Failures in monitoring of self-generated actions are thought to underlie the positive symptoms in schizophrenia. It has been hypothesized that these deficits may be caused by a dysfunction of N-methyl-d-aspartate receptors (NMDARs). Here we recorded the activity of prefrontal neurons in monkeys performing an antisaccade task, while we administered a subanesthetic dose of the noncompetitive NMDAR antagonist ketamine. Many neurons discriminated between correct antisaccades and response errors in their postresponse activity. Ketamine increased the activity for the neurons' nonpreferred response, thereby decreasing the neurons' performance selectivity. Ketamine also affected the monkeys' behavior after an error, consistent with a deficit in error detection. The results show that NMDARs play an important role in action monitoring in primates. The decrease in performance selectivity of prefrontal neurons after ketamine can help to explain the deficits in action monitoring found in humans after ketamine administration and provides support for the hypothesis that an NMDAR dysfunction underlies self-monitoring deficits and psychotic symptoms in schizophrenia.
Everyday life requires the ability to detect when one's actions are erroneous and to correct them if necessary. Electrophysiological recordings in nonhuman primates have shown that both lateral prefrontal cortex (LPFC) and ACC neurons are active when such an action monitoring is required (Amiez, Joseph, & Procyk, 2005; Ito, Stuphorn, Brown, & Schall, 2003; Niki & Watanabe, 1979). According to one prominent theory, ACC detects the presence of conflict and monitors the outcome of actions and then provides a signal to areas like the LPFC, which then adjust the level of cognitive control for future actions (Carter & van Veen, 2007; Yeung, Botvinick, & Cohen, 2004; Botvinick, Braver, Barch, Carter, & Cohen, 2001; Miller & Cohen, 2001). The strongest support for a role of ACC in action monitoring comes from studies of the error-related negativity (ERN) in human participants (Yeung et al., 2004), an ERP component that occurs after errors in RT tasks. However, the LPFC clearly interacts with ACC in action monitoring as LPFC lesions abolish differences between error trial ERN and correct trial activity and affect corrective behavior after a response error (Gehring & Knight, 2000).
Blunted brain responses to errors are characteristic for schizophrenic patients as indicated by reduced error-related scalp potentials and altered activation in the LPFC and ACC on response error trials in patients (Mathalon, Jorgensen, Roach, & Ford, 2009; Polli et al., 2005, 2008; Kopp & Rist, 1999; Gehring, Coles, Meyer, & Donchin, 1995). In fact, it has been proposed that failures in self-generated action monitoring and internal monitoring of speech output contribute to delusions of alien control (Frith & Done, 1989) and formal thought disorder (Feinberg & Guazzelli, 1999), respectively. Deficits in action and internal speech monitoring (Stone et al., 2011), together with other positive, negative, and cognitive symptoms typical for schizophrenia, also occur following acute subanesthetic doses of ketamine (Gunduz-Bruce, 2009; Domino, Chodoff, & Corssen, 1965; Luby, Cohen, Rosenbaum, Gottlieb, & Kelley, 1959), a noncompetitive N-methyl-d-aspartate receptor (NMDAR) antagonist. Therefore, it has been suggested that NMDAR dysfunction may underlie self-monitoring deficits and psychotic symptoms in schizophrenia (Stone et al., 2011) and that ketamine is a promising pharmacologically induced model of schizophrenia in nonhuman primates (Blackman, MacDonald, & Chafee, 2013; Gil-da-Costa, Stoner, Fung, & Albright, 2013; Skoblenick & Everling, 2012; Condy, Wattiez, Rivaud-Pechoux, & Gaymard, 2005).
A link between pFC function and the action of ketamine has been found in rats (Jackson, Homayoun, & Moghaddam, 2004) and nonhuman primates (Skoblenick & Everling, 2012), where low systemic doses of NMDAR antagonists potentiate the firing of most neurons. We have shown recently that LPFC neurons also lose their selectivity for the task rule following administration of a single subanesthetic dose of ketamine (Skoblenick & Everling, 2012), consistent with the hypothesis that NMDAR antagonists disrupt frontal lobe function by decreasing the signal-to-noise ratio of LPFC neurons (Jackson et al., 2004).
Here we examined the effects of ketamine on postresponse activity and intertrial activity in the LPFC by recording single neuron activity in macaque LPFC neurons before and after the injection of a subanesthetic dose of ketamine, while the animals performed randomly interleaved trials of prosaccades and antisaccades (Munoz & Everling, 2004). Prosaccade trials required the animals to simply look toward a peripheral stimulus, whereas antisaccade trials required the animals to suppress a saccade toward the stimulus and instead to look away from it to the opposite direction. Antisaccade errors, which are elevated in patients with prefrontal lesions (Ploner, Gaymard, Rivaud-Pechoux, & Pierrot-Deseilligny, 2005; Pierrot-Deseilligny, Rivaud, Gaymard, & Agid, 1991; Guitton, Buchtel, & Douglas, 1985), schizophrenia patients (Sereno & Holzman, 1995; Fukushima et al., 1988), and nonhuman primates following ketamine administration (Skoblenick & Everling, 2012; Condy et al., 2005), elicit robust error-related scalp potentials in healthy human participants (Nieuwenhuis, Ridderinkhof, Blom, Band, & Kok, 2001). Analysis of error-related activity was restricted to antisaccade trials because nonhuman primates (Bell, Everling, & Munoz, 2000), like human participants (Dafoe, Armstrong, & Munoz, 2007), produce few prosaccade errors in this paradigm.
