Cognitive control requires the selection and maintenance of task-relevant stimulus–response associations, or rules. The dorsolateral prefrontal cortex (DLPFC) has been implicated by lesion, functional imaging, and neurophysiological studies to be involved in encoding rules, but the mechanisms by which it modulates other brain areas are poorly understood. Here, the functional relationship of the DLPFC with the superior colliculus (SC) was investigated by bilaterally deactivating the DLPFC while recording local field potentials (LFPs) in the SC in monkeys performing an interleaved pro- and antisaccade task. Event-related LFPs showed differences between pro- and antisaccades and responded prominently to stimulus presentation. LFP power after stimulus onset was higher for correct saccades than erroneous saccades. Deactivation of the DLPFC did not affect stimulus onset related LFP activity, but reduced high beta (20–30 Hz) and high gamma (60–150 Hz) power during the preparatory period for both pro- and antisaccades. Spike rate during the preparatory period was positively correlated with gamma power and this relationship was attenuated by DLPFC deactivation. These results suggest that top-down control of the SC by the DLPFC may be mediated by beta oscillations.
As our environment changes, certain behaviors become more appropriate than others. Our ability to flexibly engage in goal-directed behavior depends on cognitive control. A critical component of this control is the ability to select and maintain task-relevant stimulus–response associations, or rules. In particular, the dorsolateral prefrontal cortex (DLPFC) is thought to encode representations of rules and goals, and bias other brain areas to achieve the desired outcome (Miller and Cohen 2001).
The antisaccade task, where subjects suppress a saccade toward a peripheral stimulus (prosaccade) and generate a saccade in the opposite direction (Hallett 1978; Munoz and Everling 2004), is a useful paradigm for investigating the neural basis of cognitive control. Prosaccades and antisaccades are realized by distinct stimulus–response associations, and the DLPFC has been implicated in their performance. Patients with DLPFC lesions have been found to have longer reaction times and produce more errors for antisaccades (Guitton et al. 1985; Pierrot-Deseilligny et al. 1991, 2003; Ploner et al. 2005), and functional imaging studies in humans suggest DLPFC involvement in antisaccade preparation (Sweeney et al. 1996; Desouza et al. 2003; Ford et al. 2005; Brown et al. 2007). In particular, single-unit recordings in monkeys have found task-selective activity for prosaccades and antisaccades in DLPFC neurons (Funahashi et al. 1993; Everling and DeSouza 2005; Johnston and Everling 2006b; Johnston et al. 2007). The DLPFC also contains neurons that project directly to and send task-selective signals to the superior colliculus (SC) (Goldman and Nauta 1976; Leichnetz et al. 1981; Johnston and Everling 2006b), a critical component of the oculomotor system that is strongly modulated by the antisaccade task (Everling et al. 1999; Munoz and Everling 2004).
Deactivation of the DLPFC with cooling has been shown to affect single-neuron activity in the SC. First, unilateral DLPFC deactivation increases preparatory activity in the contralateral SC on prosaccade and antisaccade trials, and increases stimulus-related activity in the contralateral SC on antisaccade trials (Johnston et al. 2014). Second, bilateral deactivation decreases preparatory activity on prosaccade and antisaccade trials, and increases stimulus-related activity and decreases saccade-related activity on antisaccade trials (Koval et al. 2011). Both studies linked DLPFC-induced changes in SC activity to behavior and, together with other studies that manipulated DLPFC activity (Condy et al. 2007; Wegener et al. 2008), suggest that the DLPFC exerts an excitatory influence on the SC (for review see Everling and Johnston 2013). However, the mechanisms by which the DLPFC biases the SC and other brain areas remain unclear.
A complementary approach to examining spiking activity is to examine local field potentials (LFPs), which reflect the average synaptic activity of a group of neurons (Buzsaki et al. 2012). LFPs and neuronal oscillations have been implicated in neuronal communication and top-down control (Varela et al. 2001; Fries 2005; Siegel et al. 2012), and may help elucidate how the DLPFC communicates task-relevant signals to target areas. Here, we report the effects of bilateral DLPFC deactivation on LFP activity in the SC. LFP activity was collected together with the spiking activity previously reported by Koval et al. (2011).
