Covert spatial selection in primate basal ganglia

Abstract

The basal ganglia are important for action selection. They are also implicated in perceptual and cognitive functions that seem far removed from motor control. Here, we tested whether the role of the basal ganglia in selection extends to nonmotor aspects of behavior by recording neuronal activity in the caudate nucleus while animals performed a covert spatial attention task. We found that caudate neurons strongly select the spatial location of the relevant stimulus throughout the task even in the absence of any overt action. This spatially selective activity was dependent on task and visual conditions and could be dissociated from goal-directed actions. Caudate activity was also sufficient to correctly identify every epoch in the covert attention task. These results provide a novel perspective on mechanisms of attention by demonstrating that the basal ganglia are involved in spatial selection and tracking of behavioral states even in the absence of overt orienting movements.

Fig 1. Change-detection task, behavioral performance, and recording sites.

(A) Task sequence in the covert attention task. While fixating a central spot and pressing down a joystick, a peripheral spatial cue (ring) was flashed for 0.2 s to indicate which part of the visual field the monkey should monitor. After a blank of 0.5 s, two motion patches were presented: one at the location previously cued (cued location) and the other one diametrically opposed (foil location). The monkey should detect when the motion direction changed at the cued location by releasing the joystick and otherwise keep holding the joystick down if the motion direction changed at the foil location or if no change occurred. Inserts at the bottom show the joystick voltage traces for one experimental session aligned on cue onset, motion stimulus onset, and cue or foil change (from left to right, respectively; vertical lines show the zero). Black traces indicate the individual trial joystick voltages. (B) Behavioral performance in the task for both monkeys. Each dot represents the percentage of correct change trials (Hits) and FAs for a single behavioral session (n = 105 for Monkey R, n = 60 for Monkey P), separately for trials with the cue contralateral (“contra”) or ipsilateral (“ipsi”) to the site of recording. Error bars indicate 95% CIs of the mean. (C) Location of neuronal recording sites in the left caudate nucleus for both monkeys. Each dot represents one recording site. Positive values on the y-axis indicate locations anterior to the AC. MRI images represented the coronal section at the AC (AC+0). Dots represented the locations of the recorded neurons within the caudate nucleus (gray surfaces) for this coronal section. Scale bar indicate 10 mm. Underlying data available in S1 Data. AC, anterior commissure; FA, false alarm.

Fig 2. Caudate neuronal activity and cue-related modulation during the CD task.

(A) Example caudate neurons. Activity of four different caudate neurons (a, b, c, and d) aligned on the onset of the motion patches (solid vertical line) when the cue was contralateral (“contra,” orange) or ipsilateral (“ipsi,” blue) with respect to the recording site. The dashed vertical lines indicate when the spatial cue was presented. (B) Normalized activity (“Norm. activity”) for our complete population of caudate neurons (n = 227). Each row represents the normalized activity of a single neuron for contralateral presentation of the spatial cue (left) or ipsilateral presentation (right) aligned on motion stimuli onset. We computed normalized PSTHs by dividing the raw values from each time bin by the maximum firing rate (peak of each neuron’s PSTH) across all conditions (i.e., either contralateral or ipsilateral conditions, whichever was higher). Neurons were ranked according to the time of their peak activity across both contralateral and ipsilateral cue conditions. The colored sidebar on the right indicates whether each neuron had maximal activity for contralateral (orange) or for ipsilateral (blue). Solid white lines indicate onset of motion stimuli; dashed lines indicate spatial cue onset. Neurons were grouped according to the timing of the peak activity (labels in pink): before the spatial cue onset (Pre-cue), between the spatial cue and the motion stimuli onset (Post-cue), after the motion stimuli onset (Visual), and during the delay period prior to motion-direction change (Delay). (C) Spatial cue modulation in caudate nucleus. The histograms display the distribution of attention cue modulation index values for the periods for each group of neurons. The p-values were corrected with the Holm’s variant of the Bonferroni method. White bars indicate the cells within the appropriate category (pre-cue, post-cue, visual, or delay), whereas gray bars illustrate the cells out of the category. Colors indicate significant side preference (orange for contralateral, blue for ipsilateral, white for neither; Wilcoxon rank sum test, p < 0.05). Underlying data available in S1 Data. CD, change-detection; PSTH, peristimulus time histogram; sp/s, spikes per second.

Fig 3. Dependence of caudate visual activity on the presence of distracters.

