Abstract
Cognitive control is engaged by working memory processes and high-demand situations like antisaccade, where one must suppress a prepotent response. While it is known to be supported by the frontoparietal control network, how intra- and interareal dynamics contribute to cognitive control processes remains unclear. N-Methyl-d-aspartate glutamate receptors (NMDARs) play a key role in prefrontal dynamics that support cognitive control. NMDAR antagonists, such as ketamine, are known to alter task-related prefrontal activities and impair cognitive performance. However, the role of NMDAR in cognitive control-related frontoparietal dynamics remains underexplored. Here, we simultaneously recorded local field potentials and single-unit activities from the lateral prefrontal (lPFC) and posterior parietal cortices (PPC) in two male macaque monkeys during a rule-based antisaccade task, with both rule-visible (RV) and rule-memorized (RM) conditions. In addition to altering the E/I balance in both areas, ketamine had a negative impact on rule coding in true oscillatory activities. It also reduced frontoparietal coherence in a frequency- and rule-dependent manner. Granger prediction analysis revealed that ketamine induced an overall reduction in bidirectional connectivity. Among antisaccade trials, a greater reduction in lPFC–PPC connectivity during the delay period preceded a greater delay in saccadic onset under the RM condition and a greater deficit in performance under the RV condition. Lastly, ketamine compromised rule coding in lPFC neurons in both RV and RM conditions and in PPC neurons only in the RV condition. Our findings demonstrate the utility of acute NMDAR antagonists in understanding the mechanisms through which frontoparietal dynamics support cognitive control processes.
- cognitive control
- functional connectivity
- lateral prefrontal cortex
- local field potentials
- posterior parietal cortex
- single unit
Significance Statement
A low dose of ketamine is known to induce a transient cognitive control deficit in healthy humans and animals, but it remains unclear whether this deficit is related to a frontoparietal disconnection. In macaque monkeys performing a rule-based pro- and antisaccade task, we found that ketamine impaired information coding in frontoparietal neurons, local oscillations, and interareal synchrony in a rule- and frequency-dependent manner. Notably, under the antisaccade rule, the amount of impairment in task performance could be predicted by the loss in frontoparietal connectivity in the period just before the monkeys responded. The observations support the utility of NMDAR antagonists like ketamine as a tool to understand the role of frontoparietal dynamics in cognitive control.
Introduction
Cognitive control serves to actively maintain a representation of both the goals and the means to achieve them (Miller and Cohen, 2001). It is therefore engaged in working memory tasks, when we maintain and use incoming information to guide the choice of appropriate responses. It is also critical when the task demand is high, such as when the appropriate response must compete against stronger alternatives. One well-known example is the antisaccade task, when one chooses to suppress the prepotent response of looking toward the target and looking toward the opposite direction instead. Cognitive control processes are supported by a frontoparietal network encompassing the lateral prefrontal cortex (lPFC) and posterior parietal cortex (PPC), among other areas (Dosenbach et al., 2006; Duncan, 2010, 2013). This network is activated by working memory (Owen et al., 2005; Yeo et al., 2015; Owens et al., 2018), preparation for antisaccades (DeSouza et al., 2003; Brown et al., 2007), and many other demanding tasks (Cabeza and Nyberg, 2000; Duncan and Owen, 2000; Cole and Schneider, 2007; Farooqui et al., 2012; Badre and Nee, 2018). In contrast, frontoparietal dysconnection has been linked to impaired cognitive performance (Eryilmaz et al., 2016; Godwin et al., 2017; Nielsen et al., 2017). This literature supports the view that frontoparietal communication plays a key role in cognitive control, but the underlying mechanism remains unknown.
Thanks to the slow decay of its current, the N-methyl-d-aspartate glutamate receptor (NMDAR) plays a key role in recurrent excitation in cortical neurons, which is additionally supported by α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptor excitation, and can serve as the foundation for working memory (Wang, 1999; Durstewitz et al., 2000; Seamans et al., 2003) and other cognitive control functions (Durstewitz and Seamans, 2008; Cavanagh et al., 2020; Lam et al., 2022). Indeed, in healthy human participants, a low dose of ketamine, an NMDAR antagonist, is known to impair working memory (Morgan et al., 2004; Driesen et al., 2013) and alter frontoparietal activity associated with “online” manipulation of information (Morgan et al., 2004; R. A. Honey et al., 2004; G. D. Honey et al., 2005). Animal studies have replicated these effects. In macaque monkeys, the same low doses of NMDAR antagonists impaired performance in tasks involving working memory (Tsukada et al., 2005; Ma et al., 2015, 2018; van Vugt et al., 2020; Busch et al., 2024 ), antisaccades (Condy et al., 2005; Skoblenick and Everling, 2012), and cognitive flexibility in context processing (Blackman et al., 2013; Zick et al., 2018). In rats, NMDAR blockade impaired performance in tasks involving working memory (Aultman and Moghaddam, 2001; Homayoun et al., 2004; Enomoto and Floresco, 2009), attention (Condy et al., 2005; Pehrson et al., 2013), and cognitive flexibility (Stefani and Moghaddam, 2003, 2010; Darrah et al., 2008). Thus, all evidence points toward a critical role of NMDARs in cognitive control.
However, the coherence of this literature is somewhat disrupted by the fact that one of the NMDAR blockers, ketamine, increases glutamate transmission via (AMPA) receptors and boosts activities in populations of frontal cortical neurons in humans (R. A. Honey et al., 2004; Rowland et al., 2005), macaque monkeys (Skoblenick and Everling, 2012; Wang et al., 2013; Ma et al., 2015), and rats (Jackson et al., 2004). Thus, instead of a simple loss of task-related excitation, ketamine likely induces its behavioral effects through shifted excitatory/inhibitory (E/I) balance and altered connectivity in the frontoparietal network, which in turn lead to a compromised signal-to-noise ratio in both individual neurons and neuronal ensembles (Ma et al., 2015) and a weakened task representation in local field potentials (LFPs; Ma et al., 2018). Previous reports have demonstrated that shifting the E/I balance, e.g., via NMDAR blockade (Cavanagh et al., 2020), can impair perceptual decision-making both within (Lam et al., 2022) and across areas in the frontoparietal network (Murray et al., 2017). Furthermore, systemic treatment with another NMDAR antagonist, phencyclidine, both weakened neural interactions within the lPFC and profoundly changed the communication between the lPFC and PPC in macaque monkeys during a cognitive control task (Kummerfeld et al., 2020). Thus, NMDAR blockers can serve as a useful tool to probe the role of intra- and interareal E/I balance and connectivity in cognitive control processes.
Here, we recorded LFPs and single-unit activities from both the lPFC and PPC in macaque monkeys performing a rule-based pro- and antisaccade task (Skoblenick and Everling, 2012). This task included randomly interleaved, rule-visible (RV), and rule-memorized (RM) trials (Fig. 1A,B). In each trial, a color cue was presented which mapped onto the pro- or antisaccade rule. In the RV trials, the color cue remained visible until the target was presented. In the RM trials, a delay was inserted between the cues and target onset, when the monkeys fixated on a white dot. Given our previous findings (Skoblenick and Everling, 2012; Ma et al., 2015), we expected to find a weakened rule representation in the PPC and lPFC in both LFPs and single-unit activities. Importantly, we hypothesized that NMDA antagonism would result in dysconnectivity in the frontoparietal network, which would correlate with impaired cognitive control.
