Introduction

Motivation is our ability to exert effort to obtain reward, and can be dramatically impacted in psychiatric and neurological disorders. One simple index of motivation is response vigour: an increase in movement speed with reward (Dudman & Krakauer, 2016; Shadmehr et al., 2010). Motivation and the lack of motivation (apathy) have been linked to dopamine (McGuigan et al., 2019; Walton & Bouret, 2019), and accordingly, vigour is reduced in Parkinson’s disease (Beierholm et al., 2013; Da Silva et al., 2018; Mazzoni et al., 2007; Zénon et al., 2016).

Striatal dopamine levels ramp up before potentially rewarding actions. Animal work demonstrates that acetylcholine controls these dopaminergic effects, both via firing rates (Forster & Blaha, 2000; Mark et al., 2011), as well as local release in the striatum (Cachope & Cheer, 2014; Shen et al., 2007). These provide two mechanisms for acetylcholine to affect motivation (Collins et al., 2019; Hoebel et al., 2007), with muscarinic receptors in nucleus accumbens playing an important role in facilitating reward-related vigour (Collins et al., 2016), possibly by potentiating dopamine release and affecting the frequency-sensitivity of striatal dopaminergic terminals (Collins et al., 2016; Threlfell et al., 2010). However, the ultimate effect of systemic muscarinic drugs on motivation is complex. Muscarinic antagonism has impaired motivation in some animal studies (Collins et al., 2016; Ostlund et al., 2014; Pratt & Kelley, 2004), while improving it in others (Hailwood et al., 2019; Nunes et al., 2013), and even less is known of how they affect humans. Pro-cholinergic drugs may ameliorate clinical apathy, a disabling symptom seen in around 50% of patients with Parkinson’s disease (Devos et al., 2014; Fahed & Steffens, 2021; Pagonabarraga et al., 2015), yet cholinergic blockers are commonly used to treat the motor symptoms. This frequently puts clinicians in a dilemma when treating patients with both tremor, which responds well to anti-cholinergic drugs such as trihexyphenidyl, but also apathy and cognitive impairment, which both respond to pro-cholinergic drugs (Bohnen et al., 2022). Distinguishing the mechanisms by which acetylcholine receptors contribute to movement and motivation will be critical for selecting appropriate treatments in these patients. A first step would be to test whether the effects of muscarinic blockade on motivation, observed in animals, generalise to humans.

Motivation influences action selection and movement invigoration based on the motivational state that is set up before the action. One potential mechanism of this may be preparatory activation of frontal premotor areas, which may be indexed on EEG by the contingent negative variation (CNV; Walter, Cooper, Aldridge, McCallum, & Winter, 1964), a slow negative potential that appears between a warning stimulus and a prompt to act (Brunia et al., 2012). Early models of the CNV proposed that it reflected cholinergic activity, modulated by dopamine, noradrenaline and GABA (Timsit-Berthier, 1991). Supporting this, muscarinic antagonists were found to disrupt the CNV in rodents (Ebenezer, 1986; Papart, Ansseau, & Timsit-Berthier, 1997; Pirch, Corbus, Rigdon, & Lyness, 1986; Papart et al., 1997). This anticipatory signal reflects preparatory activation of supplementary motor area and anterior cingulate, that can be amplified by reward signals from ventral striatum (Nagai et al., 2004; Plichta et al., 2013). As a marker of motivation, the CNV is of great clinical interest, being decreased by depression (Ansseau et al., 1985) and Parkinson’s disease (Ikeda et al., 1997), and conversely, increased by dopaminergic medications (Linssen et al., 2011). Accordingly, the CNV is strengthened by monetary incentives dependent on performance (Berchio, Rodrigues, Strasser, Michel, & Sandi, 2019; Novak, Novak, Lynam, & Foti, 2016; Novak & Foti, 2015) and reward contingency (Frömer et al., 2021). Changes in the CNV could therefore provide a mechanistic handle on the effect of drugs on motivation.

Here we ask whether blocking muscarinic receptors would reduce motivation and increase distractibility in humans, and further, whether this is mediated by preparatory activity in medial frontal areas. To test this, we measured the CNV in healthy adults after administration of an M1r antimuscarinic acetylcholine antagonist (Trihexyphenidyl; THP) or placebo (double-blinded), while they performed an incentivised eye movement task. We used a task that independently measured action selection and energisation, which may involve different neural mechanisms.