Materials and Methods
All procedures were conducted in accordance with the Canadian Council on Animal Care Policy on the Use of Laboratory Animals, and a protocol approved by the Animal Use Subcommittee of the University of Western Ontario Council on Animal Care. Details of the surgical procedures, behavioral task, and reversible cryogenic deactivation have been previously described by Koval et al. (2011) and are summarized below.
Two adult male macaque monkeys (Macaca mulatta) weighing 11 and 16 kg were prepared for LFP and single-neuron recordings in the SC using previously described techniques (Johnston and Everling 2006b). Stainless steel cryoloops were implanted bilaterally into the posterior principal sulci (Fig. 1A) for reversible deactivations of parts of the DLPFC (posterior parts of area 46 and portions of 9/46d, 9/46v). The technical details of the cryoloop surgery and deactivation method have been previously described (Lomber et al. 1999).
During the experiment, monkeys performed a randomly interleaved prosaccade and antisaccade task (Fig. 1B). Each trial began with the presentation of a colored central fixation point. For one monkey, a green fixation point signaled a prosaccade trial and a red fixation point signaled an antisaccade trial. For the other monkey, the color instructions were reversed. Monkeys were required to fixate on the fixation point within a 0.5° × 0.5° window for 1000–1200 ms at the beginning of each trial. On half the trials, the color cue remained visible throughout the trial (rule visible/overlap task), while on the other half of trials, the color cue changed to yellow 500–700 ms before stimulus presentation (rule memorized task). Subsequently, a peripheral white visual stimulus (0.15°) was pseudorandomly presented with equal probability in either the response field (RF) of an isolated SC neuron or at the mirror location. Monkeys were required to generate a saccade toward the stimulus on prosaccade trials and away from the stimulus on antisaccade trials, and rewarded when saccade endpoints were within a 5° × 5° window.
Reversible Cryogenic Deactivation
To deactivate the posterior principal sulcus, methanol was chilled using an ice bath containing dry ice and passed through a cryoloop to deactivate adjacent cortical tissue (Fig. 1A). Evoked neural activity is absent when cortical tissue is cooled below 20°C (Adey 1974). Given that the effective spread of cooling is restricted to ∼2 mm on either side of the cryoloop (Payne and Lomber 1999) and that each cryoloop measured 6 × 3 × 2 mm, the volume of cortex deactivated can be approximated by the volume of a box with dimensions of 10 × 7 × 4 mm, or 280 mm3. Accordingly, cooling of the cryoloops affected the posterior half of the principal sulci, corresponding mainly to parts of area 46 and area 9/46 (Petrides and Pandya 1999).
Each experimental session began with a precool period where the pumps were turned off. The pumps were subsequently turned on to start the cool period. The first 4 min after the pumps were turned on were excluded from all data analysis to ensure that the cortical tissue adjacent to the cryoloops was cooled below 20°C and that neurons were deactivated. At the end of the cool period, the pumps were turned off. The first 3 min after the pumps were turned off were excluded from all data analysis to ensure that the cortical tissue adjacent to the cryoloops returned to normal body temperature.
The activity of saccade-related neurons and the accompanying LFP activity were recorded in the intermediate layers of the caudal SC (saccade amplitudes 5–12°) using standard electrophysiological techniques (Johnston and Everling 2006b). To be considered a saccade-related neuron, an isolated cell had to discharge over 100 spikes/s for prosaccades into its RF, 10 ms before to 10 ms after saccade onset. Neural activity was amplified, filtered, and stored by a Plexon multichannel acquisition processor (MAP) system using a headstage with unit gain (Plexon Inc., Dallas, TX, USA). The LFP was extracted with a passband filter (0.7–170 Hz), and further amplified and digitized at 1 kHz. We only included those LFP sites in our analysis where a saccade-related neuron was recorded at the same time. The powerline artefact was removed from 10 s long data segments using a discrete Fourier transform filter that has been previously described (Womelsdorf et al. 2006). Eye movements were recorded at 500 Hz with high-speed infrared video eye tracking (Eyelink II, Kanata, ON, Canada).