Schematics for the design of (A) the rule-visible and (B) the rule-memorized pro- and antisaccade task. (C) The placement of recording chambers and (D) structural magnetic resonance images that verify the chamber placement in Monkey A and (E) Monkey C.
Materials and Methods
Animals and surgeries
Two male rhesus monkeys (Macaca mulatta), weighing 8 kg (Monkey A) and 10 kg (Monkey C), were used in the study. All training, surgical, and testing procedures performed were approved by the Animal Care Committee of the University of Western Ontario Council on Animal Care and in accordance with the Canadian Council of Animal Care policy on laboratory animal use.
As previously described (DeSouza and Everling, 2004), animals were implanted with a plastic head restraint and trained to perform the task. Once trained, they were implanted with recording chambers over the posterior third of the principal sulcus (lPFC chamber) and over the intraparietal sulcus (PPC chamber; Fig. 1C–E). Both implantation surgeries were conducted following previously published procedures under aseptic conditions in a dedicated operating room (DeSouza and Everling, 2004). Briefly, monkeys were sedated with ketamine (10–15 mg/kg). They were intubated to allow for active ventilation. General anesthesia was initiated with a bolus of propofol (2.0 mg/kg) and maintained with propofol (0.2 mg/kg/min) and midazolam (0.35 mg/kg/min) via an intravenous catheter. Heart rate, blood pressure, respiratory rate, and body temperature were monitored closely for the duration of the surgery. For head post implantation, the plastic head post (Neuronitek) was anchored to the skull using ceramic screws and dental cement, which was shaped to create a head cap. For chamber implantation, which took place several months after the head postsurgery, two 19 mm craniotomies were created. Polyether ether ketone plastic chambers (Neuronitek) were aligned to the craniotomies and stabilized using dental cement with no need for additional skull screws. The chambers were then protected with dust caps secured with set screws. Postsurgical recovery was monitored closely by university veterinarians who directed the administration of analgesics and antibiotics as appropriate.
Behavioral task
The task has been described in detail in previous papers from our group (Skoblenick and Everling, 2012, 2014; Ma et al., 2015, 2018). On each trial, animals were required to fixate a small white spot at the center of the monitor display. After 500 ms of fixation, a color cue replaced the white fixation spot and lasted for 400 ms. On rule-visible trials (RV), the cue presentation lasted for another 1,000 ms period (Fig. 1A); on rule-memorized (RM) trials, the cue was extinguished after 400 ms, followed by a 1,000 ms delay with a small jitter, during which the screen remains dark (Fig. 1B). These two trial types were randomly interleaved. For Monkey A, a red cue indicated that a prosaccade was required for the subsequent postdelay peripheral stimulus, while a green cue indicated an antisaccade. For Monkey C, this color contingency was reversed. The peripheral stimulus appeared at 8° to either the left or right of the fixation spot at its offset. If the saccade landed within a window of 4° centered on the target location, the animal received a liquid reward. Trials involving prosaccades to the left and right and those involving antisaccades to the left and right were randomly interleaved. Adjacent trials were spaced by a random intertrial interval between 1 and 1.5 s. The animals’ eye positions were recorded and digitized at 1,000 Hz using an EyeLink 1000 infrared pupillary tracking system (SR Research).
Recording
At the start of a session, 4–6 tungsten recording electrodes were advanced through a grid placed over each brain region and removed afterward. The locations of the electrodes were adjusted manually via screw microdrives and optimized for spiking activities within each session. The final adjustment to any of the electrodes took place at least 20 min before the onset of the recording session. Neural activities, including LFP and spike trains, as well as eye-tracking data were recorded using the Plexon OmniPlex system (Plexon). Spiking activities were sorted manually using Offline Sorter (Plexon), using 2-D and 3-D principal component analysis. A portion of the local field potential data (LFPs) recorded from the lPFC of Monkey A have been used in the analysis of a previously published work (Ma et al., 2018).
Drug administration
Each experimental session began with a pre-injection baseline period lasting at least 10 min. The animals received a single intramuscular injection of either 0.4 ml of ketamine or 0.4 ml of 0.9% sterile saline. Ketamine at this dosage elicited cognitive deficits with minimal anesthetic effects in rhesus monkeys (Condy et al., 2005; Stoet and Snyder, 2006; Shen et al., 2010; Blackman et al., 2013). Both monkeys started at the dose of 0.4 mg/kg. Monkey A developed tolerance over time, whereas Monkey C was unable to perform the task at this dose. Hence, we analyzed the data after every session to confirm that (1) sufficient trials were performed and (2) a significant change had been induced by ketamine in performance from 5 to 30 min post-injection compared with baseline. If this was not the case, the session was discarded, and the dose was increased/decreased by 0.1 mg/kg for the next session. For Monkey A, the ideal dose was between 0.4 and 0.8 mg/kg, and for Monkey C, this was 0.1–0.15 mg/kg. The outcome of this titration is shown in Figure 3 and described below in Results. After the injections, the monkeys continued with the behavioral session for at least another 30 min. Ketamine treatments were spaced by at least 72 h to slow any development of tolerance.
Experimental design and statistical analysis
To quantify the drug effect accurately, we used a within-subject design, in which the animals’ neural activities as well as behavioral performance with ketamine on board were compared with the pre-injection baseline period within the same recording session. For behavioral performance and saccadic reaction times, we used mixed-model ANOVA, with treatment (pre- vs post-injection), rule, and task as the within-subject variables and drug type (ketamine vs saline) as the between-subject variable. Drug type could not be a within-subject variable, due to the difference in the number of sessions with ketamine versus saline treatment. For LFP data, we focused on ketamine sessions and used repeated-measures ANOVA, with treatment (pre- vs post-injection), rule, and task as within-subject factors. The “subject” here refers to the channel, or channel pair, as is the case for coherence and Granger prediction (GP) analyses. This is because, for the duration of a session, the electrode consistently recorded signals from the same volume of brain tissue. The following subsections provide the details of each statistical analysis we performed.
LFP preprocessing
From the two animals, a total of 323 channels had single-unit activities and thus were included in the LFP analyses (lPFC and PPC: Monkey A, n = 109 and 117; Monkey C, n = 43 and 54). Because the behavioral effects of ketamine took no more than 5 min to develop and returned to baseline 30 min after injection, we compared the period from 5 to 30 min after injection to the 10 min baseline preceding the injection. Single-unit waveforms were manually sorted using Offline Sorter (Plexon) and analyzed with custom software. LFP data were analyzed in MATLAB (MathWorks, RRID:SCR_001622) using the FieldTrip toolbox (http://fieldtrip.fcdonders.nl/, RRID:SCR_001622) developed at the Donders Institute for Brain, Cognition, and Behaviour (Oostenveld et al., 2011). The continuous signal was divided into discrete trials based on event timestamps. Trials in which delay-period LFP power exceeded eight standard deviations from the average were excluded from the analysis.
The LFPs were then low-pass filtered at 100 Hz, and line noise was removed at 60 Hz using a discrete Fourier transform.
