Apathy scores in Parkinson’s disease relate to EEG components in an incentivized motor task

Abstract Apathy is one of the most prevalent non-motor symptoms of Parkinson’s disease and is characterized by decreased goal-directed behaviour due to a lack of motivation and/or impaired emotional reactivity. Despite its high prevalence, the neurophysiological mechanisms underlying apathy in Parkinson’s disease, which may guide neuromodulation interventions, are poorly understood. Here, we investigated the neural oscillatory characteristics of apathy in Parkinson’s disease using EEG data recorded during an incentivized motor task. Thirteen Parkinson’s disease patients with apathy and 13 Parkinson’s disease patients without apathy as well as 12 healthy controls were instructed to squeeze a hand grip device to earn a monetary reward proportional to the grip force they used. Event-related spectral perturbations during the presentation of a reward cue and squeezing were analysed using multiset canonical correlation analysis to detect different orthogonal components of temporally consistent event-related spectral perturbations across trials and participants. The first component, predominantly located over parietal regions, demonstrated suppression of low-beta (12–20 Hz) power (i.e. beta desynchronization) during reward cue presentation that was significantly smaller in Parkinson’s disease patients with apathy compared with healthy controls. Unlike traditional event-related spectral perturbation analysis, the beta desynchronization in this component was significantly correlated with clinical apathy scores. Higher monetary rewards resulted in larger beta desynchronization in healthy controls but not Parkinson’s disease patients. The second component contained gamma and theta frequencies and demonstrated exaggerated theta (4–8 Hz) power in Parkinson’s disease patients with apathy during the reward cue and squeezing compared with healthy controls (HCs), and this was positively correlated with Montreal Cognitive Assessment scores. The third component, over central regions, demonstrated significantly different beta power across groups, with apathetic groups having the lowest beta power. Our results emphasize that altered low-beta and low-theta oscillations are critical for reward processing and motor planning in Parkinson’s disease patients with apathy and these may provide a target for non-invasive neuromodulation.


Graphical Abstract Introduction
Apathy is one of the most debilitating non-motor symptoms of Parkinson's disease (PD), manifesting as little or no goaldirected behaviour, along with multifaceted emotional and behavioural symptoms including reduced interest, lack of concern, emotional indifference and decreased initiation of activity. 1 The ability to assess the value of potential rewards is a crucial aspect of motivation and goal-directed activity, and thus, the presence of apathy has a profoundly negative impact on people with PD, reducing their overall functioning and quality of life.While apathy in the non-PD population is mostly seen in the setting of depression, in PD apathy can be seen independently. 2 PD-related apathy, which can affect between 17% and 70% of people, 1,3 can result in withdrawal from physical activities, hobbies and social interactions, cognitive decline and poor engagement in rehabilitative treatments and impose a high burden on caregivers.
Currently, apathy is assessed by employing clinical assessment instruments such as the neuropsychiatric inventory, Lille apathy rating scale (LARS), apathy evaluation scale or apathy scale.While these apathy scales provide useful and reliable psychometric properties in PD, 4 neuroimaging studies have provided additional insights into the anatomical and functional substrates implicated in apathy.][7][8][9][10] For instance, PET studies have shown that apathy in PD is associated with reduced metabolic activity in the VS and medial frontal brain regions, including the ACC. 6,11,12PD patients with apathy also exhibit diminished dopamine receptor binding capacity in the bilateral VS 13 and blunted dopamine release in the ACC, orbitofrontal cortex, dorsolateral PFC, thalamus and globus pallidus internal. 14In addition to the alterations in the frontostriatal circuits, several neuroimaging studies suggest that apathy in PD may stem from a severe dopamine depletion state in the mesocorticolimbic circuitry. 15Overall, these changes in neural systems have cascading effects on other brain regions, impacting the evaluation of costs and benefits (here, the 'costs' are assumed to reflect the effort associated with taking an action, and the 'benefit' involves the valuation of the expected potential rewards from the action 5,16,17 ).For example, the activity in the ACC and its functional connectivity with the supplementary motor area have been found to predict behaviour apathy scores. 18everal studies have investigated abnormality in the costbenefit valuation underlying motivational modulation of motor behaviour of PD patients with apathy.Heron et al. 19 demonstrated that when participants were asked to make decisions on whether to accept or reject monetary rewards for exerting different levels of physical effort, PD patients with apathy made more rejections of offers (i.e.lower willingness to perform the physical effort) compared with both HCs and PD patients without apathy.This was predominantly observed in the trials with a low-level reward as the patients with apathy were found to be willing to exert effort to the same level as the patients without apathy if the reward was sufficiently high.This suggests that diminished 'drive' by the potential rewards for the action acts as a key factor in motivated behaviour in PD patients with apathy rather than an enhanced sensitivity to effort costs.Altered reward processing in PD patients with apathy was also observed in their blunting of pupillary dilation responses to incentives in saccadic eye movement tasks. 10,20Together, these studies suggest that alterations in processing incentivizing value of rewards play a crucial role in behavioural changes in PD patients with apathy (Fig. 1).
EEG and magnetoencephalography (MEG) studies offer valuable insights into the neuronal mechanisms underlying the reward processing and behaviours.One common approach involves analysing beta band oscillations (13-30 Hz) in motor and pre-motor areas. 22In healthy individuals, it has shown that the extent of beta suppression following reward cues is closely linked to motivational intensity 23 and reflects the expected reward value. 24Despite a substantial body of evidence demonstrating abnormal beta oscillations in PD, [25][26][27][28][29][30][31][32][33] their exact relationship with reward evaluation remains poorly understood.Notably, there is a scarcity of EEG studies investigating the neurophysiological mechanisms of apathy in PD, despite its high prevalence.To the best of our knowledge, only two EEG studies have examined reward processing in PD patients with apathy.One study 34 measured time-locked event-related potentials (ERPs) to gain or loss feedback stimuli while the participants performed a modified Gehring's gambling task.The study discovered significantly decreased amplitude of feedback-related negativity in PD patients with apathy compared with PD patients without apathy and HCs.The reduced amplitude was more notable for high-win than high-loss trials, suggesting a more selective impairment of neural processes under reward than punishment in PD patients with apathy.Another EEG study investigated alterations in spectral power and its relation to motivational motor behaviour in an incentivized motor task paradigm. 35ompared with PD patients without apathy and healthy individuals, PD patients with apathy were found to exhibit greater desynchronization in theta (4-7 Hz) and alpha (8-12 Hz) frequency bands during movement initiation and execution.The authors conjectured that it might reflect a compensatory phenomenon to counteract higher baseline theta and alpha power in the patients with apathy.
Here, we focused on apathy as a reduced motivation for goal-directed behaviour.We aimed to investigate characteristics of cortical oscillatory activity associated with incentive processing and assess whether there are changes associated with PD patients with apathy.Particularly, we introduce a novel data-driven method that can extract time-frequency domain features in multichannel EEG data that are highly relevant to the task and maximally correlated across participants.The proposed method has the advantage of (i) taking into account differences in spatial locations of event-related EEG dynamics across participants 36 and (ii) obviating the necessity of confining power spectral analysis to specific frequency bands and EEG channels a priori.With this new  21 (licenced under CC BY 4.0, http://creativecommons.org/licenses/by/4.0/).For a given reward (e.g.$1), people integrate information about the expected benefit (solid blue) and cost (solid red) to determine the expected value of motor output (solid purple), and then, they allocate the intensity of motor output (solid black arrow).For the same task and same amount of reward at stake, people with apathy may have diminished 'drive' by the potential benefit (dashed blue), which consequently alters the expected value of control (dashed purple) and intensity of effort (dashed black arrow).EVC, expected value of control. [39]

