Abstract
Neural activity in the lateral intraparietal cortex (LIP) correlates with both sensory evaluation and motor planning underlying visuomotor decisions. We previously showed that LIP plays a causal role in visually-based perceptual and categorical decisions, and preferentially contributes to evaluating sensory stimuli over motor planning. In that study, however, monkeys reported their decisions with a saccade to a colored target associated with the correct motion category or direction. Since LIP is known to play a role in saccade planning, it remains unclear whether LIP's causal role in such decisions extend to decision-making tasks which do not involve saccades. Here, we employed reversible pharmacological inactivation of LIP neural activity while two male monkeys performed delayed match to category (DMC) and delayed match to sample (DMS) tasks. In both tasks, monkeys needed to maintain gaze fixation throughout the trial and report whether a test stimulus was a categorical match or nonmatch to the previous sample stimulus by releasing a touch bar. LIP inactivation impaired monkeys' behavioral performance in both tasks, with deficits in both accuracy and reaction time (RT). Furthermore, we recorded LIP neural activity in the DMC task targeting the same cortical locations as in the inactivation experiments. We found significant neural encoding of the sample category, which was correlated with monkeys' categorical decisions in the DMC task. Taken together, our results demonstrate that LIP plays a generalized role in visual categorical decisions independent of the task-structure and motor response modality.
SIGNIFICANCE STATEMENT Neural activity in the lateral intraparietal cortex (LIP) correlates with perceptual and categorical decisions, in addition to its role in mediating saccadic eye movements. Past work found that LIP is causally involved in visual decisions that are rapidly reported by saccades in a reaction time based decision making task. Here we use reversible inactivation of LIP to test whether LIP is also causally involved in visual decisions when reported by hand movements during delayed matching tasks. Here we show that LIP inactivation impaired monkeys' task performance during both memory-based discrimination and categorization tasks. These results demonstrate that LIP plays a generalized role in visual categorical decisions independent of the task-structure and motor response modality.
Introduction
Decision-making involves evaluating task-relevant sensory stimuli and selecting appropriate motor responses. The posterior parietal cortex (PPC), particularly the lateral intraparietal area (LIP), has been identified as a hub for mediating visuomotor decisions (Shadlen and Newsome, 1996; Platt and Glimcher, 1999; Gold and Shadlen, 2007; Shadlen and Kiani, 2013; Zhou and Freedman, 2019). Previous studies have shown that neural activity in LIP closely correlates with the motor planning aspects of saccade-based decisions (Shadlen and Newsome, 2001; Roitman and Shadlen, 2002; Hanks et al., 2006; Yang and Shadlen, 2007; Kiani and Shadlen, 2009), and it has been proposed that perceptual decisions evolve as a competition between representations of potential motor plans (Gold and Shadlen, 2000, 2007). However, this model has been challenged by recent LIP inactivation studies, which have found that inactivating neural activity in LIP produced no impairment in monkeys' performance during perceptual decision tasks (Katz et al., 2016).
On the other hand, a series of studies have shown that neural activity in PPC, including LIP, correlates with the abstract (i.e., independent of motor response) behavioral relevance of visual stimuli when the stimuli are shown in neurons' receptive fields (RFs; Freedman and Assad, 2011, 2016). This encoding persists across tasks that do not require an oculomotor response (Freedman and Assad, 2006; Fitzgerald et al., 2011; Swaminathan and Freedman, 2012; Rishel et al., 2013; Swaminathan et al., 2013; Zhou et al., 2021) and in oculomotor tasks which dissociate saccade direction from the categorical decision (Mohan et al., 2021). In addition, we previously showed that LIP is causally engaged in visual motion categorization and discrimination tasks in which monkeys immediately report their decisions with a saccadic eye movement (Zhou and Freedman, 2019). In that study, we found that LIP inactivation caused a greater deficit when the visual stimulus, rather than saccade target, was placed in the inactivated visual field, indicating that LIP preferentially contributes to sensory evaluation compared with motor planning aspects of visually-based decisions. Although the monkeys reported their decisions with saccades, the directions of the saccades were not rigidly linked to the two categories or directions since each were associated with a specific target color, and the colors of each target was randomly alternated across trials. While that study suggests that the behavioral deficits from inactivation were because of disrupting decisions about the stimulus, it remains unclear whether LIP's causal role in visual decisions identified in that study generalizes to tasks in which decisions are reported without saccadic eye movements.
Here, we examined the impact of LIP inactivation on monkeys' performance during delayed match to category (DMC) and delayed match to sample (DMS) tasks. In both tasks, monkeys maintained fixation across the entire trial and indicated whether sequentially presented sample and test motion stimuli matched or did not match in category (DMC) or direction (DMS) by releasing or holding a manual touch bar. In the DMC task, monkeys indicated whether the test stimulus was a categorical match to a previously viewed sample stimulus. In the DMS task, monkeys indicated whether the test stimulus was an exact match to the sample direction, where nonmatch stimuli consist of opposite motion directions. Delayed matching paradigms have been widely used for studying sensory and memory related functions (Bauer and Fuster, 1976; Fuster and Jervey, 1982; Miller et al., 1991; Miller and Desimone, 1994; Rainer et al., 1998; Romo et al., 1999; Freedman and Assad, 2006; Sarma et al., 2016). Delay-based matching tasks dissociate the subjects' decisions about sample stimuli from the motor responses used to report those decisions. This is because the subjects cannot plan their motor response (hold vs release) until the first test stimulus appears after the delay. Thus, any difference in neural activity for different stimuli during the sample and delay periods of delayed-matching tasks can be attributed to the neural encoding of the direction and/or category of sample stimuli.
Using the DMC task, we previously showed that neural activity in the prefrontal cortex (PFC), lateral intraparietal cortex (LIP), medial superior temporal cortex (MST) and medial intraparietal cortex (MIP) all reflect learned motion categories during the stimulus presentation and memory delay epochs of the task (Freedman and Assad, 2006; Swaminathan and Freedman, 2012; Swaminathan et al., 2013; Zhou et al., 2021, 2022). Furthermore, previous work examining the strength, timing, and choice probability of category encoding across these areas suggests that LIP is the most closely involved in mediating categorical decisions about visual motion compared with MST and MIP (Swaminathan and Freedman, 2012; Swaminathan et al., 2013; Zhou et al., 2021, 2022). In this study, we show that reversibly inactivating LIP significantly impairs the monkeys' ability to perform both the DMC and DMS tasks, producing deficits in behavioral accuracy and an increase in reaction times (RTs). Taken together, these results demonstrate that LIP plays a general and causal role in perceptual and categorical decisions, even in tasks that do not involve saccadic eye movements.
Materials and Methods
Animal preparation, stimuli display, and behavioral testing
Two male monkeys (Macaca mulatta, 8–12 kg) were implanted with a titanium head post, trained on four different behavioral tasks (detailed below), and then implanted with an magnetic resonance imaging (MRI)-guided recording chamber targeting LIP. One monkey (Monkey M) was involved in the previous LIP inactivation study from our group (Zhou and Freedman, 2019). Our surgical, behavioral, and neurophysiological methods were described in detail in previous studies (Zhou and Freedman, 2019; Zhou et al., 2022). In brief, stereotaxic coordinates for chamber placement were determined from magnetic resonance imaging (MRI) scans obtained before chamber implantation. LIP chambers were centered on the intraparietal sulcus, 4.0 mm posterior to the intra-aural line and 10.0 mm lateral from the midline for Monkey M, and 2.0 mm posterior to the intra-aural line and 10.0 mm lateral from the midline for Monkey Q. Both chambers were oriented perpendicular to the horizontal plane. Monkeys were housed in individual cages under a 12/12 h light/dark cycle. Behavioral training and experimental recordings were conducted during the light portion of the cycle. Monkeys sat comfortably while head-fixed in a custom primate chair inside a dark experiment rig. Task stimuli were displayed on a 21-inch color CRT monitor (1280 × 1024 resolution, 75-Hz refresh rate, 57-cm viewing distance). A computer-controlled solenoid-operated reward system was used to deliver juice rewards to the monkeys on correct trials. Monkeys' eye positions were monitored by an optical eye tracker (EyeLink; SR Research) at a sampling rate of 1000 Hz and stored for offline analysis. Stimulus presentation, task events, rewards, and behavioral data acquisition were conducted using an Intel-based PC equipped with PCI-based National Instruments DAQ interfaces, and used the MonkeyLogic toolbox running in MATLAB 2013a (http://www.monkeylogic.net; Asaad et al., 2013). All experimental and surgical procedures were in accordance with the University of Chicago Animal Care and Use Committee and National Institutes of Health guidelines.
