Abstract
Sustained attention describes our ability to keep a constant focus on a given task. This ability is modulated by our physiological state of arousal. Although lapses of sustained attention have been linked with dysregulations of arousal, the underlying physiological mechanisms remain unclear. An emerging body of work proposes that the intrusion during wakefulness of sleep-like slow waves, a marker of the transition toward sleep, could mechanistically account for attentional lapses. This study aimed to expose, via pharmacological manipulations of the monoamine system, the relationship between the occurrence of sleep-like slow waves and the behavioral consequences of sustained attention failures. In a double-blind, randomized-control trial, 32 healthy human male participants received methylphenidate, atomoxetine, citalopram or placebo during four separate experimental sessions. During each session, electroencephalography (EEG) was used to measure neural activity while participants completed a visual task requiring sustained attention. Methylphenidate, which increases wake-promoting dopamine and noradrenaline across cortical and subcortical areas, improved behavioral performance whereas atomoxetine, which increases dopamine and noradrenaline predominantly over frontal cortices, led to more impulsive responses. Additionally, citalopram, which increases sleep-promoting serotonin, led to more missed trials. Based on EEG recording, citalopram was also associated with an increase in sleep-like slow waves. Importantly, compared with a classical marker of arousal such as α power, only slow waves differentially predicted both misses and faster responses in a region-specific fashion. These results suggest that a decrease in arousal can lead to local sleep intrusions during wakefulness which could be mechanistically linked to impulsivity and sluggishness.
SIGNIFICANCE STATEMENT We investigated whether the modulation of attention and arousal could not only share the same neuromodulatory pathways but also rely on similar neuronal mechanisms; for example, the intrusion of sleep-like activity within wakefulness. To do so, we pharmacologically manipulated noradrenaline, dopamine, and serotonin in a four-arm, randomized, placebo-controlled trial and examined the consequences on behavioral and electroencephalography (EEG) indices of attention and arousal. We showed that sleep-like slow waves can predict opposite behavioral signatures: impulsivity and sluggishness. Slow waves may be a candidate mechanism for the occurrence of attentional lapses since the relationship between slow-wave occurrence and performance is region-specific and the consequences of these local sleep intrusions are in line with the cognitive functions carried by the underlying brain regions.
Introduction
Fluctuations of sustained attention are omnipresent (Esterman and Rothlein, 2019) and lead to unresponsiveness, sluggishness, and impulsivity (Oken et al., 2006; O'Connell et al., 2009; Riley et al., 2017). Physiologic arousal appears as a major determinant of these attentional failures. While diminishing arousal following extended periods of wakefulness typically leads to behavioral instability (Doran et al., 2001; Riley et al., 2017), hyperarousal can also lead to attentional failures (Arnsten, 2000; Swann et al., 2013). Thus, not all sustained attention failures have the same underlying origin and there is a need to identify and parse the distinct processes that are at play.
The same neuromodulators regulate physiological arousal and sustained attention (Oken et al., 2006; Thiele and Bellgrove, 2018). Noradrenaline release leads to broad cortical activation via the inhibition of sleep-promoting systems and the activation of wake-promoting systems (Jones, 2005). Noradrenaline is also released in response to the presentation of salient stimuli, improving their encoding and processing (Robbins, 1997; Sara, 2009). Accordingly, a pharmacologically-induced increase or decrease in noradrenaline results in stronger (Dockree et al., 2017) or weaker sensory responses (Gelbard-Sagiv et al., 2018), respectively. However, too much noradrenergic activity can impair executive functions (Arnsten, 2000, 2010), suggesting a U-shaped relationship between noradrenaline and performance (Sara and Bouret, 2012).
Dopamine's influence on performance also follows an inverted U-shape (Maffei and Angrilli, 2018) and is difficult to separate from that of noradrenaline (Robbins, 1997; Lee and Dan, 2012). Both neuromodulators innervate related areas, share biosynthetic and intracellular signaling pathways, and are occasionally released simultaneously (Ranjbar-Slamloo and Fazlali, 2020). An integrative model posits that dopamine decreases the amount of noise whereas noradrenaline enhances signal, both contributing to an increased signal-to-noise ratio (SNR; Arnsten and Rubia, 2012).
Serotonin also plays a role in the regulation of arousal (Jones, 2005). It indexes the build-up of sleep pressure (Jouvet, 1999; Borbély et al., 2016) and could facilitate sleep onset (Oikonomou et al., 2019). Serotonin can modulate attention directly and indirectly, by inhibiting the cholinergic system which plays a key role in the maintenance of attention (Jones, 2005; Sparks et al., 2018). Serotoninergic effects appear region and dose-specific (Thiele and Bellgrove, 2018) and low tonic activity in serotonin could improve attention by reducing impulsivity whereas high tonic activity could decrease the activation of prefrontal areas commonly activated in a sustained attention task (Sturm and Willmes, 2001; Wingen et al., 2008).
In parallel, electrophysiological [electroencephalography (EEG)] markers such as α oscillations have been used to investigate the relationship between physiological arousal and attention (Cajochen et al., 1995; Oken et al., 2006; Peylo et al., 2021). A suppression of α oscillations has often been used as a proxy for cortical excitability (Romei et al., 2008). However, the relationship between α oscillations and arousal is not monotonic, as α oscillations also decrease with drowsiness (Kalauzi et al., 2012). Although increases in α power precede attentional lapses (O'Connell et al., 2009), the occurrence in wakefulness of sleep-like slow waves, a hallmark of sleep, could represent a more reliable marker of attentional lapses under low arousal (Andrillon et al., 2019). Indeed, sleep and wakefulness are not mutually exclusive states (Nobili et al., 2012), and, after an extended period of wakefulness, an increase in the power of slow (δ: [1, 4] Hz and θ: [4, 7] Hz) oscillations (Cajochen et al., 2002), or the occurrence of individual high amplitude slow waves (<7 Hz) can be detected in scalp EEG (Hung et al., 2013; Bernardi et al., 2015). The neurophysiological characterization of these slow waves is still incomplete and previous studies focused on waves within the θ range (Hung et al., 2013; Bernardi et al., 2015), δ range (Quercia et al., 2018) or a combination of both (Vyazovskiy et al., 2011; Andrillon et al., 2021). At the neuronal level, these waves have been associated with periods of neuronal silencing in rodent intracranial recordings (Vyazovskiy et al., 2011), but a similar observation in human intracranial recordings is missing. At the behavioral level, these waves have been associated with errors (Hung et al., 2013; Bernardi et al., 2015) and lapses of attention (Andrillon et al., 2021). Here, we aimed to test whether the occurrence and modulation of these slow waves could expose the relationship between arousal and attention.
To do so, monoamine (dopamine, noradrenaline and serotonin) agonists were administered to participants while they performed a sustained attention task (Fig. 1). We investigated the differential impact of both arousal-promoting (methylphedinate [MPH], atomoxetine [ATM]) and arousal-reducing (citalopram [CIT]) drugs. We also examined a diverse set of EEG markers of attention and arousal, including α oscillations and EEG markers of sleep intrusions, and their relationship to behavioral performance.