Participants
Patients diagnosed with idiopathic PD and age-and sexmatched HCs under the age of 85, with normal or corrected-to-normal vision, were recruited for this study.PD patients were recruited from the Movement Disorders Clinic at the University of British Columbia and prescribed with a stable dosage of an antiparkinsonian medication for at least 2 months before study enrolment.They were on their regular dopaminergic medication during the study visit.HCs were either spouses of patients or recruited from the community.The presence of depression and apathy was assessed using Beck's depression inventory (BDI) and Starkstein apathy scale (SAS), respectively.The disease severity of the PD patients was assessed using the Movement Disorder Society-Unified Parkinson's disease Rating Scale (UPDRS) Part III (motor examination).PD patients were classified into two groups, PDA+ (PD with apathy) and PDA− (PD without apathy), based on their SAS scores (see Demographics).All participants provided written informed consent before participation.The study protocol was approved by the Clinical Research Ethics Board at the University of British Columbia.

Study protocol
Participants performed an incentivized motor task using a grip force transducer (Hand Dynamometer Logger Sensor NUL-237, NeuLog, USA) while seated comfortably in front of a 19 inch computer screen.
Before beginning the motor task, we determined each participant's maximum voluntary contraction (MVC) to calibrate the target level of force used in the actual motor task to individuals.In the calibration process, participants were asked to squeeze the grip force transducer as hard as they could over three times using their dominant hand.The maximum of the three grips was recorded as the MVC.Participants were then asked to squeeze the grip force transducer to reach the red line on a graduated scale on the computer screen with real-time feedback, where the red line represented either 40%, 80% or 120% of the previously recorded MVC.The 120% grip force level was to ensure that the previously recorded MVC was the actual maximum participants could squeeze.For a total of nine trials, each target level of force was randomly presented three times.If participants squeezed harder than the recorded MVC in any of the three trials where the red line was 120% of the MVC, the maximum of the three 120% trials became the new recorded MVC.However, if participants, on any trial, reached or exceeded 120% of the first recorded MVC, then they were asked to restart the whole calibration process.
The task consisted of showing participants a graduated scale on the computer screen and asking them to squeeze the grip force transducer for a monetary reward proportional to how hard they squeezed.Each trial consisted of the following steps (Fig. 2): first, participants were shown the maximum monetary reward they could earn for this trial for 2 s.The maximum monetary reward was one of $1, $10 or $50.Then, they were shown a graduated scale for 4 s and asked to squeeze the grip force transducer to earn the reward.The reward amount they could earn was proportional to how hard they squeezed while their grip force was shown on the graduated scale.At the top of the graduated scale was the maximum monetary reward shown at the start of the trial.To avoid ceiling effects, the graduated scale was set such that the 50% level was the participants' MVC, as recorded in the calibration process.Then, participants were shown how much they had earned from the trial along with their current total winnings for 2 s.Finally, there was a 9 s rest before the subsequent trial.A total of 45 trials were performed, with the maximum monetary rewards of $1, $10 or $50 being presented 15 times each in randomized order.

