Clinically established early Parkinson ’ s disease patients do not show impaired use of priors in conditions of perceptual uncertainty

The ability to use past learned experiences to guide decisions is an important component of adaptive behavior, especially when decision-making is performed under time pressure or when perceptual information is unreliable. Previous studies using visual discrimination tasks have shown that this prior-informed decision-making ability is impaired in Parkinson ’ s disease (PD), but the mechanisms underlying this deficit and the precise impact of dopaminergic denervation within cortico-basal circuits remain unclear. To shed light on this problem, we evaluated prior-informed decision-making under various conditions of perceptual uncertainty in a sample of 13 clinically established early PD patients, and compared behavioral performance with healthy control (HC) subjects matched in age, sex and education. PD patients and HC subjects performed a random dot motion task in which they had to decide the net direction (leftward vs. rightward) of a field of moving dots and communicate their choices through manual button presses. We manipulated prior knowledge by modulating the probability of occurrence of leftward vs. rightward motion stimuli between blocks of trials, and by explicitly giving these probabilities to subjects at the beginning of each block. We further manipulated stimulus discriminability by varying the proportion of dots moving coherently in the signal direction and speed-accuracy instructions. PD patients used choice probabilities to guide perceptual decisions in both speed and accuracy conditions, and their performance did not significantly differ from that of HC subjects. An additional analysis of the data with the diffusion decision model confirmed this conclusion. These results suggest that the impaired use of priors during visual discrimination observed at more advanced stages of PD is independent of dopaminergic denervation, though additional studies with larger sample sizes are needed to more firmly establish this conclusion.


Introduction
Parkinson's disease (PD) is a neuropsychiatric condition that combines a broad range of motor and non-motor symptoms from the early stages (Weintraub et al., 2022;Weintraub and Burn, 2011;Weintraub and Mamikonyan, 2019).Beyond the prototypical motor triad of PD that includes bradykinesia, resting-tremor and rigidity, PD patients exhibit cognitive alterations leading to reduced quality of life (Williams-Gray et al., 2009;Robbins and Cools, 2014;Carceles-Cordon et al., 2023).These alterations result from a complex interplay between dopaminergic and non-dopaminergic denervation within cortico-basal circuits and dopamine replacement therapy (DRT; Rowe et al., 2008;Vaillancourt et al., 2013;Meder et al., 2019).
Decision-making has been extensively evaluated in PD over the past 20 years, as this cognitive ability is critical for goal-directed behavior, and depends on the integrity of dopaminergic networks (Pessiglione et al., 2006;Frank et al., 2004Frank et al., , 2007;;Friston et al., 2014).The bulk of studies have used value-based decision tasks such as the Iowa gambling task, and have reported significant decision-making deficits in PD compared to healthy controls.These deficits have been attributed to either the disease-related neurodegenerative process or DRT according to the so-called "dopamine overdose hypothesis" (Frank et al., 2004(Frank et al., , 2007;;Evens et al., 2016;Perugini et al., 2018;Poletti et al., 2010;Brand et al., 2004;Xi et al., 2015;Cools et al., 2003;Shiner et al., 2012).This hypothesis predicts altered performance on tasks that engage intact dopamine-mediated neural circuits following DRT (Rowe et al., 2008;Vaillancourt et al., 2013).Beyond value-based decision tasks, recent studies have probed decision mechanisms in PD using perceptual decision tasks (Huang et al., 2015;Perugini et al., 2016;Perugini and Basso, 2018;Herz et al., 2017;Servant et al., 2018;Vilares and Kording, 2017).These tasks allow for a mechanistic investigation of other elementary aspects of the decision process, such as the ability to trade speed for accuracy or use previously learned information (priors) to guide decisions under various conditions of perceptual uncertainty.These elementary aspects have been theorized within the framework of sequential sampling models, according to which the brain forms decisions by repeatedly sampling and accumulating noisy sensory evidence until a threshold quantity has been obtained for a particular choice (Gold and Shadlen, 2007;O'Connell and Kelly, 2021;Ratcliff and Smith, 2004).The starting point of the evidence accumulation process is adjusted as a function of the prior probability of each possible choice (White and Poldrack, 2014;Ratcliff and McKoon, 2008;Mulder et al., 2012;Leite and Ratcliff, 2011), while the threshold controls the trade-off between the speed and accuracy of the decision (a higher threshold producing longer but more accurate decisions; Ratcliff and Smith, 2004;Ratcliff and McKoon, 2008;Ratcliff and Rouder, 1998;Ratcliff et al., 2001).Empirical evidence suggests a role of the basal ganglia system in the regulation of both mechanisms.Specifically, adjustments of the starting point appear to engage cortico-striatal networks (Forstmann et al., 2010;Wang et al., 2018;Ding and Gold, 2012), while threshold adjustments depend on cortico-subthalamic networks (Herz et al., 2016(Herz et al., , 2017(Herz et al., , 2022;;Cavanagh et al., 2011 ).
Separate investigations of speed-accuracy regulation and priorinformed perceptual decisions have been conducted in PD patients.While patients and healthy controls exhibit similar threshold adjustments between speed and accuracy instructions, regardless of DRT status (Off vs. ON; Huang et al., 2015;Herz et al., 2017;Servant et al., 2018), mixed findings have been reported regarding PD patients' ability to combine prior and current perceptual information to inform decisions.One study reported no impairment in a multiple-choice visual estimation task that required estimating the location of a hidden dot on a screen.The uncertainty of both prior and current perceptual information was manipulated.Although DRT had an impact on performance at the beginning of the task, with PD patients OFF medication relying more on priors compared to patients ON medication and healthy control (HC) subjects, this difference vanished with practice (Vilares and Kording, 2017).However, two studies reported a strong impairment in a two-choice visual discrimination task, in which subjects had to report the direction (leftward vs. rightward) of glass pattern stimuli under various conditions of perceptual uncertainty.The prior probability of leftward vs. rightward stimuli was manipulated.In the first study, PD patients under DRT failed to use these priors to guide their decisions, contrary to healthy controls.This deficit was observed regardless of the nature of priors (implicitly vs. explicitly learned) and effector used to respond to the task (manual responses vs. saccadic eye movements).Importantly, the deficit was more pronounced when the direction of the prior corresponded to the side of motor symptoms (Perugini et al., 2016), suggesting a dopaminergic cause.Surprisingly, a subsequent study using the same task showed that PD patients were equally impaired at using priors both OFF and ON DRT, suggesting that the deficit is independent of dopaminergic tone (Perugini and Basso, 2018).However, ON and OFF motor scores in this study were very close, raising concerns about the condition under which PD patients were assessed and the ability to demonstrate either a hypodopaminergic or overdosing effect.Moreover, in both studies, chronic exposure to dopaminergic drugs that lead to synaptic plastic changes (Zhuang et al., 2013) may have contributed to the observed deficit.
The primary goal of the present study was to test whether the impaired use of priors observed during two-choice visual discrimination in PD is caused by a dopaminergic deficit.To this end, we used a sample of clinically established early PD (CEE PD) patients following highly specific criteria recently proposed by the Movement Disorder Society (Berg et al., 2018), and compared behavioral performance to a sample of HC subjects matched in age, sex, and education.CEE PD constitutes a relatively pure model of dopaminergic denervation, as the onset of the prototypical motor triad has been shown to be associated with a 60-80% dopaminergic loss (Brück et al., 2006(Brück et al., , 2009;;Kaasinen and Vahlberg, 2017;López-Aguirre et al., 2023;Braak et al., 2004).Although half of CEE PD patients were on DRT, they were treated for too short of a time (M = 4.2 months, range 1-6 months) using levodopa equivalent daily doses (LEDDs) that were too low (M = 238 mg, range 100-300 mg) for DRT to be optimally effective.Therefore, if the impaired use of priors during visual discrimination in PD evidenced by previous studies is due to a dopaminergic deficit, it should be observed in CEE PD patients as well.Conversely, if the impairment is due to overdosing/chronic effects of DRT, or to disease progression including non-dopaminergic neurotransmission systems (Barone, 2010;Miguelez et al., 2020), this impairment should not be observed in CEE PD.A secondary goal of the study was to provide additional evidence for a normal regulation of decision thresholds in PD.Based on previous findings, we expected speed-accuracy instructions to produce similar modulations of behavioral performance in CEE PD patients and HC.