Behavioral tasks
Delayed match to category task (DMC)
The DMC task (Fig. 1A) used in this study has been used and described in a previous study from our group (Zhou et al., 2022). Briefly, monkeys were trained to release a manual touch-bar when the categories of sequentially presented sample and test stimuli matched, or to hold the touch-bar when the sample and test categories did not match. Stimuli consisted of 10 random-dot motion directions (15°, 35°, 55°, 75°, 135°, 195°, 215°, 235°, 255°, and 315°) grouped into two categories separated by a learned category boundary oriented at 45° (Fig. 1B). In this manuscript, we refer to the directions closest to the category boundary (35°, 55°, 215°, and 235°) as “near-boundary” directions, and other directions (15°, 135°, 75°, 195°, 255°, and 315°) as “center” directions. Trials were initiated by the monkey grasping the touch-bar and maintaining central gaze fixation. Throughout the trial, monkeys were required to maintain fixation within a 2.5° radius of a fixation point. After 500 ms of fixation, a sample stimulus was presented for 650 ms, followed by a 1000-ms delay and a 650-ms test stimulus. If the categories of the sample and test stimuli matched, the monkeys needed to release a touch-bar during the test period to receive a juice reward. Otherwise, the monkeys needed to continue holding the touch-bar throughout a second delay period (150 ms). Following the second delay, a second test stimulus was presented, which was always a categorical match to the sample, requiring a release of the touch-bar. Therefore, monkeys concluded all trials with the same motor response (touch-bar release). At the start of each trial, and up to the presentation of the first test stimulus, the monkeys could not predict whether that trial would be a match or nonmatch trial, so they could not plan differential motor responses until the first test stimulus is shown. The motion stimuli were full contrast, 9° diameter, random-dot movies composed of 190 dots per frame that moved at 12°/s with 100% coherence. Identical task stimuli were used for both inactivation and electrophysiology experiments.
Behavioral tasks and stimulus configurations. A, Sequence of events in the delayed match to category (DMC) and delayed match to sample (DMS) tasks. Monkeys needed to either release a touch-bar when the category or direction of sample and test stimuli matched, or hold the bar if they did not match. On nonmatch trials, the first test stimulus was followed by a second delay and test stimulus, which always matched the sample (and required a bar release). Throughout each trial, monkeys had to maintain gaze fixation on a central fixation point. The orange dashed circle indicates an example LIP RF location during the recording sessions. B, Motion directions in both DMC (top) and DMS (bottom) tasks. In the DMC task, monkeys needed to group 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 DMS task, two motion directions (the center directions of the two categories from the DMC task) were shown at three coherence levels. In some sessions, the two boundary directions in the DMC task were tested in the DMS task (see Materials and Methods). C, Two spatial stimulus configurations used in the inactivation sessions. The dark shaded area and the dashed circle represent the inactivated visual field and the position of motion stimulus, respectively.
Delayed match to sample (DMS) task
The timings and spatial positions of the motion stimuli in the DMS task were identical to the DMC task (Fig. 1A). To receive a juice reward in this task, monkeys were required to release a touch-bar when the directions of sequentially presented sample and test stimuli matched, or hold the touch-bar when the sample and test directions did not match. Two opposite motion directions (135° and 315°, the center directions of each category in the DMC task) with three different coherence levels (9%, 18%, 36% for Monkey M; and 13%, 25%, 50% for Monkey Q) were tested during all of the control and inactivation sessions of Monkey Q and half of the control and inactivation sessions of Monkey M. In the remaining control and inactivation sessions for Monkey M, two different directions (the directions of the category boundary in the DMC task: 45° and 225°) were tested at 9%, 18%, and 36% coherence levels. Task difficulty was determined by the coherence level (i.e., signal vs noise) of the motion stimuli, defined as the percentage of dots moving coherently in one direction. The direction of motion for each noncoherent dot was chosen randomly, and each moved in a consistent direction during the entire viewing of the stimulus. As in the DMC task, the motion stimuli in the DMS task were full contrast, 9° diameter, random-dot movies composed of 190 dots per frame that moved at 12°/s.
Free-choice saccade task (Fig. 2A)
A free-choice saccade task was used to assess the efficacy of LIP inactivation as described in previous studies (Zhou and Freedman, 2019). Both monkeys performed the free-choice task at the beginning of each control and inactivation session. In this task, monkeys freely chose to saccade to either of two visually identical saccade targets. Trials were initiated when the monkeys grasped the lever and maintained central fixation (within 2.5° radius). After a 500-ms fixation period, the fixation point disappeared and two full-contrast saccade targets appeared either simultaneously or asynchronously for 60 ms, at opposite directions relative to the fixation point and with equal eccentricities (8°). On each trial, the target onset asynchrony time was randomly chosen from seven possible values with equal probability (left target relative to right target: −120, −80, −40, 0, 40, 80, and 120 ms). The monkeys were free to saccade to either the left or right targets immediately. On each trial, the reward probability for each target was determined independently and at random. In order to motivate the monkeys to saccade to the earlier target, a higher reward probability was used for the earlier than later appearing target (0.85 vs 0.6). On 20% of trials, only a single target appeared and the monkeys were required to saccade to it. The reward probability on single-target trials was equal to that of the earlier target in the two target trials (0.85).
Memory guided saccade task (MGS)
We used an MGS task to identify LIP within the intraparietal sulcus and map neurons' receptive fields (RFs). Trials were initiated when monkeys grasped the touch-bar and maintained central gaze fixation. After a 500-ms fixation period, a visual target (0.2°) flashed for 200 ms at one of eight possible locations, placed at evenly spaced 45° angular positions with eccentricity ranging from 6° to 12°. Following stimulus offset, monkeys were required to maintain central fixation for 1 s during the delay period. Then, the fixation point disappeared and the monkeys were required to make a single saccade toward the remembered target location within 500 ms, and fixate on that location for 200 ms to receive a juice reward.
Both the inactivation experiment and electrophysiological experiments were performed after monkeys fully learned the tasks. The inactivation sessions were conducted after finishing the electrophysiological recording experiments for both monkeys. There were temporal gaps between the recording and inactivation sessions so that monkeys could practice performing with the blocked design with the different spatial configurations in the inactivation experiment, which were approximately three weeks and three months for Monkeys M and Q, respectively.
Stimulus configurations in the DMC and DMS tasks
For each monkey, the location of the sample motion stimuli on the display (for each of the tested spatial conditions) was the same for all inactivation and control sessions. The sample stimulus locations were chosen with respect to the inactivated visual field location, which was determined from RF mapping during prior recording sessions. In the inactivation experiment, we arranged the motion stimulus in two different spatial configurations in different blocks of trials (Fig. 1C). Similar to our previous study (Zhou and Freedman, 2019), we recorded LIP neurons' activity during the MGS task and mapped their RF positions before the inactivation experiments. Our LIP recordings targeted different hemispheres in the two monkeys (Monkey M: left hemisphere; Monkey Q: right hemisphere). For each inactivation session, we targeted the muscimol injections to LIP locations where most recorded neurons exhibited either visually-driven or persistent activity in response to the visual target in the middle-contralateral visual field within a 6°−12° eccentricity range in the MGS task. Because our RF mapping showed that most of the LIP neurons recorded from nearby positions showed overlapping receptive fields, it is very likely that we primarily inactivated neurons with receptive fields in the middle-contralateral visual fields during inactivation sessions. The middle-contralateral visual field was therefore defined as the inactivated visual field. To verify the position of the inactivated visual field, we used the free-choice saccade task to assess inactivation effects in the middle-contralateral visual field before testing with the categorization task. The inactivated visual field was located in the right and left visual fields for Monkeys M and Q, respectively. The two spatial configurations were defined according to the locations of motion stimuli relative to the inactivated visual field (Fig. 1C). In the Stimulus-IN (SIN) condition, the motion stimuli (sample and test) were within the inactivated visual field location; in the Stimulus-OUT (SOUT) condition, the motion stimuli were presented in the ipsilateral visual field (opposite to the inactivated visual field) at the same eccentricity as the SIN condition. SOUT was used as a within-session control to monitor monkeys' general behavioral state after LIP inactivation, accounting for factors such as the animals' motivation and arousal. The stimulus configuration was fixed within each block, and the same stimulus configurations were used in the inactivation and control sessions. The order and duration of different blocks are described below.