Results

After fixating on one of three circles, participants heard an incentive cue stating the maximum reward available for the trial (50p or 0p), and 1500ms after a preparation cue they made a saccade towards the circle that dimmed, and received a proportion of the maximum reward depending on their speed. On half of the trials, the non-target circle lit up, presenting a high-salience distractor. Participants completed this task once after receiving Trihexyphenidyl (THP; M1r antimuscarinic acetylcholine antagonist) and once after placebo (double-blinded), and we measured saccade initiation times, velocity and the pull of the distractor on the saccade trajectory, along with EEG.

Acetylcholine modulates invigoration by incentives

We measured vigour as the residual peak velocity of saccades within each drug session (see Figure 1c & Methods/Eye-tracking), which is each trial’s deviation of velocity from the main sequence. This removes any overall effects of the drug on saccade velocity, while still allowing incentives and distractors to have different effects within each drug condition. As expected, residual peak velocity was increased by incentives (Figure 1d; β = 0.1266, p < .0001), while distractors slightly slowed residual velocity (single-trial linear mixed effects regression: β = -0.0158, p = .0294; see Figure 1 – Figure supplement 1 for full behavioural statistics). THP decreased the effect of incentives on velocity (incentive * THP: β = -0.0216, p = .0030), indicating that muscarinic blockade diminished motivation by incentives. Separate two-way analyses indicated this was driven mainly by trials when the distractor was absent (incentive*THP when distractors were: absent β=-.0268, p = .0071; present β=-.0164, p = .1224); although effects were in the same direction, and the three-way interaction was not significant (p > .05), suggesting that the distractor-present trials had the same effect but weaker (Figure 1d).

Figure 1.

Saccadic RT (time to initiation of saccade) was slower when participants were given THP (β = 0.0244, p = < .0001), faster with incentives (Figure 1e; β = -0.0767, p < .0001), and slowed by distractors (β = 0.0358, p < .0001). Again, THP reduced the effects of incentives (incentive*THP: β = 0.0218, p = .0002). Separate two-way analyses showed that this was driven by trials where the distractor was present (incentive*THP β=.0293, p = .0006), while this was not significant when distractors were absent (β=.0143, p = .0819) – although as the three-way interaction was not significant and the direction of effects was the same in the two, it suggests that distractors may simply have weakened this effect slightly. Additionally, the THP*Incentive interactions were correlated between saccadic RT and residual velocity at the participant level (Figure 1 – Figure supplement 2).

Cholinergic blockade increases distractibility

We measured distractibility as the angular deviation of the eye position away from the target’s orientation towards the distractor’s location, at the start of the saccade, (Figure 2a), which indicates the pull of the distractor. The distribution of distractor pull is bimodal, with saccades directed towards either the target or distractor locations (Figure 2c). When the distractor did not light up, the mean distractor angle was negative (i.e. away from the distractor angle), indicating repulsion away from the angle of the distractor, whereas when the distractor lit up and was salient the mean angle was biased towards the distractor (Figure 2b; β = 0.2446, p < .0001; see Figure 1 – Figure supplement 1 for full statistics). THP increased the pull (main effect of drug, β = 0.0283, p < .0001), but only when the distractor was salient (THP*distractor, β = 0.0226, p = .0012, pairwise drug effect: distractor absent: p > .3; present: β=.0511, p < .0001). Unlike in previous work, we found no effect of incentives on distraction (β=.0023, p = .7444), although speed-accuracy trade-off curves (Figure 2 – Figure supplement 1) showed that incentives sped up responses in such a way that distraction was reduced for a given RT.

Figure 2.

The drug-related increase in distraction could be due to either greater pull or reduced repulsion. To distinguish these possibilities, we plotted the distribution of distractor pull across trials where the distractor was present (Figure 2c). THP reduced the probability of repulsion (Figure 2d) around -30°, indicating that THP reduced the repulsion away from the distractor’s location. This suggests weaker attentional suppression of the distractor. Incentives had little effect on the distribution (yellow line) in keeping with the averages in Figure 2b.

Therefore, acetylcholine antagonism reduced the invigoration of saccades by incentives, and increased the pull of salient distractors. We next asked whether these effects were coupled with changes in preparatory neural activity.