Participants
Thirteen PD patients and 13 HC subjects matched in age, sex and education participated in the study.These sample sizes should offer sufficient power to detect a potentially large impairment in the use of priors to inform perceptual decisions in PD patients, as the original demonstration of this impairment relied on 12 PD patients and 12 HC (Perugini et al., 2016).All participants had normal or corrected-to-normal vision and were right-handed.Each PD patient had a diagnosis of CEE PD according to highly specific criteria from the Movement Disorder Society (Berg et al., 2018).Although DRT does not represent an exclusion criterion for CEE PD, we conservatively excluded patients with a DRT duration of more than 6 months and/or LEDD levels >300 mg/day to minimize the impact of DRT on our findings.Other criteria for exclusion comprised cognitive impairment (score on the Montreal cognitive assessment MoCA <24) and depressive symptoms (score on the Beck depression inventory BDI >21).The final sample of CEE PD patients comprised 7 patients that were not on DRT and 6 patients that were on DRT.Patients on DRT took their last medication within 2 h of the study.Before the experiment, each CEE PD patient was examined by a neurologist in order to quantify the severity of motor symptoms (Unified Parkinson's disease rating scale UPDRS part III) and clinical disability (Hoen and Yahr scale).Sociodemographic and clinical  1.This study was approved by the ethical committee for research of the university (agreement n • CERUBFC-2021-07-01-020).

Perceptual decision-making task
Participants performed a two-choice random dot motion task (leftward vs. rightward motion discrimination, left/right manual responses) classically used to investigate perceptual decision-making mechanisms (Gold and Shadlen, 2007;O'Connell and Kelly, 2021).The task featured within-subject manipulations of perceptual discriminability, speed-accuracy instructions, and prior probability of each choice.

Apparatus and stimuli
Subjects were individually tested in a quiet room, either at the hospital (CEE PD patients) or at home (HC).Participants were invited to sit on a comfortable chair at a distance of 80 cm from a 34.6 × 19.4 cm laptop computer screen (refresh rate: 60 Hz; resolution: 1920 x 1080).The height of the screen was adjusted so that its center was on the same horizontal plane as the subject's eyes.The experiment was programmed in Python, using functions from the Psychopy toolbox.We used a response pad designed by the Black box toolkit company to achieve high temporal resolution (1000 Hz).The response pad featured large buttons that required a pressure of only 0.54 N (55 g) in order to minimize the impact of motor symptoms during button presses (Fig. 1A).
In each trial, the stimulus consisted of white dots (0.05 • square) presented within a virtual circular aperture centered on a 24.8 • × 13.9 • black field.Dots moved at a speed of 8 • /s, and dot density was 16.7 dots/deg 2 /s.The random dot motion was controlled by a white noise algorithm: from each frame to the next, a percentage p of dots was selected to move the signal direction (leftward vs. rightward), while remaining dots were plotted in random locations (Fig. 1B).Parameter p is classically referred to as motion coherence, and determines the perceptual difficulty of the decision.If motion coherence is low, a small proportion of dots will move in the signal direction, making perceptual decision-making difficult (and conversely).We used four motion coherences that span a large range of perceptual difficulty levels (p = 0, 6, 12, 50%).A 0% coherence condition was chosen to generate maximal perceptual uncertainty and promote the use of priors, in line with previous work (Perugini et al., 2016;Perugini and Basso, 2018).Leftward/rightward stimuli for the 0% condition were pseudorandomly determined, with the constraint of complying with choice probability instructions (see below).