EEG recording and preprocessing
EEG data were recorded from 34 scalp electrodes using a 64-channel Quik-Cap (Neuroscan, VA, USA) and a Neuroscan SynAmps 2 acquisition system (Neuroscan, VA, USA) at a sampling rate of 500 Hz.Recording electrodes were positioned according to the International 10-20 placement standard with one ground electrode (AFz) and one reference electrode located between Cz and CPz.Impedances were kept below 15 k Ω using Electro-Gel (Electrode-Cap International, OH, USA).
All EEG data were preprocessed offline using custom MATLAB scripts and functions from the open-source EEGLAB (https://sccn.ucsd.edu/eeglab)toolbox.Continuously recorded EEG signals were first bandpass filtered between 1 and 55 Hz using a two-way finite impulse response (FIR) filter (the 'eegfilt' function in EEGLAB) and then were re-referenced to average reference.Stereotypical artefacts, including ocular artefacts (EOG) and muscle tension, were removed using an automatic artefact rejection method 40 based on independent component analysis (ICA). 41The EEG data were then segmented into non-overlapping 6 s epochs, each spanning 1 s of rest (i.e.baseline) before the beginning of the trial, 2 s of reward cue and 3 s of squeezing.The segmentation resulted in 45 epochs per participant, with 15 epochs per reward value ($1, $10 or $50).

EEG time-frequency analysis
After EEG segmentation, a continuous wavelet transform was applied to the 6 s EEG epochs using a seven-cycle, complex Morlet wavelet for a frequency range of 1-55 Hz (1 Hz steps).Spectral power for each frequency and timepoint was estimated by multiplying the complex signal from the wavelet transform by its complex conjugate.For each epoch, the mean baseline power (P 0 ) was computed by averaging the power during the baseline period (i.e. the first 1 s rest preceding a reward cue).Baseline-normalized event-related spectral perturbation (ERSP) was then computed by taking 10*log10 of the ratio of the spectral power to the mean baseline power.
We then discarded the first 1 s and used the remaining 5 s task portion (i.e. 2 s reward cue + 3 s squeezing) of the ERSPs for further analysis (Fig. 3A).

Multiset canonical correlation analysis
Multiset canonical correlation analysis (MCCA) is an extension of canonical correlation analysis (CCA) to allow for the joint analysis of more than two data sets.MCCA optimizes an objective function to achieve the maximum overall correlation of the canonical variates across the multiple data sets. 42,43In this study, we exploited MCCA to extract highly correlated time-frequency features in the baselinenormalized ERSPs across all trials and participants.Since the ERSPs represent task-related brain activities induced by the events (Fig. 2) that were identically timed across all trials and participants, we used MCCA to extract canonical variates of the ERSPs that are highly consistent across the trials and participants.
Figure 3 shows a schematic diagram explaining how MCCA was applied to the ERSP data.We first downsampled the ERSPs from the original sampling rate of 500-10 Hz (Fig. 3A).This resulted in an ERSP matrix with a dimension of 55 × 50 (frequency × time) per EEG channel and per epoch.Then, we vectorized the matrix and concatenated the ERSP vector for each channel along the row dimension, creating a 34 × 2750 ERSP matrix per epoch (Y k ; Fig. 3A).
Subject-level MCCA (Fig. 3B) was performed on the 45 ERSP matrices to extract the most consistent spectral patterns across all trials: Each MCCA component (i.e.canonical variate) consisted of a pair of weight (each row in U k ∈ R 20×34 ; k = 1, 2, . . ., 45 trials) and common profile (each row in ).The weight and common profiles are sorted in descending order according to the overall correlation across the 45 trials, such that the first row presents the MCCA component with the maximum overall correlation.The original ERSP data (Y) and common profile (Q) were then averaged across all trials, respectively.
A multiple linear regression was performed with each channel of Y s as the dependent variable and Q s as the independent variable.Finally, a mean ERSP matrix (X s ∈R 34×2,750 ) was obtained Squeeze: a graduated scale was displayed on the computer screen.The participants were instructed to squeeze a grip force transducer to increase the bar to the top and told that the amount of reward they would earn was proportional to how hard they squeezed within a 4 s window.Feedback: feedback on the reward they earned for the trial and total accumulated reward was given for 2 s.Rest: the participants were given 9 s to rest before the subsequent trial began.
for each subject s (s = 1, 2, . . ., 38) by projecting the original mean scalogram (Y s ) onto the common profile (Q s ): The subject-level MCCA was performed based on the assumption that the components maximally correlated across different trials are more likely to be associated with task-relevant brain activities ('signal') than transient brain activities that appear in only one epoch or a small subset of the epochs or random noise.Therefore, we expected to obtain an ERSP matrix (X s ) with an improved signal-to-noise ratio by projecting the original mean scalogram (Y s ) onto the subspace of the MCCA components.X s obtained from the subject-level MCCA was subsequently used as inputs to a group-level MCCA (Fig. 3C) to extract the time-frequency patterns that are maximally correlated across the participants: Each common profile in P s can be understood as a linear combination of the time-frequency features in X s weighted by the corresponding row in W s indicating the contribution of each channel.In this study, the most correlated time-frequency patterns across the participants were summarized into three components, which explained 90.1% variance of the input data X.Each common profile from the group-level MCCA was averaged across the participants in each of the three groups (PDA+, PDA− and HC) for visualization purposes.The scalp maps were created by averaging the corresponding weights (W) across the participants.

