Participants came to the lab three times on different days (see Table 1). Once for screening (duration 2.5 h), and twice for an experimental session in which either placebo or 1.5 mg of cabergoline was administered orally (duration 4.5 h each), as part of a double-blind crossover design. Previous studies found cognitive and neural effects of cabergoline using a dosage of 1–1.5 mg (Cavanagh et al. 2014; Cohen et al. 2007; Frank & O’Reilly, 2006; Nandam et al. 2013; Norbury et al. 2013; Yousif et al. 2016; Fallon et al. 2017; Broadway et al. 2018). As with our sample size, we chose a dosage of 1.5 mg to be on the upper end of previously employed dosages. There was at least a day in between screening and the first session, and at least a week between both sessions.
The first lab visit took place anywhere between 09:00 and 17:30. After providing written informed consent, participants answered a series of questions concerning potential medical conditions. Next, we conducted the M.I.N.I., a structured screening interview for DSM-IV axis-I disorders (Sheehan et al. 1998). We subsequently measured participant’s weight, height, BMI, blood pressure (BP), and heart rate. Participants were included in the experiment only if these measures fell within pre-established bounds (BMI 18–30, diastolic BP < 50 or > 90 mmHg, systolic BP < 95 or > 140 mmHg). Next, six external electrodes were attached to the participant’s face and ears in order to measure spontaneous eye blink rate (sEBR) at rest. Finally, participants completed an operation span (OSPAN) working memory task (Unsworth et al. 2005), and a titration procedure for two behavioral tasks to be completed during both experimental sessions (backward masking and probabilistic discrimination; see below).
All placebo and cabergoline sessions (the second and third visit) took place between 08:30 and 14:00. Participants were instructed to abstain from drug and heavy alcohol use, the day before and during the day of the session. Also, participants were instructed to abstain from caffeine and nicotine the morning of the session. Compliance to the instructions was checked by the examiner on arrival; in case of non-compliance, the session was postponed. Female participants completed a midstream pregnancy test. Breakfast was offered, in order to avoid cabergoline intake on an empty stomach. Blood pressure and heart rate were measured using an Omron® M3 comfort sphygmomanometer, and participants filled in a visual analog scale (VAS; see below) three times during the session: on arrival, at around 1.5 h after placebo or cabergoline intake, and at the end of the session. After the initial blood pressure/heart rate/VAS measurement, participants were administered either placebo or cabergoline in a double-blind fashion (order randomized across participants). After a 40-min break, a BioSemi ActiveTwo system (BioSemi Inc., Amsterdam, The Netherlands) EEG cap and electrodes were fitted. Drug plasma levels have been found to reach maximum concentration after approximately 1.5–3 h (Persiani et al., 1996; Agúndez et al. 2013). Approximately 1 h and 20 min after drug intake, participants completed 6 min of sEBR recordings, followed by the backward masking task around 1.5 h after drug intake (see below), during which EEG was recorded. After this task, the EEG setup was removed from the participant’s head. After a 30-min lunch, participants proceeded to the attentional blink task, a simple reaction time task, and the probabilistic discrimination task (see below). At the end of the experiment, one final blood pressure/heart rate and VAS measure was undertaken. At the end of the final session, participants were asked to indicate in which session they believed they had received cabergoline.
Physiological and subjective state measures
Heart rate and blood pressure
Physiological measurements were taken once during screening, and three times during both sessions; namely, on arrival, at around 1.5 h after drug intake, and on completion of testing (± 3.5 h after drug intake) (see Table 1). These measurements were obtained using an Omron® M3 comfort sphygmomanometer.
A set of sixteen VAS measures were used (Bond and Lader 1974), to assess the subjective state of the subject before medication intake, at around 1.5 h after drug intake, and on completion of testing (± 3.5 h after drug intake). Each scale consisted of a 100-mm horizontal line, anchored by contrasting states of mind (e.g., happy versus sad). Subjects were asked to regard each line as a continuum and to rate their feelings at the time by moving a vertical slider across each line. The scales could then be scored by measuring the length in millimeters from the positive end of each line to the subject’s marked location. These sixteen VAS measures were summarized as three categories: contentedness, calmness, and alertness (Bond and Lader 1974).