Task sequence within each session
For all sessions, monkeys were first tested with the free-choice saccade task for ∼250 trials (15–20 min), and then tested on the DMC and DMS tasks. For both of the delayed matching tasks, there were two different blocks corresponding to the two spatial configurations of stimuli (SIN and SOUT). The four experiment blocks (two configurations × two tasks) were randomly interleaved without replacement within each session. In each block, monkeys performed 60 or 100 (60 for DMS and 100 for DMC) correct trials before advancing to the next block. In inactivation sessions, monkeys were tested with the free-choice saccade task beginning ∼30 min after completing muscimol infusion, while during the control sessions, monkeys idly waited for an equivalent duration as the inactivation session before performing the free-choice saccade, DMC, and DMS tasks. For each trial within either DMC or DMS task blocks, the motion stimuli were randomly chosen from the pool of possible stimuli for that task. The DMC stimulus pool included 30 full coherence motion stimuli, with three unique stimuli (of the same direction) for each of the 10 motion directions. The DMS stimulus pool consisted of 18 motion stimuli, including three coherence levels for each of the two motion directions. There were three unique stimuli for each motion direction at each coherence level. This ensured that the task conditions were randomly ordered and the trial counts for each stimulus condition were balanced
We performed analysis to ensure that the monkeys attempted sufficient numbers of both easier trials (i.e., stimuli far from category boundary or high coherence) and more difficult trials (i.e., stimuli close to category boundary or low coherence). For example, on difficult trials the monkeys could have adopted a strategy of breaking fixation immediately following presentation of the difficult sample stimulus, thus aborting those difficult trials. For every control or inactivation session, both monkeys completed at least two blocks of trials for each task condition. For both monkeys during the DMC task, we found a nonsignificant trend for more completed trials for the easier (center directions) than the more-difficult (near-boundary directions) sample directions (normalized ratio: easier vs more-difficult, Monkey M = 1.04:1, p = 0.163, Monkey Q = 1.05:1, p = 0.27, paired t test). During the DMS task, both monkeys completed more trials for the easier (high coherence) than the more-difficult (low coherence) conditions (normalized ratio: easier vs more-difficult, Monkey M = 1.32:1, p < 0.0001, Monkey Q = 1.19:1, p < 0.001, paired t test). However, this modest bias still yielded a sufficient number of completed trials in each of the conditions to support our analyses. Furthermore, the difference in the numbers of attempted trials between the easier and the more-difficult conditions did not significantly vary between control and inactivation sessions (DMC: Monkey M: p = 0.359, F = 0.86, Monkey Q: p = 0.932, F = 0.01; DMS: Monkey M: p = 0.938, F = 0.01, Monkey Q: p = 0.844, F = 0.04; interaction term, two-way ANOVA). Averaged across all sessions, Monkey Q completed 57 ± 8 (mean ± standard deviation) and 55 ± 12 trials in the most difficult conditions of DMC and DMS task respectively. Monkey M completed 42 ± 5 and 40 ± 4 trials in the most difficult condition of DMC and DMS task respectively.
Muscimol infusion
Our protocol for muscimol infusion into LIP was described in detail in the previous inactivation study from our group (Zhou and Freedman, 2019). Briefly, the GABAA agonist muscimol (Sigma) was dissolved in phosphate-buffered saline to concentrations of 8 μg/μl. In each session, two infusion cannulae (32 gauge) were lowered into grid locations and depths in which we had previously recorded LIP neurons with RFs located in the middle-contralateral visual field. The two cannulae were lowered by motorized microdrives (NAN Instruments), similar to lowering recording electrodes, separated by 1–3 mm on the recording grid (NAN Instruments). Infusions were performed using a syringe pump (Harvard Apparatus) using two 10-μl microsyringes (Hamilton) and polyethylene tubing (PE20, inner diameter = 0.38 mm, outer diameter = 1.09 mm) directly connected to the cannulae. Across both cannulae, we delivered a total of 6–9 μl (3–4.5 μl for each cannula) of muscimol solution in each inactivation session, corresponding to a total drug delivery of ∼55 μg (mean = 57.3 μg) of muscimol. Similar to our previous study, the muscimol solution was infused at five different depths along a single cannula track to minimize pressure damage and maximize the inactivated cortical area. For each depth, we infused 0.5–0.9 μl muscimol solution (constant rate of 0.2 μl/min, 2.5–4.5 min), and the injection cannulae remained at that depth for at least 2 min after completing infusion. We visually monitored and tracked the movement of the solution in the polyethylene tubing to ensure that the drug was successfully delivered without leaks. Each session typically required 40–50 min to complete the infusion process. Cannulae were left at the last infusion depth in the cortex until the end of the session. Behavioral testing began 30–35 min after completing the muscimol infusion, and was completed 2.5–4 h following infusion. Control data were obtained on the day prior and 2 d following an inactivation session. The same tasks, stimuli, and parameters of the inactivation session were used in the control sessions.
Usually, a control session between two inactivation sessions was used as both the “after session” for the earlier inactivation session and the “prior session” for the later inactivation session in cases where there was not a long interval (e.g., <6 d) between the two inactivation sessions. In several cases (two and three for Monkeys M and Q, respectively), the two consecutive inactivation sessions were separated by longer (e.g., one week or more) time intervals (such as holidays). We then ran an additional control session before the next inactivation session to ensure that the animal's behavioral performance recovered.
Electrophysiological recordings
The electrophysiological recording approaches were described in detail in a previous study from our group (Zhou et al., 2022). While neural data from the same sessions was presented in that study, the current analysis focuses on different blocks of trials in which the monkeys performed the DMS and DMC tasks. Neuronal activity was recorded using 75-μm diameter tungsten microelectrodes (FHC; ∼1 MΩ) or 16-channel linear arrays (Plexon V-Probes). All recordings from Monkey M were conducted using single 75-μm tungsten microelectrodes (FHC). While 24/38 recording sessions on Monkey Q were conducted using 16-channel (Plexon) linear V-Probes after we identified the locations of target brain regions using single electrodes. Neurophysiological signals from single or multichannel recordings were amplified, digitized, and stored for offline spike sorting (Plexon Inc.) to verify the quality and stability of neuronal isolations. Before inactivation sessions, we recorded neuronal activity in the MGS task to map LIP RFs in separate recording sessions for both monkeys. Neuronal recordings during the DMC task were recorded from both monkeys in previous experimental sessions.
We identified locations within the intraparietal sulcus which we classified as LIP based on the evoked neuronal activity during the MGS task (e.g., spatially selective activity during stimulus presentation and/or the delay) observed at those recording locations. All neurons included in this study were recorded from the same grid holes and similar depths (5–10 mm from the cortical surface), where we encountered spatial selectivity in the MGS task. LIP neurons were also identified based on anatomic criteria, such as the location of each electrode track relative to that expected from the MRI scans, the pattern of gray–white matter transitions encountered on each electrode penetration, and the depths of each neuron relative to the dura mater. The placement of motion stimuli differed slightly between single-channel recording and multichannel recording sessions. For single-channel recording, motion stimuli for the DMC task were always placed in LIP neurons' receptive fields (RFs). The typical eccentricity of stimulus placement was ∼6.0−10.0°. For each multichannel recording session, we first identified one task responsive neuron according to the above criteria, and then placed the motion stimuli according to the RF of the identified neuron. If we were able to isolate and test RFs of multiple neurons, we placed the stimulus at a location which offered the best fit for those neurons taken together.