Experimental protocol. a, Each experimental session started with participants being administered one out of four treatments (PLA: placebo; MPH: methylphenidate; ATM: atomoxetine; CIT: citalopram) and completing the first Visual Analog Scale [VAS] (time = 0). The administration of the treatments was randomized across sessions (within-subject design). A second VAS was completed before starting the continuous temporal expectancy task (CTET), 90 min after treatment administration. Finally, after the CTET (and another task not analyzed here) and 180 min after treatment administration, participants completed a third VAS. High-density EEG was recorded during the CTET. b, The CTET task was used to test sustained attention. In the CTET, participants are instructed to monitor a stream of visual stimuli (black and white checkboards) displayed on a screen placed in front of them. There are two types of stimuli: target stimuli for which participants are instructed to press a response button, and nontarget stimuli that do not require a response. The only distinguishing factor between the nontarget and target stimuli is the duration of the stimulus presentation (nontarget: 800 ms; target: 1120 ms).
Materials and Methods
We re-analyzed a dataset of human participants who completed the continuous temporal expectancy task (CTET; Dockree et al., 2017), a monotonous task wherein participants are required to monitor visual checkerboard stimuli for intermittent targets defined by their longer presentation (O'Connell et al., 2009). Methylphenidate (MPH) was used to increase both noradrenaline and dopamine levels globally across the cortex while atomoxetine (ATM) was used to increase noradrenaline and dopamine more specifically in the prefrontal cortex (Bymaster, 2002; Chamberlain et al., 2009; Koda et al., 2010; Farr et al., 2014; Kowalczyk et al., 2019). Lastly, citalopram (CIT) was administered to increase serotonin. This re-analysis was approved by the Monash University Human Research Ethics Committee (project number 4663). Only data obtained under MPH and placebo have been previously reported (Dockree et al., 2017). This previous report included the analysis of some behavioral variables (misses, RT) and EEG indices [α power, steady state visually evoked potentials (SSVEP), event-related potential (ERP)] that we also analyzed here.
Participants
Forty (N = 40) healthy participants were recruited in accordance with the ethical guidelines of the University of Queensland (project number 2008000463). Participants were all right-handed white males, aged 18–45 years (μ = 24.3 years, σ = 5.6 years) who were nonsmokers, had no history of neuropsychiatric disorder, and were not under the influence of recreational drugs or psychoactive medication at the time of the study. They were screened for psychiatric disorders using the Mini-International Neuropsychiatric Interview Screen (Sheehan et al., 1998) and completed the Conners' Adult ADHD Rating Scales (Conners et al., 1999). The mean for the total ADHD symptoms in the group was 49.73 (σ = 11.89), below the threshold for ADHD diagnosis (threshold: 65). Eight (N = 8) participants were discarded from our analyses (contraindications to the study medications: 4; data loss: 1; technical issues: 3) leaving 32 participants for our analyses. A total of 30/32 participants completed all four versions and the remaining two had one missing session (but completed the placebo session: PLA: N = 32; CIT: N = 31; ATM: N = 31; and MPH: N = 32).
Experimental design
A randomized, double-blinded, placebo-controlled paradigm was used in which participants partook in four different experimental sessions. Participants were required to fast for at least 1 h before each session. They were randomly administered a different drug, 90 min before the commencement of cognitive testing (Fig. 1a). These drugs included ATM (60 mg), MPH (30 mg), CIT (30 mg), or a placebo (dextrose). ATM inhibits the reuptake of noradrenaline and dopamine, and thus, is commonly used to increase noradrenaline and dopamine levels (Bymaster, 2002; Ding et al., 2014; Bédard et al., 2015). MPH inhibits noradrenaline and dopamine transporters, resulting in an increase in cortical noradrenaline and striatal dopamine levels (Volkow et al., 2012; Farr et al., 2014; Kowalczyk et al., 2019). CIT, a selective serotonin reuptake inhibitor (SSRI), increases serotonin levels (McKie et al., 2005). Doses were selected according to the reported effects on cognition from past studies (Aron et al., 2003; Graf et al., 2011; Nandam et al., 2014). To avoid drug interactions, each session took place at least a week apart to allow the washout of drugs. Sessions took place at the same day of the week for each participant, and at the same time of the day (average: 12.6 h from midnight, range [10.0, 15.6] h) with no difference in timing between treatment conditions (one-way ANOVA: F(3) = 0.11, p = 0.95).
After ingesting the medication, participants waited for 60 min. They were then equipped with a high-density EEG cap (64 active electrodes cap, ActiveTwo BioSemi system), which took ∼30 min. This ensured a 90-min delay between medication administration and the commencement of testing, allowing the three treatments to reach peak plasma levels (Sauer et al., 2005; Chamberlain et al., 2009; Nandam et al., 2014; Kowalczyk et al., 2019). EEG recording lasted for ∼90 min while the participants completed two tasks, the order of which was counterbalanced: a CTET (O'Connell et al., 2009; Dockree et al., 2017) and a standard Eriksen flanker task (Barnes et al., 2014). We focused here on the CTET data.
For each session, participants first completed a visual analog scale (VAS; Bond and Lader, 1974) to report their sleepiness, calmness, and contentedness (Norris, 1971) across three intervals: immediately before drug administration, before cognitive testing (+90 min), and after cognitive testing (+180 min). We analyzed here only the results from the sleepiness scale (Fig. 2).
CTET
The CTET measures sustained attention. Participants were instructed to monitor a centrally presented patterned stimuli with a gray background (Fig. 1b). The stimuli consisted of a square with an 8 cm2 dimension. This was further divided into a 10 by 10 grid of squares, each diagonally split in half into white and black. Each time the frame changed, the stimuli randomly altered its orientation by 90° in a clockwise or counter-clockwise direction, producing four different arrangements. A white cross was present at the center of the frame that participants were instructed to fixate on to limit eye movements. Each pattern was presented for a duration of either 800 ms (standard stimuli) or 1120 ms (target stimuli). Between two target stimuli, 7–15 nontarget stimuli were pseudo-randomly presented. This equated to 5.6–12 s between each target presentation. Thus, this task required participants to continuously concentrate on the stream of visual information presented to them. In addition to the transition between patterns, each pattern flickered at 25 Hz during the stimulus presentation window. This allowed the generation of SSVEP. Participants were required to monitor the duration of each stimulus, pressing a button for each target stimulus detected. Before each session, all participants were required to exhibit 100% accuracy during an initial practice session, consisting of two separate practice blocks. In the first of these blocks, three targets were randomly interspersed among 25 standard stimuli without the 25-Hz flicker. For the second practice block, the flicker was added. If participants missed one or more target stimuli, the practice was performed again. If the participant still failed to identify all the targets, they were excluded from the experiment. Following the practice, participants then completed 10 experimental blocks, with each block containing 225 frames, 18–22 target stimuli presentations and lasting ∼3 min. Participants were given a 1-min break in between blocks.