Baseline dopamine proxies
Both sEBR and OSPAN are widely used measures that have been related to baseline dopamine levels (Cools and D’Esposito 2011; Jongkees and Colzato, 2016). Both measures have been used in combination with cabergoline in order to account for individual differences in baseline dopamine (Broadway et al. 2018; Cavanagh et al. 2014). Eye blink rate is defined as the number of spontaneous eye blinks per minute. The measure has high test-retest reliability (Kruis et al. 2016) and is an often-used biomarker of baseline dopamine D2 receptor functioning (Jongkees and Colzato 2016; Karson, 1983; Taylor et al. 1999, but see Sescousse et al. 2018). Subjects were asked to look at a central fixation cross on a computer screen in a relaxed state for 6 min while we measured eye activity from a set of vertical and horizontal electrodes, in order to detect eye blinks. This procedure was employed during all three lab visits.
Eye blinks were established in two ways. First, through a fully automatic procedure implemented in the python module MNE (create_eog_epochs; Gramfort et al. 2013). Second, eye blinks were established through a custom semi-automatic procedure using EEGLAB for MATLAB (Delorme & Makeig, 2004). If mean sEBR per minute differed more than 3 blinks between both methods, the semi-automatic procedure was repeated, and an average was taken of both semi-automatic attempts as the final value. Prior to any repetition of the semi-automatic method, correlations between the automatic and semi-automatic method exceeded .95 for measurements during all three lab visits.
OSPAN is a working memory task with high test-retest reliability (Unsworth et al. 2005), in which participants are instructed to remember letters, while solving simple arithmetic problems in between letter presentation (Unsworth et al., 2005). Sets of 3–7 letters were presented successively at fixation. The OSPAN score was calculated through partial credit scoring, so that each correctly recalled letter in the appropriate location was counted as correct, regardless of whether the entire sequence was recalled correctly or not. Scores could range from zero to 75. OSPAN was measured only during screening.
To assess the effects of cabergoline on alertness, we administered a 40-trial simple reaction time (RT) task (Brown et al. 2016). In this task, participants had to respond as quickly as possible by pressing the spacebar whenever a white circle (subtending approximately 3.1° of visual angle) appeared at the center of the computer screen against a black background. Stimulus onset asynchrony was jittered between 500 and 1250 ms, with a mean of 1000 ms. This task lasted less than 2 min.
Main experimental paradigms
All stimuli were presented on an ASUS VG236H 23-in. LCD screen (refresh rate = 100 Hz, resolution 1920 × 1080). Participants viewed the screen at a distance of 80 cm.
In the backward masking task, adapted from Van Opstal et al. (2014), participants had to indicate whether briefly presented masked digits (1, 4, 6, or 9) were smaller or larger than 5 and rate the confidence in their response (Fig. 1). Each trial started with the presentation of a central fixation cross (30 point Courier New), which increased in size (106 point Courier New, 150 ms duration), cueing the impending target. The target stimulus (30 point Courier New) then appeared for 10 ms at one of two positions centered at the vertical midline (top or bottom, 2.29° from fixation). Both stimulus locations were equally probable. A mask followed the target (200 ms duration) at a variable stimulus onset asynchrony (SOA). Due to the employed refresh rate of 100 Hz, the SOA could vary from 10 ms to 100 ms in 10 ms steps. By making the delay between cue and target dependent on SOA, the delay between cue and mask was held constant at 800 ms. The mask (30 point Courier New) was composed of two letters “E” and two letters “M”, tightly surrounding the target location without superimposing or touching it. All stimuli were black and presented on a white background, using the Psychophysics toolbox for MATLAB (Brainard, 1997). The central fixation cross was visible throughout the experiment.
Participants were instructed to indicate by button press whether the presented digit was smaller or larger than 5, while simultaneously indicating the confidence in their response (sure/unsure), resulting in four possible responses (<5 sure, <5 unsure, >5 unsure, >5 sure). In previous research, it was found that the D2 agonist pergolide affected response confidence (Lou et al. 2011). Responses were given by means of a response box attached to the arm rests of the participant’s chair. Response buttons were counterbalanced across participants, who were instructed to guess one of two “unsure” buttons if they did not see the target.
If the participant’s reaction time exceeded 1 s, a message was presented indicating that their response was too slow for the duration of 1 s, urging a faster response. An individual threshold for awareness was established during the screening session (see above), by fitting a logistic model (threshold defined as SOA corresponding to 75% accuracy; mean threshold = 52.93 ms, min = 31 ms, max = 89 ms, sd = 14.38; Del Cul et al. 2006; Van Opstal et al. 2014). This model was fitted on the basis of 176 trials during screening, where each of 11 SOA durations (from 0 to 100 ms) was presented 16 times.