Procedure
The task consisted of 18 blocks of 96 trials, with self-paced breaks between blocks.In each trial, participants were instructed to press the left button with their left index finger if motion direction was leftward, or the right button with their right index finger if motion direction was rightward.Motion coherence levels were randomized within blocks, while speed-accuracy instructions and prior probability of each choice were manipulated blockwise.Specifically, each block was defined by a factorial combination of speed-accuracy instructions (emphasis on response speed vs. response accuracy) and choice probability (75% leftward vs. 25% rightward 75L:25R; 25L:75R; 50L:50R), resulting in six different block types (Fig. 1C).Each block type was presented three times.The order of blocks was pseudorandomized, with the constraint of presenting the six block types in succession.This pseudorandomization scheme was chosen to minimize learning, fatigue, and sequence effects across subjects.
Within each block, trials were defined by a factorial combination of motion direction (leftward vs. rightward) and motion coherence (0, 6, 12, 50%), and were presented in a random order.The relative proportion of leftward vs. rightward motion stimuli within each coherence level was determined by the choice probability instructions.These probabilities were communicated to participants at the beginning of each block, along with speed-accuracy instructions.In speed blocks, participants were instructed to respond as quickly as possible.In accuracy blocks, participants were instructed to respond as accurately as possible.Each trial started with the presentation of the random dot motion stimulus, which remained on screen until the response.Participants had unlimited time to respond.However, in speed blocks, an implicit response time (RT) deadline was set to 1500 ms.If the RT was superior to this implicit deadline, the feedback "Please respond faster" was displayed in red for 1500 ms after the response (Fig. 1D).No feedback on accuracy was provided.In accuracy blocks, feedback on accuracy was provided for 1000 ms after each response ("Correct response" displayed in green vs. "Incorrect response" displayed in red; Fig. 1E).No feedback on RT was provided.Feedback was introduced to enforce speed-accuracy instructions.The interval between trials was 600 ms.
Participants had to answer two questions at the end of each block to probe their memory for task instructions.The first question concerned speed-accuracy instructions.Participants had to press the 'A' button on the response pad if the block was an accuracy block, or the 'B' button if the block as a speed block.The second question concerned the probability of each choice.Participants had to press the 'A', 'B' or 'C' button if these probabilities were 50L:50R, 75L:25R, or 25L:75R respectively.No feedback was provided for either question.The 'A', 'B', and 'C' buttons were different from the left/right responses used for the random dot motion task (Fig. 1A).Participants were informed at the beginning of the experiment that these questions would be asked at the end of each block.The questions had two main purposes: (i) encourage participants to pay attention to task instructions and maintain them in working memory to guide performance; (ii) assess whether CEE PD patients correctly memorized prior probabilities, which constitutes a prerequisite for using them to guide perceptual decision-making.
The 18 experimental blocks were preceded by two practice blocks of 48 trials each.Each practice block comprised an equal proportion of leftward and rightward motion stimuli in each of the four motion coherence levels, presented in a random order (choice probabilities were introduced at the beginning of the experiment).The first practice block was an accuracy block, while the second practice block was a speed block.Participants were told that the discrimination of motion direction would be very hard on some trials, but that they had to give a response on each trial, even though they sometimes had the impression that this response was random.All instructions were presented by the computer program to minimize variability in explanation.On average, the duration of the task (including instruction time and practice trials) was 65 min for HC and 75 min for CEE PD patients.

Data analyses
Permutation tests (100,000 permutations, two-sided) were used to compare group means (CEE PD vs. HC) for sociodemographic (age, years of education) and clinical (MoCA, BDI) variables.We used permutation tests instead of conventional t-tests because the data frequently showed violations of the normality assumption (assessed by Mauchly's test).We used permutation tests instead of Mann-Whitney U tests because our interest was in the difference in means. 1 Behavioral performance was analyzed using mixed-design analyses of variance (ANOVAs).The sphericity assumption for within-subject factors comprising at least three levels was evaluated using Mauchly's test.When sphericity was violated, the degrees of freedom were adjusted using the Greenhouse-Geisser procedure.All tests used a threshold of α = 0.05 for statistical significance.
1 When the data are normally distributed, p-values from permutation tests are equivalent to p-values from t-tests.For simplicity, we thus used a permutation test for each pairwise comparison.

Computational modeling
We decomposed and quantified each hypothetical information processing component presumably engaged in the random dot motion task using a parsimonious and powerful sequential sampling model of twochoice perceptual decisions, the diffusion decision model (Ratcliff and McKoon, 2008;Ratcliff et al., 2016).The model has received considerable support from both successful fits to behavioral data and neurophysiological investigations of the decision-making process (Gold and Shadlen, 2007;O'Connell and Kelly, 2021;Ratcliff et al., 2016).Its architecture is illustrated in Fig. 3A.In each trial, samples of perceptual evidence accumulate from a starting point (parameter z) to one of two decision thresholds (one for each choice; the separation between thresholds is represented by parameter a).If there is no bias for a particular choice, the starting point z is located at 0, halfway between the two thresholds.The predicted decision time is determined by the latency between accumulation onset and the first hit of a threshold, and the predicted choice is determined by which of the two thresholds was hit.Samples of perceptual evidence are assumed to be corrupted by random biological noise, making the accumulation process stochastic.This phenomenon primarily explains the trial-to-trial variability of decision times and choices.The rate of evidence accumulation, or drift rate (parameter v), depends on the quality of the perceptual evidence extracted from the stimulus.A lower stimulus discriminability is associated with a smaller drift rate, increasing decision times and the probability of errors.The mean latency of processes upstream and downstream of decision-making, such as stimulus encoding and motor output processes, is represented by parameter Ter.In each trial, the predicted RT corresponds to the sum of the predicted decision time plus