Behavioral data
For the CTET, we computed the proportion of missed targets (target trials which were not associated with a response) and false alarms (nontarget trials associated with a response). We also computed the reaction times of correctly detected target trials. A target stimulus was considered missed if a response was not registered from 800 ms after target onset and 1600 ms after the target offset (minimal reaction time: 0.8 s; maximal reaction time: 2.72 s). A false alarm was defined by the presence of a response to a nontarget stimulus without any target stimulus in the two preceding trials. Reaction times for target trials were computed from the stimulus onset. These behavioral variables were compared across drug treatments (Fig. 3).
Indices of participants' performance were extracted for each CTET block as follows: percentage of missed targets, percentage of false alarms and average reaction times for correctly detected targets (in seconds). To integrate the performance on both target and nontarget trials in one single variable, we also computed the d′ at the block level, following the Signal Detection Theory (Macmillan, 1993).
EEG
Data acquisition
EEG data were recorded using 64 active scalp electrodes and an ActiveTwo BioSemi amplifier. EEG was sampled at 1024 Hz. Eye-movements were recorded with pairs of electrodes (electrooculogram; EOG) placed above and below the left eye (vertical eye-movements), and at the outer canthus of each eye (horizontal eye-movements). EEG data were subsequently analyzed using the Fieldtrip toolbox (Oostenveld et al., 2011; version 9bfe2f49f) and custom code.
Preprocessing of EEG data
Raw EEG data were epoched on a [−0.5, 1.8] s window around stimulus onset. The average voltage computed across the entire epoch was removed and EEG data were lowpass-filtered below 40 Hz (Butterworth filter at the fourth order) and notch-filtered at 50 Hz using a discrete Fourier transform to remove line noise. EEG data were then down-sampled at 256 Hz and baseline corrected ([−0.5, 0] s). These preprocessed data were then visualized to detect faulty electrodes and artefacted trials (based on the signal variance). The former were interpolated using the “weighted” method from Fieldtrip and the latter were discarded. Finally, EEG data were re-referenced to the average of all electrodes and an independent component analysis (ICA) decomposition was performed to visually identify components associated with ocular artefacts (blinks and saccades). These components were removed from the data epoched on trials.
In parallel, we also epoched raw EEG data on experimental blocks (from –2 s after the onset of the first trial to 1s after the offset of the last trial of the block). Data were demeaned, lowpass-filtered below 40 Hz (Butterworth, fourth order), highpass-filtered above 0.1 Hz (Butterworth, fourth order) and notch-filtered at 50 Hz (discrete Fourier transform). Data were then resampled at 256 Hz and baseline corrected ([−0.5, 0] s from block onset). Faulty electrodes detected on trial-epoched data were interpolated and ocular ICA components (obtained from the data epoched on trials) were removed. Before ICA removal, EEG data were re-referenced to the average of all electrodes.
Upon visual inspection, an average of 31 ± 2.5 trials were rejected in the PLA sessions (mean ± standard error of the mean [SEM]), 28 ± 2.6 in the MPH, 38 ± 4.1 in the ATM and 35 ± 3.8 in the CIT sessions. Consequently, an average of 2250 ± 0.1 trials were included in the PLA sessions, 2249 ± 0.2 in the MPH, 2243 ± 6.5 in the ATM, and 2250 ± 0.1 in the CIT sessions. No segments were rejected based on visual inspection in the block-epoched data.
We used trial-epoched data for the analysis of Event Related Potentials (ERPs) and block-epoched data for the analysis of the power spectrum and the detection of slow waves. For topographical analyses, peripheral electrodes (Iz, P9, and P10) were removed, resulting in 61 electrodes being analyzed.
ERPs
ERPs were extracted for target and nontarget trials separately. Data were re-aligned according to stimuli's offset ([−0.2, 0.8] s) to take into account the different duration of target and nontarget trials. The EEG data were baseline-corrected from −0.2s to 0s before the offset. For each participant, the corresponding ERPs were averaged across trials for the different drug treatments. In Figure 4, we show the corresponding ERPs for electrodes Fz and Pz. Based on these ERPs and previous findings (Dockree et al., 2017), we extracted the average amplitude of the P3 potential typically elicited by Target trials (in comparison with nontarget trials). The P3 amplitude was defined as the difference in amplitude between the ERP associated with Target trials and the ERP associated with nontarget trials averaged on a [0, 0.3] s window postoffset (Fig. 4a).
Spectral analysis
We computed the power spectral density (PSD) of each EEG electrode using data epoched on experimental blocks (Fig. 5 and 6). This was done using Welch's method with windows of 10s, a 50% overlap between windows and a frequency resolution of 0.1 Hz between 2 and 40 Hz. The PSD was log-transformed (base-10 log) for further analysis. We extracted the α power by averaging for each participant, session, block, and electrode the power within the [8, 11] Hz frequency band. This estimation of the strength of α oscillation was compared across drug treatments and used to predict behavioral performance.
We also extracted the strength of the SSVEP frequency tag by computing the SNR of each frequency compared with its neighbors. This was done to identify sharp peaks in the power spectrum (associated with the frequency tagging), departing from the 1/f trend. In practice, the PSD for each frequency was divided by the average PSD on a [−0.5, −0.1] and [0.5, 0.1] Hz window around this frequency. This ratio was then log-transformed for further analysis (base-10 log). We extracted the SNR at 25 Hz (frequency of the SSVEP) for each participant, session, block, and electrode. This estimation of the strength of the frequency tagging was then compared across drug treatments.
Sleep-like slow waves
The presence of sleep-like slow waves in the EEG was assessed as previously shown (Andrillon et al., 2021), based on established algorithms used to detect slow waves in sleep (Riedner et al., 2007). In practice, EEG data re-referenced to the average of the electrodes closest to the mastoids (TP7 and TP8, labeled as M1 and M2 in the raw data) were bandpass filtered between 1 and 10 Hz using a type-2 Chebyshev filter. For each block and electrode, we then centered the filtered signal around 0 by removing the average voltage computed across the entire block. Individual waves were then detected by locating all the negative peaks between two negative zero-crossings, which were, respectively, defined as the beginning and end of the wave. The positive peak of the wave was defined as the maximum voltage between this beginning and end. For each wave, we computed its duration, frequency, and peak-to-peak amplitude. Waves with a frequency above 7 Hz (i.e., outside the δ/θ range: [1, 7] Hz), a positive peak above 75 µV or within 1s of a high-amplitude event (absolute amplitude >150 µV) were discarded.