Parameterization of aperiodic and periodic LFPs
Following preprocessing, we separated the aperiodic and periodic components of LFPs using the FOOOF algorithm (Donoghue et al., 2020) as incorporated in the FieldTrip toolbox (http://fieldtrip.fcdonders.nl/, RRID:SCR_001622; Oostenveld et al., 2011). This approach is illustrated in Figure 4. For the aperiodic component, we then analyzed the exponent—the slope of the power curve in log-log plots and offset—the y-axis intercept and compared these parameters from before and after ketamine injections and across tasks (Fig. 5). For the periodic or oscillatory component, we analyzed the central frequency and amplitude of the peaks (Fig. 6). Extremely narrow-band peaks between 58.5 and 61.5 Hz were deemed electrical noise and removed from further analysis.
Debiased weighted phase lag index
We used the debiased weighted phase lag index (dWPLI) to quantify the phase coherence between LFP channels within and across brain regions. This estimate is based on the imaginary component of the coherency between LFP series and independent of LFP power, with additional improvements: compared with the phase locking value, the dWPLI is less affected by phase delays; and compared with the phase lag index (PLI), it is less affected by noise and has enhanced statistical power (Vinck et al., 2011). Lastly, the “debiasing” refinement removed the sample size bias (Vinck et al., 2011).
We used the FieldTrip toolbox to first calculate the cross-spectra using Hanning window with a width of 4 cycles at each frequency. We then used the ft_connectivityanalysis.m and set the method to “wpli-debiased” to calculate dWPLI for each time–frequency pair. To examine ketamine-induced changes in coherence, we z-score normalized the delay-period dWPLI from after the injections against the pre-injection dWPLI (Figs. 8, 9).
Granger prediction (GP)
To analyze the task-related and ketamine-induced effects on directional connectivity in the frontoparietal network, we calculated Granger prediction using the Multivariate Granger Causality toolbox (Barnett and Seth, 2014). For each pair of channels, the time-domain data from the delays and inter-trial intervals (ITIs) were separately fit with a vector autoregressive (VAR) model, and then the VAR coefficient and residual covariance matrices were used to calculate the multivariate Granger prediction (GP). We then compared the Granger prediction using repeated-measures ANOVA, with direction, task, and treatment as within-subject variables (Fig. 10). Additionally, we analyzed how ketamine-induced changes in response accuracy and saccadic reaction time correlated with changes in lPFC-led and PPC-led GP (Figs. 11, 12).
Rule selectivity index
To quantify the rule information contained in the activities, we calculated two types of rule selectivity: (1) For a parameter derived from the LFPs, this was calculated as the absolute difference between pro- and antisaccade trials, divided by the sum of the two:
(2) For individual neurons, this was calculated as the absolute d′:
Results
On average, monkeys completed 528 trials per session (excluding omitted or no-response trials), including randomly interleaved pro- and antisaccade trials in the RV and RM conditions, with sessions ranging from 41 to 65.5 min in duration. Ketamine had a significant effect on percentage correct responses compared with saline administration (mixed-model ANOVA, main effect of drug: F1,156 = 5.4, p = 0.024). Figure 2A shows the effect of ketamine in Monkey A (left) and Monkey C (middle) and the effect of saline injection in control sessions from both monkeys combined (right). Within ketamine sessions, the effect of treatment was significant for both Monkey A (repeated-measures ANOVA, F3,99 = 16.3, p = 1.1 × 10−8) and Monkey C (F3,30 = 7.98, p = 0.00047) with effect sizes of 0.33 and 0.44, respectively (measured as ηp2). In RM trials under both rules (Fig. 2A, dashed lines), ketamine reduced performance in the first 10 min post-injection interval (Monkey A, p ≤ 0.0036; Monkey C, p < 0.0088). In the RV task, ketamine only affected antisaccade trials for Monkey A (p = 0.00015; Fig. 2A, gray solid line) and prosaccade trials for Monkey C (p= 0.017; Fig. 2A, black solid line). By the third 10 min interval, performance in each trial type showed recovery, to a level either significantly higher than either the first or the second 10 min interval after ketamine injection (p ≤ 0.0079) or statistically equivalent to the pre-injection period (p ≥ 0.44). On the effect of task, in neither animal was the effect of ketamine stronger in the RM task compared with RV. Although Monkey C's performance in the second and third 10 min intervals was lower in the RM than the RV task (p < 0.029), this was due to a lower baseline performance in the RM task; across both tasks, performance dropped by a similar 15–16.5% on average. On the effect of rule, for Monkey A, ketamine had a greater impact on antisaccade performance (p = 0.99 pre-injection vs p = 0.00012 in third 10 min interval), whereas for Monkey C, this was not the case (p > 0.78 in all intervals). By contrast, the performance across tasks and rules did not change following a saline injection (rmANOVA, F3,24 = 1.78, p = 0.18; Fig. 2A, right panel). Thus, performance was negatively affected by ketamine in both monkeys, even though this effect differed somewhat between tasks and rules.
Effects of ketamine and saline injection on the percentage of (A) correct response and (B) saccadic reaction times. Performance on RV trials are shown as solid lines and RM trials as dashed lines. Prosaccade trials are shown in blue and antisaccade trials in red. Error bars indicate the standard deviation of the mean.
Furthermore, ketamine, but not saline, significantly increased the saccadic reaction times (SRTs; mixed-model ANOVA, effect of drug, F1,150 = 6.26, p = 0.016; Fig. 2B). This was true for all trial types (post hoc Tukey's test: ketamine, p = 2.8 × 10−5; saline, p ≥ 0.98). Within ketamine sessions, drug administration significantly increased the SRTs in both Monkey A (repeated-measures ANOVA, F3,96 = 101.7, p < 4.9 × 10−324) and Monkey C (F3,27 = 18.5, p = 1.0 × 10−6), with large effect sizes (ηp2 = 0.76 and 0.67, respectively). In both tasks and under both rules, ketamine increase SRTs in the first 10 min post-injection interval (Monkey A, p = 0.00015; Monkey C, p < 0.0089; Fig. 2B, left and middle panels). By the third 10 min interval post-injection, SRTs showed recovery compared with the second 10 min interval in Monkey A (p = 0.00015) but not in Monkey C (p > 0.83). We found no effect of task in SRT at baseline under either rule (Monkey A, p ≥ 0.27; Monkey C, p ≥ 0.86), only a transient difference in the second 10 min interval post-injection in Monkey A (p = 0.021). In both monkeys, the SRTs were longer in antisaccade trials (A, F3,96 = 714.9, p < 4.9 × 10−324; C, F3,27 = 25.3, p = 0.00071; Fig. 2B, blue vs red curves). In Monkey A, the effect of ketamine was also stronger under the antisaccade rule (interaction: F3,96 = 14.3, p = 9.0 × 10−8), which was not the case for Monkey C (F3,27 = 1.19, p = 0.33).
Because the dose of ketamine was titrated for each monkey, we asked whether and how it affected the amount of change in their performance. We plotted the change in performance (Fig. 3A) and SRT (Fig. 3B) as a function of dose for each monkey and used lighter shades of gray to indicate later sessions at the same dose, to visualize any development of tolerance. In Monkey A, more severe deterioration in performance was associated with higher doses (Fig. 3A, left). In Monkey C, the amplitudes of drug-induced deficit were similar to that of Monkey A, even though smaller doses were used (Fig. 3A, right); the impairment–dose relationship was not significant at the current sample size. In both monkeys, the first session (black dots) at some doses saw a greater drop in performance post-injection than later sessions (light gray dots), although this was not the case at every dose. Larger doses are also associated with greater increases in SRTs in Monkey A (Fig. 3B, left), and a similar trend is visible in Monkey C (Fig. 3B, right). Early sessions (black and dark gray dots) were not consistently associated with a greater increase in SRTs compared with later sessions (light gray dots) at the same dose. Overall, the monkeys were similarly impaired by ketamine despite the difference in dosage. In some cases, tolerance seemed to develop in later sessions at the same dose, e.g., at 0.5 and 0.8 mg/kg in Figure 3A, left, but this trend was far from consistent across doses and monkeys.
