Recent evidence has demonstrated that spatiotemporal patterns of spontaneous activity reflect the patterns of activity evoked by sensory stimuli. However, few studies have examined whether response profiles of task-evoked activity, which is not related to external sensory stimuli but rather to internal processes, are also reflected in those of spontaneous activity. To address this, we recorded activity of neurons in the lateral intraparietal area (LIP) when monkeys performed reaction-time and delayed-response visual-search tasks. We particularly focused on the target location-dependent modulation of delay-period activity (delay-period modulation) in the delayed-response task, and the discharge-rate persistency in fixation-period activity (baseline-activity maintenance) in the reaction-time task. Baseline-activity maintenance was assessed by the correlation between the spike counts of 2 separate bins. We found that baseline-activity maintenance, calculated from bins separated by a long interval (200–500 ms), was correlated with delay-period modulation, whereas that calculated from bins separated by a short interval (∼100 ms) was correlated with trial-to-trial fluctuations in baseline activity, suggesting a link between the capability to hold task-related information in delay-period activity and the degree of baseline-activity maintenance in a timescale-dependent manner.
Different neurons often respond in different ways even when they are located in the same cortical area and subjects are performing the same behavioral task. Although this cell-to-cell difference depends on a number of factors, such as sensory stimuli, cognitive processing, and motor preparation/generation, it may also be due to differences in the intrinsic response properties of individual neurons. Indeed, fluctuations in spontaneous activity reflect the intrinsic functional architecture of the brain in human imaging studies (for reviews, see Fox and Raichle 2007; Greicius 2008) and in animal electrophysiological studies (Arieli et al. 1996; Tsodyks et al. 1999; Kenet et al. 2003; Petersen et al. 2003; Fiser et al. 2004; Curto et al. 2009; Luczak et al. 2009; Ringach 2009; Sakata and Harris 2009). Thus, it is possible that cell-specific differences may produce different neuronal responses not only during task-evoked activity but also during spontaneous activity.
Previously, we particularly focused on the discharge-rate persistence in fixation-period activity (baseline-activity maintenance) in a visual-search task. Because fixation-period activity is less influenced by voluntary eye movements and responses to visual stimuli compared with the activity during cognitive processes or behavioral actions, it may be suitable to infer the intrinsic response properties of a neuron. When comparing the degree of baseline-activity maintenance between 2 cortical areas, the frontal eye field (FEF) and area V4, we found that the ability to maintain the instantaneous discharge rate was greater for FEF neurons than for V4 neurons, suggesting a difference in the intrinsic storage capability across cortical areas (Ogawa and Komatsu 2010). However, that study did not answer the question of whether the degree of baseline-activity maintenance for a neuron actually correlates with the ability of that neuron to hold task-related signals in its activity during a task. More specifically, a neuron with a higher degree of baseline-activity maintenance could exhibit a greater ability to hold information about the target position in its delay-period activity in a memory-guided saccade task (Hikosaka and Wurtz 1983).
The lateral intraparietal area (LIP) is known to be a region in which robust and spatially tuned activity is observed during the delay period in a memory-guided saccade task (e.g., Gnadt and Andersen 1988; Barash et al. 1991a, 1991b). Thus, the LIP is a suitable site to assess the above possibility. Here, we recorded neuronal activity from the LIP of monkeys performing reaction-time and delayed-response visual-search tasks, and examined the correlations between the degree of baseline-activity maintenance during fixation in the reaction-time task and the strength of target location-dependent modulation during delay in the delayed-response task (delay-period modulation). We found significant co-variability across neurons between the degree of baseline-activity maintenance and the strength of delay-period modulation. Our results suggest that the response properties observed in the baseline activity of an LIP neuron may be linked with intrinsic functional abilities that shape cell-specific responses in the task-evoked activity of that neuron.
Materials and Methods
Animal Preparation and Apparatus
Data were collected from 2 female macaque monkeys (Macaca fuscata, monkeys Y and S, 5.0 and 6.3 kg, respectively). A plastic head holder and recording chambers were secured to the skull using dental acrylic resin (Unifast II, GC, Aichi, Japan) and ceramic screws (Thomas Recording, Geissen, Germany) under deep anesthesia and sterile conditions. The recording chambers (inner diameter, 22 mm) were placed at stereotaxic coordinates (P2 and L21.5 for monkey Y; P4.1 and L18, and P0 and L21 for monkey S) above the intraparietal sulcus (IPS) with the assistance of magnetic resonance imaging (MRI) obtained before surgery (see below). An eye coil was surgically implanted beneath the conjunctiva of 1 eye (Judge et al. 1980). The monkeys were allowed to recover for 2 weeks prior to training and recording. All procedures for animal care and the experimental protocols were in accordance with the Guide for the Care and Use of Laboratory Animals of the National Research Council (1996) and were approved by the Animal Care and Use Committee of Kyoto University.
The experiments were under the control of 2 Windows XP-based computers that presented the stimuli, recorded neural signals and eye positions, and controlled the task schedule. They were developed using LabVIEW (National Instrument Japan, Tokyo, Japan) and C++ Builder (Borland Software Corporation, Scotts Valley, CA, USA). Visual stimuli were generated using a video-signal generator (ViSaGe; Cambridge Research Systems, Cambridge, UK) and presented on a video monitor with a 100-Hz refresh rate and 800 × 600 resolution (RDF223H; Mitsubishi, Tokyo, Japan). They were viewed binocularly from a distance of 42 cm in a dark room and subtended a visual angle of 51.5 × 40.0°. The luminance of stimuli and background was measured by a ColorCAL colorimeter (Cambridge Research Systems).
Single neurons were recorded using an epoxylite-insulated tungsten electrode (Frederick Haer & Co, Bowdoinham, ME, USA) with an impedance >2 MΩ measured at 1 kHz (model IMP-1, Bak Electronics, Germantown, MD, USA). Extracellular activity was amplified using a microelectrode AC amplifier (Model-1800; A-M Systems, Carlsborg, WA, USA) and stored on a computer equipped with a multichannel analog-to-digital board at a sampling rate of 50 kHz (PCI-6143, National Instrument Japan). Eye position was monitored and recorded using the scleral search coil technique (Fuchs and Robinson 1966) (eye position detector DSC-2001; Datel, Tokyo, Japan). Precise spikes were discriminated off-line using a template-matching method. Eye position signals were recorded at a sampling rate of 50 kHz but analyzed at 1 kHz resolution.
The recording site was determined using a guide tube and a set of plastic grids that had holes spaced 1.0 mm apart and which were offset from each other by 0.5 mm. The guide tube (23 gauge) was lowered just above the dura matter surface, and the electrode penetrated the cortex through the dura using an oil hydraulic micromanipulator (MO-97A-S, Narishige, Tokyo, Japan). Under microscopic examination (OPMI-pico-i, Zeiss, Tokyo, Japan), a duratomy was made using fine forceps (Dumont No. 5) and a 25-gauge needle with the tip bent at a right angle. Only a small region of the tissue just below the selected grid hole was removed to avoid breaking the electrode. The chamber was filled with agarose (3%, A9793; Sigma, St. Louis, MO, USA) to promote recording stability. After each recording session, the dura matter surface was covered with anti-infective/anti-inflammatory ointment (Chlomy-P ointment, Daiichi-Sankyo, Tokyo, Japan).
