Posterior Parietal Cortex Plays a Causal Role in Perceptual and Categorical Decisions

Abstract

Posterior parietal cortex (PPC) activity correlates with monkeys’ decisions during visual discrimination and categorization tasks. Yet recent work questioned whether decision-correlated PPC activity plays a causal role in such decisions. That study focused on PPC’s contribution to motor aspects of decisions (deciding where to move), but not sensory evaluation aspects (deciding what you are looking at). We employed reversible inactivation to compare PPC’s contributions to motor and sensory aspects of decisions. Inactivation affected both aspects of behavior, but preferentially impaired decisions when visual stimuli, rather than motor response targets, were in the inactivated visual field (IVF). This demonstrates a causal role for PPC in decision making, with preferential involvement in evaluating attended task-relevant sensory stimuli compared to motor planning.

One Sentence Summary:

Using reversible inactivation and electrophysiological recordings, we demonstrate that the primate PPC plays a causal role in perceptual and categorical decisions, with a preferential role in evaluating sensory stimuli compared to motor planning.

Decision-making requires evaluating task-relevant sensory stimuli to select appropriate motor responses. The primate PPC is well suited to mediate decision making because of its anatomical position at a midpoint in the sensory-cognitive-motor cortical hierarchy(1–3). Indeed, there is a correlation between PPC activity and monkeys’ decisions during visual discrimination and categorization tasks(4–12). Much of this work has focused on the lateral intraparietal area (LIP) (13–15). Yet, PPC’s role in sensory stimulus evaluation and motor planning aspects of decisions has been debated. Recent work showed that reversible inactivation of LIP does not affect motor aspects of decisions during a visual motion discrimination task(16). This leaves open the possibility that LIP is instead more engaged in processes related to evaluating sensory stimuli (e.g. stimulus feature processing, perceptual or abstract categorization, and selective attention), as opposed to selecting motor responses, during decision making. Furthermore, neuronal recordings have demonstrated correlates of more flexible categorical or rule-based tasks in PPC(13), raising the possibility that the primate PPC may be more engaged in decisions requiring greater cognitive flexibility or abstraction than in the classic perceptual decision tasks.

We directly compared LIP’s role in sensory stimulus evaluation and motor planning functions during flexible visual perceptual and categorical decisions. We reversibly inactivated LIP in one hemisphere with intracortical injections of the muscimol during several variants of motion-direction categorization (MDC) and motion-direction discrimination (MDD) tasks. Within each task, monkeys reported their decisions about the category membership (MDC task) or motion direction (MDD task) of a sample stimulus with a saccade to a target of the associated color (Fig. 1A). The mappings between motion categories/directions and target color were arbitrarily assigned at the start of training (and fixed across the study), with the two categories (MDC) or directions (MDD) assigned to red and green targets, respectively. Because the locations of the red and green targets were randomly interleaved across trials, there was no fixed mapping between motion category/direction and saccade direction. In the MDC task, 10 motion directions were shown at three angular distances from the category boundary (Fig. 1B). In the MDD task, two directions of motion were shown each at three levels of coherence (i.e. signal to noise). To assess LIP’s contribution to stimulus evaluation and motor planning, each task was tested in blocks of trials using three spatial configurations of motion stimuli and saccade targets with respect to the location of the IVF contralateral to muscimol injection (Fig. 1C). In the “Stimulus IN” (S_IN) condition, the motion stimulus was shown within the IVF, with both saccade targets outside the IVF. In the “Target IN” (T_IN) condition, the motion stimulus was outside the IVF and one of the colored saccade targets was shown in the IVF. In the “Both OUT” (B_OUT) condition, neither the motion stimuli nor saccade targets were in the IVF.

Fig. 1.

Behavioral task. (A) Monkeys reported their categorical (MDC task) or directional (MDD task) decisions about visual motion stimuli by choosing either the green or red saccade target. The positions of red and green targets were randomly chosen on each trial. Using an RT design, monkeys could initiate their saccade as soon as they had made their decision. (B) The motion direction categorization task (MDC) required grouping ten motion directions (indicated by the direction of the arrows) into two categories (indicated by the color of the arrows) defined by a learned category boundary (black dashed line). In the motion direction discrimination task (MDD), two motion directions (the two category center directions from the MDC) were shown at three coherence levels. (C) Three spatial stimulus configurations tested LIP’s role in sensory evaluation (S_IN), saccade planning (T_IN), and a control condition assessing non-spatial aspects of the tasks (B_OUT). The dark shaded area and the dashed circle represent the inactivated visual field and the position of motion stimulus, respectively. The red and green spots indicate the positions of the saccade targets, and the yellow spot indicates the position of fixation.