Finally, we focused on the largest-amplitude slow waves. To do so, we computed the amplitude threshold of the top 10% of the largest waves recorded in the placebo session using their peak-to-peak amplitude. This threshold (31.8 ± 1.4 µV, mean ± SEM across participants) was computed for each individual and each electrode and was used to define large-amplitude slow waves in all sessions. Thus, the number of slow waves can be compared within-subject, using the placebo session as a reference. This approach allowed us to focus preferentially on the within-subject differences induced by the administration of the different treatments but can possibly reduce interindividual differences. Setting the threshold per electrode also allows to take into account, in the detection of slow waves, the average signal amplitude of each electrode and to avoid the dominance of frontal electrodes for example, which can show higher amplitudes notably because they are far from the reference electrodes (here, mastoids).
We decided to focus on waves detected both in the δ and θ range, in keeping with the previous literature (Vyazovskiy et al., 2011; Hung et al., 2013; Bernardi et al., 2015; Quercia et al., 2018; Andrillon et al., 2021). We will thus refer here to slow waves as the combination of δ and θ waves. Including both δ and θ waves allows us to take into account possible variations in the duration of off periods between sleep and wakefulness. Nonetheless, after selecting the waves with the largest amplitude, we extracted the frequency of each wave (1/duration of the wave) and computed the proportion of waves in the δ ([1, 4] Hz) and θ ([4, 7] Hz) range. The vast majority of detected slow waves were in the δ range (90.2 ± 0.7%, mean ± SEM across participants). We finally computed the average number of these slow waves per minute for each experimental block of each session (Fig. 7).
Statistical analyses
VAS sleepiness scores were analyzed with a two-way repeated measure ANOVA, with time points (t = 0, +90, and +180 min after drug administration; Figs. 1a, 2) and treatment (MPH, ATM, CIT, and PLA) coded as within-subject effects. We discarded two participants who did not have all four experimental sessions from this analysis. The ANOVA was performed with the 'anova_test' function from the 'rstatix' package in R. We then examined separately post hoc comparisons by time points and within treatments, and corrected the corresponding p-values for multiple comparisons using the Bonferroni method.
Effect of treatments of subjective sleepiness. Participants were asked to complete a VAS about their subjective sleepiness at three time-points during each recording session: at drug administration (t = 0 min), before the CTET (t = 90 min), and after the CTET (t = 180 min). The raincloud plot (see Materials and Methods) for each drug treatment (placebo PLA: gray, methylphenidate MPH: orange, atomoxetine ATM: purple, citalopram CIT: green) and time is shown.
Linear mixed effect models (LMEs) were implemented in R using the 'lme4' package (Bates et al., 2015) to determine the effects of treatments (categorical fixed effect) on several variables of interest: proportion of misses, proportion of false alarms, reaction times on target trials and slow wave density (averaged across all electrodes). These variables were computed for each block of each session. Subject identity was considered a random categorical effect. We report the estimate and p-values for the comparison of the different drugs (MPH, ATM, and CIT) with the placebo session.
When testing the influence of the time spent on the task (Fig. 3), we added a fixed effect of block number as well as an interaction component with the effect of neuromodulation. The significance level of the interaction was assessed by comparing the model with and without the interaction component using the R Studio 'anova' function from the 'stat' package. We report the likelihood ratio test outcomes (χ2 and p-values) for these tests.
Examining the influence of treatments on α power ([8, 11] Hz log-PSD), frequency tagging (SNR at 25 Hz) and ERPs ([0, 0.3] s ERP amplitude after offset) for each electrode leads to the multiplication of nonindependent statistical tests. We used a cluster-permutation approach to mitigate this issue (Maris and Oostenveld, 2007). For each electrode, we computed the t value and p-value corresponding to the paired comparison (t test) for the variable of interests (α power, frequency tagging and ERPs) along contrasts of interest (MPH vs PLA, ATM vs PLA, and CIT vs PLA). To examine the effect of treatment on slow-wave density at the electrode level, we used a similar approach but using LMEs to consider the non-normal distribution of slow-wave density. t Values were derived from the LMEs (with treatments as a categorical fixed effect and subject identity was considered a random categorical effect) for the same contrasts (MPH vs PLA, ATM vs PLA, and CIT vs PLA). Clusters of electrodes were defined as neighboring electrodes with p-values below 0.05 (cluster α). For each of these clusters, we computed the t value of the cluster by summing the t values of the electrodes included in the cluster. Once the clusters were defined and these t values were computed, this procedure was repeated with permuted datasets (N = 1000 permutations). Following these permutations, we compared the t values of the original clusters (nonpermuted dataset) to the distribution of the t values of the permuted clusters and computed the Monte Carlo p-value (pcluster) of each cluster. Clusters with a pcluster below a threshold of 0.05 (one-sided test) were identified as the significant clusters.
When analyzing the effect of slow-wave density and α power on behavior (proportion of misses, proportion of false alarms, and reaction times on target trials), we implemented LMEs. Slow-wave density or α power were used as a fixed effect and subject identity was coded as a categorical random effect. We fitted one model per electrode, and we extracted the t value and p-value estimated by the model for the effect of slow-wave density or α power. Clusters of neighboring electrodes were defined as electrodes with p-values below 0.05 (cluster α). Once the clusters were defined, a comparison was again performed with permuted datasets, and we used a Monte Carlo p-value threshold of 0.05 (one-sided test) to identify the significant clusters (pcluster).
Finally, to be able to interpret null results obtained for VAS scores or behavioral indices of performances (misses, false alarms and reaction times), we computed the Bayes factor for the null hypothesis (BF01) using Bayesian ANOVAs (fixed effect of drug treatment) implemented in JASP (JASP 0.16). A BF01 above three is interpreted as significant evidence for the null hypothesis.
Graphical representation
In Figures 3 and 7a, we used “raincloud plots” to show the distribution of our variable of interests at the sample level. These plots show a combination of boxplots, smoothed density and individual datapoints. For boxplots, the central thick horizontal line shows the median, the lower and upper limit of the box show the first and third quartiles, respectively, and the lower and upper limit of the vertical lines (whiskers) show the minimum and maximum values, respectively. On the left-hand side of the boxplots, individual averages are shown as circles. A smoothed density distribution is also shown on the right-hand side of the boxplots. These plots were created with the raincloud plot package for R (Allen et al., 2021).
Code accessibility
All code used for this study are available at: https://github.com/andrillon/LS_NeuroM.