Dose–response plots for (A) performance and (B) saccadic reaction times. The black dot denotes the first session at the given dose. Subsequent sessions were plotted in increasingly lighter shades of gray. A, Higher doses tended to be linked to greater impairment in the performance—a correlation significant in Monkey A (left) but not C (right). Tolerance to ketamine treatment would be visible as lighter gray dots being higher than darker dots, which was not always the case. B, Higher doses tended to be linked to a greater increase in SRT, which was significant in Monkey A (left) but not C (right). Tolerance would be visible as darker dots being higher than lighter gray dots, which was generally not the case.
Ketamine altered LFP powers in both the lPFC and PPC
In 45 sessions involving ketamine injections, we recorded from a total of 152 and 171 channels in the lPFC and PPC, respectively, from which we isolated 262 and 248 single units. Here, we consider only electrodes from which well-isolated units were detected. Using the recently developed FOOOF toolbox (Donoghue et al., 2020), we analyzed separately the effects of ketamine on the aperiodic and oscillatory components of LFPs. The aperiodic component of the LFP is not truly oscillatory and can be captured by a simple 1/f function, with two parameters: offset and exponent. In an example channel (Fig. 4A), ketamine reduced both the offset and exponent of the LFPs. The difference in offset is demonstrated by a lower point of intersection between the y-axis and the post-injection power curve (solid line), compared with the pre-injection curve (dashed line). The difference in exponent is visible in the “flatter” slope in the post-injection curve, compared with the pre-injection one (solid vs dashed line). Thus, if we were to calculate the LFP power using this aperiodic component, we would see bidirectional changes with reduced low-frequency power and increased high-frequency power, which does not reflect any change in neural oscillations. Instead, this may reflect a shift in the excitatory–inhibitory balance following ketamine injections (Gao et al., 2017). Additionally, in this channel, a single oscillatory component was identified in the beta band (Fig. 4B), which displayed a ketamine-induced shift in central frequency (dashed vs solid curves), without any change in the amplitude (i.e., the height of the curves). In the curve that combines aperiodic and oscillatory components (Fig. 4C), low-frequency LFP power between 3 and 20 Hz appeared to drop significantly following ketamine treatment (dashed vs solid curves). However, instead of reflecting a change in oscillatory power, this effect was due to changes in the aperiodic component and the shift in central oscillatory frequency (Fig. 4A,B). This example demonstrates the necessity of examining the effect of ketamine on these components separately.
An example of the effects of ketamine on the aperiodic and oscillatory components of the LFPs. A, Ketamine reduced both the exponent (visible as the slope) and the offset of the aperiodic component of the LFP power (visible as the lower position of intersection with the y-axis; dashed, pre-injection; solid, post-injection). B, In this case, ketamine increased the central frequency of the oscillatory component of the LFP, without changing its amplitude. C, Total LFP power with both the aperiodic and oscillatory components combined.
We went on to quantify changes in the aperiodic component of LFPs across all sessions and both animals. We found that ketamine administration resulted in a significant reduction in both the y-axis offset (Fig. 5A,B; empty vs filled bars) and the exponent (Fig. 5C,D). Specifically, the offset of the aperiodic component was reduced in both the lPFC and PPC, in both RV and RM tasks, and for both monkeys (mixed-model ANOVA, main effect of drug, Monkey A, F1,224 = 121.88, p < 4.9 × 10−324; Monkey C, F1,95 = 11.25, p = 0.0011; post hoc Tukey's test: p ≤ 0.003 in each subject, brain region and task). Interestingly, we found that the offset was lower in the lPFC than PPC (main effect of area, Monkey A, F1,224 = 4.48, p = 0.036; Monkey C, F1,95 = 13.31, p = 4.3 × 10−4)—an effect independent of the drug effect (no drug–area interaction: Monkey A, F1,224 = 0.43, p = 0.51; Monkey C, F1,95 = 1.90, p = 0.17). In parallel to the offset, the exponent of the aperiodic component also decreased significantly following ketamine injection, in both animals, brain regions, and tasks (mixed-model ANOVA, main effect of drug, Monkey A, F1,224 = 104.66, p < 4.9 × 10−324; Monkey C, F1,95 = 33.92, p = 7.8 × 10−8; post hoc Tukey's test: p ≤ 5.6 × 10−4 in each subject, brain region, and task). Similarly, we found that the exponent of the LFP power curve was lower in the lPFC than in the PPC (main effect of area, Monkey A, F1,224 = 8.74, p = 0.0034; Monkey C, F1,95 = 10.04, p = 0.0021), independent of ketamine injections. Together, these findings demonstrate that ketamine systematically alters the aperiodic component of the LFPs in the frontoparietal network. This effect can at least partially account for the bidirectional changes in LFP power previously observed, in which power at gamma-band activity was enhanced at the cost of activities at lower frequencies (Ma et al., 2018).
The effect of ketamine on the aperiodic component of LFPs. In both (A) Monkey A and (B) Monkey C and in both RV and RM tasks, ketamine injection led to a reduction in the offset of the aperiodic LFP. In both monkeys, the offset was significantly higher in the PPC than lPFC. Similarly, in both (C) Monkey A and (D) Monkey C and in both tasks, ketamine led to a reduction in the exponent of the aperiodic LFP. The exponent was also significantly higher in the PPC than lPFC.
We then examined the true oscillatory peaks that were separated from the aperiodic component. These peaks formed a multimodal distribution shown in Figure 6A. We therefore divided these peaks into three categories: theta band (3–8 Hz), mid-frequency oscillations (8–45 Hz), and high-frequency oscillations (45–100 Hz). According to widely used boundaries, the mid-frequency activities would include alpha, beta, and low gamma bands, and high-frequency activities here are equivalent to high gamma. We did not further divide the mid-frequency category, as this would render the sample size too small for inferential testing in Monkey C. Notably, however, the central frequencies in this broad band varied between brain regions in Monkey C. Before ketamine injections, while 73% of all peaks in the PPC were centered at 20 Hz or lower in the alpha/low beta range, 91% of peaks in the lPFC were in the high beta/low gamma range (20–45 Hz) instead. This difference did not exist in Monkey A (95 vs 93%). Hence, by grouping the peaks from 8 to 45 Hz, we only intend to explore the drug effect and do not imply that these activities were uniform across areas and subjects.
Effects of ketamine on the oscillatory components of LFP. A, Distribution of central frequencies of oscillatory components across all channels from both monkeys. Based on this plot, we divided all oscillations into three frequency bands: theta (8–10 Hz), mid-frequency (10–45 Hz), and high-frequency (45–100 Hz) oscillations. B, Central frequency of theta band did not change following ketamine injection in either region, in either animal. C, Similarly, the amplitude of theta oscillation did not show any effect of ketamine. D, Central frequency of mid-frequency oscillations increased in both areas in Monkey A (left panel) and in the PPC of Monkey C (right panel). In Monkey C, the central frequency had a regional difference. E, Amplitude of mid-frequency oscillations did not change except in the PPC of Monkey A. In both animals, the mid-frequency amplitude was greater in the PPC. F, The central frequency of high-frequency oscillations reduced following ketamine injection in the PPC of Monkey A and the lPFC in Monkey C. G, Amplitudes of high-frequency oscillations were not affected by ketamine. *p < 0.05, **p < 0.01.
