Article Figures & Data
Figures
- Figure 1.
Event sequence of the DYSC task and locations of neural recording in parietal (PAR) and prefrontal (PFC) cortex. Each trial, we presented a small circular sample stimulus and a line serving as a boundary cue. We varied the order of presentation of these stimuli, using either a sample–boundary sequence or a boundary–sample sequence. A, Categorizing stimuli according to the left/right (LR) rule under the sample–boundary sequence. Trials began with the presentation of a central gaze fixation target (gaze fixation was required throughout the trial until the response was made). A sample stimulus was presented at one of 8 or 12 randomly selected locations for 400 ms, followed by a 900 ms delay period (delay 1). The LR rule was instructed by the presentation of the boundary cue in a vertical orientation for 400 ms, followed by a 900 ms delay period (delay 2). The monkey was required to determine the spatial category of the sample position stored in working memory by evaluating its spatial relationship to the boundary cue. The sample on this trial belongs to the spatial category “right.” After delay 2, two choice stimuli were sequentially presented for 700 ms each in random order, one in the opposite spatial category as the sample (choice 1; “left”) and one in the same spatial category (choice 2; “right”). The DYSC task is a delayed category match-to-sample design. The monkey was rewarded (with a drop of juice) if it pressed the response key during the period of time that the matching choice was visible (located in the same spatial category as the sample; choice 2 on this trial). B, Categorizing stimuli according to the above/below (AB) rule under the sample–boundary sequence. The boundary cue is presented in a horizontal orientation instructing the AB rule on this trial, and choice 1 matches the spatial category of the sample (“above”). C, Categorizing stimuli according the LR rule under the boundary–sample sequence. In this case, the sample stimulus is assigned to a category based on a boundary cue stored in working memory. D, E, Locations of neural recordings in parietal and prefrontal cortex of monkeys 1 and 2 relative to positions of the principal sulcus (PS), central sulcus (CS), and intraparietal sulcus (IPS) as reconstructed from structural MRI images. The perspective is a top-down view of the left cerebral hemisphere. Anterior (Ant), posterior (Post), medial (Med), and lateral (Lat) directions are as indicated by the arrows. The larger open circles indicate inner diameter of recording chambers over parietal and prefrontal cortex. The smaller filled circles within each cortical area indicate regions sampled by electrode penetrations during neural recording. F, Eight-position sample array. G, Twelve-position sample array.
- Figure 2.
Behavioral performance of monkeys 1 (A–D) and 2 (E–H) in the DYSC task. The horizontal axis of each plot indicates the position of the sample stimulus, in degrees angle counterclockwise from the direction to the right of the fixation target (defined as 0°; sample eccentricity was fixed at 13°). The red- and blue-shaded regions indicate spans of sample angle corresponding to the spatial categories right and left (A, C, E, G), above and below (B, D, F, H), respectively. The lines and symbols indicate the probability (proportion of trials) that monkeys selected choices that were located in the right (red lines) or left (blue lines), above (blue lines) or below (red lines) spatial categories for each sample position and rule. The vertical dashed lines indicate angles corresponding to the category boundary under a given rule. A, B, Performance of monkey 1 under the LR rule, when the boundary cue was vertical. The solid lines indicated performance on trials using the boundary–sample sequence, and the dashed lines performance on trials using the sample–boundary sequence. A, Under the LR rule, monkey 1 selected left choices with high probability (blue symbols and lines) when the sample category was left (blue shading), and right choices with high probability (red symbols and lines) when the sample category was right (red shading). B, Under the LR rule, the vertical category of the choices selected by monkey 1 (above or below) did not relate systematically to the vertical category of the sample. C, D, Performance of monkey 1 under the AB rule, when the boundary cue was horizontal. C, Under the AB rule, the horizontal category of the choices selected by monkey 1 (left or right) did not relate systematically to the horizontal category of the sample. D, In contrast, under the AB rule, monkey 1 selected above choices with high probability (blue symbols and lines) when the sample category was above (blue shading), and below choices with high probability (red symbols and lines) when the sample category was below (red shading). E–H, Corresponding data for monkey 2.
- Figure 3.
Rasters and spike density functions illustrate the activity of single neurons in parietal cortex (A, C, E) and prefrontal cortex (B, D, F) varying significantly (p < 0.05) as a function of sample position (A, B), rule (or boundary orientation) (C, D), and spatial category (E, F). Neural activity associated with the preferred position, rule, or category of each neuron is illustrated in the upper raster of each panel and activity on nonpreferred trials is illustrated in the lower raster (the thin blue line in the lower rasters shows activity on preferred trials for comparison).
- Figure 4.
Proportion of task-related neurons in parietal cortex (black) and prefrontal cortex (gray) in which firing rate related significantly (p < 0.05) to the main effects of position, rule, and category in the ANOVA/ANCOVA (relative to the total number of neurons exhibiting any significant effect) in monkey 1 (A) and monkey 2 (B). Counts of parietal neurons (black) and prefrontal neurons (gray) exclude neurons significant for multiple main factors.