Results
Subjective sleepiness
To verify that the pharmacological interventions impacted subjective sleepiness, participants completed a VAS three times during each experimental session: at drug administration (t = 0 min), before the start of the sustained attention task (t = 90 min), and after the task (t = 180 min). A two-way repeated measures ANOVA was conducted to examine the effects of treatment (PLA, MPH, ATM, or CIT) and time (0, 90, and 180 min after drug administration) on subjective sleepiness. Thirty participants were included in this analysis after two subjects were excluded because of missing data (one experimental session missing in each participant; see Materials and Methods). This ANOVA revealed a main effect of time (F(1.49,43.28) = 22.4, p < 0.001) and treatment (F(3,87) = 6.3, p < 0.001) on sleepiness as well as a significant two-way interaction between treatment and time (F(6,174) = 6.5, p < 0.001). Pairwise post hoc comparisons by time points (t tests, Bonferroni-corrected) indicated that there were no significant differences in sleepiness across drug treatment at the beginning of the protocol (drug administration: F(3,87) = 0.36, padjusted = 1) but there were significant differences before and after the task (F(3,87) = 5.0, padjusted = 0.009 and F(3,87) = 10.6, padjusted < 0.001, respectively), confirming the effectiveness of the sustained attention protocol in engendering sleepiness. Pairwise post hoc comparisons within treatments (t tests, Bonferroni-corrected) indicated that this effect of treatment was driven by an absence of an increase in sleepiness for MPH, in keeping with the stimulant effect of MPH (F(2,58) = 2.0, padjusted = 0.59), compared with PLA, CIT, and ATM (all padjusted < 0.01). However, the administration of ATM or CIT did not significantly alter the subjective experience of sleepiness across the protocol compared with PLA.
Sustained attention performance
We next sought to examine the influence of the pharmacological manipulations on behavioral measures of sustained attention. To do so, we examined behavioral performance during the CTET, a task designed to monitor participants' ability to pay careful attention to a continuous stream of visual information (O'Connell et al., 2009). We focused on three behavioral variables: the proportion of missed target trials (misses), the proportion of incorrect responses to nontarget trials (false alarms) and the reaction times for target trials (Fig. 3).
Pharmacological modulations of behavioral performance. Left panels show the sample and individual averages computed across all CTET blocks (N = 10 blocks) for each treatment (placebo PLA: N = 32; citalopram CIT: N = 31; atomoxetine ATM: N = 31; and methylphenidate MPH: N = 32) in the form of raincloud plots (see Materials and Methods). Right panels show the sample averages for each drug treatment and block separately. a, Percentages of missed target trials (misses). b, Percentages of false alarms. c, Averaged reaction times on correctly detected targets. On the left panels, stars denote the significance level (ns: nonsignificant, *p < 0.05, **p < 0.01, ***p < 0.001) of the difference between placebo and the three treatments used in this protocol (MPH vs PLA, ATM vs PLA, CIT vs PLA), as determined by linear mixed-effect models. On the right panels, stars denote the significance level (ns: nonsignificant, *p < 0.05, **p < 0.01, ***p < 0.001) of the interaction between block number and drug treatment, as determined by linear mixed-effect models.
To examine the performance across the entire CTET task (10 blocks), we fitted LMEs to predict each of these variables of interest, with treatment as a categorical fixed effect and subject identity as a categorical random effect (see Materials and Methods). We compared these models with a null model, including only subject identity as a random effect using the likelihood ratio test, to test the effect of treatment on these behavioral variables. These model comparisons revealed a strong effect of treatment on misses (χ2(3) = 207.2, p < 0.001), false alarms (χ2(3) = 52.1, p < 0.001), and reaction times (χ2(3) = 82.0, p < 0.001).
Focusing on the winning models revealed that MPH, compared with PLA, was associated with an improvement in behavioral performance across all three variables: less misses (t = −10.5, p < 0.001), less false alarms (t = −3.5, p < 0.001), and faster reaction times (t = −7.2, p < 0.001). This means that, under MPH, participants were faster and better at detecting target trials without becoming too impulsive and responding to nontarget trials. We applied the signal detection theory (Macmillan, 1993) to combine responses on target and nontarget trials in one single index of behavioral performance (d′). MPH led to an increase in d′ compared with PLA (t = 8.8, p < 0.001), indicating enhanced sensitivity.
Although ATM is also used to reduce ADHD symptoms, in contrast to MPH, it is a nonstimulant. Results indicated that ATM was associated with more false alarms (t = 3.6, p < 0.001) compared with PLA, whereas misses and reaction times did not differ compared with PLA (t = −0.7, p = 0.45 and t = 0.0, p = 0.99, respectively). Thus, ATM decreased performance by increasing impulsive responses (false alarms). However, given the low proportion of false alarms, this did not result in a significant decrement of overall performance compared with placebo, as assessed by d′ (t = −1.7, p = 0.09).
We observed an increase in the proportion of misses under CIT compared with PLA (t = 4.0, p < 0.001), whereas the proportion of false alarms and the average reaction times did not significantly differ (t = −1.5, p = 0.13 and t = 0.8, p = 0.41, respectively). Thus, contrary to MPH, CIT led to a decrease in sustained attention performance, which was reflected in a decrease of the d′ (t = −2.5, p = 0.01). In addition, whereas ATM resulted in impulsive responses, CIT resulted in the opposite behavioral effect with more misses. Thus, the three pharmacological probes had unique effects on behavioral performance during the CTET, either enhancing or impairing discrete aspects of sustained attention.
Finally, we also examined the effect of time-on-task on behavioral performance and its interaction with the drug treatments. To do so, we fitted LMEs including block number and treatment as fixed effects, along with their interaction component, and subject identity as a categorical random effect. We tested the significance of the interaction component by comparing this model with the same model minus the interaction component, using the Likelihood Ratio Test. This approach showed that there was an interaction between time and treatment for misses (Likelihood Ratio Test for the model comparison: χ2(3) = 9.15, p = 0.03). This interaction was driven by MPH, which stabilized performance and prevented the increase of misses with time-on-task (Fig. 3a). CIT, on the other hand, resulted in a constant increase in misses throughout the task compared with placebo. There was no interaction between time and treatment for false alarms or reaction times (χ2(3) = 2.38, p = 0.50 and χ2(3) = 4.34, p = 0.23, respectively). This analysis confirmed the opposing effects of MPH and CIT on attentional lapses across time.
EEG indices of visual processing: ERPs and SSVEPs
To further understand the impact of pharmacological treatment on the physiology of sustained attention, we first focused on two EEG indices of visual processing: ERPs (Fig. 4) and SSVEPs (Fig. 5). Comparing the ERPs associated with the offset of the target and nontarget trials revealed a strong P3 effect (Fig. 4a), a larger positivity for Target trials, starting at stimulus offset and maximal over parietal electrodes (Fig. 4b). We thus extracted the amplitude of this P3 component and compared it across drug treatments using a cluster-permutation approach (see Materials and Methods). This analysis revealed an increase of the P3 amplitude following MPH administration compared with PLA (Fig. 4c), in line with previous findings on this same dataset (Dockree et al., 2017). Interestingly, now examining the impact of ATM and CIT as well, the amplitude of the P3 was not significantly impacted by the administration of these drugs compared with PLA (Fig. 4c).