We collected data in two male adult macaque monkeys (Macaca mulatta) following guidelines of the Canadian Council of Animal Care policy and a protocol approved by the Animal Use Subcommittee of the University of Western Ontario Council on Animal Care. All experimental procedures have been described in detail previously (Skoblenick & Everling, 2012).
Monkeys were seated in a primate chair within a sound-attenuating chamber with their heads restrained and a juice spout placed at their mouth for computer-controlled reward delivery. Stimuli were presented on a 21-in. CRT monitor 42 cm in front of the animals. Behavioral control and stimulus generation were accomplished with the CORTEX and MonkeyLogic software package, and eye movements were recorded at 120 Hz with an infrared video eye tracker (ISCAN, Boston, MA). Each trial started with the presentation of a white fixation point (FP), which changed color to either red or green in a pseudorandom order after an initial 100-msec fixation period. In Monkey O, a red FP instructed an antisaccade and a green FP a prosaccade in response to the ensuing peripheral stimulus. The mapping was reversed for Monkey T. In half of the trials, the initial central FP changed back to white after a 200-msec period. This manipulation was included for a previous study in which we investigated ketamine effects on working memory for the task rule (Skoblenick & Everling, 2012). A peripheral stimulus appeared 8° to either the left or right following a 1000- to 1200-msec fixation period. To obtain a juice reward, the animal was required to look to the correct target location within a target window, that is, toward the stimulus on prosaccade trials and diametrically opposite of the peripheral stimulus on antisaccade trials. The reward was delivered as soon as the monkey looked into target window (8° circular window in MonkeyLogic and 8° × 8° in cortex). Error trials were defined as those in which the monkey performed the incorrect saccade: a prosaccade during a trial with the antisaccade rule or vice versa.
Extracellular recordings were conducted with a semichronic screw microdrive system (Neuronitek, London, Ontario, Canada) equipped with 32 individually moveable tungsten microelectrodes (FHC Inst., Bowdoinham, ME), guided by anatomical MR images to ensure targeting the area in and around the posterior principal sulcus. Initially, all individual electrodes were lowered through both a thin silicone membrane and the monkey's dura until multiple neurons were well isolated on a maximal number of electrodes. On each subsequent recording day, the 32-channel headstage (Plexon Inc., Dallas, TX) was reconnected, and electrodes were moved to isolate new neurons for that day's experimental session. To ensure a relatively unbiased sampling of LPFC neural activity, we did not prescreen neurons for task-related responses. The microdrive system remained implanted for up to 2 weeks before it was removed for cleaning and maintenance.
Data acquisition and filtering were performed with a multiacquisition processor system (Plexon Inc., Dallas, TX). Recorded neurons were sorted offline using 2-D and 3-D PCA using Offline Sorter (Plexon Inc., Dallas, TX). Horizontal and vertical eye positions and the occurrence of behavioral events (e.g., start of trial, onset of fixation, stimulus presentation) were also stored in the Plexon multiacquisition processor system.
Each experimental session began with a 10-min (Monkey T) or 15-min (Monkey O) period of baseline activity during which the animal performed blocks of trials in a pseudorandomized order. After the baseline period, the experiment was briefly paused, and the animal received either a single subanesthetic intramuscular injection of 0.4 ml ketamine (0.4 mg/kg, diluted in sterile saline), which previously has been found to elicit cognitive deficits with minimal anesthetic effects in rhesus monkeys (Blackman et al., 2013; Shen, Kalwarowsky, Clarence, Brunamonti, & Pare, 2010; Stoet & Snyder, 2006; Condy et al., 2005), or 0.4 ml of sterile saline into their right tricep brachii muscle. The injection process only interrupted the animal's experimental session for less than 15 sec after which the monkeys would continue the paradigm. We recorded neural activity for at least 40 min following ketamine or saline injection throughout these sessions. Ketamine injection sessions were separated by at least 4 days to avoid cumulative dosing effects and potential neuronal injury (Liao et al., 2011). Saline injections were performed on some of the days on which no ketamine was administered.
Data analysis was performed using custom-designed software programmed in Matlab (Mathworks, Natick, MA). Saccade onset of the initial saccade was defined as the time at which the horizontal velocity exceeded 30°/sec after peripheral stimulus onset, and the end of the saccade was defined as the time when the velocity fell below 30°/sec. The onset of the return or correction saccade was defined as the time when the horizontal velocity exceeded 30°/sec after the end of the initial saccade. An antisaccade error was defined as an initial saccade toward the peripheral stimulus. Fixation breaks or no responses were not included in error rate calculations. All trials were visually inspected after automatic saccade detection and corrected if necessary.