We examined the impact of LIP inactivation on evaluating visual motion in the S_IN condition (Fig. 2). LIP inactivation produced a significant behavioral impairment in both the MDC and MDD tasks, shown by decrease in overall accuracy (Fig. 2, A,E,G and K; MDC: p_(accuracy) =2.2e-08, t(37) = -7.1; MDD: p_(accuracy) = 5.8e-11, t(37) = -9.1, inactivation vs control, unpaired t-test) and increase in reaction time (RT) (Fig. 2, B,F,H and L) (MDC: p_(RT) = 0.0062, t(37) = 2.9; MDD: p_(RT) = 0.0011, t(37) = 3.5, unpaired t-test). In both tasks, a greater behavioral impairment was observed for one of the motion categories/directions (fig. S1, MDC: p = 6.8e-04, t(16) = 4.2; MDD: p = 0.0048, t(16) = 3.3, paired t-test). Psychometric curve fitting showed a significant change in bias and an increase in psychophysical threshold during both tasks (Fig. 2, M to P and fig. S2, MDC: R²_{(inactivation)} = 0.90±0.034, R²_(control) = 0.94±0.044, p_(bias) = 2.7e-04, t(37) = 4.0, p_(threshold) = 1.9e-08, t(37) = 7.1; MDD: R²_{(inactivation)} = 0.91± 0.069, R²_(control) = 0.98± 0.025, p_(bias) = 9.5e-4, t(37) = 3.6, p_(threshold) = 1.2e-7, t(37) = 6.5, unpaired t-test). Behavioral impairments in the S_IN condition of both tasks were significantly greater than in B_OUT (Fig. 2, E,F,K and L, MDC: p_(accuracy) =7.5e-06, t(16) = -6.5, p_(RT) = 0.0011, t(16) = 4.0; MDD: p_(accuracy) = 1.1e-09, t(16) = -12.6, p_(RT) = 1.9e-04, t(16) = 4.8, S_IN vs. B_OUT, paired t-test), in which we did not observe a global impact on accuracy (Fig. 2, C,E,I and K; MDC: p_(accuracy) = 0.17, t(37) = -1.4; MDD: p_(accuracy) = 0.12, t(37) = -1.6, inactivation vs. control, unpaired t-test) or RT (Fig. 2, D,F,I and L; MDC: p_(RT) = 0.60, t(37) = 0.53; MDD: p_(RT) = 0.71, t(37) = 0.37, unpaired t-test).

Fig. 2.

Causal evidence for decision-related sensory evaluation in LIP. (A) Psychometric curves for the S_IN condition of the MDC task. Task accuracy pooled across both monkeys is plotted as the proportion of choosing the primary category, defined as the category for which each monkey showed a greater decrease in accuracy (on average across all sessions) following LIP inactivation (see Fig. S6 and S7 for data shown for each monkey separately). (B) Chronometric curves are shown for the S_IN condition of the MDC task. (C-D) The psychometric and chronometric curves in the B_OUT condition of the MDC task, (same format as A and B). (E-F) Comparisons of the averaged behavioral deficits following LIP inactivation in S_IN and B_OUT conditions of the MDC task. (G-L) Monkeys’ behavioral performance in MDD task is plotted for inactivation and control sessions. Psychometric and chronometric curves for S_IN (G-H) and B_OUT (I-J) conditions are shown in the same format as the MDC task (A-D). Monkeys showed a significantly greater deficit in S_IN than B_OUT conditions in the MDD task (k-L). (M-P) Paired comparisons between inactivation and control sessions for choice bias and threshold in the S_IN conditions of the MDC (M-N) and MDD tasks (O-P). The open and filled symbols denote the inactivation sessions in which the majority of the recorded neurons at the targeted cortical locations preferred the primary (open) and non-primary (filled) category/direction (see Supplementary Materials). The black stars indicate the statistical significance (*: p<0.05, **: P<0.01, ***: p<0.001/, unpaired t-test, multiple tests in A-D and G-J are Bonferroni corrected). The error bars denote ±SEM. P: primary, NP: non-primary, cat: category, dir: direction, H: high, M: middle, L: low, S_IN: stimulus-in, B_OUT: both-out.