Delayed-Response Visual-Search Task
This task was a variation of a memory-guided saccade task (Hikosaka and Wurtz 1983) (Fig. 1A). We analyzed neuronal activity of 94 LIP neurons obtained during this task. Each trial began with the appearance of a fixation spot at the center of a gray background (1 cd/m2). The monkeys had to fixate on that spot within a window for typically 1000 ms, except for 13 cells (1,500–2,000 ms). The size of the window was ±1.2–1.6°, except for 17 cells (±2.5°). After sustained fixation, a cue stimulus was displayed at 8.5° eccentricity to indicate the to-be-remembered target location (cue period, 500 or 700 ms). The target location was set with equal probability in the receptive field of the neuron under study or in the diametrically opposite location in the visual field. After fixation for a variable delay (800–1,800 ms), the fixation point was extinguished (“go” signal), and the monkeys were required to make a saccade toward the location at which the target had appeared during the cue period. If the monkeys made a single saccade landing inside a square window (±3.0 × 3.0°) centered on the target, another fixation point appeared at the target position. After 600 ms of fixation on this point, the monkey received a juice reward (successful trial). An intertrial interval (300–2,000 ms) was imposed before the beginning of a next trial after a trial ended. If the monkey made a saccade toward the location outside the target-centered window or could not keep the fixation within the window for 600 ms, the trial was immediately terminated with no reward (erroneous trial).
A cue stimulus was either an isolated stimulus (isolated-stimulus trial) or an array stimulus (array-stimulus trial). In the isolated-stimulus trials, a circle (10 cd/m2; 2.24° in diameter) was presented as a target. Its color was either white (CIE chromaticity coordinate: x = 0.29, y = 0.32), orange (x = 0.43, y = 0.47 for monkey Y; x = 0.44, y = 0.46 for monkey S), yellowish-orange (x = 0.41, y = 0.48 for monkey Y; x = 0.42–0.43, y = 0.47 for monkey S), green (x = 0.24, y = 0.41 for monkey Y; x = 0.23, y = 0.37–0.39 for monkey S), or bluish-green (x = 0.23, y = 0.38 for monkey Y; x = 0.22, y = 0.34 for monkey S). In the array-stimulus trials, 1 singleton element (target), unique in color, and 5 additional identical elements (distractors) were simultaneously presented. All elements were isoluminant circles (10 cd/m2; 2.24° in diameter) with the same colors as used in the isolated-stimulus trials. Target and distractor colors were set to either orange/green, green/orange, yellowish-orange/bluish-green, or bluish-green/yellowish-orange. Each color pair was presented at the equal probability. Of the 94 cells studied, data from 73 were recorded during both the isolated and array-stimulus trials, and data from the remaining 21 were recorded only during the isolated-stimulus trials.
Reaction-Time Visual-Search Task
The procedure for this task was the same as that in the delayed-response visual-search task, except that no delay period was imposed (Fig. 1B), the number of color pairs for array stimuli was increased, and target-absent trials were introduced. After fixation, either an isolated or an array stimulus was displayed, and the monkeys were required to make a saccade toward the target without an artificial delay. When the computer detected a saccadic eye movement, the visual stimuli and the fixation spot were immediately extinguished. In the isolated-stimulus trials, an orange, yellowish-orange, green or bluish-green circle (10 cd/m2) was presented as a target. In the array-stimulus trials, 1 color-singleton target was presented with 5 identical distractors. All 6 elements were isoluminant (10 or 1 cd/m2). Target- and distractor-color pairs were selected so that their color similarity was either small (orange/green, green/orange, yellowish-orange/bluish-green, or bluish-green/yellowish-orange) or large (yellowish-orange/orange, orange/yellowish-orange, bluish-green/green, or green/bluish-green). The stimulus colors in both trials were the same as those used in the delayed-response task. In addition to these target-present trials, target-absent trials were also interleaved. If all 6 elements had one of the 4 colors, the monkeys had to maintain their fixation (800–1,800 ms) until the end of the trial (target-absent trials).
Trials of delayed-response (17–23%) and reaction-time (77–83%) visual-search tasks were implemented pseudo-randomly. The eccentricity of the visual stimuli were fixed at 8.5° in the periphery to minimize inter-neuron differences in task difficulty across recording sessions, because behavioral performance during visual search varies with retinotopical eccentricity (e.g., Carrasco and Yeshurun 1998; Wolfe et al. 1998; Meinecke and Donk 2002). The task type was instructed by the color or shape of the fixation spot. A white circle (0.39–0.55° in diameter) was presented during the reaction-time task, whereas either a yellow circle or white square (0.39–0.55° in diameter) was presented during the delayed-response task. For a minority of cells (13/94 = 14%), trials of both tasks were conducted separately.
Single-cell activity was recorded by advancing electrode into the lateral bank of the IPS. Once a neuron was isolated, we initially assessed the location of the receptive field using the delayed-response task. An isolated stimulus was presented at one of the six evenly separated directions on an imaginary circle (eccentricity = 8.5°), because the stimulus-location eccentricity was fixed at 8.5° during the visual-search tasks. We manually adjusted those 6 directions so that 1 direction evoked the strongest activity.
Before starting the present experiment, we specified the IPS location based on the response properties; the medial bank of the IPS tended to exhibit activity related to somatosensory stimuli, whereas the lateral bank exhibits visual and saccade-related responses (Mountcastle et al. 1975; Barash et al. 1991b; Maimon and Assad 2006). After specifying the IPS locus, we started to record neurons in the lateral bank of the IPS from a region regarded as the LIP in which neurons exhibit robust, spatially tuned responses during the delay period of a memory-guided saccade task (Gnadt and Andersen 1988; Barash et al. 1991a, 1991b; Colby et al. 1996; Shadlen and Newsome 2001). To ensure that our samples were in area LIP rather than area 7a, neurons recorded at a depth shallower than 3 mm from the dura surface were excluded from the present analysis (Andersen et al. 1990; Linden et al. 1999; Gifford and Cohen 2004). Neurons were typically recorded at >5 mm depth from the dura surface (85%), and the majority of our neurons (56%, see Results) exhibited significant modulations in delay-period activity (determined by whether activity 100–700 ms before the go signal differed significantly according to target location; Mann–Whitney U-test, P < 0.05). This fraction was comparable with that reported for the LIP in previous studies (e.g., Barash et al. 1991a; Maimon and Assad 2006; Falkner et al. 2010).