Ter.
Given a set of values assigned to parameters (z, a, v, Ter) for a specific experimental condition, the diffusion decision model predicts the full behavioral performance measures: a RT distribution and a proportion for each response.Each parameter has a specific influence on these predictions.By fitting the model to data (i.e., tuning parameter values to minimize a loss function quantifying the discrepancy between data and model predictions), it is thus possible to quantify each processing component.This model-based analysis constitutes a major advantage compared to traditional analyses of behavioral performance, because it allows for a separation and quantification of processing components within the decision process, as well as a separation and quantification of decision and nondecision latencies.We applied this method to data from both CEE PD patients and HC in order to identify potential processing differences between the two groups at a fine-grained level.
We modeled left and right responses, corresponding to the upper and lower thresholds respectively, and fixed the diffusion coefficient (amplitude of within-trial noise) to 0.1 to satisfy a scaling property of the model (one parameter must be fixed in order to identify the other parameters).We assumed that motion coherence, priors, and speedaccuracy instructions modulate the drift rate v, threshold separation a, and starting point z respectively (Ratcliff et al., 2016).Note that parameter z was only free to vary in 75L:25R and 25L:75R conditions (it was constrained to be halfway between the two thresholds in the 50L:50R condition).We further let parameter z free to vary between speed and accuracy instructions, in order to capture a potential strategic difference in the use of priors between instructions.Our model thus comprised 11 free parameters: threshold separation for the speed and the accuracy condition, drift rate for each motion coherence level, From each frame to the next, a percentage p of dots was selected to move in the signal direction (leftward vs. rightward, red arrows), while the remaining dots were plotted in random locations (blue arrows).Subjects had to discriminate the signal direction and press the bottom left vs. bottom right button on the response pad (using their left or right index finger) to communicate their choice.Parameter p, referred to as motion coherence, was manipulated across four levels (0, 6, 12, 50%) to provide different conditions of perceptual uncertainty.These conditions were randomized within each block of 96 trials.C) Illustration of the six block types defined by a factorial combination of speed-accuracy instructions (emphasis on response speed vs. response accuracy) and choice probability (75% leftward vs. 25% rightward 75L:25R; 25L:75R; 50L:50R).Each block type was presented three times in a pseudorandom order.D) Structure of a trial in speed blocks.When the response time (RT) to the stimulus was > 1,500 ms, the feedback "Please respond faster" was displayed for 1,500 ms.No feedback on accuracy was provided.E) Structure of a trial in accuracy blocks.Feedback on accuracy was provided after each response for 1,000 ms.No feedback on RT was provided.
starting point for the 75L:25R in the speed and the accuracy condition, starting point for the 25L:75R in the speed and the accuracy condition, and mean nondecision time.Although between-trial variability in drift rate, starting point, and nondecision time are commonly incorporated in fits of the model to data, recovery issues and tradeoffs with the main model parameters have been highlighted (Boehm et al., 2018;Ratcliff and Tuerlinckx, 2002), so we did not incorporate them in our modeling to maximize the probability of detecting processing differences between CEE PD patients and HC.
We used the rtdists package for R to get analytical solutions from the diffusion decision model, and fit the model to the behavioral data from each subject using a standard quantile-based method (Servant and Evans, 2020;Smith and Ratcliff, 2009).Specifically, we minimized a likelihood ratio chi-square statistic (quantifying the discrepancy between observed and predicted RT distributions and proportions for each response across conditions) by first running a differential evolution global optimization algorithm (Storn and Price, 1997).The population initialization used Sobol sequences in order to provide a uniform coverage of the parameter space.We then polished the results by means of a limited memory BFGS-B local optimization algorithm (Byrd et al., 1996).Upper and lower boundaries for each parameter were based on a survey of parameter values estimated in empirical studies (Matzke and Wagenmakers, 2009).

Memory of task instructions
Percentages of correct responses to questions relative to speedaccuracy instructions and choice probabilities are shown in Fig. 2A and B respectively.A mixed-design ANOVA conducted on the percentage of correct responses to questions relative to speed-accuracy instructions with speed-accuracy instructions as a within-subject factor and group as a between-subject factor revealed a lower memory performance in the speed (M = 85.04%) compared to the accuracy condition (M = 94.44%;F 1,24 = 6.14, p = 0.021).Studies on the effects of stress on memory may help explain this finding.Speed pressure may have acted as a stressor, and the associated increase in glucocorticoid levels may have impaired memory processes (Klier and Buratto, 2020;de Quervain et al., 2009;Shields et al., 2017).The main effect of group and the interaction between the two factors were not significant (p = 0.69 and p = 0.66 respectively).
A mixed-design ANOVA conducted on the percentage of correct responses to questions relative to choice probability instructions with choice probability instructions as a within-subject factor and group as a between-subject factor suggested a possible main effect of group, reflecting a lower memory performance for CEE PD patients (M = 85.47%) compared to HC (M = 94.02%;F 1,24 = 3.85, p = 0.062).There was also a main effect of choice probability instructions (F 2,48 = 4.72, p = 0.014).Post-hoc pairwise comparisons corrected with Holm's procedure showed a lower memory performance for 25L:75R (M = 84.62%)compared to 50L:50R (M = 94.23%;p = 0.025), but no difference between 75L:25R (M = 90.38%)and 50L:50R (p = 0.22).Although the retrieval of unequal probabilities and their association with leftward/ rightward motion directions is arguably more challenging compared to the 50L:50R condition, the memory advantage for the 75L:25R over the 25L:75R condition is consistent with empirical studies showing leftward biases in the perceptual and attentional domains in healthy individuals (Marzoli et al., 2014).The interaction between choice probability instructions and group was not significant (p = 0.51).To allow for a direct comparison of experimental findings with previous work (that did not probe explicit memory for task instructions; Perugini et al., 2016;Servant et al., 2018), we did not discard blocks of trials for which participants gave at least one incorrect response to the two corresponding memory questions in the following behavioral and computational analyses.

Behavioral performance in the random dot motion task
To facilitate understanding of the effect of priors on decision-making performance and simplify statistical analyses, the behavioral data from each choice probability condition were reorganized to consider the three following conditions: "prior -" (e.g., there are 75% of leftward stimuli in the block, and a rightward stimulus is presented), "prior +" (e.g., there are 75% of leftward stimuli in the block, and a leftward stimulus is presented) and equal priors.Anticipations (RTs <150 ms; M CEE PD = 0.25%; M HC = 0.03%; permutation test: p = 0.64) and abnormally slow responses (RTs >5000 ms; M CEE PD = 0.53%; M HC = 0.17%; permutation test: p = 0.11) were discarded from analyses.Percentages of trials in which participants responded after the implicit 1500 ms deadline in speed blocks were small and not significantly different between CEE PD patients (M = 3.29%) and HC (M = 1.51%; permutation test: p = 0.13).We did not remove these trials from analyses, because they are part of subjects' normal performance.