Conventional EEG spectral analysis
Additionally, we analysed the original ERSPs using a conventional method to compare the results with those obtained using the proposed MCCA approach.Conventionally, EEG electrodes of interest are pre-defined before any group-level analysis, often based on prior studies or domain knowledge about the brain regions involved in the neural processes of interest.][46][47] For each participant, we computed the grand mean ERSP by firstly taking the mean of the baseline-normalized ERSPs of 45 trials and then taking the mean over the selected electrodes.We also computed a mean ERSP over the 15 trials per incentive level to investigate whether the ERSP were modulated by the reward condition ($1, $10 or $50).The mean ERSPs were further used for statistical and correlation analyses.

Statistical analysis
The group mean common profiles from MCCA were first visually inspected to determine the frequency bands and time periods that show clear differences between groups.The selected frequency bands and time periods were as follows: Common For each of the frequency bands and time periods of interest listed above, the mean spectral power was computed per participant.We performed a one-way ANOVA with group (PDA+, PDA−, HC) as a between-subject factor to test whether the mean power differs significantly across the groups.In addition, Tukey's honestly significant difference (HSD) tests were conducted for post hoc comparisons.
To test whether the spectral power listed above varied with different reward levels, we computed the mean spectral power for each reward level per participant.In cases where the data were not normally distributed, we conducted a Friedman test to investigate differences in spectral power between different reward levels ($1/$10/$50) within each group.Post hoc comparisons were calculated using Wilcoxon signed-rank tests.

Demographics
Twenty-seven PD patients and 13 HCs took part in the experiment.All participants were right-handed.The presence of apathy was determined by SAS scores ≥14.Accordingly, 13 out of the 27 PD patients were classified into PDA+ group and the remaining 14 PD patients were classified into PDA− group.All HCs had SAS scores below 14.One PDA− patient was excluded from the data analysis due to device malfunction, and one HC participant was excluded due to excessive artefacts in the EEG data.The demographic and clinical information of the remaining 13 PDA+, 13 PDA − and 12 HC participants are displayed in Table 1.

Subject-level analysis
To investigate the effects MCCA has on the ERSP data, we performed qualitative and quantitative comparisons between the mean ERSP obtained from the original data (Y s ) and MCCA (X s ). Figure 4A shows the qualitative comparison of the mean ERSP of channel FP1 obtained from a representative participant.It can be seen that the MCCA-derived ERSP (right panel in Fig. 4A) preserves the task-relevant brain activities, demonstrating the beta desynchronization and subsequent gamma synchronization occurred during squeezing (2-5 s).The other spectral changes appeared to be relatively indistinct, making the MCCA-derived ERSP less noisy compared with the original ERSP.Quantitative comparisons are shown in Fig. 4B where the percentage of the variance of the original ERSP explained by the MCCA-derived ERSP is shown.We found that the MCCA-derived ERSP preserved in average 89.8 ± 3.7% (mean ± standard deviation) of the variance contained in the original ERSPs.

Group-level analysis
We investigated whether the ERSP obtained from the grouplevel MCCA (P s ; s = 1, 2, …, 38) had stronger betweensubject correlations compared with the original ERSP (Y s ; s = 1, 2, …, 38).The between-subject correlations were computed as the correlations of ERSP between every pair of two participants for each channel (Supplementary Fig. 1A) or component (Supplementary Fig. 1B).It was found that the between-subject correlations of the original ERSP ranged from r = 0.35 ± 0.16 to r = 0.45 ± 0.15 across channels.The grand mean of the between-subject correlations of all channels was r = 0.40 ± 0.18.On the other hand, the between-subject correlations of the first MCCA ERSP were found to be r = 0.74 ± 0.08 (Supplementary Fig. 1B), which was significantly higher than those of the original ERSP (D = 0.83, P < 0.001; two-sample Kolmogorov-Smirnov test).The second MCCA ERSP had the betweensubject correlations of r = 0.54 ± 0.16, which was also significantly higher than those of the original ERSP (D = 0.32, P < 0.001; two-sample Kolmogorov-Smirnov test).In contrast, the between-subject correlations for the third MCCA ERSP were significantly lower than those of the original ERSP (D = 0.39, P < 0.001; two-sample Kolmogorov-Smirnov test).