Preparatory neural activity is modulated by incentive and acetylcholine

We examined EEG activity in the delay period between the preparation cue and the target (and distractor) onset, first using three time-windows of interest, then a cluster-based permutation approach. There was an early fronto-central positive ERP with a peak around 220ms, consistent with the P3a (Figure 3a), which was then followed by a growing negative potential centrally, consistent with the CNV. From the grand-average ERP over all conditions, we chose 200-280ms at Cz for the early ERP, and 1200-1500ms at Cz for the CNV. Note that both these periods began > 1.5 seconds after the incentive was presented.

Figure 3.

Prior to the preparation cue (900 to 1100ms after incentive cue, baselining at the incentive cue; green shaded area in Figure 3a), THP strengthened negativity (Figure 3b, β= -.0597, p < .0001), but incentives had no effect or interaction (p > .05). After the preparation cue, the early ERP (Figure 3c) was significantly smaller for high incentive trials (β= -0.01868, p = .0142; see Figure 3 – Figure supplement 2 for full ERP statistics). The subsequent CNV was strengthened (i.e. more negative; Figure 3d) by incentive (β = -.0928, p < .0001) and THP (β = -0.0502, p < .0001), with an interaction whereby THP decreased the incentive effect (β= 0.0172, p = .0213).

This suggests that while incentives strengthened the incentive-cue response and the CNV, muscarinic antagonism strengthened the P3a and the CNV, and the incentive*THP interaction was only seen on the CNV in the same direction as that seen with the vigour and RT – THP reduced the incentive-effect. Thus, although the drug and reward both increased the CNV buildup, the drug reduced the reward effect.

Neural preparation predicts RT and distraction

We regressed the mean amplitude of the pre-preparation activity, P3a and CNV against velocity, RT and distractor pull (including incentive, distractor and drug conditions as covariates), and Bonferroni-corrected the p-values for 9 multiple comparisons. Pre-preparation voltage and P3a did not predict any behavioural variable (Figure 3e-f), while CNV predicted RT and distractor pull (Figure 3g, p < .0001; see Figure 3 – Figure supplement 3 for statistics) but not vigour.

As vigour was not associated with our a priori ERPs, we ran a window-free analysis to find spatial or temporal regions of the EEG activity that was associated with vigour, as well as RT and distractor-pull. This is an exploratory analysis, and given the small sample size should be interpreted with caution. We regressed each electrode and time-point against the three behavioural variables separately, while controlling for effects of incentive, distractor, THP, the interactions of those factors, and a random effect of participant. This analysis therefore asks whether trial-to-trial neural variability predicts behavioural variability. To assess significance, we used cluster-based permutation tests (DMGroppe Mass Univariate toolbox; Groppe, Urbach, & Kutas, 2011), shuffling the trials within each condition and person, and repeating it 2500 times (FWER=0.05). Velocity, RT and distraction were predicted by preparatory EEG voltages before the onset of the target, each with distinct patterns (Figure 4). Residual velocity was significantly predicted by frontal electrodes from about 280ms after the preparation cue, which was strongest on electrode AF8. This did not encompass electrode Cz. RT was strongly predicted by EEG voltage over a very large scalp area, centred on Cz from about 250ms onwards. Distractor pull was predicted by frontal electrodes from about 120ms after preparation cue, and this spread posteriorly to reach central and posterior electrodes by about 650ms.

Figure 4.

To check that these associations were not confounded by correlations between the saccadic measures themselves (e.g., RT is negatively correlated with residual velocity; r = -.0681, p < .0001), we re-ran this analysis while controlling for the other two saccadic measures. This did not materially change the results, indicating that preparatory EEG predicts these aspects of performance independently (Figure 4 – Figure supplement 1).

Preparatory activity explains incentive effects on RT but not movement speed

We have found that neural preparatory activity can predict residual velocity and RT, and is also affected by incentives and THP. Finally we ask whether the neural activity can explain the effects of incentives and THP, through mediation analyses (Muller et al., 2005).

As distractor pull was not affected by incentive, we only examined residual velocity and RT, and used linear mixed effects modelling to test the mediation. We used the electrodes with the strongest associations for each measure (AF8 for residual velocity and Cz for RT), and averaged them over 1200 to 1500ms. As Figure 5 shows, Cz partially mediated the incentive effect on RT but not residual velocity. AF8 activity did not mediate either residual velocity or RT. This suggests preparatory negativity is a marker for invigoration of RT by incentives, but not movement energisation.

Figure 5.