Effect of Previous Trial Performance
Several studies have reported posterror slowing of antisaccades (Klein et al., 2007; Polli et al., 2006). For this analysis, we only included correct antisaccade trials that were preceded by a correct response (pro- or antisaccade) or an error (pro- or antisaccade) on the previous trial. Correct antisaccade trials that were preceded by a broken fixation or no response trial were excluded from this analysis. The low number of correct antisaccades that were preceded by errors required us to pool the RTs from all 10 sessions. To do this, we first z-normalized the RTs of correct antisaccades before and after ketamine for each session and then combined the z-normalized RTs from the individual sessions. There were not enough trials for this analysis in the five saline control sessions.
Neurons were classified as performance selective, based on a two-way ANOVA on neural activity in the interval 200–700 msec after saccade onset during the preinjection period, evaluated at p < .05. The factors were Performance (correct or error antisaccades) and Saccade Direction (ipsilateral or contralateral to recorded hemisphere). Only neurons that showed a main effect of Performance but no significant effect of Saccade Direction and no significant Performance × Saccade Direction interaction were classified as performance related. A neuron's preferred performance was defined as the performance (correct or error) that was associated with the maximal spiking activity. As a measure of performance selectivity (preferred vs. nonpreferred performance), the selectivity index was defined as S = (p − np)/(p + np), where p = activity on preferred trials in the interval 200–700 msec after saccade onset and np = activity on nonpreferred in the same interval.
Selectivity during Intertrial Period
According to several cognitive control models (Carter & van Veen, 2007; Yeung et al., 2004; Botvinick et al., 2001; Miller & Cohen, 2001), the activity in the LPFC is increased after response conflict or response errors to increase cognitive control for the following trial. To test whether ketamine affected the activity during the intertrial period, we classified neurons as outcome selective during the intertrial period if they exhibited significant differences (t test, p < .05) in the period 1500–2500 msec following saccade onset between correct and error trials. This period was just before the FP was presented for the following trial. A neuron's preferred outcome was defined as the outcome (correct or error) for which the neuron displayed the maximal spiking activity during this period.
Receiver Operating Characteristic Analysis
To evaluate how well LPFC neurons could discriminate between correct and error responses, we computed receiver operating characteristic (ROC) values for the discharge during the performance monitoring and the intertrial periods for selective neurons. An ROC analysis measures the degree of overlap between two distributions. For each neuron, the activity on correct trials was compared with the activity on error trials, yielding two distributions of neuronal activity. Each point on the ROC curve was created by plotting the proportion of the distribution with the higher mean activity against the proportion of the distribution with the lower mean activity. To generate the entire ROC curve, the criterion level was incremented from zero to the maximal discharge rate in 20 steps. The area under the ROC curve is a quantitative measure of the separation of the two distributions. A value of 0.5 indicates that the two distributions completely overlap. Values of 0 and 1.0 indicate that the two distributions are completely separate.
As a measure of neural reliability, we computed the fano factor (variance divided by mean) for the discharge rates during the performance-monitoring period and the intertrial period.
To test whether ketamine had different effects on narrow-spiking (putative interneurons) and broad-spiking neurons (putative pyramidal neurons), we computed mean trough-to-peak times for the extracellular waveform of each neuron using previously described procedures (Johnston, De Souza, & Everling, 2009; Mitchell, Sundberg, & Reynolds, 2007). On the basis of the results of the previous study from our laboratory where we used the same recording system and the same type of microelectrodes in the LPFC, we defined neurons with waveform durations of shorter than 270 μsec as narrow-spiking neurons and any neurons with waveform durations of longer than 270 μsec as broad-spiking neurons.
We administered subanesthetic doses of ketamine in 10 experimental sessions (five in each monkey) and control injections of saline in five sessions (four in Monkey O and one in Monkey T), while the animals performed randomly interleaved pro- and antisaccade trials. We recorded the activity of 343 LPFC neurons (111 in Monkey O, 232 in Monkey T) during the ketamine sessions and of 99 LPFC neurons during the saline injections (51 in Monkey O, 48 in Monkey T).
RTs and Error Rates
Figure 1 shows the effects of ketamine and saline injections on error rates and RTs of prosaccades (Figure 1A) and antisaccades (Figure 1B). In line with several previous reports, both animals had longer RTs for antisaccades than prosaccades and made very few prosaccade errors before ketamine administration but generated a 10–20% error rate on antisaccade trials (Bell et al., 2000; Amador, Schlag-Rey, & Schlag, 1998). Following ketamine administration (thick lines), RTs of prosaccades (Figure 1A, top) and antisaccades (Figure 1B, top) increased. At the same time, error rates on antisaccade trials (Figure 1B, bottom) and to a lesser degree on prosaccade trials (Figure 1A, bottom) increased. RT and performance effects started about 5 min following ketamine injections. Performance returned to preinjection levels after about 20 min, whereas the effects on RTs persisted for the duration of the sessions. Saline injections (thin lines) neither influenced RTs nor performance of pro- or antisaccades. For all following behavioral and neural analyses, we compared a preinjection period (10-min interval before injection) with a postinjection period (5–20 min after injection).