LIP’s role in motor aspects of decisions was assessed in the T_IN condition (Fig. 3), which is the spatial configuration nearly always tested in past studies of LIP with the MDD task(7, 8, 12, 14, 15, 17) (but see (18)). Although the motion stimulus was outside the IVF, note that the T_IN condition did require evaluating the color of the in-RF saccade target. After LIP inactivation, the monkeys’ saccadic choices were significantly biased away from the IVF in both tasks (Fig. 3, A,C,G and I; MDC: p = 1.3e-07, t(16) =8.9, MDD: p = 3.0e-04, t(16) = -5.4, paired t-test), and a greater RT increase was observed on trials when the saccade was directed toward vs away from the IVF (Fig. 3, B,D,H and J; MDC: p =7.0e-06, t(16) = -6.5, MDD: p = 6.0e-05, t(16) = 4.6, paired t-test). Although LIP inactivation produced an ipsilateral saccade bias, there was no significant difference in mean accuracy between inactivation and control sessions for either task (Fig. 3, A,E,G and K, MDC: p = 0.13, t(37) = -1.5; MDD: p = 0.56, t(37) = -0.59, unpaired t-test), and accuracy on both tasks was less influenced by inactivation in T_IN than S_IN (Fig. 3, E and K; MDC: p = 1.6e-08, t(16) = -10.4; MDD: p =1.2e-06, t(16) = -7.5, paired t-test). The magnitudes of RT effects between T_IN and S_IN were statistically indistinguishable (Fig. 3, F and L; MDC: p = 0.40, t(16) = 0.86; MDD: p = 0.093, t(16) = 1.8, paired t-test). Psychometric curve fitting showed a significant bias in saccades away from the IVF, but no increase in threshold in T_IN in both tasks (Fig. 3, M to P and fig. S3, MDC: R²_{(inactivation)} = 0.92±0.070, R²_(control) = 0.94±0.047, p_(bias) =4.0e-07, t(37) = 6.1, p_(threshold) = 0.15, t(37) = 1.5; MDD: R²_{(inactivation)} = 0.98±0.010, R²_(control) = 0.98±0.010, p_(bias) =0.0013, t(37) = 3.5, p_(threshold) = 0.33, t(37) =0.99, unpaired t-test). These effects were consistent across monkeys (figs. S4 and S5).

Fig. 3.

Causal evidence for decision-related saccade planning in LIP. (A) Psychometric curves for T_IN condition of MDC task. The choice accuracy is plotted as the proportion of contralateral saccades relative to the inactivated hemisphere. Data from both monkeys were pooled based on target location. (B) Chronometric curves for T_IN conditions of MDC task. (C-D) Comparisons of behavioral impairments following LIP inactivation between ipsilateral and contralateral target trials in T_IN condition. Monkeys’ saccade choices were biased toward the ipsilateral target following LIP inactivation, shown by both accuracy(C) and RT(D). (E-F) Comparisons of overall behavior deficits following LIP inactivation between S_IN and T_IN conditions of MDC task. Monkeys showed significantly greater behavioral impairment in the S_IN than T_IN condition, in their accuracy(E), but not RT(F). (G-L) Monkeys’ behavioral performance in T_IN condition of MDD task. Psychometric and chronometric curves (G-H) are in the same format as the MDC task (A-B). Monkeys showed consistent saccadic choice biases following LIP inactivation in both MDD and MDC tasks (I-J). In the MDD task, a greater deficit was observed following LIP inactivation in the S_IN than T_IN condition in accuracy but not RT (K-L). (M-P) Paired comparisons between inactivation and control sessions for choice bias and threshold in the T_IN conditions of the MDC (M-N) and MDD tasks (O-P).

We examined how distinct components of the decision process were affected by inactivation. We fitted our results with fixed-boundary drift diffusion models, which captured monkeys’ decision behavior across different task conditions in both MDD and MDC tasks, including both accuracy and the RT distributions (figs. S6 and S7). This revealed that LIP inactivation significantly slowed the evidence accumulation process (drift rate) only in S_IN, while significantly changing the saccade choice bias (starting point of the diffusion process) and decision boundary only in T_IN, across monkeys and tasks (tables S1 to S4). Inactivation also did not produce substantial impairments of either gaze position, microsaccades, or peak saccade velocity (figs. S8, S9 and S10).