ERPs. a, ERPs locked on stimulus offset for target (continuous lines) and nontarget (dashed lines) trials and split by drug treatment (placebo PLA: gray, methylphenidate MPH: orange, atomoxetine ATM: purple, citalopram CIT: green). ERPs are averaged across participants (PLA: N = 32; CIT: N = 31; ATM: N = 31; and MPH: N = 32). ERPs computed on two electrodes are shown: Fz (Frontal, left) and Pz (parietal, right). Colored lines show the sample average and colored areas the SEM. The yellow area ([0, 0.3] s postoffset window) shows the interval of the archetypal P3 used to compute the P3 amplitude (see Materials and Methods). b, Topographical maps of the P3 amplitude (difference of the ERP amplitude between target and nontarget trials and average on a [0, 0.3] s window) for each drug treatment. c, Topographical maps of the statistical differences in P3 amplitude between MPH and PLA (left), ATM and PLA (middle), and CIT and PLA (right). The t values obtained with paired t tests for each electrode are shown. Significant clusters of electrodes (pcluster < 0.05) are shown with black dots (cluster-permutation approach; see Materials and Methods). A significant cluster was found only for MPH.
The rapid flickering of the visual checkerboard stimuli generated a SSVEP, allowing us to examine the impact of drug treatment on an EEG index of visual processing. We observed a clear peak at the flicker frequency of checkerboard (25 Hz) in the power spectrum [Fig. 5a showing the SNR or power at each frequency bin normalized by neighboring bins; see Materials and Methods]. We thus focused on the SNR at 25 Hz and compared the strength of this entrainment for each electrode across drug treatments (Fig. 5b,c). This analysis revealed an increase for MPH compared with PLA, maximal over central electrodes (Fig. 5c). Neither ATM nor CIT significantly modulated the strength of SSVEPs compared with PLA (Fig. 5c).
SSVEPs. Visual stimuli were flashed at a 25-Hz rate on the screen, entraining neural activity at the same frequency (frequency tagging) and generating Steady-State Visually Evoked Potentials (SSVEPs). This frequency tagging can be observed when computing the power spectrum of the EEG signal and extracting the SNR of the frequency tag (see Materials and Methods). a, SNR by frequency and drug treatment, computed for electrode Cz and across participants (placebo PLA: N = 32; citalopram CIT: N = 31; atomoxetine ATM: N = 31; and methylphenidate MPH: N = 32). A clear peak at 25 Hz (highlighted in yellow) is present for all treatments (PLA: gray, MPH: orange, ATM: purple, CIT: green). b, Topographical maps for the SNR at 25 Hz for each drug treatment. c, Topographical maps of the statistical differences in 25 Hz SNR between MPH and PLA (left), ATM and PLA (middle), and CIT and PLA (right). The t values obtained with paired t tests for each electrode are shown. Significant clusters of electrodes (pcluster < 0.05) are shown with black dots (cluster-permutation approach; see Materials and Methods). A significant cluster was found only for MPH.
EEG indices of arousal: α power and sleep-like slow waves
To better understand the impact of the drug manipulations on physiological arousal, we examined a classical marker of arousal, α power. Inspection of the EEG power spectrum revealed a bump, or deviation from the 1/f trend, in the α frequency band (8–11 Hz; Fig. 6a). We thus extracted the average power in this band and compared this α power across treatments at the electrode level (Fig. 6b). A cluster permutation analysis revealed that MPH was associated with reduced levels of α power in comparison to PLA, consistent with previous reports (Janssen et al., 2016; Dockree et al., 2017) and the idea that α desynchronization is associated with improved attention. However, no differences were observed for ATM and CIT compared with PLA (Fig. 6c). Of note, peaks can be observed in the power spectrum at 1.25 Hz and its harmonics. This is because of the visual stimulation during the CTET, which is dominated by the short-duration nontarget trials (presented for 800 ms) and results in a visual stream of stimuli with a quasi-stable presentation rate of 1.25 Hz.
Power of α oscillations. We computed the (PSD) to analyze the impact of treatments on α oscillations, a common marker of physiological arousal. The power of α oscillations was obtained by averaging the log power of the PSD on a [8, 11] Hz window and for each electrode, participant, and drug treatment (placebo PLA: N = 32; citalopram CIT: N = 31; atomoxetine ATM: N = 31; and methylphenidate MPH: N = 32). a, Average PSD for electrode Cz and for each drug treatment (PLA: gray, MPH: orange, ATM: purple, CIT: green). The α power window is highlighted in yellow. b, Topographical maps of α power for each drug treatment. c, Topographical maps of the statistical differences in α power between MPH and PLA (left), ATM and PLA (middle), and CIT and PLA (right). The t values obtained with paired t tests for each electrode are shown. Significant clusters of electrodes (pcluster < 0.05) are shown with black dots (cluster-permutation approach; see Materials and Methods). A significant cluster was found only for MPH.
Since only MPH demonstrated effects compared with PLA for the different EEG indices explored thus far (ERPs, SSVEPs, and α power), these EEG markers are unable to account for the detrimental effects of CIT or ATM on behavior (Fig. 3). In addition, α power has a complex relationship with arousal, decreasing both when arousal increases and when individuals approach sleep. We therefore extracted another emerging EEG marker of low arousal, sleep-like slow waves, which has been proposed to index the transition of local networks toward sleep (Andrillon et al., 2019). We applied established algorithms to extract the number of sleep-like slow waves across electrodes and drug treatments (see Materials and Methods).
We compared the density of sleep-like slow waves (averaged across all electrodes) across sessions by fitting a linear fixed effect model comprising a fixed categorical effect of drug treatment and a random categorical effect of subject identity. We then compared this model with a model with only subject identity as a random effect (see Materials and Methods). A likelihood ratio test revealed that drug treatment had a significant effect on the density of sleep-like slow waves (χ2(3) = 32.79, p < 0.001). Examining the winning model revealed that this modulation was driven by an increase in slow-wave density for CIT compared with PLA (t = 5.13, p < 0.001; Fig. 7a). No significant differences were observed for MPH and ATM (t = 0.19, p = 0.85 and t = 1.54, p = 0.12, respectively). We then fitted a model including a fixed categorical effect of drug treatment and a fixed effect of the block number (time-on-task). A likelihood ratio test revealed that this model fitted the data better (χ2(3) = 4.04, p = 0.044). Examining this model showed a positive effect of the block number and thus the time spent on task (t = 2.01, p = 0.044). We finally explored the interaction between block number and treatment by comparing a model including treatment and experimental block as fixed effects, with a model including both fixed effects and their interaction component. A likelihood ratio test did not reveal a significant interaction between block number and treatment (χ2(3) = 1.15, p = 0.77, Fig. 7b). Rather, CIT was associated with more slow waves throughout the task, seemingly independent of any time-on-task effects.