To verify the recording positions, we acquired postoperative structural MRIs for both monkeys on a 0.2 Tesla open whole-body scanner (Signa Profile; General Electric, Milwaukee, WI, USA). The advantage of this low-magnetic field system is the reduced distortion of the original cortical structures (Petersch et al. 2004). We used a 3-dimensional spoiled, gradient-recalled pulse sequence with the following parameters: TR = 29 ms, TE = 8.3 ms, flip angle = 40°, field of view = 16 cm, 256 × 256 matrix, 100 slices, voxel size = 0.63 × 0.63 × 1.0 mm. To improve image signal-to-noise ratio, we aligned and averaged 2 or 3 acquisitions during post-processing. Figure 2 shows images from 1 monkey (monkey Y). We used 5 plastic tubes filled with an 84–87% glycerin solution (Glycerin P, Kenei, Tokyo, Japan) to be visible in the MRI to determine the plane and trajectory of the electrode penetrations (Fig. 2A). These tubes were embedded in a plastic base attached to the recording chamber so that 1 was the center and the remaining 4 were arranged in the x–y coordinate (8 mm apart from the center) of the recording grid (Fig. 2B). The penetration sites on the brain surface were reconstructed based on the coordinates determined by the positions of the reference tubes and recording grids. The number of cells recorded from each grid location is illustrated in Figure 2C. Figure 2D shows the MRIs that matched the plane of the recording penetrations. The zone of the recording sites (arrow heads), which was reconstructed from a combination of grid sites and recording depth, is superimposed on the MRIs. Figure 2E shows coronal MRIs representing the most anterior to the most posterior recording positions. We similarly verified recording positions for another monkey (monkeys S) by acquiring postoperative MRIs. Thus, our samples were presumably recorded from the LIP based on the physiological properties of the recorded neurons and verification by the MRIs.
We constructed interspike interval (ISI) distribution histograms to confirm adequate spike isolation. Neurons were included in the analysis if they maintained stable spike isolation throughout the recording session and if their distribution histograms showed no ISIs < 2 ms (the refractory period). Unless otherwise indicated, only data from successful trials were analyzed. Spike-density functions were constructed with 1 kHz resolution by convolving spike trains with a Gaussian function (SD = 10 ms) (Richmond et al. 1987). Population average spike-density functions were constructed by averaging the spike-density functions from individual neurons.
Analysis of Saccadic eye Movements
Horizontal and vertical eye velocity signals were calculated by digitally differentiating horizontal and vertical eye position signals, respectively. Saccades were detected using a computer algorithm that identified the initiation and termination of each saccade based on a velocity-threshold criterion. Change in eye velocity was recognized as a saccade if it was greater than the threshold of 120°/s for at least 10 ms. The initiation of a saccade was defined as the time at which the velocity increased to >30°/s.
Analysis of Discharge-Rate Maintenance in Baseline Activity
To assess the persistence of the instantaneous discharge rate in baseline activity, we focused on prestimulus activity during fixation (the activity in an 800-ms interval prior to stimulus onset). Because the monkeys were required to maintain their gaze directed toward the central spot on the display during the fixation period and had no information about the forthcoming visual stimuli, baseline activity was virtually free of any voluntary eye movements and response to task-related visual stimuli except the fixation spot. This allowed us to examine response properties of individual neurons under conditions in which the effects of visual sensory stimuli and eye movements may be at a minimum. We examined the prestimulus baseline activity obtained from both the reaction-time and delayed-response tasks. In particular, we were more interested in the baseline activity in the reaction-time task, because it was possible that baseline activity in the delayed-response task could be modulated by the readiness to the task requiring the monkeys to remember the spatial location of a target. Indeed, previous studies have demonstrated that prestimulus baseline activity in the LIP is modulated by the demands of the task at hand (Toth and Assad 2002; Stoet and Snyder 2004). Taken together, by examining the fixation-period activity in the reaction-time task, we could access an automatic process for activity maintenance rather than an active process driven by task demands. To ensure sufficient spike data for analysis, we examined only neurons in which the number of trials was >100 and the mean activity during the 800-ms interval before array presentation was >5 spikes/s. Of 119 neurons we recorded, 94 neurons fulfilled these criteria. The results were essentially the same when the spike-rate threshold varied between 1 and 10 spikes/s.
To intuitively illustrate baseline-activity maintenance, we separated the trials into 3 groups (high, medium, and low activity) according to the spike counts during the 100-ms interval 400–500 ms before array presentation. The distribution of spike counts during each trial was fitted to a Poisson distribution. The thresholds for trial classification were determined based on the one-third and two-third levels of the cumulative probability of the distribution. Then we calculated the spike-density function for each of the trial groups. We expected that the difference in spike-density functions at a given time would persist for a long period if a neuron stored its instantaneous discharge rate for the long term; if not, the differences in spike density would diminish after a short interval.
To quantitatively characterize baseline-activity maintenance, we calculated the temporal correlations between spike counts per bin within a trial (temporal correlation analysis), as described in Ogawa and Komatsu (2010). We first divided the 800-ms fixation period prior to stimulus onset into 8 successive 100-ms time bins and counted spikes in each bin of individual trials. The mean spike count across trials in a given bin was subtracted in advance from the spike counts of individual trials in that bin. Next, we produced spike-count pairs of bins separated by intervals of 100, 200, 300, 400, 500, 600, or 700 ms. Then we calculated the Pearson’'s correlation coefficient using all pairs of bins with each separation interval. For example, to calculate the temporal correlation with a 300-ms separation from a dataset of 100 trials, the spike-count data from 5 pairs of bins (1st–4th, 2nd–5th, 3rd–6th, 4th–7th, and 5th–8th bins) for each trial were pooled across all trials (500 spike-count pairs = 5 pairs/trial × 100 trials), and then 1 correlation coefficient value was computed using these 500 spike-count pairs. The coefficients were used as the index of baseline-activity maintenance, termed the within-trial baseline discharge-rate maintenance (within-trial BM) index. We also calculated an across-trial BM index (i.e., shift predictor) to estimate the strength of the correlations resulting from intertrial changes in baseline activity, which indicates the correlation between spike counts in 1 bin during a given trial and that in the other bin during the following trial. As described above, because the mean spike count across trials in each bin was subtracted from the spike counts of individual trials in that bin, the within-trial BM index reflected the correlation between the spike-count deviations from the mean spike count in 1 bin and those in another bin. Consequently, if a neuron has the high ability to hold its discharge rate, the stronger discharge rate would persist for a while when the discharge rate is stronger than its mean value in a given bin, whereas the weaker discharge rate would persist if the discharge rate is weaker than its mean value in a given bin. Therefore, this index can robustly measure the degree of the discharge-rate persistency, even if the mean discharge rate is not constant but rather has an unstable temporal pattern during fixation.
If the discharge rate of baseline activity slowly drifted across trials, this slow drift could generate quasi-significant correlations in the temporal correlation analysis even in the absence of actual correlations between the spike counts of the separate bins within a trial. To avoid this effect on the temporal correlation analysis, we analyzed only the baseline activity of those trials in which the mean discharge rates of the fixation-period activity were not statistically different across trials. To this end, we chose 100 successive trials (a 100-trial window) during a recording session, divided them into 4 successive 25-trial groups, and determined significant differences in the strength of activities across the 4 groups using the Kruskal–Wallis test. By moving a 100-trial window in 1-trial steps from the first to the last trial in each recording session, we determined the longest successive set of trials in which activity strength did not differ significantly (Kruskal–Wallis test, P > 0.01). Applying this statistical pretreatment, 138–818 trials (mean ± SD = 365 ± 160 trials; 27.7–100% of all trials, mean ± SD = 67.9 ± 22.9%) were chosen and used for the current analyses. As we will show, this manipulation was not essential for obtaining our main findings.