We also investigated whether either electrode could explain the cholinergic reduction in motivation (THP*incentive interaction) on RT – i.e., whether CNV mediated the THP moderation (Figure 5c). CNV mediated this moderation for RT, with both the influence of incentive on CNV, and CNV on RT being moderated by THP. This indicates cholinergic blockade changes how incentives affect preparatory negativity, and how this negativity reflects RT, which can explain some of the reduced invigoration of RT. However, this was not observed for saccade velocity.

Discussion

When incentivised, participants initiated movements faster, and with faster velocity, but these motivational effects were reduced by blocking muscarinic acetylcholine receptors with THP (Figure 1d-e). THP also reduced repulsion away from a salient distractor (Figure 2b). The CNV, a fronto-central signal believed to reflect premotor preparatory activation, was stronger when incentives were present, and with THP (Figure 3d), and crucially, THP reduced the incentive benefit on CNV, mirroring the behavioural effects. Neural preparation predicted RT and distractibility (Figure 3g), with distinct scalp distributions (Figure 4a-c). The CNV partially mediated the incentivisation of RT (Figure 5b), and could explain the drug-induced reduction in incentivisation of RT via a mediated moderation (Figure 5c). No such associations were seen in the window before the preparation cue onset, suggesting these effects relate specifically to preparing motivated action, rather than simply an incentive effect (Figure 3b & e). These results demonstrate a role of acetylcholine, and specifically muscarinic M1 receptors (M1rs), in motivation in humans.

Our behavioural measure of vigour was residual peak saccadic velocity, calculated separately within the drug and placebo sessions, and thus accounting for any overall drug effects on saccadic velocity or amplitude (or the linear relationship between the two, i.e., the main sequence). Previous studies have found this measure highly sensitive to incentivisation and motivation (Grogan et al., 2020; Manohar et al., 2017; Muhammed et al., 2016, 2018), with higher peak velocities than expected for a saccade of that amplitude, the same pattern seen here and that was reduced when M1rs were antagonised. As antagonising M1rs reduced the incentivisation of residual peak velocity and saccadic RT, we can assume that M1rs normally play a facilitating role. This mirrors some studies using antimuscarinics in animal striatum to impair motivation (Collins et al., 2016; Ostlund et al., 2014; Pratt & Kelley, 2004), although those studies used scopolamine which antagonises M1-like and M2-like receptors, so effects may have been due to M2r antagonism. This complements previous work demonstrating that nicotine increases reward responsiveness (Wang et al., 2020) by enhancing striatal anticipatory activity. Basal ganglia outputs, along with cholinergic activity from the pedunculopontine nucleus, project to the superior colliculus to disinhibit saccade initiation (Kobayashi & Isa, 2002; Naicker et al., 2017) and also modulate ongoing instantaneous firing that controls velocity during saccades (Smalianchuk et al., 2018). This offers potential mechanisms for muscarinic receptors to influence movement vigour, together with other neurotransmitters such as dopamine, implicated in incentivisation (Manohar et al., 2017).

The three different ERPs used here reported differing effects; the incentive-cue response was affected by the drug but not by the incentive itself; the P3a was decreased by higher-incentives but not affected by the drug; the CNV was the only one affected by both incentives and the drug, and showing an interaction of the two. The specificity to the CNV suggests that M1r antagonism is reducing incentivisation during the action-preparation process, and not affecting overall sensitivity to the incentive. The CNV in the lead up to the target appearing was strengthened by the incentives, replicating previous work (Frömer et al., 2021; B. K. Novak et al., 2016; K. D. Novak & Foti, 2015; Schevernels et al., 2014). M1r antagonism reduced the effect of incentives on CNV, mirroring the reduced incentivisation on RT and residual velocity. This suggests that cholinergic transmission is needed at an early stage of motivational preparation. CNV amplitude predicted RT, along with distractor pull, and frontal activity in this same time-window predicted residual velocity, aligning with recent studies linking CNV to RT and accuracy (Frömer et al., 2021), and to incentivisation of effort (Berchio et al., 2019). Our associations held up even when controlling for the other behavioural variables, suggesting they were not due to factors such as the negative correlation of RT and velocity. Together these findings suggest that incentives determine a baseline level of motivation, dependent on acetylcholine, that is integrated while we decide to act, thus influencing the speed of action initiation.