Overall, the effect of ketamine on oscillations differed by frequency band. In theta-band activities, ketamine injections had no effect on either central frequency or amplitude, in either brain region and in either animal (Fig. 6B,C, empty vs filled bars). In the mid-frequency band, ketamine increased the central frequency (Fig. 6D; mixed-model ANOVA) in both areas in Monkey A (F1,1167 = 54.3, p = 3.3 × 10−13, Tukey's test, p = 3.5 × 10−5 and 8.1 × 10−6) and in Monkey C in the PPC (F1,386 = 13.0, p = 3.5 × 104; Tukey's test, p = 0.0099) but not the lPFC (p = 0.10). Thus, ketamine affected oscillations across areas and subjects in the low beta band (15–20 Hz), but not in low gamma (30–35 Hz) in the lPFC of Monkey C. In contrast, ketamine caused no change in the amplitude of mid-frequency oscillations, except a decrease in the PPC of Monkey A (p = 7.8 × 10−4; Fig. 6E). Both subjects had stronger mid-frequency oscillations in the PPC than in lPFC—an effect that was independent of drug treatment (main effect of area, A, F1,1167 = 9.5, p = 0.0022; C, F1,386 = 12.8, p = 3.9 × 10−4). In high-gamma oscillations, ketamine had a somewhat opposite effect compared with mid-frequency activities (A, F1,236 = 7.3, p = 0.0073; C, F1,113 = 6.8, p = 0.010; Fig. 6F): It reduced the central frequency in the PPC of Monkey A (p = 0.018) and lPFC in Monkey C (p = 0.019). Lastly, ketamine did not affect the amplitude of high-frequency oscillations (Fig. 6G).
Taken together, while there was some heterogeneity in LFP oscillations across areas and subjects, it was evident that ketamine significantly increased the central frequency of cortical oscillations in the low beta range. In contrast, ketamine tended to reduce the central frequency in high-gamma oscillations.
We then asked how task-related oscillations changed following ketamine injection. This was quantified as a rule selectivity index (RSI), defined as the absolute difference in the oscillatory component (Fig. 3B) associated with pro- and antisaccade trials, divided by the sum of the two (Eq. 1, see Materials and Methods). We then examined these indices within each frequency band of interest (Fig. 7).
The effect of ketamine on rule selectivity in LFP oscillations across regions and tasks, in Monkey A and Monkey C, in theta band (A, B), mid-range frequencies encompassing alpha, beta, and low gamma bands (C, D), and in high gamma band (E, F). The empty bars denote pre-injection rule selectivity, and the filled bars represent post-injection values. The lPFC is shown in orange, and the PPC is in purple. *p < 0.05.
In theta oscillations, ketamine had an overall impact on RSI in Monkey A (F1,224 = 6.2, p = 0.014); in Monkey C, ketamine did not have a main effect but instead had an interaction with task and brain region (F1,95 = 9.9, p = 0.0023). Specifically, in Monkey A, rule selectivity reduced following ketamine treatment in the RV but not RM task (p = 0.036; Fig. 7A, orange empty vs filled bars). In Monkey C, RSI in RV was also numerically lower after treatment (Fig. 7B, orange empty vs filled bars), although this effect was not significant (p = 0.16). Instead, RSI was greater in RV than RM task before (p = 0.00054; Fig. 7B, left vs right orange empty bars) but not after ketamine treatment (p = 0.99; Fig. 7B, left vs right orange filled bars). The drug did not have any notable effect on task-related oscillations in the PPC in the theta band in either animal. Nor did it affect RSI in mid-range oscillations encompassing alpha, beta, and low gamma bands (Fig. 7C,D). In the high gamma band, ketamine reduced RSI in both brain regions combined in the RM task in Monkey A (p = 0.015; Fig. 7E, right). In Monkey C, ketamine reduced RSI in the lPFC with both tasks combined (p = 0.0046; Fig. 7F, empty orange bars vs filled orange bars); this effect was not significant in the PPC (p = 0.18).
Ketamine reduces coherence between the lPFC and PPC in a task-dependent manner
We then asked if ketamine altered the interactions between the lPFC and PPC, both nondirectionally as in coherence and directionally as in Granger's prediction. For coherence, we calculated the debiased weighted phase lag index (dWPLI; Vinck et al., 2011) for delay-period activities and subtracted the pre-injection dWPLI from the post-injection value to quantify the effect of ketamine. Within the lPFC, ketamine did not affect coherence across recording sites in the RV task (Fig. 8A,B, left panels) and had inconsistent effects in gamma-band coherence in the RM task (right panels, horizontal bars indicate p < 0.05 after familywise error correction). Within the PPC, ketamine reduced coherence in the beta band in the RV and RM tasks and under both pro- (blue) and antisaccadic (red) rules (Fig. 8C) in Monkey A. In Monkey C, coherence was also reduced across rules and tasks, and the effect was associated with theta, alpha, and gamma bands in addition to the beta frequencies. Thus, ketamine had a deleterious effect on within-PPC coherence. We then examined interareal coherence and found a similar deleterious effect across tasks and rules (Fig. 9). Notably, ketamine-related reduction in high-gamma (50–70 Hz) coherence was observed in antisaccade trials across tasks and subjects (Fig. 9A,B, red horizontal bar). In prosaccade trials, a loss of beta-band coherence at ∼20 Hz was also consistently observed in both monkeys in the RM task (blue horizontal bar, right panels). This effect was less consistent in the RV task (left panels). It should be noted that changes in coherence as measured by dWPLI are independent of changes in LFP power (Vinck et al., 2011). The drug-related reductions in beta and gamma coherence cannot be accounted for by either the bidirectional change in aperiodic power (Fig. 4) or the bidirectional shifts in beta and gamma oscillatory bands (Fig. 5). Taken together, ketamine compromised within-PPC and cross-areal nondirectional connectivity in a task- and rule-dependent manner.
Changes in coherence within the lPFC and PPC, measured as debiased weighted phase lag index (dWPLI), following ketamine treatment. A, Within the lPFC, in Monkey A: Left, In the RV task, delay-period coherence did not increase (positive numbers) or decrease (negative numbers) significantly following ketamine injections in Monkey A, in either pro- (blue) or antisaccade (red) trials. Right, In the RM task, intra-lPFC coherence decreased in the gamma band ∼50 Hz under both rules. The blue and red horizontal bars indicate p < 0.05 for pro- and antisaccade rules, respectively, after family-wise error adjustment. B, Within the lPFC, in Monkey C: Left, In the RV task, ketamine had no effect. Right, In the RM task, under the prosaccade rule, ketamine increased the coherence in high gamma. C, Within the PPC, in Monkey A: under both rules and tasks, ketamine reduced coherence in beta frequencies. It also increased coherence in the low gamma band in RV prosaccade trials. D, Within the PPC, in Monkey C: ketamine reduced coherence in both tasks and rules across several frequency bands.