Analysis of DMs of Task-Evoked Activity
In the delayed-response task, animals had to remember the spatial location of a previously presented target to make a saccade to that location after the delay. Many studies have demonstrated that LIP neurons exhibit spatially tuned responses during the delay period (e.g., Gnadt and Andersen 1988; Barash et al. 1991a, 1991b; Colby et al. 1996; Shadlen and Newsome 2001; Bisley and Goldberg 2003, 2006). Delay-period activity is thought to be related to the activity that is evoked by internal (endogenous) factors, such as memory, spatial attention, and/or motor preparation, rather than the activity that is directly driven by sensory (exogenous) factors. In this context, we consider delay-period activity as “task-evoked” activity. To quantitatively evaluate the strength of target direction-dependent modulation in the task-evoked activity for each neuron, we computed a DM index defined as |Rin– Raway|/(Rin+ Raway), where Rin and Raway refer to the mean discharge rate during trials in which the target appeared inside the receptive field and those in which the target appeared outside the receptive field, respectively. A value close to 1 indicated that the difference in the discharge rate between the 2 stimulus conditions was substantial (i.e., strong delay-period modulation), whereas a value close to 0 indicated little difference between the 2 stimulus conditions (i.e., weak delay-period modulation). To reduce the possibility that the DM index was substantially influenced by a decay in visually evoked activity or saccade eye movement-related activity, the activity was measured as the spike count per trial during a 600-ms interval from 100 to 700 ms before the disappearance of the fixation spot: starting at a mean of 544 (±217 SD) ms after target disappearance and ending at a mean of 283 (±43 SD) ms before saccade onset.
Analysis of Trial-to-Trial Fluctuations in Baseline Activity
We calculated Fano factors by dividing the variance in the spike count by its mean to assess trial-to-trial fluctuations in baseline activity. We measured the Fano factors for each neuron from the spike counts in baseline activity during each consecutive 100-ms bin during the 800 ms before search-array presentation and averaged them across bins.
Analysis of Response Types of Neurons
Manifestations of visually responsive and saccade-burst activity were examined using data sets from the isolated-stimulus trials during the delayed-response visual-search task. A neuron was defined as exhibiting visually responsive activity when the activity occurring from 50 to 150 ms after array presentation was significantly greater than the prestimulus activity occurring from 200 to 400 ms before the array presentation (Mann–Whitney U-test, P< 0.05). A neuron was defined as exhibiting saccade-burst activity if its activity during a 100-ms interval prior to saccade initiation was significantly greater than its preceding delay-period activity during a 100-ms interval starting 300 ms before the initiation of a saccade (Mann–Whitney U-test, P < 0.05) (Lawrence et al. 2005). Then we classified neurons into 3 types: neurons with only visually responsive activity (visual neurons), those with only saccade-burst activity (movement neurons), and those with both responses (visuomovement neurons).
Recordings were made from 3 hemispheres of 2 macaques (1 hemisphere in monkey Y and both hemispheres in monkey S). Only neurons that maintained stable spike isolation throughout the recording session and showed ISI distribution histograms with no ISIs<2 ms were included in the analysis. In total, 119 neurons were classified as being in the LIP based on the anatomical and physiological criteria (see Materials and Methods). Of them, 94 neurons (31 from monkey S and 63 from monkey Y) fulfilled the criteria for appreciable baseline activity during the 800 ms before stimulus onset in the reaction-time task (≥5 spikes/s, 5.0–56.6 spikes/s, mean ± SD = 17.3 ± 11.3 spikes/s) and for sufficient trials in both the reaction-time task (≥100 trials, 138–818 trials, mean ± SD = 365 ± 160 trials) and the delayed-response task (≥4 trials under each stimulus condition, 4–99 trials, 23 ± 15 trials). We used these 94 neurons for data analysis in this study.
Baseline-Activity Maintenance and Delay-Period Modulation of Single LIP Neurons
Figure 3A–D shows the response profiles from 1 LIP neuron. Figure 3A shows neuronal activity during the reaction-time visual-search task from 693 successful trials. To intuitively show the ability of baseline-activity maintenance, we separated trials into 3 groups according to the activity strength in a brief interval and compared these groups. Figure 3B shows 3 spike-density functions of the high- (red), medium- (green), and low-activity (blue) groups classified based on the 100-ms interval 400–500 ms before array presentation (gray rectangle in Fig. 3B). A significant activity difference between the high- and low-activity groups persisted not only during the 100 ms used for trial separation but also during the remaining fixation period (horizontal black line; Mann–Whitney U-test, P < 0.01), indicating strong maintenance of the instantaneous discharge rate during baseline activity. Similar results were observed when the trials were grouped according to the activity strength of the other 100-ms bins during the fixation period. To quantitatively evaluate the ability of baseline-activity maintenance, we computed the baseline discharge-rate maintenance indices (BM indices) that reflected the strength of the correlation between the spike counts in two 100-ms bins. Figure 3C shows the values of the within-trial (circles) and across-trial (squares) BM indices as a function of the time separating 2 bins. Although the strength of the within-trial BM indices gradually decreased as separation time increased, it was significantly greater than zero at all separation times except 600 and 700 ms (black circles; Pearson's correlation, P < 0.01), indicating that this cell possessed the ability to retain its instantaneous discharge rate up to 500 ms. In contrast, the across-trial BM indices were close to zero (−0.01 to 0.017) and not significantly different from zero at all separation times (white squares; Pearson's correlation, P > 0.4), indicating that the maintenance of baseline activity was not the result of neural modulation across trials but rather of neural modulation within a trial.
Then we examined the strength of delay-period modulations in the delay-response visual-search task. Figure 3D shows the activity when an isolated target appeared inside the receptive field (black trace) and when it appeared outside the receptive field (gray trace). We found a profound difference in delay-period responses depending on the location of the target (Mann–Whitney U-test, P < 0.0001). To quantify the ability of this neuron to signal the target location during the delay, we calculated a delay-period modulation index that reflected the difference in activity between trials when the target appeared inside and outside the receptive field. The DM index of this neuron was 0.83.
Figure 3E–H shows the response profile of another LIP neuron. The degree of baseline-activity maintenance was considerably weaker in this neuron (Fig. 3F). When trials (454 successful trials) were separated into 3 groups according to the strength of instantaneous activity, the significant difference between the high- and low-activity groups was restricted to the period in and around the 100-ms interval used for trial separation (horizontal black line; Mann–Whitney U-test, P < 0.01). The temporal correlation analysis of baseline activity consistently revealed that the values of the within-trial BM indices were not different from zero (Fig. 3G; Pearson's correlation, P > 0.02) except for a 200-ms separation time (Pearson's correlation, P < 0.01). Moreover, although delay-period modulation was significant (Fig. 3H, Mann–Whitney U-test, P = 0.024), its magnitude was very small (DM index = 0.12).
Thus, 1 neuron with strong baseline-activity maintenance exhibited substantial delay-period modulation, whereas another neuron with weak baseline-activity maintenance exhibited poor delay-period modulation. These results suggest that the neuron-to-neuron variability in baseline-activity maintenance during fixation is potentially associated with that in delay-period modulation of task-evoked activity. In the following analyses, we sought to assess this potential link at the population level.