(A) Firing rates for two sample caudate neurons aligned on presentation of visual stimuli for the two-patch condition (top row) and single-patch condition (bottom row). The orange/blue code indicates respectively the spatial cue location or the single patch location for contralateral (“contra”)/ipsilateral (“ipsi”) side. Gray boxes demarcate the 0.4-s time period used for computing the AMI for the two-patch condition and SMI for the single-patch condition. Asterisks indicate significant values for AMI/SMI (Wilcoxon rank, p < 0.05). (B) Scatter plot of AMI and SMI values computed for each caudate neuron with visual activity (n = 70). AMI values on the x-axis greater (less) than 0 indicate preference for contralateral (ipsilateral) hits in the two-patch condition. SMI values on the y-axis greater (less) than 0 indicate preference for contralateral (ipsilateral) for single-patch condition. Each dot represents one caudate neuron from the "visual" subpopulation (Fig 2B, n = 70). Color indicates the neuron’s group assignment based on the AMI/SMI values: AMI for two patches different for one side (green), both different from chance (purple), SMI for single patch different for one side (blue), and neither different from chance (gray). (C) Side preference for the motion CD task, MGS task, and joystick task in the population of 160 caudate neurons tested with the three tasks. Each row represents a single neuron across different visual epochs: post-cue and visual for CD task and visual periods for MGS and joystick mapping task. Neurons were sorted according their side preference for post-cue period. Colors indicate side preference (orange for contra, blue for ipsi, gray for neither) when the activity was significantly greater than the baseline (Wilcoxon rank sum test, p < 0.05), when no significant activity was reported (Wilcoxon rank sum test, p ≥ 0.05; white), or when activity was significantly lower than the baseline (Wilcoxon rank sum test, p < 0.05; black). We used all trials (hits, misses, false alarms, and correct rejects) for the CD trials and all correctly performed trials for the MGS and joystick tasks. Underlying data available in S1 Data. AMI, Attention Modulation Index; CD, change-detection; MGS, memory-guided saccade; SMI, Side Modulation Index; sp/s, spikes per second.

Fig 4. Response-choice activity in caudate nucleus.

(A) Firing rates for two sample caudate neurons (#1 and #2) aligned on the joystick release. The upper row shows the response-choice activity for contralateral hits (“contra,” orange) and ipsilateral hits (“ipsi,” blue); the lower row shows activity for hits (red) and misses (gray) pooled across stimulus locations. Gray boxes demarcate the 0.3 s time period used for the ROC analysis. A top asterisk indicates that the AROC value was significantly different from chance level (0.5). For miss trials, which did not contain joystick releases, data for each neuron were aligned on the median reaction time during the recording session. The median reaction times were 0.56 s (IQR = 0.18, contralateral trials) and 0.53 s (IQR = 0.14, ipsilateral trials) for neuron #1 and 0.55 s (IQR = 0.13) and 0.52 s (IQR = 0.13) for neuron #2. (B) Scatter plot of detect probabilities and sensory ROC values computed for each caudate neuron that showed response-choice activity. AROC values on the x-axis greater (less) than 0.5 indicate preference for contralateral (ipsilateral) hits. Each dot represents one caudate neuron (n = 80). Color indicates the neuron’s group assignment based on the AROC values: hits versus misses AROC different from chance (“motor choice,” red), contralateral hits versus ipsilateral hits different from chance (“sensory,” blue), both different from chance (“sensorimotor,” green), and neither different from chance (black). (C) Firing rate for one caudate example aligned on the joystick release. The upper row shows the response-choice activity in the presence (pink) or absence (blue) of the MC; the lower row shows activity for hits (red) and misses (black). For the MC-present trials (pink), the change event could happen at either location (cued or foil). The left column shows responses for the contralateral trials; the right, for the ipsilateral trials. Gray boxes demarcate the 0.3-s time period used for the ROC analysis. The top asterisk indicates that the AROC value is significantly different from chance level (0.5). For miss trials and some MC-absent trials, which did not contain joystick releases, data for each neuron were aligned on the median reaction time during the recording session. The median reaction times were 0.48 s (IQR = 0.14, contralateral trials) and 0.50 s (IQR = 0.12, ipsilateral trials). (D) Scatter plot of the detect probabilities as a function of the neural sensitivity (AROC "sensitive") for the caudate neurons with response-choice activity. Only neurons with at least five occurrences for each type of conditions were used for analysis (n = 75/80). The color code indicates the location of the spatial cue; contralateral (orange) and ipsilateral (blue). AROC values on the x-axis greater (less) than 0.5 indicate a preference for presence (absence) of the MC; AROC values on the y-axis greater (less) than 0.5 indicate a preference for hits (misses). Underlying data available in S1 Data. AROC, area under the receiver operating characteristic curve; MC, motion change; ROC, receiver operating characteristic; sp/s, spikes per second.