Custom Matlab (The MathWorks Inc., Natick, MA) code using the FieldTrip toolbox (Oostenveld et al. 2011) was used for data analysis. Only recording sessions and neurons that did not show any significant differences in neuronal firing in the 500 ms period before stimulus onset between precool and postcool trials (t-test, P > 0.05) were analyzed to ensure that stability of the recording did not change during the experimental session.
LFP activity related to the generation of correct and erroneous prosaccades and antisaccades was examined using data from the precool period to eliminate the potential for deactivation effects. To examine the effects of DLPFC deactivation on LFP activity, the precool and postcool data when the DLPFC was active (noncool) was compared with the cooling data when the DLPFC was deactivated (cool). This comparison included both correct and error trials.
To evaluate the event-related LFP, the LFP signal was zero-phase filtered using a third order, low-pass Butterworth filter with a cutoff frequency of 30 Hz. Only the LFP signal before saccade initiation was used to avoid ocular artifacts, and a 100 ms baseline window that started 600 ms before stimulus onset was subtracted from the LFP. The event-related LFPs of prosaccade and antisaccade trials were averaged separately, and aligned to stimulus and saccade onset.
To compare event-related LFPs to spiking activity, continuous spike density functions were constructed. The activation waveform was obtained by convolving each spike with an asymmetric function that resembled a postsynaptic potential (Hanes and Schall 1996; Thompson et al. 1996). The advantage of this function over a standard Gaussian function (Richmond and Optican 1987) is that a spike exerts an effect forward in time, but not backward.
Sliding receiver operating characteristic (ROC) analyses were conducted to highlight differences in event-related LFPs and spiking activity over time. The prosaccade and antisaccade task, and the noncool and cool conditions were compared. The ROC value was calculated for a 10 ms window (centered around the time point) starting 200 ms before stimulus onset and repeated in 1 ms increments up to 150 ms after stimulus onset. Bootstrap analyses were used to test the significance of the ROC values. The following procedure was repeated 10 000 times: for each recording site or neuron, the 2 active conditions (prosaccade and antisaccade, or noncool and cool) were randomly exchanged or unchanged with equal probability (50%), and a single average ROC timecourse was calculated. The 97.5th and 2.5th percentile values of the distribution of 10 000 average ROC values at each time point were used to indicate the 5% significance criterion.
Fourier transformations in 334 ms time windows calculated every 50 ms with Slepian sequences as tapers and 4.5 Hz frequency smoothing were used to calculate LFP power from 3 to 60 Hz. LFP power from 60 to 150 Hz was calculated with 24 Hz frequency smoothing. Analyses were conducted on periods of −800 to 200 ms from stimulus onset and −300 to 800 ms from fixation cue onset. For stimulus onset analyses, LFP power was normalized by subtracting the mean power in a 200 ms baseline window that started 1000 ms before stimulus onset and dividing by the standard deviation. For fixation cue onset analyses, LFP power was normalized by subtracting the mean power in a 200 ms baseline window that started 500 ms before fixation cue onset and dividing by the standard deviation.
To test whether power differed significantly between 2 task conditions, cluster-based nonparametric permutation tests were conducted (Maris and Oostenveld 2007). T-values were calculated for each time–frequency sample as the test statistic. T-values above a threshold of alpha = 0.05 were then clustered based on temporal and spectral adjacency and summed for each cluster. The significance of each cluster was determined using a permutation test with 10 000 repetitions. To determine whether LFP power could predict the outcome (correct or error) of a trial, ROC analyses were conducted. ROC values were calculated for each time point using mean power in the theta (5–8 Hz), alpha (8–13 Hz), low beta (13–20 Hz), high beta (20–30 Hz), low gamma (30–60 Hz) and high gamma (60–150 Hz) frequency bands. The significance of the ROC values was tested using the same method used for event-related LFPs and spiking activity.