Data analysis
Behavioral task performance
Only sessions in which monkeys performed enough trials in both the DMC and DMS tasks (at least two blocks for each stimulus configuration) were used for further analysis. For Monkey M, we collected data from nine inactivation sessions in total. However, one inactivation session from Monkey M was excluded from further analysis as the infusion line was clogged for one injection cannula, and the animal did not perform enough trials. In total, we collected and analyzed 12 control sessions and 8 inactivation sessions for Monkey M. Within these sessions, 5 control sessions and four inactivation sessions were tested with 45° and 225° (the boundary directions in the DMC task) motion stimuli in the DMS task, while the remaining control and inactivation sessions were tested with 135° and 315° motion stimuli in the DMS task. Since Monkey M's behavioral results were qualitatively similar between the two versions of the DMS task, we pooled the data from two different task versions to quantify the effect of LIP inactivation for all analyses except for Figure 6F. In Figure 6F, we measured whether LIP inactivation preferentially affected Monkey M's ability to categorize directions from different motion categories using the DMS task. Thus, we only included the data sessions in which the center direction of each category (135°, 315°) were used in the DMS task, as the directions tested in the other data sessions (45° and 225°) did not belong to either of the two motion categories. For Monkey Q, we collected data from eight inactivation sessions and 11 control sessions in total. As the data in one free-choice saccade inactivation session for Monkey Q was lost because of an operator error while saving the file, we also did not include for analysis the two free-choice saccade control sessions corresponding to that inactivation session. For the DMC and DMS tasks, only the correct and error trials which concluded with a lever release (i.e., trials which the monkey attempted) were included for analysis; break-fixation and aborted trials were excluded. Because the second test stimulus was always a match to the sample, monkeys did not need to categorize the second test stimulus or decide whether it was a match to sample to generate the correct hand movement. We therefore only include match trials for the RT analysis. We included both correct and error trials for calculating monkeys' mean RTs for both inactivation and control sessions.
Microsaccade analysis
To examine whether LIP inactivation produced significant attentional deficits when monkeys performed both the DMC and DMS tasks, we analyzed the monkeys' microsaccadic eye movements across the trial. Microsaccades were detected based on velocity criteria similar to previous studies (Herrington et al., 2009; Martinez-Conde et al., 2000). We used both horizontal and vertical eye positions (sampled at 1000 Hz) to calculate an instantaneous eye movement velocity. Velocity vectors were smoothed (10-ms boxcar) to reduce noise. Subsequently, the following criteria were used to detect microsaccades: (1) eye movement velocity >10°/s; and (2) 10 ms < eye movement duration <100 ms; and (3) eye movement amplitude ≥0.05°; and (4) a new microsaccade could not be initiated within 20 ms of a previous microsaccade. Furthermore, a “rate-of-turn” criterion, the saccade direction could change no more than 30° every 5 ms during the microsaccade, was also used to determine the end of microsaccades. Furthermore, we visually inspected raw eye movement traces to confirm the accuracy of this protocol. We focused our microsaccade analysis within a time window from 400 ms before sample stimulus onset to 250 ms after test stimulus onset (which is before than the monkeys' mean response time on match trials). We also performed the analysis in specific task periods (fixation, sample, delay), and observed similar results.
Identifying category-selective neurons
Similar to previous studies (Zhou et al., 2022), we first identified well-isolated single units in LIP that exhibited task-modulated activity in the DMC task using the following criteria: (1), the maximum averaged firing rate during at least one of the four different task periods (sample period, early delay period, late delay period, and test period) should be at least one spike per second; and (2), the activity should exhibit at least one form of task-related modulation (such as sample category selectivity, test category selectivity, sample direction selectivity, test direction selectivity, two-way nested ANOVA test, p < 0.01) during one of the four task periods, or the mean activity during at least one of the four task periods should be significantly different from the baseline activity during the fixation period.
For each neuron, we computed a category tuning index (CTI; described below) using its activity in the DMC task. We then performed a shuffle analysis to determine whether a task-related neuron exhibited significant category selectivity by determining whether the CTI value of this neuron was significantly above chance level. To obtain a null distribution of CTIs, we shuffled the direction labels of trials within each session before calculating the CTI for 1000 iterations. We applied this method to the mean activity within each time bin during the 2000-ms task period ranging from 50 ms after sample onset to 400 ms after test onset (400-ms bin size, five time bins in total). The CTI value in each time bin was determined as statistically significant if it was >99% of values from the null distribution [equivalent to p < 0.05 with Bonferroni correction for multiple (N = 5) comparisons]. Neurons were identified as category-selective if their CTI value was greater than the significance threshold in at least one of the five time bins.
CTI
A related CTI metric was previously used to quantify the category selectivity of neuronal activity (Freedman and Assad, 2006). It was determined for each neuron by computing two values: the WCD (within category difference), i.e., the difference in firing rate between pairs of directions in the same category, and the BCD (between category difference), i.e., the difference in firing rate between pairs of directions in different categories. To prevent direction tuning from unequally affecting the WCD and BCD metrics, the WCD and BCD metrics contained an equal weighting of 20°, 60°, 120°, and 160° firing rate comparisons. Specifically, the CTI is defined as follows:
WCD = (2* | DF(75,195) | + | DF(135,195) |+ | DF(75,135) | + 2* | DF(255,15) |+ | DF(315,15) | + | DF(255,315) | + 2* | DF(55,215) | + | DF(55,75) | + | DF(195,215) | + 2*| DF(35,235) | + | DF(35,15) | + | DF(255,235) |)/16;
BCD = (2* | DF(75,15) | + | DF(75,315)|+ | DF(135,255) |+ | DF(135,15) | + 2* | DF(195,255) | + | DF(195,315) | + 2* | DF(55,35) | + | DF(55,255)| + |DF(75,235) – 0.5| + 2* | DF(215,235)| + | DF(195,35) |+ | DF(215,15) |)/16;
and DF denotes differential firing rate between two directions. Positive CTI values indicate category selectivity (smaller differences in firing within category, and greater differences between categories) with greater CTI values indicating stronger category selectivity (i.e., more binary-like category tuning), while negative CTI values indicate greater selectivity among directions in the same, as opposed to different, categories. By computing the CTI in this way, the units of the CTI is spikes/sec.
Quantifying bias in neuronal category preferences
To test whether the population preferentially encoded one motion category over the other (as reported previously (Fitzgerald et al., 2013), we calculated a CTI bias index by quantifying whether more neurons than expected had significantly elevated CTI values for one of the two categories. For each neuron, we calculated the CTI value using the mean activity within a sliding window across all the task periods (200-ms bin size stepped by 10 ms). We then set any negative CTI values to zero to exclude the influence of pure direction selectivity on the CTI values. In order to define the preference of category encoding for each neuron, we set the CTI value to be positive if the neural activity was greater for category 1 than category 2; otherwise we set the CTI value to be negative if the neural activity showed the opposite trend. We finally averaged the signed CTI values across all neurons that exhibited significant category selectivity. These steps allow us to quantify the preference of neuronal category encoding at the population level, considering both the number of neurons that exhibited different preferences of category encoding and the magnitude of selectivity for each neuron. A CTI bias index significantly greater than zero indicates a bias toward preferring category 1 at the population level, whereas a significantly negative CTI bias index indicates that LIP neurons preferentially encoded category 2.
Partial correlation analysis
In order to quantify the correlation between neural activity and monkeys' categorical decisions during the DMC task, we performed a partial correlation analysis similar to previous studies (Zhou and Freedman, 2019). For each trial, we quantified three parameters: the neuronal activity, the category identity of the sample stimulus, and the monkeys' categorical choice. We assigned the stimulus category as +1 and −1 for category 1 and category 2, respectively. Their categorical choices were assigned as +1 for choosing category 1 and −1 for choosing category 2. We then calculated the partial correlation between neuronal activity and the monkeys' categorical choices, given the stimulus category:
The partial correlation analysis was calculated using each neuron's average firing rate in a 200-ms window, advanced in 10-ms steps. Because some individual neurons exhibited a different category preference in different task epochs, we used the absolute value of r-choice to track categorical encoding regardless of each neuron's categorical preference across the trial. For this analysis, we only included trials in which the sample directions were near the boundary but the test directions were far from the boundary (≥30°), as there were sufficient numbers of errors on these trials and these errors were most likely because of mis-categorizing the difficult (near boundary) sample stimulus.
Results
We inactivated LIP in one hemisphere when monkeys performed both DMC and DMS tasks in alternating blocks of trials (see Materials and Methods). In both tasks, monkeys needed to: (1) categorize (DMC) or discriminate (DMS) the sample motion stimulus during the 650 ms sample period; (2) remember the category/direction during the one second delay period; (3) categorize or discriminate the test motion stimulus; (4) decide whether the category or direction of the test stimulus matched or did not match that of the sample; and (5) report the match/nonmatch decision by releasing (match) or holding (nonmatch) a touch-bar. If the test stimulus did not match the sample, a second test stimulus was displayed after a 150-ms delay period that always matched the sample and required a touch-bar release. Thus, all correct trials concluded with the same hand movement response, and monkeys' decisions about the sample stimuli (categorization/discrimination) were not directly coupled with their manual responses. Throughout each trial, monkeys were required to maintain gaze fixation on a central fixation point. Muscimol infusions were targeted to LIP coordinates that, during previous neuronal recording sessions, contained neurons which exhibited significant category selectivity that correlated with the monkeys' decisions (see Materials and Methods).