The CNV has been linked to activity in the supplementary motor area and cingulate cortex, dependent on inputs from ventral striatum and thalamus (Nagai et al., 2004; Plichta et al., 2013), and on a range of neurotransmitters including dopamine (Linssen et al., 2011) which also modulates motivational effects on vigour (Grogan et al., 2020; Manohar & Husain, 2015). Therefore a potential mechanism for our observed effects is that M1r activation in the striatum modulates the excitability of striatal pathways, and increases dopamine release as seen in animal work (de Klippel et al., 1993; Galarraga et al., 1999; Shen et al., 2005). Striatal muscarinic receptors could signal the pre-action, pre-decision level of motivation in humans.

No cognitive or computational model currently exists of how cholinergic modulation of dopamine affects aspects of motivation and motor control (Grahek et al., 2020). Previous studies have proposed that dopamine signals reward expectation which allows more cognitive and motor costs to be paid, in the form of both vigour and cognitive precision (Ott & Nieder, 2019; Westbrook et al., 2021). In contrast, acetylcholine has often been framed as increasing signal-to-noise ratio or amplifying attention in cortical processing (Bauer et al., 2012; Cragg, 2006; Laube et al., 2017; Sarter et al., 2006). While a unified quantitative model will need considerable work (Cools & Arnsten, 2022), our data contribute by demonstrating that more than one locus of effect is likely.

The mediation analysis showed the incentivisation of RT could be partially explained by stronger CNV amplitudes, as could the antimuscarinic reduction in incentivisation. This mediated moderation was due to both reduced incentivisation of the CNV and reduced influence of the CNV on RT. These effects were rather small, which suggests the presence of an additional direct way for drug to affect incentivisation, perhaps via subcortical routes (Faure et al., 2014; Mark et al., 2011; Mena-Segovia et al., 2008) which could not be indexed by the ERPs in our study. There were no such mediations to explain the incentivisation of residual velocity, suggesting the CNV is not associated with the motivational preparation of motor speed of saccades. Animal studies link saccadic velocity to the instantaneous firing rate in the superior colliculus (Smalianchuk et al., 2018), which is inaccessible to EEG recordings, although the motivational modulation of this activity presumably arises from cortex or basal ganglia. Future investigations of other aspects of the EEG signals may illuminate us.

One potentially confusing finding is that although stronger CNV benefits RT (Frömer et al., 2021; Novak & Foti, 2015) M1r antagonism strengthened the CNV while slowing RT. The mediated moderation we found indicates that THP changes how CNV predicts RT, which suggests that THP has two effects: one upstream of the CNV strengthening anticipation, and one downstream, decreasing the coupling to RT. The former could be associated with ventral striatal drive to the CNV (Plichta et al., 2013), while the latter could reflect drug-induced changes in non-motivational cholinergic systems (Gritton et al., 2016).

While we have interpreted these effects as due to incentivisation, other closely related factors cannot be ruled out. When incentivised on this task, people expend more effort and also have a higher expectation of reward, which is linked to the effort they expend. Therefore it is possible that the CNV is measuring the greater expected reward induced by motivation, which is linked to faster saccades (Haith et al., 2012; Shadmehr et al., 2019). The fact that no associations were seen between behaviour and neural activity in the time-window before the preparation cue might suggest that factors such as expected reward or arousal are less likely to explain our results. However, these explanations cannot be fully disentangled in this paradigm. We found that M1r antagonism led to a greater bias of distractor pull angles towards a salient distractor, which was unaffected by incentives. This fits with previous studies finding cholinergic involvement in attention and distraction (Fallon et al., 2023; Gritton et al., 2016; Laube et al., 2017; Sarter et al., 2016). However, we did not observe clear motivational improvements in distractibility, which have been seen previously (Hickey & Van Zoest, 2012). This may be less surprising considering that the speed-accuracy trade-off dictates worse accuracy for faster saccades (Reppert et al., 2018) which is sensitive to the rewards available for responses (Hickey & Van Zoest, 2012). Any motivational distraction effects might be weak in this study because the incentive schedule strongly favoured faster responses, and distractors were only present on half the trials, reducing the value of greater cognitive control. The repulsive bias away from the distractor was weakened by THP (Figure 2d), which may be a behavioural manifestation of reduced reactive top-down control reported with muscarinic antagonism (Laube et al., 2017). But in our data, the distractor pull could be predicted by preparatory activity over 1 second before the target onset, and this was also affected by muscarinic blockade. The frontal signature of distractor pull was distinct to the pattern predicting RT, suggesting that cholinergic effects on proactive control and speed are dissociable.