To determine whether there was a relationship between preparatory LFP power and preparatory spiking activity, a period of −800 to −200 ms from stimulus onset was examined. This time period was chosen because it maximized the length of time in the preparatory period while excluding activity from the poststimulus period. For each neuron, the mean power in the theta (5–8 Hz), alpha (8–13 Hz), low beta (13–20 Hz), high beta (20–30 Hz), low gamma (30–60 Hz), and high gamma (60–150 Hz) frequency bands was computed and compared with the average spike rate in the same −800 to −200 ms window. Mean power at each frequency band was correlated with spike rate for each neuron using the Spearman correlation coefficient.
To determine whether there was a relationship between LFP power and saccadic reaction time (SRT), a period of −50 to 50 ms from stimulus onset was examined. This period was the same as the prestimulus period previously used for single-neuron activity (Koval et al. 2011) and can detect oscillations in the high beta, low gamma, and high gamma frequency bands. Lower frequency bands were not examined due to the lack of sensitivity of this time period to detect low frequency oscillations. For each trial, the mean power in the high beta, low gamma and high gamma frequency bands was computed for the prestimulus period. Trials and the associated SRTs were divided into 5 equal sized bins based on the magnitude of mean power. A one-way analysis of variance was used to test whether SRTs were significantly different between the power bins (Haegens et al. 2011). This analysis was also conducted for preparatory LFP power using the period −800 to −200 ms from stimulus onset.
Over a total of 52 experimental sessions, LFPs from 26 LFP sites and 35 SC neurons were included in the analyses (15 LFP sites and 20 neurons from the first monkey, and 11 LFP sites and 15 neurons from the second monkey). The effects of bilateral DLPFC deactivation on error rates and reaction times have previously been reported in detail by Koval et al. (2011). Briefly, DLPFC deactivation increased error rates on antisaccade trials, and increased SRTs on antisaccade and prosaccade trials (Table 1). In this study, we present LFP activity and spiking activity related to the LFP.
|Rule visible||Rule memorized||Rule visible||Rule memorized|
|Rule visible||Rule memorized||Rule visible||Rule memorized|
Event-related LFPs Respond to Stimulus Presentation
Figure 2 shows event-related LFPs aligned to stimulus onset for the rule visible task. On trials where the stimulus was presented into the RF of LFP sites, prosaccade trials showed a significantly greater stimulus-related response than antisaccade trials, starting 58 ms after stimulus onset (Fig. 2A, Supplementary Fig. 1A, red line). In contrast, on trials where the stimulus was presented opposite to the RF, antisaccade trials showed a significantly greater response than prosaccade trials, starting 92 ms after stimulus onset (Fig. 2B, Supplementary Fig. 1B, red line). There were no significant differences between prosaccade and antisaccade trials before stimulus onset. There were no significant differences between DLPFC cooling (blue lines) and noncooling trials (red lines).
Similar to the event-related LFPs, prosaccade trials showed significantly greater stimulus-related spiking activity than antisaccade trials when the stimulus was presented into the RF (Fig. 2C). Unlike the event-related LFPs, this difference was significant as early as 165 ms before stimulus onset (Supplementary Fig. 1C, red line). For trials where the stimulus was presented opposite to the RF, antisaccade trials showed a significantly greater response than prosaccade trials, starting 118 ms after stimulus onset (Fig. 2D, Supplementary Fig. 1D, red line). Koval et al. (2011) found that DLPFC deactivation significantly decreased prestimulus activity for both prosaccades and antisaccades. While DLPFC deactivation did not affect activity after stimulus onset on prosaccade trials, neurons remained active longer on antisaccade trials (Koval et al. 2011). Aligning event-related LFPs to saccade onset demonstrated that the event-related LFPs were dominated by stimulus presentation. The LFP decreased prior to saccade onset for both prosaccade and antisaccade trials regardless of stimulus location, whereas spiking activity showed a prominent saccade-related motor burst (Supplementary Fig. 2).
LFP Power and Task Performance
The rule memorized task required that the task instruction be held briefly in working memory and had a greater cognitive demand than the rule visible task, where the task instruction was visible throughout the task. Here, we examined whether or not this increased cognitive demand is reflected in SC LFP activity. LFP power before cooling was not significantly different between the rule visible task and rule memorized task, for prosaccades and antisaccades into and opposite to the RF of LFP sites (P > 0.05). In addition, power was not significantly different between prosaccades and antisaccades or for saccade direction, for both the rule visible and rule memorized tasks (P > 0.05).