Population Analyses of Baseline-Activity Maintenance
To determine the degree of baseline-activity maintenance at the population level, we calculated the BM indices for each of 94 LIP neurons using the baseline activity in the reaction-time task. Figure 4A shows the values of the within-trial BM index as a function of separation time. The mean within-trial BM indices (thick black line) decreased gradually as separation time increased (Spearman's correlation, ρ = −0.57, P < 0.0001). At the level of individual neurons, the degree of baseline-activity maintenance varied largely from neuron to neuron (thin gray lines). The values of the within-trial BM indices were significantly greater than zero in 95%, 95%, 83%, 76%, 50%, 38%, and 26% of the neurons at separation times of 100, 200, 300, 400, 500, 600, and 700 ms, respectively (Pearson's correlation, P < 0.01). We found no neuron in which the within-trial BM index was significantly less than zero (Pearson's correlation, P > 0.01). This result indicates that almost all neurons (>90%) maintained the discharge rate for a period ranging from 100 to 200 ms, whereas about one-third or one-fourth of neurons (< 40%) did so for up to 600 or 700 ms. Figure 4B shows the across-trial BM indices. In contrast to that of the within-trial BM index, the strength of the across-trial BM index was rather low across neurons. In only a minority of neurons, the across-trial BM index was significantly different from zero (Pearson's correlation, P < 0.01): 17%, 15%, 12%, 6%, 9%, 5%, and 3% of neurons at separation times of 100, 200, 300, 400, 500, 600, and 700 ms, respectively. Overall, the mean across-trial BM indices were significantly lower than the mean within-trial BM indices at all separation times (paired t-test, P < 0.0001).
To confirm that these observations were not limited to the BM indices obtained using a particular bin size (100 ms), we calculated the BM indices using the other bin widths (50, 200, and 300 ms). The mean within-trial BM indices decreased gradually as a function of increases in the separation time (Supplementary Fig. 1A; Spearman's correlation, ρ = −0.57, −0.50, and −0.39 for bin widths of 50, 200, and 300 ms, respectively, P < 0.0001), and the mean across-trial BM indices were close to zero irrespective of separation time (0.008–0.037; Supplementary Fig. 1B), consistent with the results observed when a 100-ms bin width was used for the calculation. Therefore, the following analyses were conducted using the BM indices calculated with a 100-ms bin width unless otherwise indicated.
Before starting the fixation in each trial, the eye position was not restricted. Therefore, there is the possibility that that differential visually evoked and/or saccade-related activity might be elicited when the monkeys made saccades to the fixation spot from different locations. If such differential activity lasted, this could yield false temporal correlations in baseline activity. In this case, it is expected that the BM indices were reduced as the fixation time increased, because visual and saccadic responses should gradually diminish after the onset of fixation. To test this possibility, we examined how the within-trial BM index changed depending on the time during the fixation period. The result shows that the magnitude of the BM indices tended to hold during fixation (Supplementary Fig. 2A), indicating that it is unlikely that visual responses and prefixation eye movements were responsible for producing the temporal correlations observed in baseline activity.
In our recordings, the fixation duration prior to stimulus presentation was typically fixed (1,000 ms), but it was variable (1,500–2,000 ms) for a minority of cells. This difference in the fixation period did not produce any difference in the degree of baseline-activity maintenance. We compared the within-trial BM indices between 2 groups of neurons separated according to whether the fixation duration was fixed (n = 81) or variable (n = 13), and found no difference between them at all separation times (Supplementary Fig. 2B; two-sample t-test, P > 0.07).
Previous studies have shown that the correlation between spike trains increases with firing rate (De la Rocha et al. 2007). To test whether the observed correlations were due to higher discharge rates of baseline activity, we examined the dependence of the BM indices on the discharge rate of baseline activity. Although the discharge rate during the 800-ms interval preceding the array presentation varied from neuron to neuron (5–57 spikes/s, mean ± SD = 17 ± 11 spikes/s), we found no significant correlation between the discharge rate and the BM indices at all separation times across neurons (Pearson's correlation, r = -0.06–0.17, P > 0.09), indicating that the BM index did not depend on the discharge rate in baseline activity.
Population Analyses of Delay-Period Activity Modulation
We calculated the delay-period modulation index for each of the 94 LIP neurons. Figure 5A shows the distribution of the DM indices, which varied across neurons (0–0.83, mean ± SD = 0.25 ± 0.22). Significant delay-period modulation was observed in 56% (53 of 94) of the neurons studied (gray bar in Fig. 5A; Mann–Whitney U-test, P < 0.05). To illustrate the relationship between the DM index and the time course of activity during the delayed-response task, the 94 neurons were divided into 3 groups based on the 2 threshold levels corresponding to one-third and two-thirds of the difference between the minimum and maximum DM indices, and the average spike-density functions were calculated separately for each group (Fig. 5B–D). Delay-period modulation was prominent for neurons with high DM indices (Fig. 5B), moderate for neurons with medium DM indices (Fig. 5C), and weak for neurons with low DM indices (Fig. 5D), assuming that the currently defined DM index is an appropriate measurement of the strength of the task-related modulation in delay-period activity.
For further confirmation to exclude the influence of decaying visually evoked activity on the DM indices, we calculated these indices using the delay-period activity from the array-stimulus trials (Supplementary Fig. 3A). Because isoluminant target or distractor elements were always presented in the receptive field of the neuron during the cue period of the array-stimulus trials, the difference between visual responses when the target appeared inside the receptive field and that when it appeared outside the receptive filed should have been substantially reduced compared with that during the isolated-stimulus trials. The cell-to-cell variation in the DM indices for the array-stimulus trials was consistent with that in the isolated-stimulus trials (Supplementary Fig. 3B), indicating that delay-period modulations were unlikely to be explained by the effects of decaying stimulus-driven activity evoked by a cue stimulus.
Effects of Trial-to-Trial Fluctuations in Baseline Activity
To know the relationship between the degree of baseline-activity maintenance (within-trial BM index) and the strength of delay-period modulation (DM index), we calculated the raw correlation between the 2 indices across our sample of 94 neurons (Fig. 6). Figure 6A shows the relationship between the DM indices and the within-trial BM indices, which were calculated from bins separated by either 100-, 300-, or 500-ms intervals. Interestingly, the correlation coefficients were significant for the BM indices with a 300- or 500-ms separation (Pearson's correlation, r = 0.44 and 0.22, respectively, P < 0.05), but not a 100-ms separation (Pearson's correlation, r = 0.18, P = 0.08). Figure 6C summarizes the dependence of the correlation between the DM and the within-trial BM indices on separation time (see also Supplementary Table 1). The strength of the correlation was strongest at a separation time of 300 ms and decreased gradually as separation time increased or decreased from that separation time (black bars in Fig. 6C).
This result seems implausible because the average within-trial BM index was strongest at 100 ms of separation and monotonically decreased as separation time increased (Fig. 4A). What mechanisms explain this discrepancy? It is generally known that the discharge rate in baseline activity fluctuates from trial to trial. Such trial-to-trial fluctuations evaluated with Fano factors in baseline activity could affect the strength of the within-trial BM index. Indeed, we found a significant raw correlation between the Fano factor (variance/mean spike count in a 100-ms window) and the within-trial BM index at 100 ms of separation (Fig. 6B; Pearson's correlation, r = 0.47, P < 0.0001), suggesting that the influence of trial-to-trial fluctuations may interfere with the correlation between the DM and the within-trial BM indices at short separation times (∼100 ms).