RTs of Correct Responses and Errors on Antisaccade Trials
We next examined the effects of ketamine on RTs of correct responses and errors separately for each monkey. Figure 2A shows cumulative distributions of correct antisaccade RTs for ketamine and saline sessions. In Monkey O (Figure 2, top) RTs increased from 193 ± 2 msec to 252 ± 5 msec (t test, p < .00001) following ketamine injections. No effects were found for saline injections (196 ± 5 msec vs. 198 ± 3 msec; t test, p = .65). Similarly, in Monkey T (Figure 2, bottom) antisaccade RTs increased from 188 ± 2 msec to 237 ± 4 msec (t test, p < .00001) after ketamine injections. Saline injections also did not increase antisaccade RTs in this animal (180 ± 5 msec vs. 178 ± 3 msec; t test, p = .18).
The cumulative RT plots in Figure 2B show that monkeys mainly generated errors in the range of express saccades (Boch, Fischer, & Ramsperger, 1984) before ketamine injections, whereas post-ketamine errors had considerably longer RTs. In Monkey O (Figure 2B, top), error RTs increased from 103 ± 4 msec to 185 ± 8 msec (t test, p < .00001), and in Monkey T (Figure 2B, bottom), error RTs increased from 96 ± 4 msec to 248 msec (t test, p < .00001). We found a small increase in error RTs following saline injections (from 87 ± 1 msec to 94 ± 3 msec; t test, p < .05) in Monkey O and no differences in Monkey T (from 199 ± 32 to 128 ± 22 msec; t test, p = .07).
Behavior after Initial Saccade
Next, we investigated whether ketamine had effects on the monkeys' behavior after the saccade. Figure 3A shows the cumulative RTs of the saccade away from the initial antisaccade fixation location. On these correct trials, the animals received a liquid reward immediately after the antisaccade entered the target window (see Methods). In both monkeys, the RTs of return saccades did not differ between the preinjection period (solid blue line) and postinjection period (solid red line) in ketamine session (Monkey O: 398 ± 13 msec vs. 360 ± 17 msec, respectively; p = .11; Monkey T: 981 ± 20 msec vs. 948 ± 20 msec, respectively, p = .24). Similarly, the RTs of return saccades did not differ before (dashed blue line) and after (dashed red line) saline injections (Monkey O: 429 ± 13 msec vs. 399 ± 13 msec, respectively; p = .09; Monkey T: 1041 ± 29 msec vs. 975 ± 27 msec, respectively, p = .11).
On antisaccade error trials (Figure 3B) on which the animal looked toward the stimulus and did not receive a reward, the RTs of the return saccades were significantly longer in the postinjection (solid red line) compared with the preinjection period (solid blue line) in ketamine sessions in Monkey O (730 ± 51 msec vs. 248 ± 20 msec, respectively; p = 3.2 × 10−12) and Monkey T (904 ± 36 msec vs. 416 ± 37 msec, respectively; p = 2.2 × 10−42). The RTs of return saccades after antisaccade errors were similar to the RTs of return saccades after correct prosaccades (solid black line) after ketamine (Monkey O: 685 ± 51 msec; Monkey T: 1057 ± 15 msec). In both cases, the animals looked toward the stimulus, but they were only rewarded on correct prosaccade trials and not on antisaccade error trials. Saline injections (dashed lines) had no effect on the RTs of the return saccade after an error (Monkey O: 225 ± 19 msec vs. 235 ± 33 msec, p = .81; Monkey T: 347 ± 54 msec vs. 557 ± 96 msec, p = .07), and the RTs of these return saccades post-saline injections were significantly shorter than the RTs of return saccades after correct prosaccades (Monkey O: 445 ± 11 msec, p < .00001; Monkey T: 1159 ± 13 msec, p = 2.2 × 10−20). These findings show that ketamine did not influence the time when monkeys made a return saccade after a correct antisaccade, but it significantly increased the RT of return saccades after an error, rendering them similar to return saccades after correct prosaccades.
Although these data suggest that ketamine impaired the animals' ability to detect that they made a mistake on antisaccade error trials, it is also possible that the presence of the stimulus on the fovea after the response errors attracted fixation and prevented the animals from looking back. In this case, ketamine would impair the disengagement of fixation and not performance monitoring. If the deficit is because of an attraction to visual stimuli following ketamine, then one would expect that errors on prosaccade trials would be corrected faster after ketamine. Our data do not support this possibility. Monkey O made a return saccade after 433 ± 41 msec before ketamine injections and after 611 ± 32 msec post-ketamine injections on prosaccade error trials (i.e., an antisaccade; p < .005). The same trend was found in Monkey T (300 ± 47 msec preinjection and 451 ± 55 msec post-ketamine injections). This was not significant because of the small number of trials (5 pre-ketamine trials and 38 post-ketamine trials). Therefore, the pattern of RTs is inconsistent with the disengagement explanation in both animals, but it cannot be definitely ruled out because of insufficient power.