Percentage of correct responses
The percentage of correct responses for each condition is displayed in Fig. 2C.The results of a mixed-design ANOVA computed on these percentages with motion coherence, priors, and speed-accuracyinstructions as within-subject factors and group as a between-subject factor are shown in Table 1.The main effect of each within-subject factor was significant: the proportion of correct responses (i) decreased as motion coherence decreased, (ii) was lower under speed than accuracy instructions, and (iii) was lower for the prior -compared to the prior + condition (with the equal prior condition being approximately halfway).Priors significantly interacted with motion coherence, reflecting a larger effect of priors for lower coherence levels, consistent with previous work (Perugini et al., 2016;White and Poldrack, 2014;Ratcliff and McKoon, 2008;Mulder et al., 2012).The amplitude of this effect was slightly reduced in the speed compared to the accuracy condition, as suggested by a possible three-way interaction between within-subject factors.The main effect of group was not significant, and the group factor did not significantly interact with any of the within-subject factors.This important finding suggests that CEE PD patients and HC subjects are equally sensitive to experimental factors.In particular, CEE PD patients appear to use priors to guide perceptual decisions in a way similar to HC.We further checked this finding by computing a second mixed-design ANOVA, focusing on the lowest motion coherence level (0%, for which the effects of priors are largest) and averaging across speed-accuracy instructions.This analysis revealed a main effect of priors (F 1.38,33.11= 107.53,p < 0.001).The main effect of group and the interaction between group and priors were not significant (p = 0.72 and p = 0.23 respectively), consistent with findings from the previous ANOVA.

Mean RT data
Mean RT data for correct trials are shown in Fig. 2D.The results of a mixed-design ANOVA computed on these data with motion coherence, priors, and speed-accuracy instructions as within-subject factors and group as a between-subject factor are shown in Table 2.The main effect of each within-subject factor was significant: mean RT (i) increased as motion coherence decreased, (ii) was faster under speed than accuracy instructions, and (iii) was faster for the prior + compared to the priorcondition.There was also a significant three-way interaction between these factors: the effect of priors was larger for lower motion coherence levels, particularly under accuracy instructions.The group factor significantly interacted with motion coherence and speed-accuracy instructions.As can be seen in Fig. 2D, the amplitude of the motion coherence effect on mean RT was larger under accuracy compared to speed instructions, but this effect was more pronounced for CEE PD patients.
Overall, these analyses suggest that CEE PD patients are not impaired in using priors to guide perceptual decisions, regardless of speed-accuracy instructions.In the next section, we use formal modeling to provide a mechanistic processing account of the data and further evaluate this hypothesis.

Analyses of best-fitting parameters from the diffusion decision model
The diffusion decision model provided a reasonable fit to data and captured its main trends (Supplementary Fig. 1), though the model overestimated the proportion of correct responses in the prior -condition for low motion coherence levels.Best-fitting values of the starting point z of the evidence accumulation decision process averaged across subjects are shown in Fig. 3B.This parameter was constrained to be 0 (halfway between the two decision thresholds) in the 50L:50R condition, and was free to vary in the 75L:25R and 25L:75R conditions.Since the upper and lower thresholds represent left and right responses respectively, a bias toward the left response should translate into a positive z, while a bias toward the right response should translate into a negative z. Fig. 3B shows that best-fitting z values are positive in the 75L:25R condition and negative in the 25L:75R condition for both CEE PD patients and HC, suggesting that participants from each group strategically adjusted the starting of evidence accumulation according to choice probabilities.These observations were confirmed by significant one-sample permutation tests against 0 for both CEE PD patients (all ps < 0.05) and HC (all ps < 0.05).We further ran a mixed-design ANOVA on parameter z using choice probabilities (75L:25R, 25L:75R) and speed-accuracy instructions as within-subject factors and group as a between-subject factor.This analysis revealed a significant main effect of choice probabilities (F 1,24 = 120.24,p < 0.001), reflecting the sign difference of parameter z between 75L:25R (positive) and 25L:75R (negative).There was also a significant interaction between choice probabilities and speed-accuracy instructions (F 1,24 = 26.04,p < 0.001), reflecting a smaller effect of choice probabilities in the speed compared to the accuracy instruction.This phenomenon explains the reduction of the effect of priors on the behavioral performance measures under speed pressure.Importantly, the main effect of group was not significant (p = 0.48), and did not significantly interact with the other factors (group × choice probabilities: p = 0.70; group × speed-accuracy instructions: p = 0.61; group × choice probabilities × speed-accuracy instructions: p = 0.10).Thus, CEE PD patients and HC appear to adjust the starting point  The plot represents evidence accumulation trajectories for 100 simulated trials and a highly discriminable leftward motion stimulus.Evidence accumulation stops when the process reaches one of two thresholds, marking the end of the decision process.The upper threshold is associated with the left response, and the lower threshold is associated with the right response.In each trial, the decision time corresponds to the latency between the onset of evidence accumulation and the first hit of a threshold.Note that evidence accumulation is a noisy process.Noise produces variability in decision times and choices.Parameter z indexes the starting point of evidence accumulation, and represents the meeting point between memory and decision systems: it is strategically adjusted as a function of the prior probability of occurrence of each choice.If the two choices are a priori equiprobable, the starting point is located at 0, halfway between the two thresholds (as illustrated).Parameter a quantifies the separation between the two thresholds, and regulates the speed-accuracy tradeoff of the decision.The rate of evidence accumulation, or drift rate, is quantified by parameter v, and depends on the quality of the perceptual evidence extracted from the stimulus.In each trial, the predicted response time corresponds to the sum of the decision time and the nondecision time.The mean nondecision time is represented by parameter Ter.B) Best-fitting parameter z averaged across subjects as a function of choice probabilities, speed-accuracy instructions, and group (CEE PD vs. HC).C) Best-fitting parameter a averaged across subjects as a function of speed-accuracy instructions and group.D) Bestfitting parameter v averaged across subjects for each motion coherence level and group.E) Best-fitting parameter Ter averaged across subjects for each group.Parameter units are arbitrary (a.u.: arbitrary units), except for Ter which is expressed is seconds (s).Error bars in plots B-E represent ±1 between-subject standard error of the mean.Notes.Each mixed-design ANOVA incorporates motion coherence, priors, and speed-accuracy instructions as within-subject factors and group as a between-subject factor.Degrees of freedom with decimals were corrected using the Greenhouse-Geisser procedure, following violation of the sphericity assumption.sai: speedaccuracy instructions; df: degrees of freedom; * indicates statistical significance at the α = 0.05 threshold.
of the evidence accumulation decision process across choice probabilities and speed-accuracy instructions in a similar way.Fig. 3C shows parameter a (separation between decision thresholds) as a function of speed-accuracy instructions and group.A mixed-design ANOVA on this parameter with speed-accuracy instructions as a withinsubject factor and group as a between-subject factor revealed a main effect of speed-accuracy instructions (F 1,24 = 33.24,p < 0.001), reflecting a smaller threshold separation under time pressure.The analysis also suggested a possible interaction between the two factors (F 1,24 = 3.85, p = 0.062).Analyses of simple effects showed a trend for a larger threshold separation for CEE PD patients than HC in the accuracy condition (p = 0.053), but no difference in the speed condition (p = 0.333).This modulation explains the three-way interaction between group, coherence, and speed-accuracy instructions observed on mean RT.
To test for additional differences in information processing between CEE PD patients and HC, we computed a mixed-design ANOVA on the drift rate parameter v, with motion coherence as a within-subject factor and group as a between-subject factor.This analysis revealed a main effect of motion coherence, reflecting a decrease of parameter v as motion coherence decreases (F 1.49,35.75= 253.44,p < 0.001; Fig. 3D).The other effects were not significant (group: p = 0.76; group × coherence: p = 0.91).Finally, we computed a permutation test on parameter Ter to determine whether the mean latency of processing components in upstream and downstream decision-making differ between CEE PD patients and HC.This test was not significant (p = 0.14; Fig. 3E).These additional analyses suggest that the quality of the evidence entering the decision process and the mean latency of nondecision processes do not differ between CEE PD patients and HC.