- Figure 5.
Plots of decoding accuracy over time indicate the proportion of trials in which sample position (A, B), rule (C, D), and spatial category (D, E) were accurately decoded from population activity patterns measured in successive 50 ms time bins in parietal cortex (red) and prefrontal cortex (blue). Data from monkey 1 (A, C, E) and monkey 2 (B, D, F) are plotted separately. Time bins in which the proportion of correctly decoded trials varied significantly between prefrontal and parietal cortex are indicated by filled circles and sections of the time courses plotted with a thicker line (z test of proportions, α level indicated in each panel).
- Figure 6.
Rasters and spike-density functions (σ = 20 ms) illustrate rule-dependent, category-selective activity of a neuron in parietal cortex (A, B) and a neuron in prefrontal cortex (C, D). The activity of both neurons was significantly influenced by the interaction between rule and category (p < 0.05). A, Activity of the parietal neuron on LR rule trials (vertical boundary cue) using the boundary–sample sequence. B, Activity of the same parietal neuron on AB rule trials (horizontal boundary cue). The neuron exhibits a moderate increase in firing rate late in the delay period following the presentation of the sample stimulus when the sample was located in the above spatial category (positions 1–6: diagram of sample positions at right). C, Activity of the prefrontal neuron on LR rule trials (vertical boundary cue) using the boundary–sample sequence. D, Activity of the same prefrontal neuron on AB rule trials (horizontal boundary cue). The neuron was strongly activated during the boundary period and the subsequent delay period on trials in which the sample stimulus was located in the above spatial category (positions 1–4; diagram of sample positions at right). The black bars labeled “S” and “B” indicate the sample and boundary periods, respectively. The red tick mark on each row of the rasters indicates the time at which the monkey depressed the response key.
- Figure 7.
Spike density functions (σ = 40 ms) illustrate the activity time course of individual neurons (4 each in parietal and prefrontal cortex) exhibiting activity that was significantly influenced (p < 0.05) by the interaction between rule and either horizontal or vertical category in the ANCOVA. A–D, Plots illustrate the activity of four parietal neurons exhibiting rule-dependent category selectivity. Activity on trials when the sample was located in the right (black) and left (gray) spatial categories is plotted separately for neurons with horizontal category preferences (B, D). Activity on trials when the sample was located in the above (gray) and below (black) spatial categories is plotted separately for neurons with vertical category preferences (A, C). In each pair of plots in each panel, activity on compatible rule trials (e.g., the LR rule for neurons preferring horizontal categories, or the AB rule for neurons preferring vertical categories) is plotted on the left, and activity on incompatible rule trials on the right. E–H, Corresponding data for four prefrontal neurons.
- Figure 8.
Average normalized population SDFs (σ = 40 ms) illustrate the activity of neurons coding rule-dependent category in parietal cortex and prefrontal cortex. Separate SDFs in each plot illustrate the activity of the same neural population on trials that the sample was assigned to the preferred (black) and nonpreferred (gray) category of each neuron. Pairs of plots comprising each panel (A–H) illustrate population activity under the LR and AB categorization rules, respectively (the number of neurons in each population is indicated in the bottom right corner of the right plot of each pair). Neurons were included if their activity during the boundary and following delay periods related significantly (p < 0.05) to horizontal category and its interaction with rule, or vertical category and its interaction with rule. Neurons were excluded if their activity related to sample position (p < 0.1) during the sample or subsequent delay periods. The vertical lines delineate the sample period (black bar labeled “S”) and the boundary cue period (black bar labeled “B”), and indicate the presentation of the first choice (“C1”). A, C, Population SDFs illustrating the activity of neurons selective for horizontal categories under the LR rule (left) and AB rule (right) recorded using the sample–boundary sequence. B, D, Population activity of neurons selective for vertical categories under the LR rule (left) and AB rule (right) recorded using the sample–boundary sequence. E–H, Corresponding data for activity recorded using the boundary–sample sequence.
- Figure 9.
Distribution of the category selectivity index for single neurons recorded using the sample–boundary sequence in parietal and prefrontal cortex. Neurons were included if their activity related significantly to either horizontal or vertical category (p < 0.05 by ANCOVA/ANOVA). The vertical white dashed line indicates the mean value of the category index in each distribution. A, Distribution of category index values in parietal cortex on trials in which the rule was compatible with the category preference of each neuron. B, Distribution of category index values in prefrontal cortex on compatible rule trials. C, Distribution of category index values in parietal cortex on incompatible rule trials (same neural sample as in A). D, Distribution of category index values in prefrontal cortex on incompatible rule trials (same neural sample as in B).
- Figure 10.