Care should be taken when interpreting a lack of effects in this study, given the reduced power due to a smaller than expected sample size. This may be more applicable to the whole-brain regressions and the mediation analysis, so future replications would be useful to verify these results.

Additionally, as this study only tested young men, it is not known how generalisable the findings are to older adults, who may have age-induced changes in their cholinergic function, or indeed to patients with Parkinson’s disease, who have additional dopaminergic deficits. Recent work showed that cholinesterase inhibitor withdrawal in patients with Parkinson’s disease impaired a range of attentional and memory functions (Fallon et al., 2023), and investigating motivational impairments would be useful too. The effects of acetylcholine on motivation are of crucial importance in Parkinson’s disease, where the balance between dopamine and acetylcholine is disrupted (Pisani et al., 2007; Schulz et al., 2018), leading to apathy or impulsivity (Devos et al., 2014). As trihexyphenidyl is often used to treat tremor in Parkinson’s disease, the finding that it impairs motivation suggests that it may worsen motivational symptoms in a population already troubled by apathy.

Conclusion

Muscarinic M1r antagonism reduced the incentivisation of saccadic peak velocity and RT, suggesting that normally M1r activity is important for motivation. The incentives strengthened the CNV, a preparatory EEG component, and this mediated the incentivisation of RT and the reduction of this incentivisation by the drug, implicating the CNV as a potential marker of muscarinic invigoration.

Methods

Design

We used a randomised, counterbalanced, double-blinded, placebo-controlled trial of Trihexyphenidyl (THP). Participants were tested twice (minimum one-week separation), once on placebo, and once on THP, making this a within-subject study. Ethical approval was granted by the University of Oxford MSIDREC (R45265/RE001).

Drugs

Participants were administered 2mg THP or 200mg lactose pills, both encapsulated in a digestible shell, and labelled A and B by a colleague not involved in the study. Participants were randomised to receive either tablet A or B on their first session, and B or A on their second. The experimenters did not know whether A or B contained the drug until all data were collected and pre-processed. On arrival, we checked participants felt well, and administered the dose. Testing began 1.5-2 hours later.

Participants

Our sample size calculations suggested 27 participants would detect a 0.5 effect size with .05 sensitivity and .8 power. We recruited 27 male participants, but due to the pandemic, we had to halt the study part-way, with 20 completed participants and 5 participants who had completed one session only. We only analysed the 20 completed participants, which achieved a post-hoc power of 0.7. The mean age was 28.15 years (SD = 8.03 years).

Participants read the information sheet, and gave written informed consent. Participants were screened for contraindications (e.g., cardiovascular disease, hypo/hypertension, cardiac arrhythmia, stroke, kidney/liver disease, psychiatric conditions, gastrointestinal haemorrhage, glaucoma, epilepsy, lactose hypersensitivity, porphyria) for the drug and placebo, and 1-lead ECG was taken to check for prolonged QTc interval of over 480ms. A medical doctor checked all this information before the participant was admitted to the study.

Task

To measure invigoration of saccades by incentives, we adapted an incentivised saccade task (Manohar & Husain, 2015), where participants had to make speeded saccades to a low-salience target in exchange for money, while avoiding a high-salience distractor. The task was run in Matlab R2018b and Psychtoolbox-3 (Kleiner et al., 2007).

Participants saw three grey circles each with a black fixation-cross (Figure 1a) on a black screen (11° apart), and had to fixate on the one that turned pink after 500ms (+0-100ms jitter) for 300ms. An audio incentive cue was played 200ms after fixation, 1100ms duration, of a voice saying ‘50p maximum’ or ‘0p maximum’ to indicate the maximum money available on this trial. Fixation was checked again after this to ensure fixation (300ms + 100ms wait), and then the black fixation-cross turned white. This was the preparation cue, which occurred 1500ms before the target onset. Then, one of the other two circles dimmed, indicating it was the target, and on 50% of trials the other circle simultaneously brightened, as a salient distractor. Participants were rewarded a proportion of the incentive based on how quickly they looked at the target circle, with an adaptive reward rule that had an exponential fall-off depending on the average reaction time in the previous up to 20 trials. This kept the rewards received roughly constant over the task. If participants looked at the distractor, they would need to make a corrective eye movement to the target, which would slow the time to reach the target, and result in less reward. Once gaze reached the target, a feedback sound was played if a medium (10-30p) or large (>30p) reward was obtained, while the value of reward earned flashed in the centre of the target circle for 800ms. A 1100ms (+0-100ms jitter) rest period followed each trial, where participants were allowed to blink. The target on one trial became the fixation location for the next trial.

