) axis and the medial (M)-lateral (L) axis.
Figure 2
Psychophysical performance on the low-high task.
Psychometric (a) and chronometric (b) functions for Monkey T (top) and Monkey A (bottom). These functions were generated from their responses on the current trial. Psychometric functions are plotted as the percentage of trials in which a monkey chose ‘high frequency’ as a function of signed coherence, in which larger negative/positive coherence values indicate more low/high frequency tone bursts. The horizontal gray lines on the psychometric plots indicate lapse rates (errors for strong stimuli, presumably reflecting lapses in attention or inappropriate application of the decision-motor mapping), which were estimated from logistic fits (solid blue lines). Chronometric functions are plotted using the mean RT, which was the time interval between stimulus onset and onset of joystick movement. Gray dots are low-frequency choices, and black dots are high-frequency choices. Solid pink curves are simultaneous fits of both the psychometric and chronometric data to a drift-diffusion model (DDM). The horizontal dashed gray lines on the chronometric plots indicate choice-dependent non-decision times (NDT) estimated by the DDM fits. Decision times (DT) were estimated as the difference between the trial-specific RT and the choice-specific NDT. (c) Psychometric functions computed separately for different sequential conditions, as indicated in the top panel. (d) Distributions of best-fitting, session-by-session beta coefficients (β0, overall choice bias; β1, sensitivity to coherence; β2, the tendency to repeat a correct choice; and β3, the tendency to repeat an erroneous choice) and lapse rates from the logistic fits. Circles indicate data from sessions using 1250 and 2500 Hz as low and high frequencies, respectively; squares indicate data from other sessions (note that the two conditions corresponded to differences in lapse rates for monkey T but little effect on the other model parameters). Filled data points indicate likelihood-ratio test, H0: regression coefficient equals 0, p<0.05. Horizontal bars indicate median values; red bars indicate Wilcoxon sign-rank test, H0: median value equals 0, p<0.05.
Figure 3
Neuronal sensitivity to choice in single vlPFC neurons.
(a–d) The left plots are raster and peristimulus-time histograms from correct trials only showing sensitivity to low-frequency (<0% coherence; red) and high-frequency choices (>0% coherence; blue). The thick lines indicate mean firing rate, and the dotted lines indicate the 95% confidence intervals. Data are aligned relative to stimulus onset. Gray circles in the raster plots indicate the time of onset of joystick movement. The middle plots show the responses of the same neurons but aligned relative to the onset of joystick movement. The arrow indicates the time of peak choice selectivity. The right plot summarizes each neuron’s firing rate during its peak firing rate ±100 ms: correct low-frequency choices are shown in red, high-frequency choices in blue, and incorrect choices in gray (only for coherences with at least five trials). Error bars indicate the standard error of the mean.
Figure 4
Population selectivity for vlPFC neurons.
(a) Summary of choice selectivity. Data from individual neurons are sorted by the onset of choice selectivity (open circles), defined as the first of three consecutive time bins with reliably different responses for the two choices (Wilcoxon rank-sum test, H0: no median difference in firing rates for the two choices, p<0.05, FDR corrected). Color indicates the ROC value of choice selectivity from correct trials (see legend). Rows show data for high (<−80% versus >+80%), middle (−80% to −20% versus +80% to+20%), and low (−20% to 0 versus 0 to +20%) coherence trials, as indicated. (b) Percentage of neurons with significant selectivity for choice or coherence (Wilcoxon rank-sum test for H0: no median difference in firing rates elicited by high- versus middle- coherence stimuli for each choice, p<0.05, FDR corrected) computed in 300 ms time bins with 10 ms steps. Choice selectivity is shown separately for high, middle, and low coherences, as indicated. Red points indicate times corresponding to a significant difference in the proportion of choice-selective neurons at each coherence level (running χ2-test for H0: proportion is the same, p<0.05, FDR corrected). Coherence selectivity is shown in dark red for preferred choices (i.e., the choice direction that elicits higher firing rates) and light red for non-preferred choices. In the leftmost panel, the horizontal bars represent the range of the inferred times of the decision commitment for high (black), middle (dark gray), and low (light gray) coherence stimuli (the range is indicated by the large vertical bar). In (a) and (b), the data in each panel are aligned relative to different task epochs (from left to right): stimulus onset, inferred decision commitment, onset of joystick movement, and time of reward delivery.
Figure 5
Choice selectivity on correct and error trials.
