Preferential encoding of visual categories in parietal cortex compared with prefrontal cortex

Abstract

The ability to recognize the behavioral relevance, or category membership, of sensory stimuli is critical for interpreting the meaning of events in our environment. Neurophysiological studies of visual categorization have found categorical representations of stimuli in prefrontal cortex (PFC), an area that is closely associated with cognitive and executive functions. Recent studies have also identified neuronal category signals in parietal areas that are typically associated with visual-spatial processing. It has been proposed that category-related signals in parietal cortex and other visual areas may result from 'top-down' feedback from PFC. We directly compared neuronal activity in the lateral intraparietal (LIP) area and PFC in monkeys performing a visual motion categorization task. We found that LIP showed stronger, more reliable and shorter latency category signals than PFC. These findings suggest that LIP is strongly involved in visual categorization and argue against the idea that parietal category signals arise as a result of feedback from PFC during this task.

Figure 1. Delayed match-to-category (DMC) task

(a) Monkeys grouped 6 motion directions into two categories (corresponding to the red and blue arrows) separated by a learned “category boundary”. Two additional directions were shown as sample stimuli that were on the category boundary and had ambiguous category membership (the two yellow arrows). (b) Monkeys performed a delayed match-to-category (DMC) task, and had to indicate (by releasing a lever) whether sample and test stimuli were in the same category. “RF” indicates the position of a neuron’s receptive field. (c,d) The monkeys’ average categorization performance (proportion of directions classified as C1) during LIP (c) and PFC (d) recordings is shown is shown as a function of distance from the category boundary.

Figure 2. Examples of category selective LIP and PFC neurons

The responses of three LIP (a–c) and three PFC (d–f) neurons are shown. The red and blue traces indicate the three directions in category 1 and category 2, respectively. The pale red and blue traces represent directions closer to the category boundary, and the dark traces represent directions in the center of each category. Each neuron shows a tendency for strong selectivity for sample category during the sample, delay and/or test epochs. Data is shown only for correct trials.

Figure 3. Strength of category selectivity across LIP and PFC populations

The strength of category selectivity was measured using (a) ROC and (b) CTI analysis. ROC values for individual neurons could vary from 0.5 to 1.0. Average fixation period ROC values greater than 0.5 are expected because raw ROC values (which can vary from 0.0 to 1.0) are rectified about 0.5, and this does not indicate any neuronal bias or anticipatory category signals (see Methods). CTI values could vary from −1.0 to 1.0. For both measures, greater positive values indicate stronger category selectivity, and mean values are shown for LIP (dark gray) and PFC (light gray) across all direction selective (according to one-way ANOVA) neurons in each epoch. During the fixation epoch, ROC and CTI values are shown for neurons that were direction selective in any epoch. Error bars indicate the standard error of the mean. Asterisks denote the level of significance of T-Test (LIP vs. PFC) significance (* = P<0.05; ** = P<0.01).

Figure 4. Time-course of LIP and PFC category selectivity

The time-course of category selectivity across direction-selective LIP and PFC populations was determined by “sliding” ROC (a) and CTI (c) analyses. The shaded gray area around the solid traces (the mean ROC or CTI value) indicates the standard error of the mean. Average fixation period ROC values greater than 0.5 are expected because raw ROC values (which can vary from 0.0 to 1.0) are rectified about 0.5, and this does not indicate any neuronal bias or anticipatory category signals (see Methods). (b,d) Cumulative latency distributions, across all neurons that showed significant category selectivity prior to 500 ms after sample onset according to (b) ROC (LIP: N = 62; PFC: N = 243) and (d) CTI (LIP: N = 39; PFC: N = 181) analysis, shows the fraction of LIP and PFC neurons that had become category selective by each time-point. (e,f) Scatter plots show the relationship between category selectivity strength and category selectivity latency for LIP (e) and PFC (f) neurons. (g,h) Scatter plots show the relationship between firing rate and category selectivity latency for LIP (g) and PFC (h) neurons. For panels e–h, linear regression fits are indicated by the dotted line.

Figure 5. Neuronal activity to on-boundary directions with ambiguous category membership

(a–d) Examples of single LIP (a–b) and PFC (c–d) neurons to ambiguous sample directions (solid red and blue traces, activity sorted according to monkeys’ reports of category membership on each trial) and non-ambiguous directions (dotted red and blue traces, activity averaged across the three directions in each category). (e,f) Population average category selectivity (ROC) on ambiguous trials is shown for LIP (e) and PFC (f). Neurons are sorted according to whether they preferred Category 1 (blue) or Category 2 (red) on non-ambiguous trials. ROC values of 0.0 and 1.0 indicate strong selectivity for Category 1 and Category 2, respectively. Error bars show the standard error of the mean.