RTs on the Following Trial
The RTs of correct antisaccades were significantly longer when they were preceded by an antisaccade error trial compared with a correct antisaccade trials (z scores, 0.39 ± 0.15 vs. −0.04 ± 0.04; p < .005, t test) before ketamine was injected. These differences disappeared after ketamine administration (z scores, 0.03 ± 0.12 vs. −0.01 ± 0.05; p = .98, t test).
Performance-related Neural Activity
Consistent with previous reports (Niki & Watanabe, 1979), we observed that many LPFC neurons had differences in postresponse activity between correct and error trials. Of the 343 neurons recorded during the 10 ketamine injection sessions, 63 neurons (18.4%) showed a main effect of performance, 30 (8.6%) showed an effect of saccade direction, 16 neurons (4.6%) showed both main effects, and 47 neurons (13.5%) showed an interaction effect. The results were comparable for the 99 neurons that we recorded in the five saline injection sessions. Here, 16 (16.2%) showed a main effect of performance, 10 (10.1%) showed a main effect of direction, 5 (5.1%) showed both effects, and 17 neurons showed an interaction effect (17.2%). For those neurons that had a main effect of saccade direction, there was a mild direction preference for contralateral saccades (54% in the ketamine sessions and 60% in the saline sessions were more active after contralateral saccades).
For the following analyses, we defined a neuron as performance related, if it had a significant main effect of performance and no main effect of saccade direction or interaction. Of the 63 performance-related neurons recorded during the ketamine injection sessions, 44 neurons (69.8%) were more active after errors and 19 neurons were more active after correct responses (30.2%). This proportion was similar for the saline injection sessions, where 13 of 16 neurons (81.3%) were more active for errors than correct responses.
Figure 4 shows single neuron examples of performance-related LPFC neurons on correct and error antisaccade trials (Figure 4A) before (left) and after the administration of 0.4 mg/kg ketamine (right). The neuron in Figure 4B was more active after errors during the preinjection period, whereas the neuron in Figure 4C was more active after correct responses during this period. Following ketamine, the differences in activity between correct and error responses disappeared in both neurons. For the population of recorded neurons, the effects were very similar for neurons that preferred errors and those that preferred correct responses (see Table 1). Therefore, we combined the responses for the two types of performance-related neurons by defining each performance-related neuron's preferred outcome as the response (correct or error) that yielded the maximal response during the preinjection period.
|Pre Activity (spikes/sec)|
|Post Activity (spikes/sec)|
|Pre Activity (spikes/sec)|
|Post Activity (spikes/sec)|
|Correct < Error||3.2 ± 0.5**||6.1 ± 0.8||7.4 ± 1.8||9.3 ± 2.8|
|6.4 ± 0.8||6.7 ± 0.9||13.8 ± 3.4||11.8 ± 2.3|
|Correct > Error||10.2 ± 2.1||11.1 ± 2.9||4.6 ± 2.0||5.0 ± 2.2|
|5.6 ± 1.5*||11.3 ± 3.1||2.7 ± 1.6||3.3 ± 1.7|
|Pre Activity (spikes/sec)|
|Post Activity (spikes/sec)|
|Pre Activity (spikes/sec)|
|Post Activity (spikes/sec)|
|Correct < Error||3.2 ± 0.5**||6.1 ± 0.8||7.4 ± 1.8||9.3 ± 2.8|
|6.4 ± 0.8||6.7 ± 0.9||13.8 ± 3.4||11.8 ± 2.3|
|Correct > Error||10.2 ± 2.1||11.1 ± 2.9||4.6 ± 2.0||5.0 ± 2.2|
|5.6 ± 1.5*||11.3 ± 3.1||2.7 ± 1.6||3.3 ± 1.7|
*p < .01.
**p < .00001.
Figure 5A shows the activity of the population of performance-related neurons before ketamine (dashed lines) and after ketamine (solid line). The neurons' activity increased before saccade onset and began to decline 100 msec after saccade onset for nonpreferred trials (blue dashed line). On preferred trials (red dashed line), the activity decreased later. Following ketamine administration (solid lines), the presaccade firing rates increased in the population of neurons, consistent with the known effects of NMDAR antagonists on LPFC neurons in primates (Skoblenick & Everling, 2012) and rodents (Jackson et al., 2004). Ketamine did not seem to affect the activity on preferred trials following the saccade (red solid line), but it clearly increased the activity on nonpreferred trials (blue solid line).