There were thus four trial types, with two maximum incentive levels (50p or 0p) and the presence or absence of a distractor. There were six of each trial in a block, giving 24 trials, and 20 blocks (480 trials in total; 120 per condition). Participants could pause as long as they liked between blocks, with a minimum 4-second break. The first trial of each block also had an extra 4-second rest period before the trial began to ensure participants were settled and ready.

Participants were given 24 practice trials during the screening visit and before the main task on each main visit. These included an extra fixation check during the 1500ms delay between the preparation cue and target onset, and if they moved their eye more than 1° during this time, the trial was flagged for repetition and the experimenter asked them to maintain fixation better. This training improved fixation during piloting.

In addition to this task, two other tasks were performed that are not reported in this paper, including a working memory task and a reversal learning task. We asked participants to rate how they felt on a visual analogue scale before and after the tasks, and participants also performed a pro- and anti-saccade task, and measurement of the pupillary light reflex.

Eye-tracking

We tracked participants’ eyes with an EyeLink 2000 (SR Research) at 1000Hz. Participants were seated 75cm from the screen, with their head on a chin-rest and forehead-rest. Nine-point calibration was performed at the start of the task, and one-point validation was done at the start of each block, with re-calibrations if necessary. Stimuli were shown on an ASUS VG248Qe3D screen (53×30cm, 1920×1080 pixels, 60Hz).

Saccades were parsed with standard criteria (velocity > 30°s^-1, acceleration > 8000°s^-2). We took the first saccade over 1° in amplitude within 100-900ms after the target onset, and calculated velocity with a sliding window of 4ms, excluding segments faster than 3000°s^-1 or where eye-tracking was lost. Saccades with peak velocities outside 50-1600°s^-1 were excluded.

Saccadic velocity is correlated with the amplitude of the saccade, an effect known as the main sequence (Bahill et al., 1975), and saccade amplitude can also be affected by reward (Grogan et al., 2020). To remove the effect of amplitude on velocity, we regressed peak velocity against amplitude within each participant and drug-session, combing across incentive-level and distractor-level, and took the residual peak velocity for each individual trial as our main measure (Figure 1c). This reflects the difference between the velocity measured and the velocity predicted by the main sequence, with positive values meaning faster than expected velocity. By calculating velocity residuals for each drug condition separately, this method removes any overall effects of the drug on velocity or amplitude, and instead measures individual trial variations within that drug condition. Velocity residuals have previously been shown to be most sensitive to reward manipulations of vigour (Blundell et al., 2018; Grogan et al., 2020; Manohar, Finzi, Drew, & Husain, 2017; Manohar, Muhammed, Fallon, & Husain, 2019).

Saccadic reaction time (RT) was taken as the time between target onset and the start of the saccade (as detected by EyeLink velocity and acceleration criteria; Figure 1b) in ms; we used log RT for the analyses but plot raw RT. Distractor pull was measured as the angular deviation of the eye from a straight line linking the fixation and target circles (Figure 2a) when it left the fixation circle; positive values reflected a bias towards the distractor, while negative values reflected a bias away from the distractor.

EEG acquisition and pre-processing

We recorded EEG with a Refa72 amplifier (TMSi, B.v Netherlands) at 1024Hz and using OpenVibe software (Renard et al., 2010). We used a 64-channel cap (TMSi). The ground was placed on the left clavicle, and we recorded horizontal EOG with bipolar electrodes placed either side of the eyes. Due to the cap, we could not place an EOG electrode above the eye, so one was placed 1cm under the left eye, and this was converted into a bipolar EOG signal as the difference from electrode FP1, which was the closest cap electrode to the left eye. Impedances were kept below 10KΩ as it was a high-impedance system.

Data were processed with custom Matlab scripts, and EEGLab and ERPLab toolboxes (Delorme & Makeig, 2004; Lopez-Calderon & Luck, 2014). Channels were referenced to the average of the two mastoid electrodes A1 + A2, and synchronised with the eye-tracking traces using the EYE-EEG toolbox (Dimigen et al., 2011). Data were band-pass filtered at 0.1-80Hz with an IIR Butterworth filter, notch filtered at 50Hz with a stop-band Parks-McClellan filter, and down-sampled to 256Hz.