Prior to the experiment, participants first completed a practice block (176 trials in screening, 88 trials during placebo and cabergoline sessions). In both drug sessions, participants completed 920 trials in total, split by seven possible SOAs between target and mask: 200 mask-only trials (0 ms SOA), 200 trials each for the main SOAs (10 ms/awareness threshold/100 ms), 40 trials surrounding the threshold (threshold minus 10 ms and threshold plus 10 ms), and 40 trials with a 70 ms SOA. The trials in between individual thresholds and 100 ms were excluded from analysis (threshold +10 ms and 70 ms), because some participants arrived at an individual threshold at or above 70 ms. This meant that trials with a SOA at threshold +10 ms and 70 ms would fall below or above participant’s individual threshold, depending on the participant. A total of 80 trials per participant were discarded for this reason, leaving 840 trials.
Participants also performed a standard AB task in which they had to identify two digits (T1 and T2) presented in a rapid stream of centrally presented distractors (letters and symbols; adapted from Slagter et al. 2012; Slagter et al., 2017; see Fig. 2). T2 followed T1 either in the time window of the AB, after 200 ms (short-interval trial), or outside the time window of the AB, after 800 ms (long-interval trial). Each trial started with a central fixation cross (1500 ms), after which the stimulus stream began, consisting of 22 stimuli. Stimuli were presented on a black background (RGB 70, 70, 70) at the center of the screen (28 point Arial; 0.85° visual angle) for 50 ms, followed by a 50 ms blank. Digits were drawn randomly (without replacement) from the set 2–9. Distractors were randomly drawn (without replacement) from the following set of 30 letters and symbols: W, E, R, T, Y, U, P, A, D, F, G, H, J, K, L, Z, X, C, V, B, N, M, @, #, $, %,}, &, <, and =. Participants were asked to indicate sequentially the identity of the targets they saw, using the numpad on a standard keyboard. If they missed a target, they were instructed to guess. Stimulus presentation was performed using Presentation (Neurobehavioural Systems).
In both sessions, participants first completed a short practice block (20 trials), in which the first 8 trials moved at half speed. Next, participants moved on to the main experiment (222 trials), spread over 6 blocks consisting of 37 trials each.
In the probabilistic discrimination task, adapted from Bauer et al. (2016, September), participants were presented continually with a central fixation cross (28 point Arial, RGB 0, 0, 0), on top of which an image (6.68° visual angle) of either a face or house was presented for 120 ms, against a gray background (RGB 128, 128, 128). Face stimuli were created on the basis of the Park Aging Mind Laboratory, University of Texas at Dallas (Minear & Park, 2004), while house stimuli were based on the Caltech University Computational Vision database (http://vision.caltech.edu/archive.html). On each trial, participants had to report the category of the image with by pressing “Q” or “P” on a standard keyboard. Responses were counterbalanced across participants, who were instructed to emphasize accuracy over and above speed. The maximum response interval was 1700 ms, after which the next stimulus was presented regardless of whether a response was given. The inter-trial interval was jittered and varied from 800 to 1200 ms.
The difficulty of stimulus discrimination was manipulated in terms of stimulus coherence (Fig. 3). Stimuli from the abovementioned databases were cropped to the outlines of faces and houses, and a 2-dimensional spatial Fourier transform (on luminance values for x-/y-coordinates) was calculated. The amplitude (power-) spectra of all 82 face and house images were averaged and subsequently applied to all individual images, such that all images (faces and houses of all coherence levels) had an identical power spectrum. In other words, none of them differed in global contrast or luminance. The phase-spectra of each individual image (that therefore provided all pictorial information) was retained and was subsequently superimposed with various levels of (uniform) random noise for each image (Bauer et al. 2016, September). To account for bias in the circular phase distribution of superimposed noise and signal phase-spectra, noise-spectra were sampled following previous suggestions (Dakin et al. 2002).
Titration consisted of two subsequent procedures, in order to establish three difficulty levels for each individual participant. First, participants completed 300 trials, spread over 10 staircase blocks, in order to establish difficulty levels corresponding to an accuracy of 75% for each stimulus category separately. During this 3-up-1-down staircase procedure, participants received a green thumbs-up (RGB R [56,154 79], border RGB [17 79 22]) or red thumbs-down (RGB [83 2 5], border RGB [251 84 84]) at the center of the screen as feedback after each trial (500 ms duration).