Sleep-like slow waves. a, Slow-wave density (wave/min averaged across all electrodes) per treatment. The raincloud plot for each drug treatment [placebo PLA: gray (N = 32), methylphenidate MPH: orange (N = 32), atomoxetine ATM: purple (N = 31), citalopram CIT: green (N = 31)] is shown in the left panel. Stars denote the significance level (ns: nonsignificant, *p < 0.05, **p < 0.01, ***p < 0.001) of the difference between placebo and the three treatments used in this protocol (MPH vs PLA, ATM vs PLA, CIT vs PLA), as determined by linear mixed-effect models. In the right panel, slow-wave density split by treatment and block (N = 10 blocks) showing the sample average (circle) and SEM (colored areas) across participants is shown. A nonsignificant interaction (ns) between block number and drug treatment was found, as determined by linear mixed-effect models (see Materials and Methods). b, Topographical maps of slow-wave density for each drug treatment. c, Topographical maps of the statistical differences in slow-wave density between MPH and PLA (left), ATM and PLA (middle), and CIT and PLA (right). The t values obtained with mixed-effect models for each electrode are shown (see Materials and Methods). Significant clusters of electrodes (pcluster < 0.05) are shown with black dots (cluster-permutation approach; see Materials and Methods).
Finally, we examined topographical differences by extracting slow-wave density across electrodes and comparing it across treatments using LMEs with a single fixed effect of treatment (see Materials and Methods). A cluster-permutation analysis revealed a widespread increase in slow-wave density for CIT. We also observed a reduction of slow wave density for MPH and ATM over predominantly central and frontal electrodes respectively. Finally, we also observed an increase in slow wave density for ATM over a small cluster of occipital electrodes (Fig. 7c).
Predicting behavioral errors with EEG indices of arousal
Given that α power and slow waves decreased following the administration of MPH (Figs. 6c, 7c) and slow waves increased following the administration of CIT (Fig. 7c), and since these two treatments also showed the strongest modulations of behavioral performance (Fig. 3), we examined whether the amount of α power or the density of slow waves at the sensor level could predict behavioral performance, regardless of drug treatment, and the spatial distributions of these effects. To do so, we fitted LMEs for each scalp electrode using the proportion of misses, false alarms, or the average reaction times as predicted variables, and either slow waves or α power as a single fixed effect. In all these models, subject identity was fitted as a random categorical effect. We then used a cluster permutation approach to identify clusters of electrodes for which α power or slow waves significantly predicted behavioral variables (Fig. 8).
Predicting behavioral errors with slow waves and α power. We used slow waves (top row) and α power (bottom row), two complementary markers of physiological arousal, to predict behavioral variables: misses (left), false alarms (middle), and reaction times (right). For each predictor (slow waves or α power) and each behavioral variable (misses, false alarms, and reaction times), we fitted mixed-effect models at the electrode level (see Materials and Methods). The t values derived from these models and estimating the influence of the predictor on the behavioral variable of interest are shown as topographical maps. Significant clusters of electrodes (pcluster < 0.05) are shown with black dots (cluster-permutation approach; see Materials and Methods). Both α power and slow waves predict behavioral errors, but slow waves do so in a region-specific fashion.
This analysis revealed that an increase in α power was predictive of an increase in misses in all electrodes. An increase in α power was also predictive of a slowing of responses over centroparietal electrodes. Finally, a decrease in α power predicted false alarms in frontal, right temporal and occipital electrodes. In all three of these cases, topographical maps were homogeneous in the sense that, across all electrodes, the association between α power and behavioral variables was always in the same direction (positive for misses and reaction times, negative for false alarms).
Slow waves, on the other hand, had contrasting effects (Fig. 8, top). An increase in posterior slow waves was predictive of more misses but an increase in slow waves over frontal electrodes was predictive of faster responses. These effects suggest that slow waves can have opposite associations with behavioral errors depending on the spatial location of their occurrence. Interestingly, when including drug treatment in the models to examine the relationship of slow waves on behavior above and beyond treatments, the negative association between slow waves and reaction times was still significant (pcluster < 0.05), but the positive association between slow waves and misses over posterior electrodes was no longer significant (pcluster < 0.1).
Discussion
Sustained attention is fine-tuned by neuromodulators
We found that MPH decreased misses and false alarms, and led to faster responses compared with placebo (Fig. 3). These results replicate previous findings obtained with this dataset (Dockree et al., 2017) and others within the literature (Solanto, 1998; Spencer et al., 2009; Nandam et al., 2011; Bédard et al., 2015; Lufi et al., 2015). The beneficial effect of MPH could stem from its combined effect on both dopamine and noradrenaline, neuromodulators involved in cortical activation (Jones, 2005), sensory processing (Sara and Bouret, 2012) and motivation (Robbins, 1997). More specifically, MPH increases noradrenaline and dopamine levels in the prefrontal cortex (Berridge et al., 2006) and the striatum (Volkow et al., 2012), brain regions involved in the execution and monitoring of action (Arnsten and Li, 2005), and action selection and motivation (Liljeholm and O'Doherty, 2012), respectively.
Our results for ATM however show that increasing the concentrations of noradrenaline and dopamine does not necessarily lead to behavioral improvements. Indeed, ATM led to similar levels of sleepiness (Fig. 2, BF01 = 5.8), misses and reaction times (Fig. 3, BF01 > 3) as placebo, and an increase in false alarms (Fig. 3). This suggests that participants tended to become impulsive under ATM, without making them more attentive to the target trials. This difference in results for ATM and MPH is somewhat surprising given that both target noradrenaline and dopamine. Since the relationship between dopamine or noradrenaline and performance follows an inverted U-shape (Graf et al., 2011; Ross and Van Bockstaele, 2021), tailored administration of ATM doses could be needed to achieve improvements in task performance. Alternatively, the failure of ATM to improve performance in contrast with MPH could be because ATM impacts the dorsolateral prefrontal cortex, in contrast to the broader effects of MPH on thalamocortical networks (Bymaster, 2002; Farr et al., 2014; Kowalczyk et al., 2019). Unlike ATM, MPH increases the concentration of dopamine in the striatum (Volkow et al., 2012), a key region involved in motivation.
Finally, the administration of CIT led to a different pattern of results with participants missing more targets, without a slowing down of reaction times or an increase in false alarms (Fig. 3, BF01 > 3 for reaction times and false alarms). Thus, CIT rendered participants less responsive to targets, which contrasts with the impulsivity following ATM administration. This interpretation is in line with previous work on another SSRI, paroxetine, which resulted in an increase in missed targets and reaction times (Schmitt et al., 2002).
Overall, these results show that optimal performance on a sustained attention task needs to be achieved by balancing activation and inhibition, impulsivity, and sluggishness. Of course, this optimal balance is achieved through the synergetic effects of various neuromodulators.
Methylphenidate boosts visual processing
We investigated drug-induced changes in visual processing through two EEG indices: ERPs and SSVEPs. For ERPs, we focused on the P3, which has been associated with the detection of rare, task-relevant events (Polich, 2007) such as the target trials in the CTET (Fig. 4a). In agreement with a previous report on the same dataset (Dockree et al., 2017), MPH robustly enhanced the P3 amplitude over central electrodes (Fig. 4b,c). Examining SSVEPs confirmed an enhancement of visual processing with MPH (Fig. 5b,c), also maximal over central electrodes. ATM and CIT had no significant effect on the P3 or SSVEP amplitude.