Epochs were from -200:1500ms around the preparation cue onset, and were baselined to the 100ms before the preparation cue appeared. We rejected trials where participants blinked or made saccades (according to EyeLink criteria above) during the epoch, or where EEG voltage was outside - 200:200μV (muscle activity). On average 104/120 trials per condition per person were included (SD = 21). A repeated-measures ANOVA found there were no significant differences in number of trials excluded for any condition (p > .2).

We took the late CNV period (1200:1500ms) at electrode Cz as our a priori region of interest, along with the cue-P3 (200-280ms). In order to see whether our results were specific to preparatory activity, we looked at activity before the preparation cue began, and looked at the late ERP to the incentive cue. We epoched the data from -200:1100ms around the incentive cue onset (which was the duration of the incentive cue), and used the same artefact rejection criteria as above.

Analysis

We used single-trial linear-mixed effects models to analyse our data, including participant as a random effect of intercept, with the formula ‘∼1 + incentive*distractor*THP + (1 | participant)’. We z-scored all factors to give standardised beta coefficients. For the difference-wave cluster-based permutation tests (Figure 3 – Figure supplement 4), we used the DMGroppe Mass Univariate toolbox (Groppe et al., 2011), with 2500 permutations, to control the family-wise error rate at 0.05. This was used for looking at difference waves to test the effects of incentive, THP, and the incentive*THP interaction (using difference of difference-waves). We adapted this toolbox to also run cluster-based permutation regressions to examine the relationship between the behavioural variables and the voltages at each electrode at each time point. On each iteration we shuffled the voltages across trials within each condition and person, and regressed it against the behavioural variable, with the model ‘∼1 + voltage + incentive*distractorPresent*THP + (1 | participant)’. This was repeated 2500 times, and the clustering used to control the family-wise error rate at .05. Given the relatively small sample size here, these whole-brain analyses should not be taken as definitive.

For the mediation analysis (Muller et al., 2005), we first checked that the behavioural variable (outcome) and ERP (mediator) were both significantly associated with incentive (treatment; behaviour ∼ 1 + incentive + (1 | participant)), and that the ERP was significantly associated with the behaviour when controlling for incentive (behaviour ∼ 1 + ERP + incentive + (1 | participant)). The mediation was measured by the reduction in the standardised beta coefficient between incentive and behaviour when the ERP mediator was included. Mediated moderation was used to see whether the effect of the drug on behaviour is mediated by the ERP. Moderation was first assessed by the significance of the interaction terms in predicting the behaviour and the ERP (behaviour ∼ 1 + incentive + drug + incentive*THP + (1 | participant), and was measured by the reduction in the beta coefficient for incentive between the previous model and the moderation model (behaviour ∼ 1 + incentive + drug + incentive*drug + ERP + ERP*THP + (1 | participant)). This measures how much of the association between incentive and behaviour was explained by allowing the ERP (mediator) to interact with the drug (moderator), i.e., mediated moderation.

Data and code availability

Anonymous data are available on OSF (https://osf.io/zuq5c/), and analysis code is on GitHub (https://doi.org/10.5281/zenodo.5141792).

Conflict of Interest

The authors declare that no competing interests exist.

Acknowledgements

The authors would like to thank Andrea Bocincova and Yongzhi Huang for help with EEG set up and analysis. The work was funded by a Medical Research Council Clinician Scientist Fellowship to SGM (MR/P00878X). MVS was funded by the Medical Research Council (MC_ST_U16042). MJG is funded by the Jon Mouton Charity Trust (Guernsey). The Refa72 was funded by the Academy of Medical Sciences, Starter Grant for Clinical lecturers (MJG). The scheme is generously supported by the Wellcome Trust, Medical Research Council, British Heart Foundation, Versus Arthritis, Diabetes UK and British Thoracic Society (through the Helen and Andrew Douglas bequest).

Author Contributions

Conceptualization: SGM, JPG. Data curation: JPG. Formal Analysis: JPG. Funding acquisition: SGM. Investigation: JPG, MR, MvS. Methodology: JPG, SGM. Project administration: JPG. Resources: MJG, ALG. Supervision: SGM. Validation: JPG. Visualisation: JPG. Writing – original draft: JPG. Writing – review & editing: JPG, SGM, MR, MvS, ALG, MJG.