Ketamine affected lPFC–PPC coherence in multiple frequency bands in both (A) Monkey A and (B) Monkey C, in both RV and RM tasks and under both pro- (blue) and antisaccade (red) rule. The blue and red horizontal bars indicate p < 0.05 for pro- and antisaccade rules, respectively, after family-wise error adjustment.
Ketamine weakened lPFC–PPC and PPC–lPFC directional connectivity in the time domain
Results from the coherence analysis beget the question of whether the task-related connectivity was directional and how ketamine might affect it. We therefore calculated time-domain Granger predictions (Barnett and Seth, 2014) with the lPFC leading the PPC and vice versa. In both monkeys, ketamine injections led to a reduction in Granger prediction in both directions (repeated-measures ANOVA: Monkey A, F1,384 = 114.3, p = 4.9 × 10−324; Monkey C, F1,207 = 52.7, p = 7.7 × 10−12; Fig. 10, light vs dark colors). This reduction was significant in both tasks and under both rules (Monkey A, p = 2.9 × 10−5; Monkey C, p ≤ 0.0093, in each comparison). Overall, there was also an effect of direction: lPFC-led Granger prediction was stronger than in the opposite direction (Monkey A, F1,384 = 59.8; p = 9.3 × 10−14; Monkey C, F1,207 = 119.8, p < 4.9 × 10−324; Fig. 10). This effect did not interact with the drug effect. Nor was there any task effect in Granger prediction in either direction (A, F1,384 = 3.30, p = 0.07; C, F1,207 = 1.82, p = 0.18).
Ketamine affected the directional connectivity between the lPFC and PPC in the time domain (Granger prediction). A, In Monkey A, Granger prediction is overall stronger in the lPFC-lead than the PPC-lead model (left vs right set of bars). Ketamine significantly reduced their connectivity in both RV and RM tasks and under both prosaccade (light vs dark red) and antisaccade (light vs dark blue) rules. B, In Monkey C, ketamine induced the same set of changes in Granger prediction. *p < 0.05, **p < 0.005, ***p < 0.0005.
Since ketamine-induced impairment in performance coincided with reduced directional connectivity in both directions, we then asked whether these two changes may be correlated from session to session. In the RV task, we found a significant positive correlation between ketamine-related changes in task performance and lPFC–PPC Granger prediction, specifically under the antisaccade (Pearson's r = 0.25 and 0.27, p = 4.7 × 10−7 and 5.1 × 10−5, respectively; Fig. 11B,D) and not the prosaccade rule (r = 0.069 and −0.0040, p = 0.18 and 0.95; Fig. 11A,C). This finding indicates that greater reductions in lPFC-led connectivity during the delay period were associated with larger deficits in response accuracy. There was no significant correlation between changes in performance and changes in PPC-led Granger prediction. Nor was there any correlation between Granger prediction and performance in the RM task that displayed a consistent pattern across monkeys.
Ketamine-related changes in the level of performance were in some cases positively correlated with ketamine-related changes in lPFC–PPC Granger prediction, such as in the RV task, under the antisaccade rule in both (B) Monkey A and (D) Monkey C. This was not the case under the prosaccade rule in either monkey, as shown in A and C. The red lines depict the linear relationship between the drug-related changes. Also shown are Pearson's r and its associated p value.
Additionally, we found a significant negative correlation between changes in session-averaged SRTs and lPFC-led connectivity, this time in the RM task and again under the antisaccade rule (r = −0.13 and −0.17, p = 0.011 and 0.017; Fig. 12B,D). This would mean that greater losses in lPFC-led connectivity in the delay period preceded larger increases in reaction time. The same effect was found for the prosaccade rule in Monkey C (r = −0.21, p = 0.0028; Fig. 12C) but not in Monkey A, for which the correlation had the opposite pattern (r = 0.12, p = 0.024; Fig. 12A). For changes in PPC-led Granger prediction, we did find a negative correlation with SRT increase in the RM task under the antisaccade rule, but only in Monkey C (r = −0.22 and −0.26, p = 0.0014 and 1.2 × 10−4) and not Monkey A (r = −0.0041 and −0.02, p = 0.94 and 0.69).
Ketamine-related changes in SRT were in some cases negatively correlated with ketamine-related changes in lPFC–PPC Granger prediction, as in the RM task under the “antisaccade” rule in both (B) Monkey A and (D) Monkey C. This was also the case under the “prosaccade” rule in (C) Monkey C but not in (A) Monkey A. The red lines depict the linear relationship between the drug-related changes. Also shown are Pearson's r and its associated p value.
Taken together, not only did we find a profound impact of ketamine on the delay-period directional connectivity between the lPFC and PPC, but we also observed a link between this loss of connectivity with impairment in task performance. Greater reduction in lPFC-led connectivity during the delay period predicted more severe deficits in either response accuracy or reaction time in RV and RM tasks. Whether a link also exists between PPC-led connectivity and performance remains unclear.
Ketamine has task-specific effects on rule coding by lPFC and PPC neurons
After characterizing the effects of ketamine on frontoparietal field potentials, we went on to quantify its effects on information coding in spiking activities. We recorded a total of 262 and 248 neurons from the lPFC and PPC, respectively, with a mean pre-injection firing rate of 3.02 ± 5.89 Hz (mean ± standard deviation) and 2.41 ± 3.20 Hz. We conducted a mixed-model repeated-measures ANOVA on the firing rates in four 10 min intervals before and after ketamine administration. There was a significant effect of the drug (F3,1524 = 56.15, p < 4.9 × 10−324) and no effect of area (F1,508 = 2.00, p = 0.16). Consistent with our previous report (Ma et al., 2015), ketamine increased neuronal activities in the lPFC marginally in the first 10 min (p = 0.055) and significantly in the second and third 10 min periods (p = 3.2 × 10−5). This effect was significant in each subject. In Monkey A, the increase in firing rates became significant in the second and third 10 min post-injection (p = 3.2 × 10−5), while in Monkey C, it reached significance in the third 10 min period (p = 0.0033). In the PPC, its effect started in the first 10 min (p = 0.0015) and remained so in the later intervals (p = 0.00032). A similar effect was found in Monkey C (p = 0.024, 0.0024, 0.0013 in three periods, respectively), whereas in Monkey A it strengthened from the first 10 min (p = 0.08) to the second and third 10 min periods (p = 3.6 × 10−5, 3.2 × 10−5). Additionally, in both areas, ketamine increased the coefficient of variation in interspike intervals, which reflects the irregularity in spiking activities (F3,1518 = 31.48, p < 1 × 10−324). Again, this effect was confirmed in both subjects (Monkey A, F3,1083 = 26.05, p = 3.3 × 10−16; Monkey C, F3,429 = 16.72, p = 2.7 × 10−10). Thus, ketamine increased the level as well as the irregularity in neuronal activities in the frontoparietal network.
We then quantified neuronal coding of the task rule by computing the d′ between the rules during fixation, cue presentation, delay, and perisaccadic epochs for single units recorded from the lPFC and PPC. Since RV trials had no delay period, the later portion of the cue period, which equaled the delay period in duration, was used instead. The significance of the d′ for each unit was determined against a distribution of d′ created from trial-shuffled firing rates for that unit (see Materials and Methods). In Monkey A, the percentage of units with significant d′ rose from 13% to a range of 17–24% in the delay epoch in both the lPFC and PPC. In Monkey C, this percentage increased from 9.5 to 15–16% in the PPC, but only from 5.6 to 7% in the lPFC. Given that we did not capture enough rule-selective lPFC neurons from Monkey C, conclusions from this analysis will be purely speculative, although it is known that nonselective neurons can still contribute to information coding at the ensemble level (Hyman et al., 2012).