Behavioral deficits in free-choice saccade task
Both monkeys performed the free-choice saccade task to assess the efficacy of LIP inactivation at the beginning of each control and inactivation session. Consistent with previous studies (Balan and Gottlieb, 2009; Wardak et al., 2002; Zhou and Freedman, 2019), both monkeys preferentially chose the ipsilateral saccade target in the inactivation sessions compared with control sessions (Monkey M: p = 0.0039, t(18) = 3.3; Monkey Q: p = 5.03e-05, t(14) = −5.75; unpaired t test; Fig. 2B,C, upper panels). Moreover, monkeys' RTs after LIP inactivation were significantly greater for contralateral saccades compared with control sessions, but were not significantly different for ipsilateral saccades (Monkey M: ipsilateral: inact. vs control = 119.2 vs 117.9 ms, p = 0.57, t(18) = −0.57, contralateral: inact. vs control = 142.5 vs 116.8 ms, p = 1.69e-04, t(18) = −4.66; Monkey Q: ipsilateral: inact. vs control = 112.4 vs 113.8 ms, p = 0.58, t(14) = 0.57,contralateral: inact. vs control = 150.7 vs 127.9 ms, p = 6.39e-06, t(14) = −6.99, unpaired t test; Fig. 2B,C, lower panels). These results are consistent with successful inactivation of LIP because of muscimol injection.
Free-choice saccade task and behavioral performance in the free-choice saccade task. A, Monkeys freely chose between two visually identical saccade targets which appeared either simultaneously or sequentially at opposite positions relative to the fixation point. B, Monkey M's accuracy and RTs in both control and inactivation sessions are shown. In the upper panel, the proportion of rightward saccades is plotted as a function of each target onset asynchrony (TOA) condition (left target onset time relative to right target onset time). In the lower panel, RTs for contralateral and ipsilateral saccades are shown. C, Monkey Q's accuracy and RTs in both control and inactivation sessions are shown in the same format as B. Note that the left hemisphere was inactivated for Monkey M and the right hemisphere was inactivated for Monkey Q. Therefore, the inactivated visual fields (contralateral) are the right and left visual fields for Monkeys M and Q, respectively. The black stars indicate the statistical significance for comparisons between control and inactivation sessions (*p < 0.05, **p < 0.01, ***p < 0.001, unpaired t test). The error bars denote ±SEM.
Behavioral deficits in the DMC task
We quantified the impact of LIP inactivation on monkeys' performance of the DMC task (Fig. 3) by quantifying and comparing their accuracy and RTs during inactivation and control sessions. Their overall DMC accuracy was significantly lower during inactivation sessions when the motion stimuli were positioned in the inactivated hemifield [SIN: Monkey M, inact. vs control = 0.799 vs 0.849, p = 0.012, t(18) = 2.80 (Fig. 3A,B); Monkey Q, inact. vs control = 0.797 vs 0.871, p = 1.94e-05, t(17) = 5.85; unpaired t test (Fig. 3G,H)], but not when the stimuli were positioned in the ipsilateral hemifield [SOUT: Monkey M, inact. vs control = 0.816 vs 0.801, p = 0.229, t(18) = −1.25 (Fig. 3A,C); Monkey Q, inact. vs control = 0.696 vs 0.717, p = 0.186, t(17) = 1.38; unpaired t test (Fig. 3G,I)]. The decrease in accuracy in the SIN condition was significantly greater than SOUT for both monkeys [Monkey M, p = 0.0055, t(7) = −3.96 (Fig. 3A); Monkey Q, p = 0.0015, t(7) = −5.03; paired t test (Fig. 3G)]. The spatially specific behavioral deficits suggest that LIP inactivation did not primarily affect global behavioral factors such as hand movements, motivation, and arousal. To examine whether inactivation effects differed according to task difficulty, we separated DMC trials into four subgroups based on the angular distance of sample and test directions from the category boundary (see Materials and Methods): (1) both sample and test stimuli were center directions (Seasy,Teasy), (2) sample stimuli were center directions and test stimuli were near-boundary directions (Seasy,Tdiff), (3) sample stimuli were near-boundary directions and test stimuli were center directions (Sdiff,Teasy), (4) both sample and test stimuli were near-boundary directions (Sdiff,Tdiff). We found that the decrease in accuracy after LIP inactivation was more prominent in the Sdiff,Teasy subgroup compared with the other subgroups for Monkey M [accuracy decrease: Seasy,Teasy = 0.030, Seasy,Tdiff = 0.041, Sdiff,Teasy = 0.111, Sdiff,Tdiff = 0.033, p = 0.030, F(7) = 3.46, one-way ANOVA, post hoc test (Sdiff,Teasy vs (Seasy,Teasy, Seasy,Tdiff and Sdiff,Tdiff)), maximum p = 0.032, paired t test Fig. 3B]. A nonsignificant trend in the same direction was also observed for Monkey Q (accuracy decrease: Seasy,Teasy = 0.043, Seasy,Tdiff = 0.059, Sdiff,Teasy = 0.106, Sdiff,Tdiff = 0.090, p = 0.1267, F(7) = 2.07, one-way ANOVA; Fig. 3H).
LIP inactivation produced significant behavioral deficits in the DMC task. A, Comparison of Monkey M's average accuracy following LIP inactivation versus control sessions in SIN (red) and SOUT (black) conditions of the MDC task. Overall accuracy in SIN and SOUT conditions of the DMC task is shown separately for the inactivation (solid line) and control sessions (dashed line). B, C, Monkey M's accuracy in four conditions based on sample and test motion direction (center vs near-boundary directions) is shown separately for both SIN (B) and SOUT (C). D, Comparison of Monkey M's averaged RT following LIP inactivation in SIN and SOUT (black) conditions. The averaged RTs are shown for the inactivation session and control sessions. E, F, Monkey M's RTs in four conditions based on sample and test motion direction (center vs near-boundary directions) are shown for both SIN (E) and SOUT (F). G–L, Monkey Q's accuracy and RTs in the DMC task in inactivation and control sessions are shown in the same format as A-F. The red and black stars indicate the statistical significance in the SIN and SOUT conditions (*p < 0.05, **p < 0.01, ***p < 0.001, unpaired t test), respectively. The error bars denote ±SEM
LIP inactivation also impacted monkeys' RTs in the DMC task, although the two subjects exhibited different patterns of results. Compared with control sessions, Monkey M's RTs in the SIN condition but not SOUT condition were significantly slower following LIP inactivation (SIN: inact. vs control = 353.0 vs 330.3 ms, p = 0.031, t(18) = −2.33; SOUT: inact. vs control = 321.5 vs 318.0 ms, p = 0.517, t(18) = −0.66; unpaired t test; Fig. 3D–F). In Monkey Q, inactivation did not affect RTs in the SIN condition but instead produced faster RTs in the SOUT condition (SIN: inact. vs control = 271.8 vs 277.0 ms, p = 0.274, t(17) = 1.13; SOUT: inact. vs control = 271.4 vs 294.4 ms, p = 4.1e-05, t(17) = 5.47; unpaired t test; Fig. 3J–L).
Behavioral deficits in the DMS task
Similar to deficits produced by LIP activation in the DMC task, overall accuracy in the DMS task was also impaired in the SIN condition after inactivation compared with control sessions for both monkeys [Monkey M, inact. vs control = 0.677 vs 0.833, p = 1.87e-05, t(18) = 5.76 (Fig. 4A,B); Monkey Q, inact. vs control = 0.752 vs 0.824, p = 0.0017, t(17) = 3.73; unpaired t test (Fig. 4G,H)] and did not significantly change in SOUT condition [Monkey M, inact. vs control = 0.802 vs 0.806, p = 0.906, t(18) = 0.12 (Fig. 4A,C); Monkey Q, inact. vs control = 0.768 vs 0.784, p = 0.512, t(17) = 0.67; unpaired t test (Fig. 4G,I)]. The decrease in accuracy was significantly greater in the SIN than SOUT condition for both monkeys [Monkey M, p = 0.0098, t(7) = −3.51 (Fig. 4A); Monkey Q, p = 0.035, t(7) = −2.61; paired t test (Fig. 4G)]. Unlike the DMC task, sample stimuli in the DMS task were shown at different levels of motion coherence (high, medium, and low; see Materials and Methods; Fig. 1B). We analyzed the impact of inactivation on behavior as a function of stimulus coherence and found that LIP inactivation produced significant behavioral deficits for all task difficulty levels in the DMS task (maximum p = 0.0385, Monkey Q, t(17) = 2.24, unpaired t test; Fig. 4B,H), and the behavioral deficits were not significantly different among different coherence levels (Monkey M: p = 0.727, F = 0.32; Monkey Q: p = 0.639, F = 0.46; one-way ANOVA).