Scatterplots showing, on a neuron-by-neuron basis, the peak ROC choice-selectivity value computed on correct versus error trials. Both values were computed from spiking data occurring at the time of peak ROC-based choice selectivity from correct trials for the given neuron. Black/gray points correspond to data from high/middle coherence stimuli. The line in each panel is the line of unity. The panels show data computed relative to different task epochs (from left to right): stimulus onset, inferred decision commitment, onset of joystick movement, and time of reward delivery. Across all epochs, error ROC values tended to be smaller than correct ROC values (Wilcoxon sign-rank test for H0: median ROC values are the same, p<0.05). Different panels have different numbers of data points because for some sessions, there were not enough trials to reliably calculate the error ROC.
Figure 6
Classifier analysis.
The ability of a linear classifier to determine from the population of vlPFC neurons the: (a) current choice (low frequency [<0%] or high frequency [>0%]), or (b) stimulus coherence in four bins (<−50%, −50–0%, 0%–+50%, or >+50%). Results were computed using correct trials only in 300 ms time bins with 10 ms steps. Thick lines represent median decoding performance; dashed lines are the interquartile range. In the leftmost panel, the horizontal bar represents the range of the inferred times of the decision commitment for high (black), middle (dark gray), and low (light gray) coherence (the range is indicated by the large vertical bars). Choice- and coherence-decoding performance is aligned relative to different task epochs (from left to right): stimulus onset, inferred decision commitment, onset of joystick movement, and time of reward delivery. We did not conduct a classifier analysis on error trials because there was not enough data to generate reliable results.
Figure 7
Choice selectivity for the current and next trial.
For Monkey T (top) and Monkey A (bottom), choice selectivity is plotted as a function of time relative to the onset of joystick movement (a) and reward delivery (b). Lines indicate ROC-based choice selectivity computed in 300 ms time bins, with 10 ms steps from pooled spiking data across all recorded neurons (z-scored per neuron) that contributed at least 121 for monkey T and 54 trials for monkey A under the given conditions. Solid/dotted lines correspond to correct/error outcomes on the current trial. Black lines indicate selectivity for repeated (ROC values > 0.5) versus switched (<0.5) choices on the next trial, relative to the choice on the current trial (i.e., values > 0.5 imply that the neuronal population tended to respond more in anticipation of a repeated choice). For reference, gray lines indicate selectivity for the preferred choices on the current trial (i.e., values > 0.5 indicate, by definition, selectivity for the choice that elicited the larger average spike rate during peak firing rate ±100 ms for each neuron). Red points, computed only for the black curves, indicate permutation test for H0: ROC value equals 0.5, p<0.05.
Figure 8
Effect of microstimulation on behavioral performance on the current and next trial.
(a and b) Single-site examples of the effects of vlPFC microstimulation on psychometric performance on the current trial for a low-choice site (a) and a high-choice site (b). Psychometric functions are plotted as in Figure 2. Red/blue symbols are for data from trials with/without microstimulation. Solid lines are logistic fits, computed separately for the two conditions. Dotted lines are 95% confidence intervals of the non-microstimulation trials that were calculated by a bootstrap procedure (Ding and Gold, 2012b). (c and d), Scatterplots showing session-by-session effects of microstimulation on the correlation between neuronal choice selectivity and the percent change in psychometric choice bias (c); Spearman’s rank correlation coefficient ρ = 0.15, p=0.43) and the change in psychometric slope (d); ρ = 0.15, p=0.42) of the current decision. (e and f) Single-site examples of microstimulation’s effects on psychometric performance on the next trial for a low-choice site (e) and a high-choice site (f). The data are formatted in the same manner as panels (a) and (b). (g and h), Scatterplots show session-by-session effects of microstimulation on the correlation between neuronal choice selectivity and the percent change in psychometric choice bias (g; ρ = 0.60, p=0.0003) and the change in psychometric slope (h; ρ = 0.11, p=0.56) of the next decision. Filled data points are significant single-session microstimulation-induced changes in the given psychometric property (permutation test, p<0.05).
Additional files
-
Transparent reporting form
- https://doi.org/10.7554/eLife.46770.010
Download links
A two-part list of links to download the article, or parts of the article, in various formats.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Post-decision processing in primate prefrontal cortex influences subsequent choices on an auditory decision-making task
eLife 8:e46770.
https://doi.org/10.7554/eLife.46770