Analysis of Partial Correlations Between Baseline-Activity Maintenance, Delay-Period Modulation, and Fano Factors
To determine whether the variables were actually correlated, we calculated the partial correlation between the DM and the within-trial BM indices, factoring out their association with the Fano factor, and similarly calculated the partial correlation between the within-trial BM index and the Fano factor, factoring out the DM index. The Fano factor was significantly correlated with the within-trial BM index, particularly at a separation time of 100 ms (gray bars in Fig. 6D, Pearson's partial correlation, r = 0.47, P < 0.05), whereas the DM index was significantly correlated with the within-trial BM index at separation times ranging from 200 to 500 ms (black bars in Fig. 6D, Pearson's partial correlation, r = 0.22–0.44, P < 0.05) (all r and P-values are summarized in Supplementary Table 1). We found no significant correlation between the DM index and the Fano factor (Pearson's correlation, r = 0.035, P = 0.73). These results suggest that the strength of delay-period modulation may be linked with the degree of longer term maintenance (200–500 ms) of baseline activity, whereas the trial-to-trial fluctuations may be linked with the degree of short-term maintenance (∼100 ms).
To confirm the robust association between the baseline-activity maintenance and the delay-period modulation, we performed the same partial correlation analysis when the delay-period modulation was measured in 2 different ways. When a partial correlation was calculated using the DM indices obtained from the array-stimulus trials instead of the isolated-stimulus trials (Supplementary Fig. 3C), or when it was calculated using an alternative measurement for delay-period modulation (the delay-period discharge rate when the target appeared inside the receptive field vs. the baseline discharge rate, Supplementary Fig. 4), the results were essentially the same.
In this study, to avoid the possibility that slow drifting activity could generate the quasi-significant BM indices, we analyzed only the baseline activity of those trials in which the mean discharge rate of baseline activity was stable across trials (see Material and Methods). Such a data-preprocessing step was not crucial to obtain the present data. Even when we performed the data analysis without this preprocessing, the results were essentially the same (Supplementary Fig 5).
Our description of the results thus far has relied on a particular bin width (100 ms) and a particular set of separation times (ranging from 100 to 700 ms, with a resolution of 100 ms) for computing the BM index. These values were not crucial for obtaining the present results. To demonstrate this, we conducted partial correlation analysis using various combinations of separation times (range: 10–790 ms, 10-ms resolution) and bin widths (range: 40–400 ms, 10-ms resolution). Figure 7A,B shows the Pearson's partial correlation coefficients between the DM and within-trial BM indices and those between the within-trial BM indices and the Fano factors. The correlation coefficient between the DM and within-trial BM indices increased when the separation time ranged from 200 to 400 ms (the strongest coefficient, 0.46, was obtained at a 310-ms separation time and a 100-ms bin width), whereas that between the within-trial BM indices and the Fano factors was strongest at separation times of 100 ms or less (the strongest coefficient, 0.75, was obtained at a 40-ms separation time and a 40-ms bin width), confirming the findings presented in Figure 6D. In addition, when the Fano factor was computed using a bin width other than 100 ms (25–800 ms), virtually the same results were obtained.
Our results showed a significant correlation between baseline-activity maintenance during fixation and delay-period modulation (Fig. 6D). However, it is possible that other factors commonly affected both baseline-activity maintenance and delay-period modulation and produced a spurious correlation between them. To assess this issue, we performed a series of control analyses to test whether other factors could explain the present results.
We have thus far focused on the degree of baseline-activity maintenance obtained from trials in the reaction-time visual-search task. Therefore, we tested whether the present results were still preserved when the BM index was evaluated based on the baseline-activity in the delayed-response task rather than the reaction-time task. Although the differences in the mean values of the within-trial BM indices were significant at all separation times except for 200 ms (paired t-test, P < 0.05), their differences were small and exhibited very similar dependencies on the separation time between the delayed-response and the reaction-time tasks (Supplementary Fig. 6A). Then we performed the same partial correlation analysis for the within-trial BM indices derived from the delayed-response task (Supplementary Fig. 6B). The DM index was significantly correlated with the within-trial BM index with longer separation times ranging from 400 to 500 ms, whereas the Fano factor was significantly correlated with the within-trial BM index with a separation time of 100 ms (Pearson's partial correlation, P < 0.05). Although we found that the separation time at which the correlation with the DM index was the strongest was longer in the delayed-response task (500 ms, Supplementary Fig. 6B) than that in the reaction-time task (300 ms, Fig. 6D), there is the possibility that this difference may be caused by the small number of cells and trials in the delayed-response task (55 cells and 163 trials on average) compared with those in the reaction-time task (93 cells and 365 trials on average). But, when the number of cells and trials were matched between the tasks, the distribution of correlation coefficients in the reaction-time task did not change substantially (compare Fig. 6D with Supplementary Fig. 6C), suggesting that the difference between tasks was not caused by the size of the data set. Nonetheless, the present results that the within-trial BM indices with longer and shorter separation times were separately correlated with the DM index and the Fano factor, respectively, were preserved even when the BM index was calculated from the baseline activity in the delayed-response task rather than the reaction-time task.
Next, we assessed the influences of motivation/behavioral performance. Previous studies have reported that neuronal activity during the fixation period varies predictably according to the expected trial state (success/error) and impending saccadic reaction time, reflecting the motivational level of a monkey (Kobayashi et al. 2002; Okada et al. 2009). Consequently, one may argue that the association between stronger baseline-activity maintenance (within-trial BM index) and stronger DM (DM index) in a given recording session simply reflects session-to-session differences in motivational level. To assess this possibility, we examined the associations of the delay-period modulation and within-trial BM indices with performance scores and saccadic reaction times. Because the average performance score and the mean saccadic reaction time across sessions differed significantly between the 2 monkeys (performance score: 73.5 ± 5.3% vs. 80.3 ± 6.1%, two-sample t-test, P < 0.0001; saccadic reaction time: 299.5 ± 18.0 ms vs. 203.5 ± 17.8 ms, two-sample t-test, P < 0.0001), these values were transformed into z-scores for each monkey. We found no significant correlation between the z-scored performance scores or saccadic reaction times and either the within-trial BM index or the DM index (Supplementary Table 1; Pearson's correlation, r = −0.13–0.15, P > 0.14). These data indicate that the present results cannot be explained simply by session-to-session differences in behavioral performance.
Next, we assessed the effects of eye movements. Although the monkeys were required to fixate on a small spot located at the center of the display throughout the fixation and/or delay period, the eye position window criteria allowed small eye movements, namely microsaccades, within the eye position window. Previous studies have found an association between such small movements and neuronal responses (Martinez-Conde et al. 2004; Gur and Snodderly 2006; Hafed et al. 2009; Hafed and Krauzlis 2010). Therefore, we assessed whether session-to-session variability in small eye movements could lead to a spurious correlation between baseline-activity maintenance and delay-period modulation. We counted the number of small eye movements (absolute eye velocity >20°/s, duration ≥10 ms) during the 800 ms prior to array presentation in the reaction-time task and during the 100–700 ms before the disappearance of the fixation point in the delayed-response task for each of the 94 neurons. In both cases, we observed no correlation between the frequency of small eye movements and the within-trial BM index or the DM index (Supplementary Table 1; Pearson's correlation, r = −0.13 to 0.08, P > 0.2), indicating that the present results could not be explained by small eye movements.