To quantify these effects, we compared the activity before and after ketamine for preferred and nonpreferred responses in the period 200–700 msec after saccade onset. On preferred trials (Figure 5B), no significant effect (p = .30; paired t test) was found between the activity before and after ketamine (7.6 ± 0.8 and 8.0 ± 1.1 spikes/sec, respectively). On nonpreferred trials (Figure 5C), performance-related activity increased from 3.9 ± 0.6 to 7.6 ± 1.1 spikes/sec (p < .0001; paired t test). For saline injections (Figure 5D), no differences in activity were found before and after injections for preferred trials (Figure 5E; 11.9 ± 2.9 and 10.6 ± 2.0 spikes/sec, respectively; p = .30, paired t test) and nonpreferred trials (Figure 5F; 6.5 ± 1.6 and 8.2 ± 2.4 spikes/sec, respectively; p = .13, paired t test).
The behavioral finding that ketamine increased RTs of error antisaccades raised the possibility that the ketamine effects on neural activity in the LPFC might have been related to the different onset times of return saccades. To test this possibility, we aligned neural activity on the onset of the return saccade (Figure 6). There was no decline in neural activity before the return saccade, ruling out the possibility that differences in RTs caused the differences in neural activity of LPFC neurons before and after ketamine.
To evaluate the effects of ketamine on the ability of neurons to discriminate between correct and error trials, we compared the ROC values before and after injections. For the population, the ROC value dropped from 0.73 ± 0.01 to 0.53 ± 0.02 (p = 6.59 × 10−16) after the administration of ketamine (Figure 7A). There was also a decrease in ROC value, that is, in the discriminability between correct and error trials, from 0.73 ± 0.02 to 0.64 ± 0.03 (p < .01) following the injection of saline (Figure 7B), which might be related to fatigue or motivation effects. The reduction in ROC value, however, was significantly larger following ketamine compared with saline injections (−0.19 ± 0.02 vs. −0.09 ± 0.03; p < .01, paired t test; Figure 7C).
Although we observed changes in the neural activity after ketamine administration, there were no differences in response variability as measured by the fano factor before and after ketamine injections for the preferred (3.01 ± 0.17 vs. 3.03 ± 0.20; p = .98, paired t test) or nonpreferred condition (3.06 ± 0.17 vs. 3.11 ± .0.24; p = .91, paired t test). There were also no changes in the fano factor for the saline sessions.
Narrow-spiking versus Broad-spiking Neurons
Of the 63 performance-selective neurons recorded during the ketamine injections, 10 were classified as narrow spiking, that is, putative interneurons, and 44 neurons were classified as broad spiking, that is, putative pyramidal neurons, based on their waveforms (see Methods). Of the 10 narrow-spiking neurons, four were more active after correct responses and six were more active after errors. Figure 8 shows the effects of ketamine on broad-spiking and narrow-spiking neurons. Broad-spiking neurons exhibited no differences in activity before and after ketamine (6.5 ± 0.8 and 6.8 ± 1.0 spikes/sec, respectively; p = .51, paired t test) for the neurons' preferred condition. For the nonpreferred condition, broad-spiking neurons increased their activity from 3.3 ± 0.5 to 6.5 ± 0.9 spikes/sec (p < .000001, paired t test). The responses of narrow-spiking neurons were similar. No significant effect (p = .41; paired t test) was found between the activity before and after ketamine (13.2 ± 3.1 and 14.7 ± 4.3 spikes/sec, respectively) for the neurons' preferred condition, but their activity increased from 7.3 ± 2.2 to 13.7 ± 4.7 spikes/sec (p < .05; paired t test) for the neurons' nonpreferred condition.
Time Course of Performance Selectivity
To investigate the time course of the ketamine effect on LPFC activity, we computed mean normalized selectivity indices (see Methods) for consecutive 10-min time bins for the ketamine and saline injection sessions (Figure 9). Performance selectivity decreased in the 10 min following ketamine administration and reached its minimum 10–30 min following the injection. Although performance selectivity slowly increased afterwards, it did not return to its preinjection level during our recording sessions. For saline injections, there were some fluctuations in performance selectivity, but the levels never fell below those after ketamine injections and performance selectivity was similar at the end and beginning of the recording sessions.
Intertrial Neural Activity
In 24.8% (85/343) of the neurons recorded during the ketamine sessions, the intertrial activity was significantly different depending on whether the previous trial was correct or a response error (p < .05, t test). The majority of the neurons (83.5%) were more active following a response error on the previous trial than a correct response. Ketamine decreased the activity for the neurons' preferred trial outcome from 8.42 ± 0.69 to 7.26 ± 0.74 spikes/sec (p < .02; paired t test). For the nonpreferred outcome, the activity increased from 4.77 ± 0.5 to 5.87 spikes/sec (p < .0001; paired t test) following ketamine administration (Figure 10A).