First group-level MCCA component
Figure 5A shows the ERSP (P) at the bottom and its weights across EEG channels (W) at the top for the first group-level MCCA component.The results are demonstrated as the group means of the participants in the PDA+, PDA− and HC groups, respectively.For all three groups, the EEG channels in the parietal region contributed most strongly to the ERSP, followed by the EEG channels in the frontal region.The spectral power changes in the ERSP were characterized by the increased power in the delta (<4 Hz, 0-5 s) and gamma (30-55 Hz, >3 s) bands and the decreased power in the beta band (12-30 Hz, >0.3 s).The beta power changes, particularly in the low-beta frequency band (12-20 Hz), were of particular interest because (i) the temporal profile corresponded well to the timing of task stimuli (i.e.greater attenuation immediately after the reward cue presented at t = 0 s and target force level presented at t = 2 s) and (ii) prior studies have indicated that low-beta oscillations are implicated in impaired movement initiation in PD. 48,49 Therefore, we investigated changes in the low-beta band power (Fig. 5A) during a reward cue (a: 0.3-1 s) and squeezing (b: 2.5-4 s) to find out whether there is a significant difference across the PDA+, PDA− and HC groups.
A one-way ANOVA analysis demonstrated a significant group difference in the beta desynchronization during reward cue presentation [F(2,35) = 3.64, P < 0.05].In addition, Tukey's HSD post hoc multiple comparison tests revealed a greater beta desynchronization in the HC group compared with the PDA+ group (P HSD < 0.05; Fig. 5B), whereas there was no significant difference between PDA+ and PDA− groups (P HSD = 0.72) and PDA− and HC groups (P HSD = 0.16).
We further evaluated the clinical significance of the diminished beta desynchronization observed in the apathy group using the SAS and LARS apathy scores (Fig. 5C).The beta desynchronization during the reward cue was positively correlated with SAS (r = 0.42, P < 0.01; N = 38) and LARS (r = 0.32, P < 0.05; N = 38) scores, respectively, indicating that the more apathetic the participants are, the less beta desynchronization they exhibit during the reward cue.We further investigated whether the beta desynchronization level was modulated by the values of the monetary incentive (i.e.$1, $10 or $50).Friedman tests showed a significant effect of the monetary value on the beta desynchronization for the PDA− (χ 2 (2) = 6.4,P < 0.05) and HC (χ 2 (2) = 6.9, P < 0.05) groups, but not for the PDA+ group (χ 2 (2) = 2.0, P = 0.37) (Figure 5D).Post hoc Wilcoxon signed-rank tests revealed that $50 reward resulted in a greater amount of beta desynchronization in both PDA− (Z = 2.34, P < 0.05) and HC (Z = 2.35, P < 0.05) participants.In contrast, $10 reward resulted in greater beta desynchronization in the HC participants only (Z = 2.27, P < 0.05).
The beta desynchronization during squeezing (2.5-4 s) did not significantly differ across the three groups

Second group-level MCCA component
The ERSP and weights of the second MCCA component are shown in Fig. 6A.The weights presented in the scalp maps showed the spectral changes in the ERSP were associated with frontal and parietal brain regions.The ERSP was featured with the increased theta (4-8 Hz) power over the entire period of the reward cue and squeezing and attenuated gamma (30-55 Hz) power during squeezing (2.5-5 s) compared with the baseline.A one-way ANOVA analysis showed a significant group difference in the theta power [F(2,35) = 3.4,P = 0.045; Fig. 6B], with a significantly higher theta power found in the PDA+ group compared with the HC group (P HSD = 0.048).The PDA− group tended to have a higher theta power than the HC group, but this difference did not reach statistical significance (P HSD = 0.10).The increased theta power observed in the PD participants in both the PDA+ and PDA− groups was found to be significantly correlated with their Movement Disorder Society-Unified Parkinson's Disease Rating Scale (MDS-UPDRS) III scores (r = 0.48, P = 0.012, N = 26), indicating that the patients with higher disease severity exhibited exaggerated theta power during the reward cue and squeezing.We also found that the theta power was negatively correlated with the Montreal Cognitive Assessment (MoCA) scores of the PD participants (r = −0.43,P = 0.030, N = 26), suggesting that the excessive theta power may be also related to cognitive decline in PD.When all participants were taken into account, the correlation between the theta power and MoCA scores became no longer statistically significant (r = 0.26, P = 0.116, N = 38).
For the gamma power during squeezing, we did not find a significant difference across the three groups [one-way ANOVA: F(2,35) = 1.16,P = 0.330; Fig. 6E].