Next, we assessed the effects of anticipation. Baseline activity of LIP neurons can be modulated by anticipation of a behaviorally relevant sensory event and its effects often emerge as ramp-like (build-up) activity prior to the onset of a behaviorally relevant stimulus (Colby et al. 1996). In fact, we observed increasing activity prior to stimulus presentation in some of the neurons we recorded (e.g., Fig. 3E). Therefore, it is possible that anticipation strongly modulates baseline and delay-period activity in some neurons, but it only weakly modulates that in other neurons. In such a case, there would be a spurious correlation between baseline-activity maintenance and delay-period modulations across neurons. To test this possibility, we classified our samples into 2 groups depending on whether neurons exhibited ramp-like activity prior to the time of stimulus onset (Supplementary Fig. 7A). Of 94 neurons, 25 exhibited ramp-like activity (25 build-up-type neurons and no build-down-type neurons). Neither the within-trial BM index nor the DM index significantly differed between the ramping and nonramping neurons (Supplementary Fig. 7B,C). Furthermore, the results were essentially the same as those shown in Figure 6D even when the ramping neurons were excluded from the partial correlation analysis (Supplementary Fig. 7D). Thus, the present results could not be explained by the effects of anticipation.
Next, we assessed the influences of fixation-related activity. Previous studies have reported fixation neurons in the FEF (Bizzi 1968; Suzuki and Azuma 1977; Bruce and Goldberg 1985; Burman and Bruce 1997; Izawa et al. 2009) and the superior colliculus (Munoz and Wurtz 1993; Everling et al. 1998) that are characterized by persistently increasing activity during active fixation, even during a brief blink of a foveal stimulus, and the cessation of discharge during saccades. Similar fixation-related activity has been found in the LIP (Ben Hamed and Duhamel 2002). If some of our neurons had such fixation-related activity, it was possible that the values of the within-trial BM and DM indices differed between the fixation-related neurons and the others and this produced a spurious correlation between the BM and DM indices across neurons. Of 94 neurons, only 7 were identified as fixation-related neurons. When the partial correlation analysis was performed after excluding these fixation-related neurons, we still observed essentially the same result as those shown in Figure 6D (Supplementary Fig. 8), indicating that the present results could not be explained by the influences of fixation-related activity.
Finally, we considered the possible influences of rhythmic oscillations in baseline activity. A previous study have found that a portion of neurons in area 7a, an adjacent area of LIP, exhibit rhythmic oscillations in their activity even during fixation periods (Joelving et al. 2007). One might argue that if baseline activity rhythmically oscillates at a cycle period on the order of the bin size (100 ms) used for analysis, such oscillations would produce high BM indices. To assess this possibility, we performed a computer simulation in which oscillatory activity was hypothesized, and calculated the within-trial BM indices from this oscillatory activity in the same way as for the actual data (Supplementary Fig. 9). Importantly, the simulation results predict that if oscillatory activity is crucial, there would be not only positive BM indices but also negative BM indices depending on separation time (Supplementary Fig. 9D). However, this is inconsistent with our observation that no neuron exhibited the within-trial BM index which was significantly less than zero (Fig. 4A). This, it is unlikely that the currently observed high BM indices were produced by rhythmic oscillations in baseline activity.
Effects of Neuron Type
LIP neurons can be classified into subpopulations based on the existence of visually responsive and/or saccade-burst activity (see Materials and Methods). It is possible that the correlation between the degree of baseline activity and the strength of delay-period modulations varies depending on neuron type. To test this, we calculated the correlation strength separately for different types of neurons. Of the 94 LIP neurons, 57 (61%), 2 (2%), and 28 (36%) were classified as visual, movement, and visuomovement neurons, respectively. We found significant partial correlations between within-trial BM and DM indices for visuomovement neurons, factoring out Fano factors, at separation times ranging from 100 to 500 ms (Supplementary Fig. 10; Pearson's partial correlation, r = 0.42–0.60, P < 0.05). On the other hand, a significant correlation was observed for visual neurons only at a 300-ms separation (Supplementary Fig. 10; Pearson's partial correlation, r = 0.34, P < 0.05). These results indicate that although the absolute magnitude of correlation coefficients was largely different between the visual and visuomovement neurons, the basic relationship (the longer term and short-term maintenance of baseline activity complementarily correlated with delay-period modulation and trial-to-trial fluctuations) was preserved irrespective of neuron type.
Relationship to BM in Other Cortical Areas
We have previously demonstrated that the degree of baseline-activity maintenance for FEF neurons is significantly stronger than that for V4 neurons, suggesting a difference in the intrinsic temporal storage capability across cortical areas (Ogawa and Komatsu 2010). Although the experimental design and animals used in this study were different from previous studies, it was interesting to compare the mean within-trial BM indices of LIP neurons in the current study with those of FEF (or V4) neurons in our previous study (Supplementary Fig. 11). The results show that the mean within-trial BM indices for LIP neurons (n = 94) were significantly lower than those of FEF neurons (n = 33) at all separation times (two-sample t-test, P < 0.001), but these values were significantly higher than those of V4 neurons (n = 40) at all separation times except for 100 ms (two-sample t-test, P < 0.05). This result is not contrary to the previously suggested view that the ability to maintain the instantaneous discharge rate in baseline activity increases with the progression in cortical hierarchy (Ogawa and Komatsu 2010).
We found that the degree of baseline-activity maintenance during fixation varied from neuron to neuron in the LIP and was significantly correlated with the strength of response modulation signaling task-related information during the delay across neurons. In particular, a significant correlation was observed in the degree of long-term (200–500 ms) baseline-activity maintenance. In contrast, the degree of short-term (∼100 ms) baseline-activity maintenance was correlated with trial-to-trial fluctuations evaluated with Fano factors in baseline activity. Thus, the present results suggest that the capability to hold task-related information in delay-period activity is associated with the degree of baseline-activity maintenance in a timescale-dependent manner.
Baseline-Activity Maintenance During Fixation
In this study, we focused on baseline activity during the fixation period to assess the intrinsic ability of a neuron to persistently maintain its discharge rate. This allowed us to examine response properties of LIP neurons under the condition in which the effects of visual sensory stimuli and eye movements may be at a minimum. Although there is the possibility that baseline activity might have been influenced by visually evoked/saccade-related activity occurring before fixation and/or small eye movements during fixation, our control analyses ruled out these possibilities (Supplementary Fig. 2A and Table S1), consistent with our previous study (Ogawa and Komatsu 2010). In addition, although the differences in the within-trial BM indices were significant between the delayed-response and the reaction-time visual-search tasks, their differences were very small (Supplementary Fig. 6A), suggesting that the preparation to the ongoing task (e.g. Toth and Assad 2002; Stoet and Snyder 2004), if it exists, only weakly influenced on the baseline activity during the fixation period. Finally, behavioral performance had little influence on the degree of baseline-activity maintenance (Supplementary Table S1). Thus, the current results show that the degree of the discharge-rate persistency during the fixation period is very robust irrespective of changes in endogenous or exogenous factors, suggesting that baseline-activity maintenance that we measured in this study may reflect one of the intrinsic response properties of LIP neurons.