A similar proportion of neurons exhibited differences in intertrial activity between previous correct and error trials in the saline sessions (28% or 28/99 neurons). Like in the ketamine sessions, more neurons were active for errors than correct response (64.3%, 18/28). The activity for the preferred condition did not differ before and after saline injection (10.29 ± 1.28 vs. 9.73 ± 1.39 spikes/sec, respectively; p = .28, paired t test), but the activity for the nonpreferred condition increased in the period after the injection (6.24 ± 0.9 vs. 7.28.12 spikes/sec, respectively; p < .05, paired t test; Figure 10C).
There was a significant decrease in the ROC for the population from 0.74 ± 0.01 to 0.58 ± 0.01 (p = 8.1 × 10−17) after the administration of ketamine (Figure 10B). Although there was also a decrease in ROC value from 0.72 ± 0.01 to 0.63 ± 0.02 (p < .001) following the injection of saline (Figure 10D), the reduction in discriminability was significantly larger for ketamine compared with saline injections (−0.16 ± 0.02 vs. −0.09 ± 0.02; p < .02, paired t test; Figure 10E). This finding shows that ketamine decreased the discriminability during the intertrial period between previous correct and erroneous response trials.
The fano factor did not differ between before and after ketamine injections for the preferred (2.21 ± 0.20 vs. 2.63 ± 0.26; p = .07, paired t test) or nonpreferred condition (1.94 ± 0.15 vs. 2.11 ± .0.17; p = .12, paired t test) during the intertrial period. The findings were similar for the saline injections for the preferred (2.29 ± 0.50 vs. 1.65 ± 0.5 p = .09, paired t test) and nonpreferred condition (1.98 ± 0.37 vs. 1.98 ± .0.3; p = .93; paired t test). This finding indicates that ketamine did not affect the reliability of neural responses during the intertrial period.
Our data demonstrate that disruption of NMDARs impairs error processing in nonhuman primates: A single systemic subanesthetic dose of ketamine reduced the differences in postresponse activity between correct and error trials of LPFC neurons in an antisaccade task. For those neurons that had higher activity following an error than a correct response, ketamine increased the postresponse activity on correct trials only. Ketamine had no effect on the activity following response errors in these neurons. Similarly, neurons that exhibited higher postresponse activity for correct trials increased their activity following ketamine on error trials but not on correct trials. The effect of these performance-dependent increases in neural activity was that performance selectivity was significantly reduced in LPFC neurons following ketamine administration. Consistent with the finding that ketamine impaired performance-related activity of LPFC neurons, we also found that animals fixated longer on the stimulus on error trials following ketamine, thereby resembling the fixation durations after correct prosaccades. The absence of any behavioral or neural effects after saline injections demonstrates that the effects were because of the action of ketamine and not related to a decrease in task performance and performance selectivity of LPFC in response to the injection process.
Consistent with previous reports (Skoblenick & Everling, 2012; Condy et al., 2005), we found that ketamine impaired the overall performance of antisaccades and, to a smaller degree, also of prosaccades. We have previously argued that this decline in performance might be related to impairments in the animal's ability to selectively maintain or apply the appropriate task set (Skoblenick & Everling, 2012). Such an impairment in context processing, that is, the ability to produce a different response to the same stimulus depending on the goal or rule (Miller & Cohen, 2001; Cohen & Servan-Schreiber, 1992), has also recently been described for the Dot Pattern Expectancy Task after ketamine injections (Blackman et al., 2013). Here we found some evidence that ketamine also affected the monkeys' behavior after a saccade. Although ketamine did not affect the behavior after a correct response, both animals maintained much longer fixation on the stimulus after a response error following ketamine administration. In fact, fixation durations after a response error on an antisaccade trial resembled the fixation duration after a correct prosaccade. After both types of responses, monkeys fixated the peripheral cue, but they were only rewarded on the correct prosaccade trials. The finding that ketamine did not affect the behavior after correct antisaccades indicates that the delivery of the reward was sufficient to signal that the trial was performed correctly even under ketamine and that ketamine did not lead to a general increase in fixation durations after a saccade. Instead, the data suggest that ketamine impaired the animal's ability to detect that their saccade was a response error, which would be consistent with its effects on performance-related activity of LPFC neurons. This would also explain the finding that posterror slowing after an antisaccade response error disappeared after ketamine administration because posterror slowing has only been found after aware, but not unaware, antisaccade errors in human participants (Klein et al., 2007).
In addition to its effect on action-monitoring activity of LPFC neurons, ketamine also altered neural activity during the intertrial period. Consistent with the conflict-monitoring model (Carter & van Veen, 2007; Yeung et al., 2004; Botvinick et al., 2001; Miller & Cohen, 2001), we found that a large proportion of LPFC exhibited differences between correct and error trials in the following intertrial period. This finding is reminiscent to the observation that some LPFC neurons maintain conflict-related signal throughout the intertrial period (Mansouri, Buckley, & Tanaka, 2007). Ketamine reduced the differences between previous correct and previous error trials by decreasing the activity for the neurons' preferred condition and increasing the activity for their nonpreferred condition.