Third group-level MCCA component
The ERSP of the third MCCA component was found to be associated with the EEG electrodes located in the central region (Supplementary Fig. 2A).The ERSP showed enhanced gamma (30-55 Hz) power during reward cue (0-2 s), followed by suppression of beta (12-30 Hz) power during squeezing (2.5-4 s).A one-way ANOVA revealed no significant difference in gamma power [F(2,35) = 3.4,P < 0.05] across the three groups.In contrast, the beta power was found to be significantly different across the groups [one-way ANOVA: F(2,35) = 3.68, P < 0.05], with PDA+ group having a significantly lower beta power compared with the HC group (P HSD < 0.05) during squeezing.
We investigated whether participants' average grip force during the experiment was associated with their apathy scores and the ERSP obtained from the group-level MCCA.Specifically, we used the low-beta ERSP of the first MCCA component during reward cue (Fig. 5).The mean grip force (across all trials) of the participants was found to be negatively correlated with both the SAS scores (r = −0.35,P < 0.05, N = 38; Fig. 7B) and low-beta power (r = −0.43,P < 0.01, N = 38; Fig. 7B) during reward cue, indicating that the participants with greater apathy severity used lower grip force and had higher low-beta power (i.e.smaller amount of beta desynchronization) during the reward cue.

ERSP results from conventional analysis
Supplementary Figure 3A shows the group mean ERSPs over the central-parietal channels (C3, CZ, C4, CP5, CP1, CPZ, CP2, CP6, P3, PZ and P4).The low-beta power was found to be suppressed during reward cue (a: 0.3-1 s) and squeezing (b: 2.5-4 s), similar to the results found in the first group-level MCCA component (Fig. 5A).One-way group ANOVA revealed a significant group difference in the beta power during reward cue [F(2,35) = 5.23, P < 0.05; Supplementary Fig. 3B].The amount of beta desynchronization in the HC group was greater than the PDA− group (P HSD < 0.01) but not the PDA+ group (P HSD = 0.080).No significant correlation was found between the beta power during reward cue and apathy scores (SAS: r = 0.04, P = 0.808; LARS: r = 0.06, P = 0.724; Supplementary Fig. 3C).We did not find a significant correlation of the beta power with the apathy scores even when we computed the beta power over every possible subset of the central-parietal channels as shown in Supplementary Fig. 3D (SAS: r = 0.04 ± 0.05; LARS: r = 0.06 ± 0.04), which was in contrast to the significant correlations found with the first MCCA component (Fig. 5C).Furthermore, it was found that the beta power did not vary according to different reward cues (Supplementary Fig. 3E).

Effects of sampling rate on MCCA
The MCCA results discussed in previous sections were based on ERSPs downsampled to 10 Hz, a step taken to reduce computational burden and data storage space.To assess the consistency of our findings at a higher sampling rate, we downsampled the ERSPs from the original 500 to 50 Hz and applied the same subject-level and group-level MCCA procedures outlined in Fig. 3 to the 50 Hz ERSP data.
The subject-level analysis results (Supplementary Fig. 4A) closely mirrored those obtained from ERSPs sampled at 10 Hz, affirming that the MCCA-derived ERSP preserves the task-relevant cortical oscillations.The MCCA-derived ERSPs explained 90.5 ± 3.4% of the variance in the original ERSPs (Supplementary Fig. 4B).The grand mean of between-subject correlations for the 50-Hz ERSPs across all channels (Supplementary Fig. 5A) was r = 0.40 ± 0.18, while the between-subject correlations for the first, second and third MCCA-derived ERSPs (Supplementary Fig. 5B) were r = 0.75 ± 0.09, 0.54 ± 0.17 and 0.17 ± 0.25, respectively.These results closely paralleled the results shown in Supplementary Fig. 1B.Supplementary Figures 6-8 show the group-level MCCA components derived from the 50 Hz ERSPs, affirming the consistency of the group-level results in comparison with the MCCA results obtained from 10 Hz ESRPs, as previously presented in Figs. 5 and 6 and Supplementary Fig. 2.