Decoding time courses illustrate fluctuation in the accuracy of decoding spatial category based on population activity patterns in parietal cortex (red) and prefrontal cortex (blue), measured in successive 50 ms time bins. Separate decoding time courses in each panel plot the results obtained using population activity on trials in which the rule was compatible with the category preference of each neuron (red and blue), and trials in which the rule was incompatible (black). Time bins in which the proportion of correctly decoded trials differed significantly between compatible and incompatible rule trials are indicated by black circles and corresponding thicker sections of the decoding time courses. Neurons in this analysis were ranked according to the p value associated with the interaction between rule and category in the ANOVA/ANCOVA, and then varying numbers of the most significant neurons were selected to include in the populations used for the decoding. Decoding results indicate the accuracy obtained when based on the activity of the most significant 70 neurons (solid lines), the most significant 200 neurons (dashed lines), or all significant neurons (dotted lines) in each cortical area and monkey. A, B, Decoding accuracy obtained when using population activity in parietal cortex (A) and prefrontal cortex (B) of monkey 1 on sample–boundary trials. The difference between decoding accuracy on compatible and incompatible rule trials measures the modulation of category signals by the rule. For the decoding analysis including all significant neurons, 55 neurons in parietal cortex and 58 neurons in prefrontal cortex contributed. C, The blue and red time courses indicate the mean posterior probability over time (averaged over trials) associated with the correct spatial category on compatible rule trials, based on neural activity in prefrontal and parietal cortex, respectively (data from monkey 1 on sample–boundary trials). The black time course illustrates the cumulative difference between the two time series. The diagonal dashed lines indicate the upper and lower confidence boundaries established by the sequential trials test. The cumulative difference function crosses the upper confidence boundary, indicating that the posterior probability is significantly larger in prefrontal cortex relative to parietal cortex (p < 0.05). D–F, Corresponding data from monkey 2 on sample–boundary trials. For the decoding analysis including all significant neurons, 62 neurons in parietal cortex and 90 neurons in prefrontal cortex contributed. G–I, Corresponding data from monkey 1 on boundary–sample trials. For the decoding analysis including all significant neurons, 104 neurons in parietal cortex and 120 neurons in prefrontal cortex contributed.
- Figure 11.
Time course of regression coefficients obtained for category and the interaction between category and rule for individual neurons in parietal and prefrontal cortex. Each row in the color plots indicates the time series of regression coefficients obtained for a single neuron normalized to the peak coefficient for that neuron. In monkey 1, data from sample–boundary and boundary–sample trials are aligned to the onset of the second stimulus (“S2”) and combined. (The two data sets shared the same duration of S2. The different durations of the first stimulus, S1, included in these data are indicated by the black and white portions of the bar marked “S1.” The different timings of the onset of the first choice are indicated by vertical lines marked “C1a” and “C1b.”) A, B, Regression coefficients for category (A) and the interaction between rule and category (B) obtained for parietal neurons in monkey 1. The white lines show the time of peak coefficients in prefrontal cortex (Prefrontal peak) of the same monkey for comparison. C, D, Regression coefficients for category (C) and the interaction between rule and category (D) obtained for prefrontal neurons in monkey 1. The white lines show the time of peak coefficients in parietal cortex (Parietal peak) of this monkey for comparison. The dark blue regions indicate time points at which coefficients were nonsignificant. E–H, Corresponding data for monkey 2.
- Figure 12.
Comparison of the strength and timing of regression coefficients obtained for the interaction between rule and category in parietal and prefrontal neurons (data restricted to regression coefficients that were significant at p < 0.01). A, B, Plots illustrate variation over time in the mean population regression coefficient for the interaction in parietal cortex (red lines) and prefrontal cortex (blue lines) of monkey 1 (A) and monkey 2 (B). C, D, Results of sequential trials tests applied to population coefficient time series in parietal and prefrontal cortex in monkey 1 (C) and monkey 2 (D), starting at the onset of the second stimulus in the trial sequence. The thick black lines indicate the cumulative difference between population coefficient time series in the two cortical areas (prefrontal − parietal). The cumulative difference functions deflect upward and cross the upper confidence boundary (at the points indicated by the vertical blue lines), indicating that the coefficient for the interaction between rule and category is significantly larger in prefrontal cortex (p < 0.05, minimum effect size of 0.8). E, F, Frequency distribution histograms of the time to the half-maximum regression coefficient for the interaction between rule and category in prefrontal cortex (blue bars) and parietal cortex (red bars) of monkey 1 (E) and monkey 2 (F). G, H, Cumulative distributions of the time to the half-maximum regression coefficient for the interaction between rule and category in prefrontal cortex (blue lines) and parietal cortex (red lines) of monkey 1 (G) and monkey 2 (H).
- Figure 13.
Relationship of population decoding accuracy to behavioral performance. In this analysis, population activity was used to decode rule-dependent spatial category defined jointly by sample position and boundary orientation and coded as a categorical variable with four levels (left, right, above, and below). Decoding is based on the 50 neurons in each cortical area and monkey having activity most significantly related to the interaction between rule and category. Decoding accuracy was significantly reduced on error trials (gray) relative to correct trials (black), both in parietal cortex and prefrontal cortex in monkeys 1 and 2 considered individually. The significance of the difference in decoding accuracy on correct and error trials, evaluated using the z test of proportions, is as indicated by asterisks in the figure. Chance decoding (given 4 categories) is 25% correct.