We hypothesized that the inactivation-related behavioral deficits in the S_IN condition of the MDD task are due to disrupting neuronal activity related to evaluating stimulus motion within the IVF. Thus, we recorded from 194 LIP neurons (Monkey M: N=78; Monkey B: N=116) during the S_IN condition, targeting the same cortical locations within the same hemispheres as in the inactivation experiments. More than half of LIP neurons (monkey M: 50/78, monkey B: 54/116) showed significant motion direction selectivity (DS) in the MDD task (one-way ANOVA, P < 0.01). DS emerged prior to the monkeys’ mean RTs (fig. S11A), and the magnitude of DS was positively correlated with motion coherence at the single neuron and population levels, but did not converge to a fixed threshold near the decision time (Fig. 4, A to D and fig. S11B). Neuronal activity on zero-coherence motion trials (in which monkeys could only guess about motion direction) reflected the monkeys’ trial-by-trial decisions, consistent with a role in the decision process. Decision-correlated activity across all motion-coherence levels was evident in single-neuron activity (Fig. 4, A and B and fig. S12), the LIP population (Fig. 4C; fig. S13 for results reported separately for each monkey), and was confirmed using a receiver operating characteristic (ROC) analysis (Fig. 4D).

Fig. 4.

LIP activity reflects decision-related sensory evaluation. (A-B) Activity is shown for each motion-coherence level for two example direction-selective LIP neurons in the MDD task. The motion stimulus but not the targets appeared in neurons’ RF. The zero coherence trials were grouped based on the monkeys’ choices. The two vertical dashed lines mark the time of target and motion stimulus onset, respectively. (C) Average normalized population activity across all direction-selective neurons is shown for each motion coherence level, aligned to stimulus onset (left panel) or saccade onset (right panel). Activity shown in the left panel is truncated at the monkeys’ mean RT for each coherence level. (D) The motion direction selectivity (DS) (determined by ROC) for different coherence levels is shown as in (C). (E) Average activity on low-coherence trials is shown for neurons’ preferred and non-preferred directions, separately for correct and error trials. (F) Neuronal selectivity on low-coherence trials is compared for the monkeys’ decisions about motion direction compared to the physical direction of stimulus motion. The black stars indicate time periods in which there was a significant difference (P < 0.01, paired t-test). (G-H) DS on low coherence trials (G) and choice selectivity on zero coherence trials (H) is compared between faster RT and slower RT trials. Shaded areas denote ±SEM. (I-J) Partial correlation analysis. (I) The value of r-stimulus (partial correlation between neuronal activity and stimulus direction, given the monkeys’ choices) and r-choice (the partial correlation between neuronal activity and monkeys’ choice, given the stimulus direction) are plotted across time. (J) Correlation between r-stimulus and r-choice. Note that most LIP neurons showed a congruent sign between r-stimulus and r-choice values.

To further test whether DS in LIP was decision related, we compared activity on correct vs. error low-coherence trials (see Methods). LIP population activity was more similar on trials in which monkeys decided that different motion directions were the same vs. different (Fig. 4E, p = 1.9e-04, t(100) = 3.9, paired t-test). Accordingly, LIP activity co-varied more closely with the monkeys’ trial-by-trial decisions about motion direction than it did with the physical motion direction (Fig. 4F; p = 6.8e-04, t(103) = -3.4, paired t-test, comparing ROC values on chosen vs physical direction of motion). Furthermore, LIP activity co-varied with the monkeys’ RTs, with most neurons showing a greater response to their preferred motion direction on shorter vs longer RT trials for all motion coherence levels (and weaker response to their non-preferred direction; fig. S14, S15). Even on low-coherence (Fig. 4G) and zero-coherence (Fig. 4H) trials, DS evolved more rapidly on shorter vs longer RT trials (comparing slope: p(low) < 0.01, p(zero) < 0.01, bootstrap). At the population level, the slope of DS negatively correlated with monkeys’ RT (fig. S15, r = -0.9455, p = 0.0044). Elevated DS was even observed before stimulus onset on low-coherence trials (Fig. 4H), suggesting that the state of LIP activity prior to stimulus onset was predictive of the monkeys’ upcoming decision on that trial. Furthermore, a partial correlation analysis examined choice- and stimulus-related components of DS (i.e., r-choice and r-stimulus; Fig. 4I and fig. S16)(19). This revealed that the choice and stimulus components of LIP DS were positively correlated (Fig. 4J, r = 0.6276, p < 0.0001).