Next, a total of 810 trials followed, across 27 blocks, in order to extrapolate the acquired difficulty level to three difficulty levels. For the second part of the titration procedure, participants received feedback in the break in between blocks, in order to counteract the development of a bias for one of two response categories. Difficulty levels were estimated using the method of constant stimuli (MOCS; Bauer et al. 2016, September). The psychometric functions obtained through this procedure were used to estimate difficulty (coherence) levels corresponding to 70, 82, and 95% accuracy.
In both the placebo and cabergoline session, the same three difficulty levels were employed that were acquired from the titration procedure during screening. Unbeknownst to participants, the prior probability of each category was manipulated in a block-wise manner (20/35/50/65/80%), spread over 25 blocks of 40 trials each, for a total of 1000 trials per session. As such, we manipulated perceptual information (difficulty) and stimulus prior probability independently of one another. Stimulus presentation for the titration procedure during the screening was performed using the Psychophysics toolbox for MATLAB (Brainard 1997), and Presentation was used to present the task in both experimental sessions (Neurobehavioural Systems).
Physiological and subjective state measures
In order to test whether physiology and subjective state changed over the course of the experiment, and whether these measures were influenced by cabergoline, we conducted a repeated-measures analysis-of-variance (RM ANOVA) for heart rate, diastolic and systolic blood pressure, and each VAS category separately, across all three time-points and both sessions. We conducted a paired-samples t-test between sessions for our simple RT task, as an additional measure to assess alertness. In order to test whether cabergoline exerted influence on sEBR, we conducted a paired-samples t-test between sEBR under placebo versus cabergoline. Kendall’s Tau correlation was employed to establish the relationship between sEBR sessions, as well as the relationship between sEBR and OSPAN, as this coefficient is more robust in the case of small samples and tied ranks (Bonett and Wright 2000). Correlations were Bonferroni corrected for multiple comparisons.
The dependent measures in the backward masking task were accuracy (0/1) and confidence (unsure/sure). For both of these measures, we computed a 2 (Drug; placebo/cabergoline) × 5 (SOA; mask-only/10 ms/threshold–10 ms/threshold/100 ms) RM ANOVA. Furthermore, we performed an additional analysis including screening sEBR and OSPAN as covariates in both of these analyses, as we predicted cabergoline effects may depend on individual baseline dopamine levels, based on previous reports (Broadway et al., 2018; Cavanagh et al. 2014; Cools and D’Esposito, 2011; Jongkees and Colzato, 2016). We repeated these analyses for the two-alternative forced choice version of the signal detection theory parameters d’ (Green and Swets 1966) and meta-d’ (Maniscalco and Lau 2012), for the three primary SOAs (10 ms, threshold, and 100 ms).
One participant confused the confidence response buttons in one session. We reversed these confidence scores manually. One participant experienced side effects only near the end of the last session. This participant thus completed the backward masking task twice without knowledge about drug condition. In order to maximize statistical power, this participant was included in all analyses concerning the backward masking task (including EEG), but not in the analyses of other tasks. It did not matter for our results whether this participant was included in the backward masking task analyses or not.
The dependent measures for the AB task were T1 accuracy and T2 | T1 accuracy. In other words, T2 accuracy was based only on those trials where T1 was correctly reported. For each of these measures, we computed a 2 (Drug; placebo/cabergoline) × 2 (Lag; 2/8) RM ANOVA. For this task as well, we computed additional analyses in order to include sEBR and OSPAN as covariates. Finally, we computed AB size, in order to investigate the relationship between AB size and our baseline dopamine measures as some (Colzato et al. 2008) but not other studies (Slagter et al. 2012) have found.
In the case of the probabilistic discrimination task, our dependent measure of interest was accuracy. As such, we computed a 2 (Drug; placebo/cabergoline) × 3 (Difficulty; easy/medium/hard) × 5 (Probability; .2/.35/.5/.65/.8) RM ANOVA. For this analysis as well, we computed additional models including screening sEBR and OSPAN as covariates.