Discussion
This study investigated the brain oscillations recorded in EEG to study the neurophysiological processes underlying motivational control of goal-directed movements and how it is associated with the clinical phenotype of apathy in PD.We investigated ERSPs induced by the reward cues and squeezing actions using a novel data-driven method utilizing MCCA to extract spectral features that are maximally correlated across trials and participants and therefore provided more robust representations of the brain dynamics related to the monetary incentive task.
An important finding of the present study was the ability of the proposed method to disentangle different spectral perturbations into separate components.The first MCCA component demonstrated strong modulation of beta oscillations during the performance of the incentivized task.The low-beta (12-20 Hz) power attenuated during reward cues, and the amount of suppression was inversely correlated with clinical apathy scores.That is, the participants with greater apathy severity showed impaired beta suppression to reward cues.Furthermore, the beta suppression was sensitive to the valence of potential reward in the HC group, but this was observed between $1 and $50 cues in the PDA− group and not observed at all in the PDA+ group.In contrast, the amount of beta suppression computed using a conventional method using the same frequency range and time segment was not found to be correlated with the apathy scores nor was it sensitive to the level of potential reward, even in HCs.These results suggest that the proposed MCCA approach could be a valuable way to derive ERSPs that are more robust to noise and task-irrelevant neural processing and reflect the essential neural processes underlying the task.
Our result of the beta power mediated by motivational salience, as revealed by the first MCCA component, is corroborated by a study 50 that investigated the effects of positive and negative monetary incentives on beta oscillations during an incentivized goal-directed reaching task.The amount of beta suppression during the reward cue was found to be significantly greater in reward trials compared with neutral and punish trials.It also scaled with the rewards at stake and significantly correlated with the participants' movement time.These results support the notion that pre-movement beta suppression reflects the neural processes underlying incentive-driven motor planning, which is disrupted in PD patients with apathy.
5][56] The sensorimotor cortex and basal ganglia are considered the principal sources for the emergence of beta desynchronization. 39Extensive studies have tried to elucidate the functional roles of beta oscillations in the sensorimotor system.8][59][60][61] In this context, beta activity needs to be inhibited to allow the initiation of motor planning and execution. 3963][64][65] In this context, our result demonstrating reduced beta desynchronization during reward cues in PD participants may partially reflect the pathological changes in the neural processing necessary for motor planning for an intended movement.
In the PD participants with apathy, we observed an overall reduction in the amount of beta suppression (Fig. 5B), and their beta suppression did not scale with the reward level compared with the PD participants without apathy and HCs (Fig. 5D).Behaviourally, the PD participants with apathy also displayed reduced sensitivity to changes in reward level (Fig. 7A).Our results are aligned with the findings from prior research that PD patients with apathy are less likely to engage in physical effort for a reward, especially when the reward is relatively small, as compared with those PD patients without apathy. 19,20A recent study 10 has investigated whether this diminished reward sensitivity is present in PD patients with apathy regardless of motor preparation.The authors found that it only manifests when actions are required to achieve rewarding goals.This suggests that the interaction between reward evaluation and the initiation of goal-directed action plays a crucial role in apathy in PD.Unfortunately, our study design does not allow us to discern whether our findings reflect impairments in sensitivity to reward (i.e.reward evaluation) or translating reward evaluation into effort (i.e.motor planning) or a combination of both.In light of the insights from the aforementioned studies, the aberrant beta suppressions we observed in PD patients with apathy may reflect pathological neural mechanisms underlying both reward evaluation and motor planning. 10To the best of our knowledge, no study to date has explicitly investigated the role of cortical beta oscillations in PD patients with apathy.Therefore, further research is warranted in the future to delve deeper into and substantiate this hypothesis.
In the second MCCA component, we observed that frontal-parietal theta power was generally higher in PD participants than HCs throughout the reward cue and squeezing periods.In contrast to the beta power of the first MCCA component, the theta power did not correlate with the incentive level or grip force responses.Based on prior studies, we conjecture that the elevated theta power we observed may be related to a general slowing of background cortical oscillatory activity in PD patients.As reported in a recent systematic review examining 19 studies with 312 PD patients and 277 controls, 66 a large body of EEG studies on PD have demonstrated slowing of cortical activity, which is reflected as an increase of spectral power of delta and theta oscillations or a slower peak frequency.0][71] In support of this notion, we found a significant correlation between the elevated theta power and lower MoCA scores in the PD participants (Fig. 6C).
The other prominent feature of the second MCCA component was the suppression of gamma oscillations.Gamma rhythms (30-50 Hz) have been implicated in various cognitive tasks such as working memory, mental arithmetic, visuomotor coordination and selective attention. 72,73Prior studies on PD patients with dementia or MCI often report EEG slowing in the patients accompanied by decreased power of gamma oscillations, [74][75][76] and similarly, we found that the theta power was inversely correlated with the gamma power (r = −0.40,P = 0.013).Since the gamma suppression was extracted together with the elevated theta power in the same MCCA component, we surmise it may also reflect some cognitive aspects of brain activity.
To conclude, the present study provides a critical link between the clinical features of apathy in PD and altered oscillatory dynamics, including beta, gamma and theta band frequencies.Given that neuromodulation can augment desynchronization in certain circumstances, 77 our results may allow for specific non-pharmacological targeting of apathy symptoms in PD.

Figure 1 A
Figure 1 A schematic diagram to illustrate allocation of behavioural effort determined by expected benefit and cost.The diagram was adapted from the EVC theory21 (licenced under CC BY 4.0, http://creativecommons.org/licenses/by/4.0/).For a given reward (e.g.$1), people integrate information about the expected benefit (solid blue) and cost (solid red) to determine the expected value of motor output (solid purple), and then, they allocate the intensity of motor output (solid black arrow).For the same task and same amount of reward at stake, people with apathy may have diminished 'drive' by the potential benefit (dashed blue), which consequently alters the expected value of control (dashed purple) and intensity of effort (dashed black arrow).EVC, expected value of control.