Taken together, our results suggest that LIP plays a significant role in visually-based perceptual and categorical decisions, with preferential involvement in stimulus evaluation compared to motor planning aspects of such decisions. This is supported by behavioral impairments on discrimination and categorization tasks caused by reversibly inactivating LIP. This is also supported by neurophysiological recordings revealing decision-correlated LIP activity during the MDD task, as well as work from multiple groups showing sensory-decision-related LIP activity in several tasks(11, 13, 20), including categorization(6, 9). Both tasks tested in this study had a flexible mapping between the decisions about motion stimuli and the direction of the saccade used to report those decisions. Thus, it will be interesting to examine whether LIP plays a similar role during tasks with fixed sensory-motor mappings.

It is well established that LIP helps mediate covert spatial attention(21–23) and saccades(1, 24, 25). However, the current study highlights LIP’s role in evaluating the abstract behavioral significance of visual stimuli during decision making. Although attention contributes toward the selection and representation of task-relevant stimuli, and disruption of attention by LIP inactivation may contribute to the current results, our inactivation results in the S_IN condition are unlikely to arise primarily from disrupting attention or saccade related functions for several reasons. First, the LIP neurons targeted for inactivation showed activity which correlated with MDD task decisions, indicating that LIP’s role extends beyond the comparatively fixed representation of motion direction in the middle temporal area(4, 6, 11, 14, 16). The partial correlation analysis suggests LIP DS may be more closely choice correlated compared to other parietal areas (19, 26), although a direct comparison across multiple areas during the same tasks will be needed. Second, inactivation produced greater impairments for specific directions and categories, rather than uniformly impacting performance, and we found similar results when we analyzed inactivation behavioral data with respect to neurons’ stimulus preferences at the targeted LIP sites (figs. S17 and S18). This may relate to biased representations of categories and directions observed in LIP during MDD and MDC tasks(27), or could reflect anatomical clustering of such representations in LIP. Third, LIP inactivation did not produce an obvious impairment of eye movements or profound spatial neglect, although we did observe a mild ipsilateral bias during the free choice saccade task (fig. S19), consistent with previous studies(16, 23, 28). Nevertheless, future work using more precise causal approaches—such as stimulating or silencing pools of neurons with specific direction/category preferences—can more precisely dissect LIP’s contributions toward decision making tasks.

A recent study came to a different conclusion regarding LIP’s role in perceptual decisions—finding no significant impact of inactivation on MDD task performance(16). However, that study only considered LIP’s contributions to motor planning (T_IN) aspects of the MDD task, but not sensory stimulus evaluation (S_IN). Another study found that LIP activity was decision-correlated in both S_IN and T_IN conditions of a MDD task, but concluded that LIP activity correlated more closely with motor aspects (but did not directly test LIP’s causal contributions)(18). Consistent with the earlier inactivation study, we found that LIP inactivation did not impair average MDD-task accuracy in the T_IN condition. However, we did observe a saccade bias away from the IVF in the T_IN condition, suggesting a deficit in mapping decisions to actions. Several factors may account for this discrepancy. First, we used an RT rather than fixed-interval version of the MDD task, perhaps prompting different strategies or speed-accuracy tradeoffs. Second, our tasks employed a flexible mapping (based on saccade target color) between motion stimuli and saccades. Thus, LIP may be more engaged by decisions requiring greater cognitive flexibility or abstraction. Furthermore, the T_IN condition in our tasks required discriminating the color of the in-RF saccade target to plan the correct saccade, so that condition had greater sensory evaluation demands than the traditional MDD task(13). Compared to the previous study, our animals learned both the MDD and MDC tasks, and the strategy used to solve the MDD task could have been affected by DMC training. The greater behavioral deficit observed in the MDD task (which employed low-coherence motion stimuli) than MDC task (which used only 100% coherent motion) is consistent with LIP integrating noisy sensory evidence(12, 14, 17), and the saccadic choice bias in the T_IN condition supports LIP’s involvement in mediating the saccades used for reporting decisions(14, 29). Thus, decision-related motion encoding might be transformed into saccadic signals via coordination between sensory and motor pools of LIP neurons, in contrast to LIP reading out sensory inputs exclusively from upstream visual areas such as MT(30).

Perceptual and categorical decisions rely on contributions from and coordination among a network of cortical and subcortical areas spanning the sensory, cognitive, and motor hierarchy(13–15, 31, 32). Although the current study establishes primate PPC as an important node in that network, along with recent work in rodents(33–35), a more complete understanding necessitates the characterization of local and long-range circuits, within different animal models and humans, which flexibly transform sensory encoding into decisions and actions, and the process by which task-related neuronal encoding emerges during learning.