LIP inactivation produced significant behavioral deficits in the DMS task. A, Comparison of Monkey M's average accuracy following LIP inactivation versus control sessions in SIN (red) and SOUT (black) conditions of the DMS task. The averaged accuracy in both SIN and SOUT conditions of the DMS task is shown separately for inactivation and control sessions. B, C, Monkey M's accuracy for high, medium, and low motion coherence conditions is shown for both SIN (B) and SOUT (C). D, Comparison of Monkey M's average RT following LIP inactivation in SIN and SOUT in the DMS task. E, F, Monkey M' RTs in both SIN and SOUT conditions are shown for three different levels of motion coherence. G–L, Monkey Q's behavioral performance in the DMS task in both inactivation and control sessions are shown in the same format as A–F. The statistics are shown in the same format as in Figure 3.
LIP inactivation also affected RTs in the DMS task. RTs in the SIN condition were significantly slower following LIP inactivation for both monkeys (Monkey M: inact. vs control = 371.8 vs 341.7 ms, p = 0.0017, t(18) = −3.68; Monkey Q: inact. vs control = 319.8 vs 303.6 ms, p = 0.0013, t(17) = −3.84; unpaired t test; Fig. 4D,E,J,K) compared with control sessions. In the SOUT condition, monkeys' RTs either did not significantly change (Monkey M, inact. vs control = 328.6 vs 337.1 ms, p = 0.378, t(18) = 0.90; Fig. 4F) or were significantly faster (Monkey Q, inact. vs control = 294.9 vs 312.3 ms, p = 0.0106, t(17) = 2.87; Fig. 4L) after LIP inactivation.
Behavioral deficits after LIP inactivation were consistent with neuronal encoding in the inactivated cortical area
We hypothesized that the observed behavioral deficits in the SIN condition after LIP inactivation were because of disrupting neuronal activity related to evaluating motion stimuli in the inactivated visual field. To test this, we analyzed LIP neural activity during DMC task performance recorded in prior experimental sessions in these animals, recorded from the same cortical locations in the same hemispheres as the inactivation experiments conducted in later sessions. In total, we identified 361 well-isolated LIP neurons (Monkey M: 89, Monkey Q: 272). Consistent with our previous studies, the majority of the LIP neurons in both monkeys exhibited significant category encoding in at least one epoch of the DMC task (Monkey M: 58/89, Monkey Q: 164/272; see Materials and Methods). Figure 5A,B shows an example category-selective LIP neuron from each of the two monkeys. The neuron in Figure 5A is category selective throughout most time periods of the DMC task, whereas the neuron in Figure 5B is category selective mainly during the delay period. The time course of category selectivity across the LIP population for both monkeys is shown by the category tuning index (CTI; Fig. 5C,D; see Materials and Methods). CTI values greater than zero indicate category-selectivity, i.e., greater differences in activity between pairs of directions in different categories compared with those from the same category (p < 0.01 throughout the sample, delay and test periods, one sample t test; Fig. 5C,D).
LIP neural activity correlated with monkeys' categorical decisions in the DMC task. A, An example neuron from Monkey Q exhibited sample category selectivity in the DMC task. The average activity for each sample direction is plotted as a function of time. Different colors represent different sample categories, and different shades indicate the angular distance from the boundary with darker colors indicating directions that were closer to the boundary. The first and second vertical dashed lines denote the timing of sample stimulus onset and offset, respectively, while the third vertical dashed line indicates the time of test stimulus onset. B, An example neuron from Monkey M exhibited sample category selectivity mostly during the delay period of the DMC task. C, Population-level sample category selectivity of neurons from Monkey Q. The category tuning index (CTI) plot shows the magnitude and time course of sample category selectivity. The shaded area denotes ±SEM. The black stars denote the time point for which the CTI values were significantly greater than zero (p < 0.01, one sample t test). D, Population level sample category selectivity of neurons from Monkey M is shown in the same format as C. E, F, The correlation between LIP neural activity and monkeys' categorical decisions were quantified by a partial correlation analysis. The values of r-choice (the partial correlation between neuronal activity and monkeys' categorical choice, given the stimulus category) of LIP neurons are shown separately for Monkey Q (E) and Monkey M (F). The correlation between LIP neural activity and the category of motion stimuli independent of the monkeys' choices (r-stimulus) is shown for each monkey. The red stars indicate the time points for which the r-choice value was significantly greater than baseline (greater than the mean r-choice value during the fixation period, paired t test, p < 0.01). The black stars indicate the time points for which the r-choice value was significantly greater than the r-stimulus value (paired t test, p < 0.01). The shaded area denotes ±SEM.
Furthermore, we examined whether the LIP sample category selectivity correlated with monkeys' trial-by-trial categorical decisions in the DMC task using a partial correlation analysis. We calculated the r-choice (the partial correlation between neuronal activity and monkeys' categorical choice, given the stimulus category) and r-stimulus (the partial correlation between neuronal activity and the stimulus category, given monkeys' categorical choice) for each category-selective neuron using both correct and error trials (see Materials and Methods). The r-choice and r-stimulus represent the contribution of the monkeys' trial-by-trial category judgment and the physical stimulus to the neuronal response respectively, with greater values indicating greater effects on the neural activity. We found that the r-choice values were significantly elevated during the sample, delay and test periods in the DMC task for both monkeys (through sample to test periods, p < 0.01, paired t test; Fig. 5E,F), compared with the baseline calculated during fixation period. Importantly, the r-choice was significantly greater than the r-stimulus in several DMC task epochs for both monkeys (Monkey Q: sample, delay and test periods; Monkey M: middle delay and test periods; r-choice versus r-stimulus, p < 0.01, paired t test; Fig. 5E,F). This is consistent with our previous findings that LIP neuronal activity correlated significantly with the monkeys' categorical decisions in the DMC task (Zhou et al., 2022).
As in previous studies that used the DMC task (Fitzgerald et al., 2013), we quantified the bias in neuronal category preferences in each animal, since it is common to find that neurons across the population tend to prefer one of the two categories. To do so, we first counted the number of neurons that significantly encoded each category during different epochs of the DMC task by comparing the CTI values to a null distribution (see Materials and Methods). In Monkey Q, a significantly greater number of neurons preferred category 2 during the test period (0–400 ms after test onset, category 1 vs category 2 = 34:65, p = 5.71e-4; χ2 test). By contrast, more neurons preferred category 1 during the test period in Monkey M (category 1 vs category 2 = 19:9, p = 0.0396; χ2 test). We were interested in whether this bias of category preferences was related to the pattern of behavioral effects observed when inactivating LIP (e.g., greater deficits for directions in the overrepresented category). In order to take both the number of neurons preferring each category and the magnitude of the category selectivity of each neuron into account, we quantified whether LIP neurons recorded from each monkey showed a biased category preference at the population level using a CTI bias index (see Materials and Methods). Positive CTI bias index values indicate that LIP neurons showed greater encoding of sample category 1, whereas negative CTI bias index values indicate a preference for category 2. Only neurons that exhibited significant category selectivity during the DMC task were included in this analysis. As shown in Figure 6A, the CTI bias index value of LIP neurons recorded from Monkey Q was significantly less than zero during the test period of the DMC task (p = 6.42e-4, t(163) = −3.48, one sample t test), indicating that LIP neurons recorded from Monkey Q preferred sample category 2 during that task epoch. LIP neurons recorded from Monkey M showed a nonsignificant bias toward preferring category 1 (throughout sample, delay and test periods: minimum p = 0.257, t(57) = 1.16, t test; Fig. 6B).