Figure 3 shows the average evolution of LFP power for correct and error trials in the prosaccade and antisaccade conditions during the precool period in the rule memorized task. The evolution of LFP power for high gamma (60–150 Hz) is shown in Supplementary Figure 1. LFP activity was prominent after stimulus onset for correct prosaccades and antisaccades into and opposite to the RF of LFP sites, in the theta (5–8 Hz), alpha (8–13 Hz), low beta (13–20 Hz), high beta (20–30 Hz), low gamma (30–60 Hz), and high gamma (60–150 Hz) frequency bands. This activity was diminished for erroneous prosaccades and antisaccades. Contrasts of these time–frequency plots were performed to examine LFP power related to correct and erroneous prosaccade and antisaccade generation.
Overall, correct saccades were associated with higher power at the time of stimulus onset than erroneous saccades. Specifically, correct prosaccades into the RF of LFP sites had significantly higher power in the theta, alpha, low beta, high beta, and low gamma frequency bands around the time of stimulus onset (P < 0.05) (Fig. 3A). Correct prosaccades opposite to the RF had significantly higher power in the low beta and high beta frequency bands after stimulus onset (Fig. 3B), and in the high-gamma frequency band around the time of stimulus onset (Supplementary Fig. 3) (P < 0.05). Correct antisaccades into the RF had significantly higher power in the theta, alpha, and low beta frequency bands after stimulus onset (P < 0.05) (Fig. 3C). Correct antisaccades opposite to the RF had significantly higher power in the alpha and low beta frequency bands from 350 to 150 ms before stimulus onset (Fig. 3D), and in the high gamma frequency band around the time of stimulus onset (Supplementary Fig. 3) (P < 0.05). These differences between correct and erroneous prosaccades and antisaccades were not observed during the cooling period in the lower frequency bands (P > 0.05) (Supplementary Fig. 4). ROC analyses showed that the outcome of a trial can be predicted by LFP power in the same frequency bands and time periods that showed significant differences based on cluster-based nonparametric permutation tests (Supplementary Fig. 5).
DLPFC Deactivation Reduces Beta and Gamma Power
Figure 4 shows the average evolution of LFP power aligned to stimulus onset for pro- and antisaccade trials during the noncool and cool periods in the rule visible and rule memorized tasks. The evolution of LFP power for high gamma (60–150 Hz) is shown in Figure 5. LFP activity was prominent after stimulus onset for prosaccades and antisaccades during the noncool and cool periods, in the theta, alpha, low beta, high beta, low gamma, and high gamma frequency bands. To examine LFP power related to DLPFC deactivation, LFP power during the noncool period was contrasted with LFP power during the cool period. DLPFC activity may be particularly important for maintaining task rules in working memory in the rule memorized task. Contrasts using only trials during the precool period were very similar to contrasts using only trials during the postcool period (Supplementary Fig. 6); therefore, trials during the precool and postcool periods were combined (noncool). Similarly, contrasts using only trials into the RF were very similar to contrasts using only trials opposite to the RF, and trials into and opposite to the RF were combined.
Prosaccade and antisaccade trials showed differences before stimulus onset (Figs 4 and 5, bottom). In particular, prosaccades and antisaccades during the cool period in both the rule visible and rule memorized tasks had significantly lower power in the high beta frequency band starting from 800 ms before stimulus onset (P < 0.05). Power was also lower in the high gamma frequency band starting from 800 ms before stimulus onset (P < 0.05). There were no significant differences in the rule visible and rule memorized tasks between the noncool and cool periods after stimulus onset in any frequency band (P > 0.05). Cooling affected LFP power in the rule visible task and rule memorized task similarly.