Figure 2
Figure 2 Monetary incentive delay task.Reward cue: participants were presented with a certain amount of reward ($1, $10 or $50) for 2 s to indicate the maximum rewardthey could earn for the trial.Squeeze: a graduated scale was displayed on the computer screen.The participants were instructed to squeeze a grip force transducer to increase the bar to the top and told that the amount of reward they would earn was proportional to how hard they squeezed within a 4 s window.Feedback: feedback on the reward they earned for the trial and total accumulated reward was given for 2 s.Rest: the participants were given 9 s to rest before the subsequent trial began.

Figure 3
Figure 3 EEG data analysis using wavelet transform and MCCA.(A) A scalogram was obtained by applying a continuous wavelet transform to EEG timeseries.The ERSP scalogram was normalized based on the mean power during the baseline period, downsampled to 10 Hz and vectorized.(B) Subject-level MCCA was performed on the ERSP matrices (Y k ∈ R 34 × 2750 ) of 45 trials to extract 20 components (Q k ∈ R 20 × 2750 ) that are maximally correlated across the 45 trials.The original ERSP matrices (Y k ) and MCCA components (Q k ) were averaged, and a multiple linear regression was performed with each column in Y Ts as a dependent variable and the MCCA components Q T s as independent variables.(C) Group-level MCCA was performed with the ERSP matrices (X s ∈ R 34 × 2750 ) derived from the subject-level analysis, resulting in three MCCA components.Each component consists of a common profile and corresponding weights on the EEG channels (depicted in dotted boxes).ERSP, event-related spectral perturbation; MCCA, multiset canonical correlation analysis.

Figure 4
Figure 4 Qualitative and quantitative comparison of the original ERSP and ERSP derived from MCCA.(A) The trial mean ERSP of channel FP1 from a representative participant.The MCCA-derived ERSP (i.e.X s in Fig. 3) preserves task-relevant spectral changes in the original ERSP (Y s ) as denoted by arrows.(B) The distribution of the variance of the original ERSP explained by the MCCA-derived ERSP computed for each participant per channel.ERSP, event-related spectral perturbation; MCCA, multiset canonical correlation analysis.

Figure 5
Figure 5 First group-level MCCA component.(A) The ERSP and its weights across EEG channels are demonstrated as a scalogram (bottom) and scalp map (top) for each group.The low-beta (12-20 Hz) frequency band during a reward cue (a: 0.3-1 s) and squeezing (b: 2.5-4 s) is denoted as dotted boxes.(B) Group comparison of the beta power during a reward cue (statistics: one-way ANOVA, Tukey's honestly significant difference test).(C) Correlations between the beta power during a reward cue and clinical apathy scores (SAS and LARS) of all the participants.(D) The beta power during the reward cue is presented per reward level for each group (statistics: Friedman test, Wilcoxon signed-rank test).(E) Group comparison of the beta power during squeezing (statistics: one-way ANOVA).*P < 0.05.ERSP, event-related spectral perturbation; HC, healthy controls (N = 12); LARS, Lille apathy rating scale; MCCA, multiset canonical correlation analysis; PDA+, Parkinson's disease patients with apathy (N = 13); PDA−, Parkinson's disease patients without apathy (N = 13); SAS, Starkstein apathy scale.

Figure 6
Figure 6 Second group-level MCCA component.(A) The ERSP and its weights across EEG channels are demonstrated as a scalogram (bottom) and scalp map (top) for each group.The theta (4-8 Hz) frequency band during the reward cue and squeezing (a: 0-5 s) and gamma (30-55 Hz) frequency band during squeezing (b: 2.5-5 s) are denoted as dotted boxes.(B) Group comparison of the theta power during the reward cue and squeezing (statistics: one-way ANOVA, Tukey's honestly significant difference test).(C) Correlation between the theta power and MDS-UPDRS Part III scores (left) and between the theta power and MoCA scores (right) of the PD participants.(D) The theta power is presented per reward level for each group (statistics: Friedman test).(E) Group comparison of the gamma power during squeezing (statistics: one-way ANOVA).*P < 0.05.ERSP, event-related spectral perturbation; HC, healthy controls (N = 12); LARS, Lille apathy rating scale; MCCA, multiset canonical correlation analysis; MDS-UPDRS, Movement Disorder Society-Unified Parkinson's Disease Rating Scale; MoCA, Montreal Cognitive Assessment; PD, Parkinson's disease; PDA+, Parkinson's disease patients with apathy (N = 13); PDA−, Parkinson's disease patients without apathy (N = 13); SAS, Starkstein apathy scale.

Table 1 Demographic comparison of the participants
Numbers in brackets represent standard deviations.All participants are right-handed.SAS, Starkstein apathy scale; LARS, Lille apathy rating scale; BDI, Beck's depression inventory; MoCA, Montreal Cognitive Assessment; MDS-UPDRS, Movement Disorder Society-Unified Parkinson's Disease Rating Scale; n/a, not applicable.*indicates significant a P-value.