Prefrontal cortex activity during flexible categorization

Abstract

Items are categorized differently depending on the behavioral context. For instance, a lion can be categorized as an African animal or a type of cat. We recorded lateral prefrontal cortex (PFC) neural activity while monkeys switched between categorizing the same image set along two different category schemes with orthogonal boundaries. We found that each category scheme was largely represented by independent PFC neuronal populations and that activity reflecting a category distinction was weaker, but not absent, when that category was irrelevant. We suggest that the PFC represents competing category representations independently to reduce interference between them.

Figure 1.

Stimuli and behavioral task. A, An image stimulus set was generated by blending prototypes along six morph lines. Monkeys were taught to group the same images under two different category schemes. B, Schematic diagram of the delayed match-to-category task. Each trial began with the monkeys fixating on a dot and holding a lever. The dot was briefly a color that cued the monkey as to which category scheme was relevant. The dot switched to white and a sample image appeared for 600 ms. Following a brief 1 s delay, a test image was presented. If the test image and sample image were of the same category, monkeys released a lever. Otherwise, monkeys continued to hold the lever through a delay until a matching image was displayed. C, Anatomical locations of recording sites of category selective neurons in both monkeys. A, Anterior; P, posterior; D, dorsal; V, ventral.

Figure 2.

Behavioral performance of both monkeys. A, B, Monkeys O (A) and L (B) were proficient at categorizing images in both schemes (scheme A, filled bars; scheme B, open bars). C, The performance of the two monkeys combined.

Figure 3.

Category sensitivity of an example PFC neuron. A, The neuron's average activity (mean ± SEM) in response to all images from the two categories under category scheme A. Insets indicate average responses to each image across the sample presentation and memory delay intervals. B, The same neuron's activity when scheme B was relevant and data were sorted by that scheme. C, The neuron carried less information about category scheme A when the animal was performing category scheme B. sp; Spikes; Fix., fixation epoch.

Figure 4.

The category effect in average population activity. A, B, Normalized activity of PFC neurons to the images in each morph level collapsed across the six morph lines, sorted by preferred category scheme when relevant (left panel), nonpreferred scheme when relevant (center panel), and preferred scheme when irrelevant (right panel) during the sample presentation (A) and memory delay (B).

Figure 5.

Distribution of category selectivity index values. A, Index values of PFC neurons when the preferred category scheme was relevant during the sample presentation (left panel) and memory delay (right panel). B, Index values for the same neurons when the nonpreferred scheme was relevant during the sample presentation (left panel) and memory delay (right panel). C, Finally, index values for the preferred category scheme when irrelevant during sample (left panel) and memory delay (right panel). D, Comparison of overall firing rate of each category-sensitive neuron for the preferred scheme and when the preferred scheme was irrelevant for sample presentation (♦) and memory delay-sensitive neurons (•).

Figure 6.

Category selectivity reflected in correlations between pairs of images during the sample presentation and memory delay. A, Activity in response to the preferred category scheme when it was relevant. Neurons respond more similarly (higher correlations) to images from the same category. To the right is a map of the image comparisons. B, The same neurons carried virtually no information about the nonpreferred category scheme. C, Correlations were weaker to the preferred category scheme when it was irrelevant. D, Comparison of average correlation values showed there was a greater effect of category relevance on category information per se.

Figure 7.

Multidimensional scaling of correlation coefficients during the sample presentation and memory delay. A, The images tended to cluster into two groups corresponding to the two categories (images 1–14 and images 15–28) of the preferred category scheme. The representation of the constituent images (dashed lines) from the two prototypes of each category overlapped. B, The clustering conveyed little information about the nonpreferred category scheme when it was relevant. C, When the preferred category scheme was irrelevant, the clustering by category diminished and clustering of same prototype and category images increased. The scree plots (right panels) indicate that the first two dimensions were sufficient to describe the data. The legends show where the images were in stimulus space for each category scheme.