Impact of DRT on starting point adjustments
As detailed in Section 2.1 and Supplementary Table 1, our sample of CEE PD comprised 7 patients that were not on DRT and 6 patients that were on DRT.UPDRS-III scores did not significantly differ between the two subgroups (DRT: M = 25.33;no DRT: M = 19.57;permutation test: p = 0.43),2 consistent with our hypothesis that patients on DRT were treated for too short of a time using LEDDs that were too low for DRT to be optimally effective.Nevertheless, DRT might still have had an impact on starting point adjustments.To evaluate this hypothesis, we computed a mixed-design ANOVA on the starting point of evidence accumulation (parameter z) using choice probabilities (75L:25R, 25L:75R) and speedaccuracy instructions as within-subject factors and CEE PD subgroup (DRT vs. no DRT) as a between-subject factor.Although this analysis must be taken with caution due to the small number of subjects in each subgroup, it gave a pattern of statistical results that are clear and similar to those previously observed.Specifically, the analysis showed a main effect of choice probabilities (F 1,11 = 92.91,p < 0.001), reflecting the sign difference of parameter z between 75L:25R (positive) and 25L:75R (negative).There was also a significant interaction between choice probabilities and speed-accuracy instructions (F 1,11 = 30.76,p < 0.001), reflecting a smaller effect of choice probabilities in the speed compared to the accuracy instruction (Fig. 4A).Importantly, the main effect of the CEE PD subgroup was not significant (p = 0.53), and did not significantly interact with the other factors (subgroup × choice probabilities: p = 0.80; subgroup × speed-accuracy instructions: p = 0.44; subgroup × choice probabilities × speed-accuracy instructions: p = 0.41).Thus, DRT does not appear to modulate starting point adjustments.

Impact of the side of motor symptoms on starting point adjustments
Previous analyses suggest that CEE PD patients and HC use priors to guide perceptual decisions in a similar way.However, it remains possible that CEE PD patients show an impairment only when the direction of priors corresponds to the side of motor symptoms (Perugini et al., 2016).To evaluate this hypothesis, we first determined the side of motor symptoms for each patient by computing the difference between lateralized motor subscores from the UPDRS III (right-left; a positive difference indicates a predominance of motor symptoms on the right side and conversely).This classification procedure resulted in six CEE PD patients with a predominance of motor symptoms on the right side and six CEE PD patients with a predominance of motor symptoms on the left side.One patient did not show any lateralization of motor symptoms, as indicated by a null difference between right and left UPDRS III subscores.
Fig. 4B displays the starting point of evidence accumulation (parameter z) for each choice probability and speed-accuracy condition as a function of the side of motor symptoms.We computed a mixeddesign ANOVA on parameter z using choice probabilities (75L:25R, 25L:75R) and speed-accuracy instructions as within-subject factors and lateralization of motor symptoms (left vs. right) as a between-subject factor.This analysis showed a main effect of choice probabilities (F 1,10 = 85.99, p < 0.001) and a significant interaction between choice probabilities and speed-accuracy instructions (F 1,10 = 25.48,p < 0.001), similar to previous findings.The effect of motor symptoms lateralization was not significant (p = 0.66), and did not significantly interact with the other factors (group × choice probabilities: p = 0.36; group × speedaccuracy instructions: p = 0.52; group × choice probabilities × speedaccuracy instructions: p = 0.48).Thus, the side of motor symptoms does not appear to modulate starting point adjustments.