Bias in neuronal category preference. The bias in neuronal preference for the two sample categories is quantified across by the CTI bias index. Results from Monkey Q (A) and Monkey M (B) are shown. The black stars denote the time points for which the CTI bias values are significantly different from zero (p < 0.01, one sample t test). LIP neurons from Monkey Q are biased toward preferring sample category 2 during the test period of the DMC task. C, D, Behavioral task accuracies following LIP inactivation are shown separately for each sample category in the DMC task. Monkey Q exhibited a greater deficit on sample category 2 trials than category 1 trials. The black stars indicate statistical significance (*p < 0.05, paired t test). The error bars denote ±SEM. E, F, Average accuracies following LIP inactivation are shown for the two sample directions in the DMS task (same format as in C and D). Monkey Q exhibited a greater deficit on trials with sample direction 2 than direction 1. Monkey M showed an opposite, but nonsignificant, trend. n.s.: nonsignificant.
If LIP neural activity directly contributed to monkeys' behavioral performance in the DMC task, we would expect a correlation between the behavioral deficits caused by LIP inactivation and the categorical preferences of LIP activity in the DMC task. Specifically, we predicted that monkeys would show a greater deficit on trials in which the sample category was from the category that was preferred by the neurons at that LIP location. Consistent with our prediction, Monkey Q exhibited a significantly greater behavioral impairment on sample category 2 trials than sample category 1 trials (category 1 vs category 2 = 0.042 vs 0.103, p = 0.0140, t(7) = 3.25, paired t test; Fig. 6C). A significantly greater behavioral impairment in the DMS task was also observed for direction 2 (the center direction of category 2) trials than direction 1 (the center direction of category 1) trials for Monkey Q (direction 1 vs direction 2 = 0.025 vs 0.111, p = 0.015, t(10) = 2.91, paired t test; Fig. 6E). However, Monkey M did not show different levels of deficits between the two sample categories in the DMC task (category 1 vs category 2 = 0.053 vs 0.048, p = 0.892, t(7) = 0.141; Fig. 6D), and only showed a nonsignificant trend between trials tested with the two sample directions in the DMS task (direction 1 vs direction 2 = 0.189 vs 0.123, p = 0.389, t(3) = 1.00, paired t test; Fig. 6F).
LIP inactivation did not affect patterns of microsaccades or gaze position during DMC or DMS tasks
Similar to the effects of inactivation on behavior during the free choice saccade task observed in the current study (Fig. 2) and previous work, human patients with damage to the posterior parietal cortex often display attentional neglect for stimuli in one visual hemifield, including a decreased probability of saccades toward the neglected hemifield. Thus, we tested whether monkeys' inactivation-induced deficits in the Sin condition of both the DMC and DMS were accompanied by differences in the patterns of microsaccades which could be consistent with spatial neglect. This approach is supported by past work showing that microsaccades are more frequently directed toward the covertly attended location compared with unattended locations in tasks involving covert spatial attention (Hafed and Clark, 2002; Engbert and Kliegl, 2003; Lara and Wallis, 2014; Lowet et al., 2018; Yu et al., 2022). We analyzed both the frequency and directions of monkeys' micrsosaccades in the inactivation and control sessions. As shown in Figure 7A,B, LIP inactivation did not significantly impact the monkeys' microsaccade frequencies in either the SIN or SOUT conditions of either task [minimum p = 0.283, t(17) = 1.11, Monkey Q (Fig. 7A); minimum p = 0.492, t(17) = 0.702, Monkey Q, unpaired t test (Fig. 7B)]. We also compared the proportion of microsaccades directed toward the inactivated visual field between control and inactivation sessions, and found that microsaccade directions in both the SIN and SOUT conditions of both tasks did not significantly shift away from the inactivated visual field in inactivation sessions compared with control sessions (minimum p = 0.298, t(17) = 0.54, Monkey Q, unpaired t test; Fig. 7C–R). Although these results do not directly support the possibility that neglect explains the main results of the study, they also cannot conclusively rule out neglect as a contributing factor.
LIP inactivation did not impact monkeys' microsaccades. A, Microsaccade frequency for each monkey in the DMC task is shown for inactivation and control sessions, separately for SIN and SOUT conditions. B, Both monkeys' microsaccade frequencies in the DMS task is shown for inactivation and control sessions. C–R, Distributions of monkeys' microsaccades in inactivation and control sessions. C–D, Distributions of Monkey M's microsaccade directions in the SIN condition of the DMC task for control (C) and inactivation (D) sessions. Note that 0° and 180° denote the rightward and leftward directions, respectively. E, F, Distributions of Monkey M's microsaccade directions in the SOUT condition of the DMC task in control sessions (E) and inactivation (F) sessions. G–J, Distributions of Monkey M's microsaccade directions in the DMS task are shown in the same format as in C–F. K–R, Distributions of Monkey Q's microsaccade directions in both the DMC and DMS tasks are shown in the same format as C–J.
Furthermore, we tested whether LIP inactivation caused deficits in the monkeys' abilities to maintain gaze fixation. First, we quantified the proportion of trials in which the monkey failed to maintain fixation during control versus inactivation sessions, and found that the frequency of fixation-break trials did not significantly increase in either the DMC or DMS task following LIP inactivation (inact. vs control, Monkey M: DMC, p = 0.31, t(18) = 1.04, DMS, p = 0.37, t(18) = 0.93; Monkey Q: DMC, p = 0.72, t(17) = −0.36, DMS, p = 0.44, t(17) = 0.78, unpaired t test). Second, we compared the monkeys' gaze positions during trials in which they maintained central fixation successfully in both control and inactivation sessions. Previous studies showed that monkeys shifted gaze position away from the inactivated visual field following inactivation of brain areas involved in attention and oculomotor control, and this was interpreted as an attentional deficit (Chen et al., 2020; Dias and Segraves, 1999; Goffart et al., 2012). If inactivation of LIP caused a strong deficit in covert spatial attention, we might expect a significant change in gaze position along the horizontal axis following LIP inactivation. However, we did not find a significant change in horizontal gaze position following LIP inactivation in either the SIN or SOUT conditions (Fig. 8A–D, minimum p = 0.092, Monkey M, t(18) = 1.78, unpaired t test, the SIN condition in Fig. 8A). In the SIN condition of the DMS task following LIP inactivation, one monkey showed a significant trend for gaze position to shift in the downward direction, which was orthogonal to the axis of the IVF and position of the motion stimulus on the display (p = 0.0097, Monkey Q, t(17) = 2.91, unpaired t test; Fig. 8H).
Monkeys' gaze position did not shift away from the inactivated visual field. A, The averaged gaze position in the horizontal direction of Monkey M is shown for inactivation and control DMC sessions, separately for SIN and SOUT conditions. The error bars denote ±SEM. The positive and negative values denote right and left directions, respectively. The inactivated visual field was located in the middle-right visual field for Monkey M. B, The averaged gaze position in the horizontal direction for Monkey M in the DMS task is shown in the same format as in A. C, D, The averaged eye position in the horizontal direction for Monkey Q is shown separately for DMC (C) and DMS (D) tasks. The inactivated visual field was located in the middle-left visual field for Monkey Q. E–H, The averaged gaze position in the vertical direction of both monkeys is shown in the same format as in A–D. The positive and negative values denote upward and downward directions, respectively. The black stars indicate the statistical significance of the comparison between control and inactivation sessions (**p < 0.01, unpaired t test).
Discussion
The goal of this study was to determine whether LIP, an area strongly associated with planning saccadic eye movements, plays a causal role in visually-based decisions that do not involve reporting decisions with saccades. We previously found that inactivating LIP impaired performance in visual motion categorization and discrimination tasks in which the animals reported their decisions with saccades. However, it was unclear whether LIP's causal role in mediating categorical decisions in those tasks would extend to tasks with a different motor response modality. In this study, we found that reversibly inactivating LIP impacts monkeys' behavioral performance during delay-based visual-motion categorization and discrimination tasks. Both tasks require monkeys to maintain gaze fixation across the entire trial, compare sequentially presented visual motion stimuli, and report their match/nonmatch decisions with a hand movement. We found that inactivation of LIP produced significant behavioral deficits in both the DMS and DMC tasks, involving both the monkeys' accuracy and RTs. Furthermore, electrophysiological recordings from the same LIP locations revealed decision-correlated category selectivity in the DMC task. Taken together, these results demonstrate that LIP plays a generalized and causal role in visually-based decisions across multiple types of task paradigms involving both saccadic and manual motor responses.