We analyzed the effects of ketamine administration, task, and epoch in rule selectivity across all well-isolated single units (Fig. 13). In both monkeys, rule selectivity rose significantly with epoch (Monkey A, F3,933 = 42.1, p < 4.9 × 10−324; Monkey C, F3,366= 2.96, p = 0.032) and reduced following ketamine injections (Monkey A, F1,311 = 45.9, p = 6.3 × 10−11; Monkey C, F1,122 = 13.1, p = 0.00044). In Monkey A, this deleterious effect of ketamine was significant in both areas (p = 7.7 × 10−6 and 0.0014; Fig. 13A,C, gray vs black lines); in Monkey C, it was significant in the PPC (p = 0.048; Fig. 13D) and had a similar trend in the lPFC (p = 0.055; Fig. 13B).
Rule selectivity (d′) during different epochs in all single units recorded from the (A) lPFC in Monkey A and (B) Monkey C and from the (C) PPC in Monkey A and (D) Monkey C. The third epoch from RV trials is the later part of the prolonged cue period, which matched the delay period in RM trials. Solid lines, RV trials; dashed lines, RM trials. pre-injection, gray; post-injection, black. Error bars indicate the standard deviation of the mean. **p < 0.005, ***p < 0.0005.
Interestingly, in Monkey A, rule selectivity among lPFC neurons rose quickly from the cue to the delay epoch in both RV and RM tasks (p = 9.7 × 10−5 and 2.9 × 10−5; Fig. 13A, difference by color in both solid and dashed lines), which was true for PPC neurons only in the RV task (p = 0.0016; Fig. 13C, difference by color in solid but not dashed lines). After ketamine treatment, all these sudden increases in rule representation were abolished (p > 0.54). Notably, the effect of ketamine was specific to the delay (p < 0.001), and not any other epochs (p > 0.31), in both areas. We speculate that in both tasks, the red/green cues must be mapped onto pro- or antisaccade rules based on memory, before the monkeys could choose the appropriate response. It is possible that this “mapping process” takes place in the delay period in both tasks, with and without the presence of the cue, hence the rise in rule selectivity in frontal neurons. It appears that PPC neurons only participated in this process in the presence of the cue. Due to a relative scarcity of rule-selective neurons from Monkey C, this delay-specific effect could not be established and will need to be verified in future studies.
Discussion
Through an analysis of LFPs and single-unit activity in macaque monkeys before and after systemic ketamine injections, we demonstrated a connection between frontoparietal dysconnectivity and impaired performance in a rule-based antisaccade task that required cognitive control (Figs. 8–12). We also found that ketamine increased the E/I ratio in both the lPFC and PPC, as reflected in changes in the aperiodic component of LFPs (Figs. 4, 5). Additionally, ketamine increased the central frequency of true oscillatory activities between 8 and 45 Hz in the PPC across subjects, and in the lPFC in one subject (Fig. 6), and weakened rule coding in a task- or area-dependent manner (Fig. 7). In one of the monkeys, this was accompanied by a task-dependent weakening in rule coding among frontoparietal neurons during the extended cue and delay period (Fig. 13).
Like in previous reports (Skoblenick and Everling, 2012; Ma et al., 2015), response accuracy was reduced, and saccadic reaction times were prolonged within 10 min following injection, then partially recovered in an additional 20 min (Fig. 2). Modeling studies have suggested that NMDA hypofunction can lead to less accurate responses in decision-making tasks accompanied by shorter reaction times (Murray et al., 2017; Cavanagh et al., 2020; Lam et al., 2022), which is different from what we observed. One possible explanation is that the preparation for antisaccades requires different neural computations at the microcircuit level. Another possible explanation, which will be discussed further, is that the synaptic effect of ketamine is more complex than NMDAR blockade alone.
Acute ketamine administration has general effects on frontal and parietal activity
As previously reported for the lPFC (Skoblenick and Everling, 2012; Ma et al., 2015) as well as in the current study, we observed an increased level of activity in the PPC. This appears to be a general effect of ketamine on the cerebral cortex. At first glance, this is not expected from NMDAR antagonism. However, similar effects have been reported for MK-801 (Jackson et al., 2004), although a third NMDAR antagonist, phencyclidine, had no effect on firing rates on average (Zick et al., 2018). We speculate that ketamine may tilt the E/I balance toward excitation through the following mechanisms. First, ketamine is known to enhance spiking activity by increasing glutamate transmission through AMPA/kainate receptor currents (Wang et al., 2013; Moran et al., 2015). Second, it may also exert a stronger NMDAR-mediated inhibitory effect on fast-spiking parvalbumin-positive (PV+) interneurons, thereby disinhibiting excitatory neurons (Homayoun and Moghaddam, 2007; McNally et al., 2011). Third, a metabolite of ketamine, (2S,6S;2R,6R)-hydroxynorketamine, stimulates AMPA receptors (Zanos et al., 2016) and can further enhance the level of activity in cortical neurons regardless of their type. Future studies will verify if this effect may apply to sensory and motor cortical areas and to subcortical structures. Consistent with its excitation-enhancing effect, and independent of its effect on true neural oscillations, ketamine tilted the balance in the aperiodic component of LFPs so that spectral power was strengthened in the gamma range and weakened in lower frequencies in both the lPFC and PPC (Figs. 4, 5). This bidirectional shift in spectral power is consistent with our previous finding on LFP powers in macaque lPFC (Ma et al., 2018), as well as reports on healthy people receiving an acute, low dose of NMDA antagonists (Plourde et al., 1997; Hong et al., 2010). A recent systematic review found that ketamine induced the same pattern of changes in EEG and MEG spectra in both patients with depression and healthy controls (Le et al., 2024). In contrast, ketamine had no consistent effect on the amplitude of LFP oscillations in any frequency bands (Fig. 6). Notably, it did induce an upward shift in the central frequency of the midrange band from 8 to 45 Hz, so a “beta-band oscillation” centered at 28 Hz may become a “low-gamma oscillation” centered at 32 Hz. This can contribute to the bidirectional change in alpha/beta and gamma spectral power in previous studies that used predetermined frequency bands (Cornwell et al., 2012; Muthukumaraswamy et al., 2015; Nugent et al., 2019; Gilbert et al., 2020, 2022), in addition to the aperiodic changes. Overall, our findings support the effect of ketamine in enhancing synaptic excitation across neuronal populations, rather than in modulating rhythmical oscillations orchestrated by a specific cell type (Lodge et al., 2009; Sohal et al., 2009; Takada et al., 2014). This increase in synaptic transmission may have a therapeutic value via synaptic potentiation, resulting in a long-lasting enhancement in sensory-evoked response hours after the end of any acute effect, only in patients responding to ketamine but not in nonresponders (Cornwell et al., 2012). We suggest that future studies searching for biomarkers that predict or correlate with the effectiveness of ketamine as an antidepressant can benefit from analyzing the aperiodic and oscillatory components of EEG and MEG separately. It will also be important to conduct recordings after longer intervals to determine how long this effect on E/I balance would last.