Overall, these results indicate that MPH boosted visual processing, paralleling the improvement in behavioral performance. These results are consistent with previous findings showing an enhancement of visual responses with pharmacological increases of noradrenaline (Gelbard-Sagiv et al., 2018). This enhancement could result from the improvement of neuronal gain and SNRs (Thiele and Bellgrove, 2018). The fact that the effect on the frequency tag was maximal over central electrodes, that is, away from the primary visual cortex generating the SSVEP, suggests that higher-level, cognitive processes are at play. An increase in dopamine could also heighten motivation (Thiele and Bellgrove, 2018; Ranjbar-Slamloo and Fazlali, 2020) and lead to similar behavioral improvements. Importantly, no significant modulation was observed for the P3 or SSVEP for ATM and CIT, despite the impact of these drugs on performance.
EEG markers of sleepiness are linked to changes in sustained attention
To better understand the neural mechanisms underlying the pharmacological effects on behavior, we examined how the different treatments impacted neural indices of arousal. We first focused on α oscillations, which were reduced following the administration of MPH and could be interpreted as a positive effect of MPH on arousal. This is in line with the effect of noradrenaline on the promotion of wakefulness (Robbins, 1997; Lee and Dan, 2012; Sara and Bouret, 2012) and the use of MPH in the treatment of narcolepsy (Mignot, 2012).
However, the relationship between α power and vigilance is not monotonic and, when individuals near sleep, α power also decreases. α Oscillations are thus a rather ambiguous index of vigilance since a desynchronization of α oscillations can be associated with both an increase and a decrease in alertness. To circumvent this limitation, we focused on another, emerging index of fatigue: sleep-like slow waves (Andrillon et al., 2019). Indeed, recent research has shown that extended wakefulness and/or engagement in a given task-set is associated with an increase in patterns of EEG activity reminiscent of sleep slow waves (Vyazovskiy et al., 2011; Hung et al., 2013; Bernardi et al., 2015). In rodents, this bistable dynamic has been associated with the occurrence of “down-states” in which neuronal assemblies are silent suggesting that these so-called “local sleep” intrusions could result from a gradual transition of cortical activity toward a bistable dynamic (Vyazovskiy et al., 2011). These neuronal lapses, when occurring during wakefulness, could impair the cognitive processes performed by the impacted neuronal networks, leading to attentional lapses and errors (Andrillon et al., 2019).
To determine whether sleep-like slow waves could partly explain the behavioral effects of medications, we isolated these slow waves from the ongoing EEG activity while participants performed the CTET. We observed an increase in the number of these sleep-like slow waves following the administration of CIT, a drug that can be associated with sleepiness although it is less sedative than other classes of antidepressants (Milne and Goa, 1991). However, in our dataset, the administration of CIT did not significantly modulate the subjective ratings of vigilance (Fig. 2). It is possible thus that CIT has an effect that is difficult for participants to perceive or communicate through standard sleepiness scales. An increase in sleep-like slow waves is in line with recent findings linking serotonin with the build-up of sleep pressure in wakefulness (Jouvet, 1999; Oikonomou et al., 2019). Thus, an increase in serotonin levels through the administration of CIT could have favored the occurrence of these slow waves by shifting cortical dynamics toward bistability. We also observed local decreases in slow-wave density following the administration of MPH and ATM, which is in line with their actions on dopamine and noradrenaline, two wake-promoting neuromodulators, and the reduction of sleepiness scores in MPH sessions (Fig. 2).
Indices of arousal can predict behavioral performance
Sleep-like slow waves are predicted to be mechanistically linked with attentional lapses and the cognitive consequences of slow waves would depend on their location and the functions performed by the brain regions they affect (Andrillon et al., 2021). Thus, in the CTET, slow waves should be (1) predictive of behavioral errors (2) in a region-specific fashion. Crucially, this is in contrast with α oscillations, which likely represent a sensor and not an effector of the transition toward sleep. Our present findings support these hypotheses as α power was predictive of only a certain type of behavioral error (e.g., misses) and in a homogeneous fashion (Fig. 8): all significant electrodes were positively associated with misses and slower reaction times, and negatively associated with false alarms. In contrast, slow waves accounted for both more misses (unresponsiveness) and shorter reaction times (impulsivity) and these relationships showed regional specificities (Fig. 8). A cluster of frontal electrodes showed a negative relationship between slow waves and reaction times, in contrast with posterior electrodes which showed a positive relationship between slow waves and misses. A similar association between frontal slow waves and impulsivity on the one hand, and posterior slow waves and sluggishness on the other hand, have been found in healthy well-rested, un-medicated individuals performing a sustained attention task (Andrillon et al., 2021). This pattern of results could be explained by a perturbation of executive functions by frontal slow waves, leading to impulsivity, and a perturbation of sensorimotor processes by posterior slow waves, leading to sluggishness or unresponsiveness. Interestingly, in the context of the CTET task, participants need to balance speed and accuracy, and a release of executive control can both lead to positive (faster responses) and negative (false alarms) outcomes. This could explain why the presence of slow waves over frontal areas was associated, paradoxically, with shorter RTs. It is important however to note that these results were obtained by pooling all sessions together, which means that the relationship between α power or slow waves and behavior could be influenced by the drug treatment.
In conclusion, sustained attention is fine-tuned by a combination of neuromodulators, possibly through the maintenance of an optimal level of physiological arousal and the balance of excitation and inhibition. The effect of MPH suggests that promoting arousal can lead to improved visual processing and behavior through the combined effect of dopamine and noradrenaline. We also found that decreasing arousal through an increase in serotonin levels led to poorer behavior in terms of increased misses. Importantly, this behavioral impairment seems associated with a state of hypo-arousal, characterized by the intrusion of sleep-like slow waves in wakefulness. These intrusions seem to result in local, region-specific consequences on behavior and could represent a simple, unitary mechanism accounting for the various behavioral outcomes of inattention. This research paves the way for future studies seeking to examine the link between arousal and attention in clinical populations. These results also reinforce the notion that the pattern of behavioral impairments observed in disorders such as ADHD, might stem from a dysregulation of arousal (Owens, 2005).
Footnotes
T.A. was supported by the Human Frontier Science Program Long-Term Fellowship LT000362/2018-L and the National Health and Medical Research Council Ideas Grant APP2002454. R.G.O. was supported by the Horizon 2020 European Research Council Consolidator Grant IndDecision 865474. P.M.D. was supported by the Irish Research Council Laureate Grant 201911. M.A.B is supported by a Senior Research Fellowship (Level B) from the Australian National Health and Medical Research Council (NHMRC). We thank Jessica Barnes for data collection.
The authors declare no competing financial interests.
- Correspondence should be addressed to Thomas Andrillon at thomas.andrillon{at}icm-institute.org