Decades of work from multiple groups has established that LIP is involved in planning saccadic eye movements, while the medial intraparietal cortex (another PPC subregion) is involved in planning hand movements (Andersen and Buneo, 2002; Cui and Andersen, 2007; Snyder et al., 1997). Using a visual search task, a previous study showed that monkeys' behavioral deficits after LIP inactivation were independent from the effector used to report the decision (contralateral vs ipsilateral hand relative to the inactivated hemisphere; Balan and Gottlieb, 2009). Another study found that saccade-related spatial encoding and nonspatial category encoding in LIP are distinct at both the single-neuron and population levels (Rishel et al., 2013). Furthermore, our previous study found that LIP plays a greater role in the evaluation of in-RF visual stimuli rather than planning saccadic eye movements used to report the animals' perceptual and categorical decisions (Zhou and Freedman, 2019). Extending beyond these previous studies, the current study provides further evidence that LIP's role in visually-based decisions generalizes across multiple types of behavioral paradigms, and is independent of the particular motor responses used to report decisions.
It is also well established that LIP is involved in mediating covert spatial attention. Previous studies have shown that covert attention is engaged to select and evaluate task-relevant stimuli (Bashinski and Bacharach, 1980; Moran and Desimone, 1985; Cohen and Maunsell, 2009; Desimone and Duncan, 1995; Knudsen, 2007; Luo and Maunsell, 2018), and LIP activity closely correlates with shifts of spatial and feature-based attention (Vallar, 1998; Colby and Goldberg, 1999; Bisley and Goldberg, 2003, 2010; Saalmann et al., 2007; Ibos and Freedman, 2014, 2016). Reversible inactivation or lesions of the posterior parietal cortex, including LIP, results in impaired attention in both humans and nonhuman primates (Smania et al., 1998; Battelli et al., 2003; Wardak et al., 2004; Liu et al., 2010). Thus, one concern is that the behavioral deficits from LIP inactivation observed in the current study could be due, at least in part, to impaired attention. However, we argue that monkeys' behavioral deficits in the SIN condition were unlikely to primarily derive from disrupting covert attention, but instead from disrupting the evaluation of (i.e., decisions about) in-RF motion stimuli. One reason for this is that neuronal recordings from the LIP neurons targeted for inactivation showed strong decision-correlated category encoding in the DMC task. Second, previous studies from our group showed that category encoding in LIP emerged with a shorter latency than other cortical areas tested in the DMC task, including PFC, MIP, and MST (Swaminathan and Freedman, 2012; Swaminathan et al., 2013; Zhou et al., 2022). In one of the monkeys in this study, we found that LIP encoding was biased toward preferring one of the two categories. In that animal, LIP inactivation produced greater behavioral accuracy deficits on trials that involved the preferred category versus the nonpreferred category. In the other monkey, LIP activity did not exhibit a significant bias for encoding one motion category, and LIP inactivation impaired that monkey's performance similarly for both categories. Thus, the behavioral accuracy deficits because of inactivation were consistent with the neuronal encoding observed in LIP in each animal. Furthermore, LIP inactivation did not produce different levels of behavioral deficits for stimuli with different levels of noise (coherence), which is inconsistent with an impairment primarily caused by an attentional deficit. This is because an attentional deficit would be expected to produce a greater impairment for the more difficult stimuli with lower signal-to-noise ratios (Dosher and Lu, 2000a, b; Reynolds and Chelazzi, 2004; Wardak et al., 2004). In addition, we compared microsaccadic eye movements and gaze positions between control and inactivation sessions. However, we did not find substantial differences in microsaccade direction or frequency between inactivation and control sessions, and the monkeys' gaze positions did not shift away from the inactivated visual field following LIP inactivation. Nonetheless, we are enthusiastic about future studies that will employ more targeted causal approaches, such as stimulating or silencing pools of neurons with specific category/direction preferences, to more precisely dissect LIP's contributions toward attention and decision-making.
In the current study, LIP inactivation resulted in modest, but not catastrophic, impairments of the monkeys' behavior in both the DMC and DMS tasks. This modest impact on behavior may be related to several factors. First, we did not inactivate the entirety of LIP in that hemisphere, although we infused the muscimol solution through two injection cannulae at multiple depths. Thus, a significant population of LIP neurons may still remain active after inactivation (including the noninactivated hemisphere), which could compensate for categorization deficits. Second, previous work from our lab found that neural activity in other brain areas including PFC, MST, and MIP encoded the monkeys' categorical decisions about sample stimuli in the DMC task (Swaminathan and Freedman, 2012; Swaminathan et al., 2013; Zhou et al., 2021, 2022). Thus, it is possible that these and other brain areas, such as FEF (Ferrera et al., 2009) and even superior colliculus, also play a role in mediating the DMC and DMS tasks and may compensate for inactivation of LIP.
One limitation of the methodology used in the current study is that the electrophysiological recordings and drug infusion were conducted in separate experiment sessions. Therefore, the RF locations and selectivity of the inactivated neurons had to be estimated from recordings made on different sessions. This introduces some uncertainty regarding our precise location and depth within the brain while recording and inactivating. In the current study, we adopted several tactics to maximize our targeting precision and to ensure that we indeed were inactivating LIP. First, we used two injection cannulae (separated by 1–2 mm) on each session to inject musimol solution at five different depths (10 injection sites in total). This allowed us to inactivate a relatively large cortical area within PPC, supported by previous MRI studies showing that musimol solution is expected to spread within a spherical area of ∼2 mm in diameter around the injection site (Arikan et al., 2002; Wilke et al., 2012). Second, we carefully mapped the RFs of hundreds of LIP neurons for each monkey before conducting the inactivation experiments, and only infused the drug to recording sites (grid positions and depths) centered on those where we typically recorded LIP neurons with RFs located in the middle-contralateral visual fields. Third, we used the exact same setup and procedures for positioning the electrode during recording sessions and the injection cannula during inactivation sessions. Finally, we used the free-choice saccade task to assess inactivation effects in the middle-contralateral visual field before testing with the categorization task. Nonetheless, it will be important for future experiments to use more precise causal techniques, such as injectrodes which allow for the simultaneous measurement of spiking activity during drug delivery or optogenetic stimulation. These approaches will better allow precise spatial and temporal manipulation of LIP activity, and will enable site-by-site analyses comparing the neural selectivity and the behavioral impacts of neuronal manipulation.
In both the DMC and DMS tasks, monkeys' decisions about the stimulus categories were not directly linked with their motor responses used to report those decisions. This is because the monkeys' motor response was used to signal their match versus nonmatch decisions, which were decoupled from their decision about the sample-stimulus identity or category by virtue of the delayed matching task structure. By contrast, in the classical motion direction discrimination task which is widely used in motion-based perceptual decision-making studies, monkeys' decisions about stimuli are usually rigidly associated with particular motor responses to report the decisions (Shadlen and Newsome, 1996, 2001; Gold and Shadlen, 2000; Roitman and Shadlen, 2002; Hanks et al., 2006; Yang and Shadlen, 2007; Kiani and Shadlen, 2009; Ding and Gold, 2010, 2012; Katz et al., 2016), for example, by making a rightward versus leftward saccade to indicate rightward versus leftward motion. This difference in task requirements, and the behavioral training required to learn tasks involving greater behavioral or cognitive demands, might result in different neural substrates mediating the decision process (Latimer and Freedman, 2023), and will be important to examine in future experimental work.
In summary, our study demonstrates that LIP plays a generalized and causal role in visual decision-making, which extends across different task contexts and motor response modalities. It will be important for future work to further test the generality of LIP's causal role in decision-making by testing sensory features beyond visual motion (e.g., shape, color, or stimuli from other sensory modalities), as well as to examine the relative roles of visuomotor areas interconnected with LIP, such as the frontal eye field and superior colliculus, in perceptual and categorical decisions.
Footnotes
This work was supported by the National Institutes of Health (NIH) Grant R01EY019041. O.Z. was supported by NIH Grants F30EY033648 and T32GM007281. We thank Kenneth Latimer, Grace DiRisio, and Matthew Rosen for their helpful comments during manuscript preparation and for expert veterinary assistance from the Animal Resources Center at The University of Chicago.
The authors declare no competing financial interests.
- Correspondence should be addressed to Yang Zhou at yangzhou1{at}pku.edu.cn or David J. Freedman at dfreedman{at}uchicago.edu