Acute ketamine administration impaired rule coding in frontoparietal neurons and oscillations
Consistent with our previous studies involving lPFC alone (Skoblenick and Everling, 2012; Ma et al., 2015), we observed a deleterious effect of ketamine on rule coding in both frontal and parietal neurons in one subject, although few rule coding neurons were detected in the second monkey. In the RM task, this could lead to a direct impact on short-term memory. In the RV task, while the cue remained visible, rule representation in the frontoparietal neurons was enhanced in the late phase of the cue period (Fig. 13). This enhanced rule coding could be correlated with the retrieval of the pro/antisaccadic rule to guide action selection. That is, a memory process was engaged even though short-term memory for the cue was not required.
Several studies have demonstrated that both the lPFC and PPC contribute to short-term working memory by enhancing activity during the delay period (Friedman and Goldman-Rakic, 1994; Chafee and Goldman-Rakic, 1998; Gruber and von Cramon, 2001; Li et al., 2021; Dang et al., 2022). Meanwhile, the two areas differ in anatomical and functional organization (Katsuki et al., 2014; Gonzalez-Burgos et al., 2019; Arion et al., 2022) and their roles in working memory (Katsuki and Constantinidis, 2012). Here, we observed an enhancement of rule encoding in the lPFC across tasks, whether the cue remained visible or not, and in the PPC only when the cue remained visible (RV task). This is consistent with previous findings that neural activity in the PPC is less generalized across tasks compared with the lPFC (Sarma et al., 2016; Masse et al., 2017). Alternatively, the lack of enhanced rule encoding at the population level may be related to the susceptibility of PPC neurons to distraction (Friedman and Goldman-Rakic, 1994; Constantinidis and Steinmetz, 1996; Chafee and Goldman-Rakic, 1998; Rawley and Constantinidis, 2009; Qi et al., 2015; Qi and Constantinidis, 2015). Our findings suggest that working memory in the intact brain is supported by cortical networks that include the lPFC and PPC (Christophel et al., 2017; Leavitt et al., 2017; Sreenivasan and D’Esposito, 2019; Mejías and Wang, 2022). We also showed that ketamine impaired rule-related memory processes through widespread effects on the frontoparietal network (Kummerfeld et al., 2020). It remains to be tested whether the effect of ketamine extends to a greater cognitive control network involving additional cortical and subcortical areas in the primate brain (Haber and Behrens, 2014; Hori et al., 2020).
Going beyond rule coding at the neuronal level, we found that ketamine reduced rule selectivity in oscillatory activities between 45 and 80 Hz, while having no consistent effect on rule information in lower frequency bands (Fig. 7). This was observed in the lPFC in both monkeys, and in the PPC only in Monkey A. Notably, this effect was completely dissociated from the general effect of ketamine on LFP oscillations, in which alpha/beta and low gamma bands were affected but not high gamma (Fig. 6). This is consistent with the theory that gamma rhythm plays a role in higher cognitive processes by organizing task-related spiking activities (Miller et al., 2018). Additionally, we found a highly consistent ketamine-induced reduction in frontoparietal coherence in the high gamma band in the delay and preparation period preceding antisaccades, but not prosaccades (Fig. 9). The same was not found within either area (Fig. 8). These findings support the view that gamma rhythm can serve as a channel for long-range communication in the brain (Gregoriou et al., 2009; Bosman et al., 2012; Fernández-Ruiz et al., 2021).
Acute ketamine administration led to frontoparietal dysconnectivity
Since cortical afferents often synapse with apical dendrites, which have a high density of NMDAR (Monaghan and Cotman, 1985; Rosier et al., 1993), ketamine is expected to disrupt frontoparietal functional connectivity (Uhlhaas and Singer, 2012). Indeed, in a human fMRI study, ketamine reduced frontoparietal connectivity during a spatial working memory task (Driesen et al., 2013). Similarly, in MEG from humans performing a visuomotor task, Muthukumaraswamy et al. (2015) found that low doses of ketamine reduced connectivity, especially in the frontoparietal direction and in the alpha and beta range, and used biophysical modeling to reveal that this reduction was likely mediated by both NMDARs and AMPARs. Consistent with these findings, we found a significant reduction in lPFC–PPC LFP coherence following ketamine injections (Fig. 9), in different frequency bands depending on the task and rule. Moreover, we observed a robust reduction in time-domain connectivity in both top-down and bottom-up directions across tasks, rules, and subjects (Fig. 10).
By contrast, Kummerfeld et al. (2020) found a reduction in frontoparietal communication accompanied by an increase in parietofrontal connectivity, using a causal discovery analysis of LFPs recorded from macaque monkeys. This discrepancy may be explained by the difference in the effect of ketamine and phencyclidine on glutamatergic transmission, going beyond NMDA antagonism, as discussed above. It may also be related to the differences in the type of brain signals (e.g., fMRI vs LFP) and analytical methods.
The need for cognitive control, as well as better performance on such tasks, is known to be correlated with enhanced frontoparietal communication (Repovs et al., 2011; Crowe et al., 2013; Sheffield et al., 2015). Reduced frontoparietal connectivity due to NMDA antagonism has been linked to errors on a contexting-processing task that required cognitive control (Kummerfeld et al., 2020). In agreement with this literature, we found a greater loss in frontoparietal connectivity during the preparation period preceded a larger drop in antisaccade performance, specifically in the rule-visible condition (Fig. 11). Similarly, in the rule-memorized condition, a greater loss in frontoparietal connectivity during the delay period preceded a bigger increase in antisaccade reaction time (Fig. 12). Together, these findings support a critical role of frontoparietal communication during cognitive control processes, in both humans and macaque monkeys.
In the context of understanding the antidepressant effect of ketamine, it should be noted that these findings pertain to healthy humans and monkeys only. Patients suffering from major depressive disorder (MDD) are known to have altered connectivity in several functional networks in the brain (Kaiser et al., 2015; Goya-Maldonado et al., 2016) and can be divided into subgroups based on their diverse changes in functional connectivity (Tozzi et al., 2024). Ketamine has been suggested to exert its antidepressant effect by enhancing glutamatergic transmission (Zanos et al., 2016) and by NMDAR-related mechanisms (Zanos et al., 2023), both of which can in turn contribute to restoring connectivity in the frontoparietal control network and other functional networks. Future clinical studies will be necessary to provide a complete understanding of any long-term effect of ketamine treatment on both cognition and functional connectivity in patients with MDD.
Conclusion
Through a comprehensive analysis of both LFPs and spiking activities in the frontoparietal network in macaque monkeys performing a rule-based antisaccade task, we demonstrated that acute ketamine treatment altered the excitatory–inhibitory balance and task-related oscillation and synchrony. Notably, ketamine-induced loss in frontoparietal communication was correlated with the deterioration in performance. Given the increasing popularity in the use of ketamine as an antidepressant (Marcantoni et al., 2020), future studies are required to better characterize any long-lasting effect on the primate cognitive control network following repeated exposure to the drug.
Footnotes
- Received May 31, 2023.
- Revision received October 19, 2024.
- Accepted October 22, 2024.
This work was supported by the Canadian Institutes for Health Research, Radboud University Nijmegen, and Canada Research Chairs (Chaires de recherche du Canada).
The authors declare no competing financial interests.
- Correspondence should be addressed to Liya Ma at liyama{at}yorku.ca.
- Copyright © 2024 the authors