Fig 5. Error trials and hand influence.

(A, B, C) Scatter plots of mean caudate activity during FAs (A), joystick breaks (“J. breaks”) (B), and joystick trial end (“J. trial end”) (C) as a function of mean activity to hits for contralateral trials (“Contra”, orange) and ipsilateral trials (“Ipsi”, blue). Filled dots (orange or blue) indicate when mean activities were significantly different (Wilcoxon rank sum test, p < 0.05, corrected with the Holm’s variant of the Bonferroni method). Dashed lines represent identity lines. Joystick breaks are defined as trials when the animals released the joystick but there was no MC event at either stimulus location and joystick trial end as joystick releases at the end of correct reject trials, when the animal was obliged to release the joystick in order to end the trial. We performed these analyses on response-choice neurons for which we recorded at least five trials for each condition (FA, joystick breaks, or joystick trial end). (D) Scatter plot of mean responses for contralateral and ipsilateral hits (orange and blue) computed during the 0.3-s time period when animals used either their ipsilateral hand (y-axis) or contralateral hand (x-axis) with the joystick. Same convention as A, B, and C. Underlying data available in S1 Data. FA, false alarm; MC, motion change; sp/s, spikes per second.

Fig 6. Influence of task context on response-choice activity.

(A) Scatter plot for each caudate neuron (n = 70) with response-choice activity comparing AROC values computed for single-patch and two-patch conditions. AROC values on the x-axis greater (less) than 0.5 indicate preference for contralateral (“contra”) (ipsilateral [“ipsi”]) hits for two-patch condition; AROC values on the y-axis greater (less) than 0.5 indicate preference for contralateral (ipsilateral) hits for single-patch condition. Color indicates the neuron’s group assignment based on the AROC values: contralateral hits versus ipsilateral hits different from chance for two patches only (Two*; green), contralateral hits versus ipsilateral hits different from chance for single patch only (Single*; blue), contralateral hits versus ipsilateral hits different from chance for both conditions (Both*; purple), and neither different from chance (Neither; gray). (B) Scatter plot of AROC values computed for each caudate neuron tested with the dimming joystick task (n = 70) with response-choice activity for two-patch and dimming joystick conditions. AROC values on the y-axis greater (less) than 0.5 indicate preference for contralateral (ipsilateral) hits for dimming joystick condition. Same conventions as A except that blue color indicates when responses to contralateral hits versus ipsilateral hits are different from chance for joystick task only (Joystick*; blue). (C) Side preference for the motion CD task for CD with two patches (CD), single patch, and joystick task in the population of 70 caudate neurons tested with these three conditions. Each row represents a single neuron across different task conditions during the response choice period (0.3 s): two patches (CD), single patch, and joystick mapping task. Neurons were sorted according their side preference for CD task. Colors indicate side preference (orange for contralateral, blue for ipsilateral, gray for neither [“No pref.”]). Underlying data available in S1 Data. AROC, area under the receiver operating characteristic curve; CD, change detection.

Fig 7. Linear classifier performance for the covert attention task.

(A) Cartoon illustrating how the linear binary classifiers were trained and tested. Each single-trial data set was generated by random draws of spike counts from each neuron for each of the 14 epochs. For training of each classifier, trials 1–120 of these data sets were used to construct a classifier that could distinguish between data from its own epoch and data pooled across the other 13 epochs. For testing of each classifier, trials 121–150 were used to test and cross-validate the classifier with data from its own epoch individually and also to test how frequently it incorrectly recognized each of the other 13 epochs individually. This procedure was repeated 1,000 times to obtain the median values shown graphically in the other plots. (B) Linear classifiers applied to caudate activity from different epochs of the attention task. The matrix illustrates the percentage of trials positively identified, as indicated by the color scale, by each of the 14 classifiers (rows) for each of the 14 trial epochs (columns). The diagonal of the matrix corresponds to the percentage of trial epochs correctly identified by each classifier; values outside the diagonal indicate the percentages of trial epochs that were incorrectly identified by classifiers trained on other epochs. Numbers overlaid on the matrix indicate percentage scores for every case that was significantly greater than chance performance, for correct identifications (black numbers along diagonal) and for incorrect identifications (white numbers). (C) Linear classifier performance for trial outcomes during the covert attention task. The matrix illustrates the percentage of trials positively identified, as indicated by the color scale, by each of the four classifiers trained on change-epoch data from correct trials (rows) for each of the 10 possible trial outcomes (columns). Numbers overlaid on the matrix indicate percentage scores for every case that was nonzero, for values greater than chance performance (black numbers), and for values less than chance performance (white numbers). FA, false alarm; J. break, joystick break.