Discussion
The primary aim of the present work was to further investigate the origin of the impaired use of priors during visual discrimination in PD evidenced by previous studies (Perugini et al., 2016;Perugini and Basso, 2018).The novelty of our work lies within the use of CEE PD patients, which allowed us to test whether the impairment is caused by a dopaminergic deficit.Specifically, we compared the behavioral performance of CEE PD patients and HC (matched in age, sex, and education) in a random dot motion task featuring manipulations of perceptual difficulty and prior probability of leftward vs. rightward motion stimuli.We further incorporated a manipulation of speed-accuracy instructions to examine whether findings are robust across different speed-accuracy regimes, as both prior-informed decision-making and speed-accuracy regulation have been shown to engage the basal ganglia system (Herz et al., 2016(Herz et al., , 2017(Herz et al., , 2022;;Forstmann et al., 2010;Wang et al., 2018;Ding and Gold, 2012;Cavanagh et al., 2011).
The behavioral data showed that CEE PD patients used choice probabilities provided at the beginning of each block to guide perceptual decisions.Compared to the equiprobable choice condition, the proportion of correct responses was higher and mean RT was faster when motion direction corresponded to the more likely choice (and vice versa), especially when stimulus discriminability was low.These modulations were observed for both speed-accuracy regimes, though to a lesser extent when subjects were under time pressure.Importantly, the effects of choice probabilities on behavioral performance were not significantly different between CEE PD patients and HC, suggesting that both groups used priors to inform perceptual decisions in a similar way.To further test this hypothesis, we directly compared processing mechanisms underlying prior-informed decision-making using computational modeling of the data with the diffusion decision model (Ratcliff and McKoon, 2008;Ratcliff et al., 2016).Consistent with previous work (White and Poldrack, 2014;Ratcliff and McKoon, 2008;Mulder et al., 2012;Leite and Ratcliff, 2011), the model explained the effects of the choice probability manipulation through a modulation of the starting point of the evidence accumulation decision process.Specifically, the starting point was adjusted closer to the threshold associated with the most likely response, producing faster decision times and higher choice probabilities for that response.Although these adjustments were generally smaller in the speed compared to the accuracy condition, possibly due to reduced cognitive control resources in the former (van Wouwe et al., 2014;Gehring et al., 1993), they did not differ between CEE PD patients and HC.Additional analyses showed that starting point adjustments were not modulated by the predominant side of motor symptoms or DRT medication.
Altogether, our behavioral and model-based analyses suggest that CEE PD patients were able to acquire and use prior knowledge to guide decisions in a two-choice visual discrimination task.Given that this population constitutes a relatively pure model of dopaminergic depletion, we conclude that the impairment evidenced by previous studies (Perugini et al., 2016;Perugini and Basso, 2018) is unlikely to be caused by a dopaminergic deficit, though additional studies with larger sample sizes are needed to more firmly establish this conclusion.This leaves three possible causes 3 : an overdosing effect of DRT (Rowe et al., 2008;Vaillancourt et al., 2013), synaptic plastic changes following chronic exposure to DRT (Zhuang et al., 2013), or alterations in non-dopaminergic neurotransmission systems that predominate at more advanced stages of the disease (Barone, 2010;Miguelez et al., 2020).The recent observation of a similar impairment in a sample of patients with dopa-unresponsive focal dystonia tends to favor the latter hypothesis (Perugini and Basso, 2018), though caution is required given the relatively poor understanding of dystonia pathophysiology (Ribot et al., 2019;Jinnah et al., 2013).More generally, our results suggest that disease duration is an important variable to consider when studying prior-informed decision-making in PD.This variable may contribute to explaining previous discrepant findings.As mentioned in the Introduction, one study has shown preserved prior-informed perceptual decision-making abilities in PD in a visual estimation task (Vilares and Kording, 2017).Interestingly, disease duration was highly heterogeneous in this study, ranging from 1.5 months to 18 years.The incorporation of CEE PD patients may have blurred potential impairments in prior-informed decision-making at more advanced stages of the disease.Differences in task (multiple-choice visual estimation vs. two-choice visual discrimination) and experimental manipulations (mean vs. variance of prior information) may have also contributed.This point should be given attention in the future.
Besides the implementational level of information processing, the precise origin of the impaired use of priors in PD at the algorithmic level has yet to be elucidated.Prior-informed decision-making requires intact learning, memory, and decision-making.Our sample of CEE PD patients showed intact evidence accumulation decision-making abilities, as revealed by our diffusion model analysis.The rate of evidence accumulation, in particular, did not differ between patients and HC, suggesting that the quality of evidence entering the decision process was similar between the two groups.In addition, nondecision latencies were not significantly different between patients and HC.Although this result may seem surprising in light of the mild motor symptoms exhibited by patients (and evidenced by the UPDRS III scale, see Table 1), we minimized their impact on performance as much as possible by using large response buttons that required a very small amount of force.Therefore, nondecision latencies essentially reflect processes upstream of decisionmaking such as sensory encoding, and these processes do not seem to be altered in CEE PD patients.However, patients did show a trend for a learning/memory impairment of choice probabilities, as suggested by a larger proportion of incorrect responses to questions probing memory for choice probability instructions at the end of each block.A mild attention/executive deficit might be responsible for this finding, as MoCA scores were significantly lower for CEE PD patients compared to HC (though above the standard cut-off for cognitive impairment).The similar memory performance between groups for questions relating to speed-accuracy instructions also favors this hypothesis, as subjects might have used trial-by-trial feedback on speed-accuracy performance as cues to retrieve instructions.Could a more pronounced learning deficit be responsible for impaired prior-informed decision-making observed at later stages of the disease?This hypothesis has previously been rejected based on two pieces of evidence.First, it has been argued that learning is not necessary when choice probabilities are communicated explicitly, and PD patients still showed impaired use of priors in this condition (Perugini et al., 2016).However, choice probabilities -even when communicated explicitly-must be encoded and stored in memory to bias perceptual decisions.Second, when choice probabilities were implicitly learned, PD and HC showed similar win-stay and loose-shift strategies in the condition of maximal perceptual uncertainty (0% coherence) following feedback on response accuracy.In particular, subjects used significantly more win-stay strategies when choice probabilities were unequal compared to when they were equal (and vice 3 One reviewer suggested that the impairment evidenced by previous studies in later stages of PD might be caused by stronger dopaminergic denervation.Although we cannot reject this hypothesis, we believe that it is unlikely, as the onset of the prototypical motor triad in PD has been shown to be associated with a 60-80% dopaminergic loss (see Introduction).If prior-informed decision-making was mediated by dopaminergic neural circuits, an alteration of performance would have been observed in the present study.versa), suggesting some awareness of the statistical structure of the task.This analysis, however, does not specifically assess learning of probabilities for each choice alternative, and is inherently based on a small number of trials.Therefore, future work may benefit from a more direct and specific learning evaluation of these probabilities, such as the one conducted in the present work.
A secondary aim of the present study was to assess the strategic regulation of decision thresholds as a function of speed-accuracy instructions in CEE PD patients, as this ability -which is critical for adaptive decision-making-has been shown to engage corticosubthalamic circuits (Herz et al., 2016(Herz et al., , 2017(Herz et al., , 2022;;Cavanagh et al., 2011).Similar to HC, CEE PD patients complied with instructions, as revealed by faster mean RTs and higher error rates in the speed compared to the accuracy condition.Diffusion model fits further revealed that decision thresholds were significantly lower under time pressure.Thus, CEE PD patients were able to strategically adjust decision thresholds as a function of speed-accuracy instructions.There was, however, a significant three-way interaction between group, motion coherence, and speed-accuracy instructions on mean RT data.Specifically, the amplitude of the motion coherence effect on mean RT was larger in the accuracy compared to the speed condition, and this effect was more pronounced for CEE PD patients.The diffusion model explained this pattern by a larger threshold separation for CEE PD patients compared to HC in the accuracy condition only.This result may seem surprising for two reasons.First, theoretical work using biophysically-based spiking network models predicted a reduced range of threshold variations in PD (OFF medication) compared to HC (Wei et al., 2015), a pattern opposite to the present findings.Second, previous studies found similar adjustments of threshold separation between speed-accuracy conditions (Huang et al., 2015;Herz et al., 2017;Servant et al., 2018).We note, however, that one of these studies reported a main effect of group on threshold separation, with a larger separation for PD patients compared to HC (Huang et al., 2015).A larger threshold separation within the diffusion model produces slower but more accurate decisions (ceteris paribus).As such, this phenomenon might be related to findings from personality studies describing PD patients as cautious, conservative and introverted, with low novelty seeking and high harm avoidance scores (Kaasinen et al., 2001;Meira et al., 2022;Poletti and Bonuccelli, 2012;Santangelo et al., 2018).Follow-up studies are needed to investigate this potential relationship.