Our titration procedure was not successful for all participants. As a result, a number of participants ended up with only two difficulty levels for one out of two stimuli. For this reason, we repeated the above analysis for both the group with all difficulty levels (N = 16), and participants who had either two or three difficulty levels as a result of the titration procedure (N = 24). In this latter analysis, we excluded the medium-difficulty trials. One participant was excluded from both analyses, because this participant ended up with only one difficulty level for face stimuli.
All covariates in the abovementioned RM ANOVAs were centered (van Breukelen and van Dijk 2007). For all repeated-measures ANOVA analyses, whenever Mauchly’s test suggested a violation of sphericity, we report Geenhouse–Geisser corrected P values, but uncorrected degrees of freedom. In order to test for order effects, we repeated each of the RM ANOVAs for our behavioral paradigms including a between-subject factor indicating whether a participant received either placebo or cabergoline in the first session.
In order to evaluate evidence in favor of our (null) hypotheses, we conducted Bayesian statistics. For each reported frequentist test, we report the Bayes factor corresponding to the inclusion of a factor or interaction within the model in question (shortened to BFincl), compared to equivalent models stripped of the effect. For example, BFincl = 10 indicates that a model including the factor in question is ten times more likely given the data compared to a model without the variable. Conversely, BFincl = .1 indicates that a model without said effect is ten times more likely given the data. All Bayesian statistics were conducted using JASP (JASP Team, 2019, version 0.10.0).
All data visualization was performed with the help of raincloud plots (Allen et al. 2019), which include the mean, individual data points, as well as the overall distribution of the measure in question.
Recording and preprocessing
EEG data, digitized at 512 Hz, were continuously recorded in both the placebo and cabergoline session during 6 min of sEBR and the backward masking task, using an ActiveTwo system (BioSemi, Amsterdam, the Netherlands), from 64 scalp electrodes placed according to the 10/20 system, four electro-oculographic electrodes placed above and below, and to the side of the eyes, and two external electrodes attached to each earlobe. EEG data were offline referenced to the average activity recorded at the earlobes, and high-pass “firws” filtered (default settings) at 0.05 Hz using a Kaiser window, following previous suggestions (Widmann et al. 2015). The continuous data were subsequently epoched from −1.5 to 1.5 s around stimulus presentation and baseline corrected to the average activity between −200 ms and 0 ms pre-stimulus. Epochs containing EMG artifacts or eye blinks surrounding stimulus presentation were rejected based on visual inspection. Extremely noisy or broken channels were interpolated. Remaining eye blink artifacts were removed by decomposing the EEG data into independent sources of brain activity using an independent component analysis, and removing eye blink components from the data for each subject individually. Epochs were low-pass filtered at 30 Hz for visualization purposes only. Preprocessing was done using the Fieldtrip toolbox (Oostenveld et al. 2011) for Matlab (The MathWorks, Inc. Natick, MA, USA) using custom-written Matlab scripts.
To determine the effect of our manipulations on ERP markers of information processing, we examined the effects of SOA and drug on the amplitude of the visual-evoked P1 and N1 components, the N2, as well as of the later P3b (Del Cul et al. 2007). In line with Del Cul et al. (2007), we epoched the ERP data to the onset of the mask. Next, we subtracted the data from the mask-only SOA condition from all other SOA conditions. Finally, we shifted ERP onset back to target onset, in order to compute target-locked ERPs. Visual inspection of the grand- and condition-average ERPs showed that the P1 and N1 components peaked over lateral occipitoparietal scalp sites (PO7, PO3, O1, PO4, PO8, O2), the N2 over centroparietal scalp regions (C1, Cz, C2, CP1, CPz, CP2), and the P3b over central parietal scalp sites (P1, Pz, P2, PO3, POz, PO4). These scalp sites were used to determine the peak amplitude and latency of these components for each condition of interest. Specifically, the largest positive voltage value between 75 and 150 ms post-target, and the largest voltage negativity within 150–225 ms were selected to determine the amplitude and latency of the P1 and N1 peaks, respectively, for each subject separately. In the case of the P2 and N2, these intervals were 175–250 ms, and 250–375 ms, respectively. For the P3, an interval of 300–450 ms was used. All average amplitude values 15 ms around the peak sample, as well as individual latencies, were entered into separate RM ANOVAs with two within-subject factors: SOA (10 ms, threshold-10 ms, threshold, 100 ms) and drug (placebo/cabergoline). Because the mask-only condition is used to acquire ERP data for the remaining four SOA conditions (see above), the SOA factor contains four instead of five levels for ERP analyses.