To determine whether the decrease in high beta power was dependent or independent of task preparation, LFP power was aligned to the onset of the colored fixation point which conveyed the task rule for the upcoming trial. Figure 6 shows the average evolution of LFP power aligned to fixation cue onset for prosaccade and antisaccade trials during the noncool and cool periods in the rule visible and rule memorized tasks. LFP activity was prominent after cue onset for prosaccades and antisaccades during the noncool period, in the theta, alpha, low beta, high beta, and low gamma frequency bands. Activity in the low beta and high beta frequency bands was diminished for prosaccades and antisaccades with DLPFC deactivation. As with stimulus onset, LFP power related to DLPFC deactivation was examined by contrasting LFP power during the noncool period with LFP power during the cool period, with trials into and opposite to the RF combined.
Prosaccade and antisaccade trials during the cool period in the rule visible and rule memorized tasks showed significantly lower power in the low beta and high beta frequency bands in the 300 ms period after fixation cue onset (P < 0.05) (Fig. 6, bottom). This coincided with the prominent LFP activity seen in Figure 6 (top). For prosaccades in the rule memorized task and antisaccades in the rule visible and rule memorized task, the significant decrease in beta power persisted beyond 300 ms after fixation onset (P < 0.05). The contrasts also show that high beta power was significantly decreased during the cool period already before fixation cue onset for pro- and antisaccades. This finding indicates that high beta power was decreased during DLPFC cooling before the task instruction was presented.
DLPFC Deactivation Reduces Correlations Between Spiking Activity and Gamma Power
LFP activity may reflect neuronal processes that influence the spike rate of neurons. Figure 7 shows the proportion of SC neurons with spike rates that were significantly positively or negatively correlated with LFP power during the preparatory period (800–200 ms before stimulus onset, see Materials and methods). During the noncool and cool periods, more neurons were significantly positively correlated than negatively correlated with spike rate at the low gamma and high gamma frequency bands (χ²(1) > 4.3 and P < 0.05 for all comparisons). DLPFC deactivation significantly decreased the proportion of positively correlated neurons at these frequency bands (χ²(1) > 4.2 and P < 0.05 for all comparisons). Significant differences between proportions of neurons with positive correlations and negative correlations were not observed for the theta, alpha, low beta, and high beta frequency bands. Overall, 77.1% of the neurons recorded (27/35) had spike rates that were significantly correlated with LFP power.
LFP Power and SRT
Previous studies (Dorris et al. 1997; Everling et al. 1999) have examined the relationship between single-neuron activity and SRT for pro- and antisaccades, but not the relationship between LFP power and SRTs. Supplementary Figure 7 shows the relationship between LFP power −50 to 50 ms from stimulus onset and SRTs. For high beta, SRTs decreased as power increased for prosaccades into the RF (P < 0.05) and opposite to the RF (P < 0.01) (Supplementary Fig. 7A). For low gamma, SRTs decreased as power increased for prosaccades opposite to the RF (P < 0.01) (Supplementary Fig. 7B). For high gamma, SRTs decreased as power increased for prosaccades and antisaccades into the RF and opposite to the RF (P < 0.01) (Supplementary Fig. 7C). During the cool period, only SRTs for antisaccades opposite to the RF decreased as high gamma power increased (P < 0.05). While increases in power at short SRTs may be explained by the infringement of saccade-evoked potentials −50 to 50 ms from stimulus onset, this is unlikely as significant relationships between power and SRT were found for saccades opposite to the RF. During the preparatory period, SRTs for prosaccades into the RF decreased as high beta power increased and SRTs for prosaccades opposite to the RF decreased as high gamma power increased (P < 0.01).
The DLPFC has been implicated to be involved in encoding rules and biasing other brain areas to execute the appropriate task (Miller and Cohen 2001). While the DLPFC has been shown to send task-selective signals to the SC (Johnston and Everling 2006b) and modulate SC neural activity (Koval et al. 2011; Johnston et al. 2014), how it does so is poorly understood. In this study, we characterized LFP activity in the SC during an interleaved pro- and antisaccade task, and investigated the effects of bilateral DLPFC deactivation on LFP activity. We found that event-related LFPs showed stimulus-related differences between prosaccades and antisaccades, and LFP power distinguished between correct and erroneous saccades. While event-related LFPs were not affected by DLPFC deactivation, DLPFC deactivation reduced beta and high gamma power during the preparatory period. In addition, the positive correlation between gamma power and spike rate during the preparatory period was attenuated by DLPFC deactivation. Overall, these results demonstrate that neuronal oscillations may mediate communication between the DLPFC and SC.