Although ketamine binds to other receptors besides the NMDAR, there is strong evidence that its behavioral effects are mediated primarily by NMDAR (Duncan, Miyamoto, Leipzig, & Lieberman, 1999; Ginski & Witkin, 1994; Byrd, Standish, & Howell, 1987). Our data show that ketamine increased the activity of LPFC neurons. This finding seems surprising given that noncompetitive NMDAR antagonists block the NMDA channel (Huettner & Bean, 1988) and decrease the firing rate of neurons in anesthetized animals (Moghaddam, Adams, Verma, & Daly, 1997). However, increased LPFC activity following a subanesthetic dose of noncompetitive NMDAR antagonists has also been found in rodent LPFC neurons (Jackson et al., 2004). The authors hypothesized that the blockage of the NMDAR may lead to a compensatory overactivation of AMPA receptors. Alternatively, ketamine has been shown to block NMDAR in fast-spiking inhibitory interneurons more effectively than in pyramidal neurons in rodents (Seamans, 2008; Homayoun & Moghaddam, 2007; Olney, Newcomer, & Farber, 1999). We have recently also found increases in LPFC activity following ketamine administration in nonhuman primates, but these effects were similar for putative interneurons and pyramidal neurons (Skoblenick & Everling, 2012). Here we also found that ketamine had similar effects on putative interneurons and pyramidal neurons. It is critical to note that we did not observe an overall increase in performance-related LPFC activity here, but a selective increase in a neuron's postresponse activity for its nonpreferred performance. The finding that some LPFC neurons were active after response errors and other neurons that were more active after correct responses supports the notion that opponent coding is a general feature of prefrontal decision-making (Lennert & Martinez-Trujillo, 2013; Kusunoki, Sigala, Nili, Gaffan, & Duncan, 2010). We also show here that these two response patterns are likely not mediated by different classes of neurons as putative interneurons and pyramidal neurons were found in both populations. This result is reminiscent to the absence of waveform differences between ipsilateral and contralateral LPFC neurons (Lennert & Martinez-Trujillo, 2013).
Although we have found that ketamine altered action-monitoring activity of LPFC neurons, we cannot assume that ketamine acted specifically on these neurons. Resting-state fMRI studies have demonstrated that ketamine leads to an increase in global connectivity throughout the brain (Driesen et al., 2013) and alters the relationship between task-positive and task-negative neural systems (Anticevic et al., 2013), indicating that it does not selectively act on the LPFC. Furthermore, single neuron recordings in monkeys (Phillips & Everling, 2012; Parent & Hazrati, 1995) combined with functional imaging studies and lesion studies in humans suggest that a cortico-BG-thalamic circuit is involved in action monitoring. In particular ACC and LPFC are closely interconnected (Paus, 2001; Bates & Goldman-Rakic, 1993), and it is possible that the effects were mediated by projections from ACC to the LPFC. Note, however, that in humans with LPFC damage, the correct trial ERN activity is equal to the error trial ERN (Gehring & Knight, 2000), which clearly demonstrates that the LPFC interacts with the ACC in performance monitoring.
According to an influential model of error processing and reinforcement learning (Holroyd & Coles, 2002), dopaminergic neurons in the ventral tegmental area and the substantia nigra pars compacta send a negative reinforcement signal to the ACC. Support for this model has come from participants suffering from Parkinson's disease, which is characterized by a severe degeneration of dopaminergic neurons in the substantia nigra pars compacta. It has been shown that individuals with Parkinson's disease exhibit a reduced ERN and deficits in performance monitoring (Jocham & Ullsperger, 2009). However, it has been argued by others that the mesoprefrontal dopamine signal lacks the temporal precision required to generate the fast ERN (Jocham & Ullsperger, 2009). In addition, more recent studies have also pointed toward a role for serotonin, norepinephrine, GABA, and adenosine in performance monitoring (Jocham & Ullsperger, 2009). NMDAR interacts with all these systems, so it might be impossible to trace the effects of ketamine on prefrontal action-monitoring activity to a single neuropharmacological mechanism.
Our results demonstrate that a low dose of ketamine alters action-monitoring activity of LPFC neurons in nonhuman primates. These changes in neural activity could explain the deficits in action monitoring found in humans after ketamine administration (Stone et al., 2011). Taken together, the findings also support the hypothesis that an NMDAR dysfunction may mediate self-monitoring deficits and ultimately leads to the psychotic symptoms in schizophrenia (Bickel & Javitt, 2009).
This research was supported by operating grants from the Ontario Mental Health Foundation, the Canadian Institutes of Health Research, a fellowship by the Center for Interdisciplinary Research, University of Bielefeld, Germany to Stefan Everling, and a PhD fellowship from the Canadian Institutes of Health Research to Kevin Skoblenick.
Reprint requests should be sent to Stefan Everling, Robarts Research Institute, 1151 Richmond Street North, London, Ontario N6A 5B7, Canada, or via e-mail: firstname.lastname@example.org.