Fig. 1 .
Fig. 1.Random dot motion task.A) Respond pad used to record responses.The bottom left and bottom right buttons were used as response buttons in the random dot motion task.Buttons 'A', 'B', and 'C' were used to probe memory for task instructions at the end of each block.B) Stimuli consisted of a field of moving dots.From each frame to the next, a percentage p of dots was selected to move in the signal direction (leftward vs. rightward, red arrows), while the remaining dots were plotted in random locations (blue arrows).Subjects had to discriminate the signal direction and press the bottom left vs. bottom right button on the response pad (using their left or right index finger) to communicate their choice.Parameter p, referred to as motion coherence, was manipulated across four levels (0, 6, 12, 50%) to provide different conditions of perceptual uncertainty.These conditions were randomized within each block of 96 trials.C) Illustration of the six block types defined by a factorial combination of speed-accuracy instructions (emphasis on response speed vs. response accuracy) and choice probability (75% leftward vs. 25% rightward 75L:25R; 25L:75R; 50L:50R).Each block type was presented three times in a pseudorandom order.D) Structure of a trial in speed blocks.When the response time (RT) to the stimulus was > 1,500 ms, the feedback "Please respond faster" was displayed for 1,500 ms.No feedback on accuracy was provided.E) Structure of a trial in accuracy blocks.Feedback on accuracy was provided after each response for 1,000 ms.No feedback on RT was provided.

Fig. 2 .
Fig. 2. Behavioral performance of clinically established early Parkinson's disease patients (CEE PD) and healthy controls (HC).A) Percentage of correct responses to the first question at the end of each block probing memory for task instructions.Error bars represent ± 1 between-subject standard error of the mean.B) Percentage of correct responses to the second question probing memory for choice probabilities.Error bars represent ± 1 between-subject standard error of the mean.C)Percentage of correct responses in the random dot motion task as a function of coherence, choice probabilities, speed-accuracy instructions, and group.For clarity, the data from each choice probability condition were reorganized to consider the three following conditions: prior against the stimulus direction (prior -), prior in favor of the stimulus direction (prior +) and equal priors.Shaded areas represent ± 1 within-subject standard error of the mean.D) Mean response time (RT) in correct trials as a function of coherence, priors, speed-accuracy instructions, and group.Shaded areas represent ± 1 within-subject standard error of the mean.

Fig. 3 .
Fig. 3. Analysis of the behavioral data from clinically established early Parkinson's disease patients (CEE PD) and healthy controls (HC) with the diffusion decision model.A) Architecture of the diffusion decision model for two-choice perceptual decisions.The plot represents evidence accumulation trajectories for 100 simulated trials and a highly discriminable leftward motion stimulus.Evidence accumulation stops when the process reaches one of two thresholds, marking the end of the decision process.The upper threshold is associated with the left response, and the lower threshold is associated with the right response.In each trial, the decision time corresponds to the latency between the onset of evidence accumulation and the first hit of a threshold.Note that evidence accumulation is a noisy process.Noise produces variability in decision times and choices.Parameter z indexes the starting point of evidence accumulation, and represents the meeting point between memory and decision systems: it is strategically adjusted as a function of the prior probability of occurrence of each choice.If the two choices are a priori equiprobable, the starting point is located at 0, halfway between the two thresholds (as illustrated).Parameter a quantifies the separation between the two thresholds, and regulates the speed-accuracy tradeoff of the decision.The rate of evidence accumulation, or drift rate, is quantified by parameter v, and depends on the quality of the perceptual evidence extracted from the stimulus.In each trial, the predicted response time corresponds to the sum of the decision time and the nondecision time.The mean nondecision time is represented by parameter Ter.B) Best-fitting parameter z averaged across subjects as a function of choice probabilities, speed-accuracy instructions, and group (CEE PD vs. HC).C) Best-fitting parameter a averaged across subjects as a function of speed-accuracy instructions and group.D) Bestfitting parameter v averaged across subjects for each motion coherence level and group.E) Best-fitting parameter Ter averaged across subjects for each group.Parameter units are arbitrary (a.u.: arbitrary units), except for Ter which is expressed is seconds (s).Error bars in plots B-E represent ±1 between-subject standard error of the mean.

Fig. 4 .
Fig. 4. Effect of dopamine replacement therapy (DRT) and motor symptom lateralization on starting point adjustments.Best-fitting parameter z (starting point of evidence accumulation) averaged across clinically established early Parkinson's disease (CEE PD) patients as a function of choice probabilities, speed-accuracy instructions, and medication (DRT vs. no DRT).B) Best-fitting parameter z averaged across CEE PD patients as a function of choice probabilities, speed-accuracy instructions, and lateralization of motor symptoms (left side vs. right side).Parameter units are arbitrary (a.u.: arbitrary units).Error bars represent ±1 betweensubject standard error of the mean.

Table 1
Sociodemographic and clinical characteristics of clinically established early Parkinson's disease patients (CEE PD) and healthy control (HC) subjects.

Table 2
Results from mixed-design ANOVAs computed on the proportion of correct responses and mean RT data in the random dot motion task.