In contrast to spiking single-neuron activity associated with saccade generation (Wurtz and Goldberg 1972; Sparks et al. 1976; Munoz and Wurtz 1995), almost nothing is known about LFPs in the SC. Whereas spiking activity represents the output of a single neuron, LFP activity reflects the synchronized synaptic activity of a group of neurons (Buzsaki et al. 2012). LFP activity consequently provides different and complementary information about neuronal processing. Accordingly, we found that although event-related LFPs like single-neuron activity showed a greater stimulus-related response for prosaccades compared with antisaccades (Everling et al. 1999), event-related LFPs were dominated by the stimulus-related response. This finding indicates that event-related LFPs in the SC primarily reflect the incoming visual signal, rather than the motor signal for the saccade.
While LFP power did not differentiate between pro- and antisaccades, higher power was observed around stimulus onset for correct compared with error trials. Correct prosaccades into the RF had greater power in the theta, alpha, low beta, high beta, and low gamma frequency bands. This is consistent with a report of LFPs in the human SC while subjects fixated on a central fixation point or made horizontal saccades (Liu et al. 2009). Although Liu et al. (2009) did not use a task that produced errors, they found that theta and alpha power increased during saccade generation. Beta activity has been implicated in processing for motor control (Engel and Fries 2010) and low gamma activity may represent local neuronal processing (Kopell et al. 2000; Fries 2005). Taken together, LFP activity may facilitate the processing of incoming information to the SC. In particular, neuronal oscillations may enable fixation-related and saccade-related neurons throughout the SC to receive the appropriate temporal and spatial information required for saccade generation. This coordination of information may be critical for determining which neurons increase spiking output, how much spiking is generated, and when changes in spiking occur. Thus, LFP activity may reflect the modulation of single-neuron activity to correctly generate a saccade toward a visual stimulus.
Compared with prosaccades, antisaccades emphasize greater cognitive control. They require the suppression of a prepotent saccade toward a stimulus and the generation of a voluntary saccade in the opposite direction (Everling and Fischer 1998; Munoz and Everling 2004). Here, higher power for correct compared with erroneous antisaccades was limited to the lower frequency bands: theta, alpha, and low beta for antisaccades into the RF, and alpha and low beta for antisaccades opposite to the RF. Indeed, alpha activity has been proposed to be involved in inhibition (Klimesch et al. 2007; Jensen and Mazaheri 2010; Haegens et al. 2011), whereas beta activity may be involved in motor set maintenance (Engel and Fries 2010). In particular, Swann et al. (2009) found that in a stop-signal task alpha and beta power decreased in motor cortex for both successful and unsuccessful stop trials, but the effect was less pronounced for successful trials. This relative increase in power was proposed to be related to increased GABA inhibition. Thus, the power increases observed for correct antisaccades could be related to inhibitory processes in the SC. Given the role of alpha activity in inhibition and beta activity in motor set maintenance, increased power may also reflect saccade generation under a general state of increased cognitive control.
The DLPFC is implicated in the cognitive control of prosaccades and antisaccades and has been shown to send task-selective signals that bias neural activity in the SC (Johnston and Everling 2006a,b; Koval et al. 2011; Johnston et al. 2014). While bilateral DLPFC deactivation affected poststimulus spiking activity in the SC (Koval et al. 2011), effects were not observed for LFP activity, thereby highlighting further differences between spiking and LFP activity. Nonetheless, deactivation decreased both spiking and beta activity during the preparatory period for prosaccades and antisaccades. This coincided with increased reaction times for both tasks and supports our recent proposal that the DLPFC does not inhibit but excite the SC (for review see Everling and Johnston 2013). The LFP results are consistent with studies that suggest a role for the DLPFC in task preparation (Desouza et al. 2003; Ford et al. 2005; Brown et al. 2007), and studies by Condy et al. (2007); Wegener et al. (2008), and Johnston et al. (2014) that manipulated the DLPFC unilaterally using pharmacological deactivation, microstimulation, and cryogenic deactivation, respectively. Taken together, these findings indicate that DLPFC deactivation impairs the ability to establish and maintain the appropriate task rule.