Relationship with Previous Studies of Spontaneous Activity
Previous studies using functional MRI techniques have shown that brain regions in which similar blood oxygen level-dependent signal modulations are observed during specific tasks or types of sensory stimulation exhibit correlated spontaneous fluctuations even in the absence of tasks or stimuli (Biswal et al. 1995; Greicius et al. 2003; Fox et al. 2005; Vincent et al. 2007; Northoff et al. 2010), indicating that spontaneous fluctuations in resting state activity manifest an intrinsic functional connectivity (inter-regional correlations in neuronal variability) of the brain. In contrast, the current study focused on single-neuron activity in a particular cortical region and revealed that cell-to-cell functional differences in task-evoked (delay-period) activity are significantly associated with that in baseline activity. Thus, the present results provide new evidence that intrinsic properties characterizing neuronal performance in task-evoked activity could be manifested in spontaneous (baseline) activity at the single-neuron level.
Previous optical imaging studies combined with single-unit recordings have shown that analogous spatial patterns in the cat visual cortex are associated with both stimulus-evoked and spontaneous activities (Arieli et al. 1996; Tsodyks et al. 1999; Kenet et al. 2003). More recently, large-scale multi-unit recording studies found similarly structured patterns of population activity in both stimulus-driven and spontaneous responses of the rat auditory cortex (Curto et al. 2009; Luczak et al. 2009). Furthermore, a single-unit recording study found that spike-time correlations in stimulus-evoked activity are generated by mechanisms common to those operating during spontaneous activity that occurs in the primary visual cortex of anesthetized primates (Jermakowicz et al. 2009). In those studies, the response patterns for stimulus-evoked and spontaneous activity were compared in the absence of any behavioral tasks. In contrast, the present study focused on task-evoked activity (delay-period activity) originating not from sensory stimuli per se but rather from such as memory, attention, and/or motor preparation processes occurring during a behavioral task. Thus, our findings emphasize that intrinsic response properties may reflect not only exogenous stimulus-driven processing but also endogenous processing at the level of single neurons.
Possible Mechanisms for the Short- and Long-Term Maintenance of Baseline Activity
The present results showed that trial-to-trial fluctuations evaluated with Fano factors in baseline activity were more strongly correlated with short-term (∼100 ms) baseline-activity maintenance, whereas task-evoked modulation of delay-period activity was more strongly correlated with long-term (200–500 ms) baseline-activity maintenance. What mechanisms could account for the 2 different time scales of the mechanisms underlying baseline-activity maintenance? It is widely assumed that persistent activity, such as delay-period activity in a memory-guided saccade task, is sustained by the reverberant activation of recurrent excitatory circuits (for reviews, see Amit 1995; Wang 2001). Moreover, model studies suggest that not only recurrent circuits but also a slow component of N-methyl-d-aspartate (NMDA) synaptic transmission, which has a decay-time constant of ∼100 ms (Hestrin et al. 1990; Spruston et al. 1995), play roles in the production of long persistent activity (Wang 1999, 2002).
We speculate that the observed short- and long-term discharge-rate maintenance in baseline activity may be accounted for by these mechanisms. If LIP neurons are equipped with slow-decay NMDA synaptic transmission, they are able to continue receiving excitatory synaptic inputs through NMDA receptors with a decay-time constant of ∼100 ms following an action potential; this may explain the potential ability of the neurons to retain their instantaneous discharge rate up to 100 ms. However, because NMDA receptors are broadly distributed over the cerebral cortex (Monaghan and Cotman 1985; Meoni et al. 1998; Scherzer et al. 1998), cell-to-cell variability in the degree of short-term (∼100 ms) baseline-activity maintenance may be characterized by the strength of the random fluctuations in input signals. However, if the LIP neurons are also involved in recurrent circuits, they could retain their discharge rate for hundreds of milliseconds or more (Wang 2001). This long persistence may produce not only long-term (200–500 ms) baseline-activity maintenance but also persistent modulation of delay-period activity.
A significant correlation between baseline-activity maintenance in the reaction-time task and delay-period modulation during the delayed-response task (Fig. 6) suggests that persistence in both activities may be produced by shared neural-circuit dynamics. Indeed, a previous in vitro study reported that recurrent synaptic reverberation in local neural circuits generates slowly varying spiking patterns in spontaneous activity (Shu et al. 2003) as well as persistent activity during a delay period (Wang 2001). Ganguli et al. (2008) have proposed a theoretical attractor network model that could explain the neuronal behaviors in the LIP during an attentional or perceptual decision-making task. In this model, when feed-forward sensory inputs to the network is absent, population dynamics are assumed to converge on a 1-dimensional space in which activity patterns are simply scaled versions of a single pattern of relative firing rates across neurons. This assumption led to the linear relationship between baseline and delay-period activities. They reported that the magnitude of baseline activity is correlated with that of delay-period activity across neurons by examining neurophysiological data recorded from the LIP. In addition to their results, our results demonstrated that temporal structures (discharge-rate persistency) in baseline activity are also correlated with the degree of response modulation of delay-period activity.
Possible Functional Properties Linked with Baseline-Activity Maintenance
The present study demonstrated significant covariance between baseline-activity maintenance and modulation of delay-period activity across LIP neurons. However, this does not mean that an LIP neuron with strong baseline-activity maintenance always exhibited strong modulation in its delay-period activity in the present experiment. Careful inspection of the results of the correlation analysis between the two (e.g., Fig. 6A, 300-ms separation time) revealed that although neurons with weaker baseline-activity maintenance (lower BM indices) tended to have only weaker delay-period modulation (lower DM indices), those with strong baseline-activity maintenance (higher BM indices) appeared to show a range of strengths in delay-period modulation (widely distributed DM indices), demonstrating that some neurons showed strong baseline-activity maintenance but only weak delay-period modulation during the delayed-response task.
What functional role do these neurons play in the LIP? One possibility is that these neurons play roles in memory-guided actions in which a body part (e.g., arm) other than the eyes is used as the effecter (Snyder et al. 1997; Quian Quiroga et al. 2006; Cui and Andersen 2007). Another possibility is that these neurons play roles in perceptual visual discrimination such as target–distractor discrimination and random dot motion–discrimination tasks (Shadlen and Newsome 2001; Roitman and Shadlen 2002; Ipata et al. 2006; Thomas and Paré 2007; Balan et al. 2008; Ogawa and Komatsu 2009), which require sensory signal integration over time for decision-making (Mazurek et al. 2003). A long integration time constant is generally essential for decision-making during various perceptual–discrimination tasks. As described above, this can be biophysically achieved using a combination of slow synaptic transmission (possibly dominated by NMDA receptors) and strong network recurrence (Wang 1999, 2002; Shen et al. 2010). Therefore, LIP neurons that had strong baseline-activity maintenance but exhibited only weak delay-period modulation in the present task potentially had the ability to play crucial roles in executing other cognitive tasks that were not assessed in the current experiments.
This work was supported by grants from the Ministry of Education, Culture, Sports, Science, and Technology of Japan (MEXT) to T.O. (20500283, 20033010, and 22120507).
The authors gratefully acknowledge the continuous support and encouragement of Prof. Kenji Kawano. They also acknowledge Nihon University and the National Bio-Resource Center for supplying the monkeys used in this study. They also thank the editor and anonymous reviewers for their valuable comments and constructive suggestions on the manuscript. Conflict of Interest: None declared.