In the DLPFC, task rules are represented in the spiking activity of single neurons (White and Wise 1999; Asaad et al. 2000; Wallis et al. 2001; Everling and DeSouza 2005), the activity of neural populations (Stokes et al. 2013), and LFP activity (Buschman et al. 2012). In particular, Buschman et al. (2012) showed that task-specific neural ensembles were formed by increases in synchrony in the high beta frequency band. The role of beta activity in task representation is supported by findings that prefrontal beta oscillations are involved in working memory (Siegel et al. 2009; Salazar et al. 2012; Spitzer et al. 2014) and the maintenance of cognitive sets (Engel and Fries 2010). Here we found that beta power was already reduced before the onset of the fixation stimulus, which conveyed the specific task instruction. Therefore, DLPFC deactivation seems to reduce high beta power in the SC independent of a specific task preparation.
Beta activity is increased in top–down control (Buschman and Miller 2007; Siegel et al. 2012) and is thought to mediate communication between distant brain areas through synchrony (Kopell et al. 2000; Fries 2005). While synchrony between the DLPFC and SC was not investigated in this study, cortical-subcortical coupling in the beta frequency band has been shown for the subthalamic nucleus (Lalo et al. 2008). Thus, persistent communication between the DLPFC and the SC may be mediated by beta oscillations.
Changes in neuronal oscillations may reflect processes that influence neural activity. Accordingly, LFP power has been correlated with neuronal spiking (Pesaran et al. 2002; Rasch et al. 2008; Manning et al. 2009). While spiking was correlated with low gamma power during the preparatory period, differences in power in this frequency band were not observed. Low gamma oscillations may represent local neuronal processing (Kopell et al. 2000; Fries 2005) in the SC, regardless of the task being performed or the input received from the DLPFC. On the other hand, spiking was correlated with high gamma power, and DLPFC deactivation decreased preparatory period power in this frequency band. Changes in high gamma power are thought to reflect changes in spiking activity (Ray et al. 2008). The decrease in prestimulus SC spiking activity for pro- and antisaccades and the decrease in the proportion of neurons correlated with high gamma power with DLPFC deactivation are consistent with this idea. In this study, only the spiking activity of saccade-related neurons was examined. Thus, it is possible that the spiking activity of other types of neurons contributed to the changes observed for high gamma power. Overall, given that DLPFC deactivation also decreased high beta power, SC neural activity may be modulated by DLPFC-associated beta activity.
Although the DLPFC sends direct projections to the SC (Goldman and Nauta 1976; Leichnetz et al. 1981), SC LFP activity may also be influenced indirectly. DLPFC deactivation has been shown to affect activity in the thalamus (Alexander and Fuster 1973). Based on its wide connectivity with other brain areas, the thalamus may actively regulate neural synchrony, particularly using low frequency oscillations (Saalmann and Kastner 2011). DLPFC deactivation may also affect the frontal eye field (FEF), supplementary eye field, and basal ganglia, which send task-related signals to the SC (Schlag-Rey et al. 1997; Everling and Munoz 2000; Watanabe and Munoz 2009). For example, the FEF, shows comparable spiking activity to the SC for prosaccades and antisaccades (Everling and Munoz 2000), and likely receives excitatory input from the DLPFC, given that the majority of corticocortical connections are between excitatory neurons (Bunce and Barbas 2011). Consequently, the FEF may also show a decrease in beta power with DLPFC deactivation. Overall, the changes observed in SC LFP activity may be due to LFP changes in a broader neural network.
In summary, our results suggest a mechanism by which the DLPFC exerts control on the SC. Decreased beta power during bilateral DLPFC deactivation suggest beta oscillations may play a critical role in mediating communication between the DLPFC and the SC.
Supplementary material can be found at: http://www.cercor.oxfordjournals.org.
This work was supported by grants from the Canadian Institutes of Health Research (CIHR) to S.E. and S.G.L., and a CIHR MD/PhD Studentship to J.L.C.
We are grateful to J. Gati for help in MR imaging prior to the surgery. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. Conflict of Interest: None declared.