Sequence learning in Parkinson's disease: Focusing on action dynamics and the role of dopaminergic medication

Highlights

• We studied the action dynamics of sequence learning in Parkinson's disease.

• The effects of dopaminergic medication on mouse movements were considered.

• Medication enhanced sequence learning as reflected in initiation time.

• Conversely, it impaired sequence learning as reflected in movement accuracy.

• These findings are discussed in light of the dopamine overdose hypothesis.

Abstract

Previous studies on movement sequence learning in Parkinson's disease (PD) have produced mixed results. A possible explanation for the inconsistent findings is that some studies have taken dopaminergic medication into account while others have not. Additionally, in previous studies the response modalities did not allow for an investigation of the action dynamics of sequential movements as they unfold over time. In the current study we investigated sequence learning in PD by specifically considering the role of medication status in a sequence learning task where mouse movements were performed. The focus on mouse movements allowed us to examine the action dynamics of sequential movement in terms of initiation time, movement time, movement accuracy, and velocity. PD patients performed the sequence learning task once on their regular medication, and once after overnight withdrawal from their medication. Results showed that sequence learning as reflected in initiation times was impaired when PD patients performed the task ON medication compared to OFF medication. In contrast, sequence learning as reflected in the accuracy of movement trajectories was enhanced when performing the task ON compared to OFF medication. Our findings suggest that while medication enhances execution processes of movement sequence learning, it may at the same time impair planning processes that precede actual execution. Overall, the current study extends earlier findings on movement sequence learning in PD by differentiating between various components of performance, and further refines previous dopamine overdose effects in sequence learning.

Introduction

Parkinson's disease (PD) is a progressive neurodegenerative disease, in which dopamine (DA) levels in the striatum decrease due to the loss of DA-producing cells in the substantia nigra. The disease is characterized primarily by clearly observable motor symptoms such as resting tremor, difficulty to initiate movement (akinesia), slowness of movement (bradykinesia), rigidity, and postural instability (e.g., Rodriguez-Oroz et al., 2009). At the same time, PD is often accompanied by problems in the cognitive domain (e.g., Barone et al., 2011; Dirnberger and Jahanshahi, 2013; Kudlicka et al., 2011), which reflects additional neuropathology in non-motor related areas of the brain. The DA deficit not only affects patients’ ability to smoothly perform well-practiced gross motor skills like walking, but additionally impairs the learning of novel movement sequences. However, previous studies on movement sequence learning in PD have produced mixed results (for a review, see Ruitenberg et al., 2015). Possible explanations for these inconsistent findings include the influence of dopaminergic medication, the disease severity of the participating PD patients, and methodological variations in the employed tasks between studies. Here we utilized mouse tracking to study various movement parameters of sequence learning in PD. Additionally, we examined whether movement sequence learning in PD is modulated by medication.

Movement sequence learning refers to the acquisition, through repeated practice and explicit instruction, trial-and-error discovery, or incidental detection of regularity, of a fixed series of movements that can eventually be performed rapidly and accurately. Typically, learning is indicated by significantly better performance on a fixed, repeated sequence of stimuli as compared to a random series of stimuli (e.g., Abrahamse et al., 2010, 2013; Shin, 2008; Shin and Ivry, 2003). Previous studies have employed various paradigms to examine such learning in PD, with PD patients often (but not always; see Ruitenberg et al., 2015) showing impairments in sequence learning as compared to healthy control participants on for example explicit trial-and-error paradigms (e.g., Mentis et al., 2003; Tremblay et al., 2010; Wilkinson et al., 2009), as well as implicit learning paradigms (e.g., Shin and Ivry, 2003; Stephan et al., 2011; Wilkinson et al., 2009). Only a small number of studies has addressed the effect of medication status on movement sequence learning. For example, Pascual-Leone et al. (1993) had PD patients perform a sequence learning task while being on their regular medication, as well as after withholding from their medication for 12–36 h. Results showed that, irrespective of medication status, sequence learning in PD patients was impaired compared to control participants. Similarly, Ghilardi et al. (2007) found no effect of medication status on either motor or visual sequence learning. In contrast, other studies suggest that dopaminergic medication modulates sequence learning (Feigin et al., 2003, Kwak et al., 2010, Kwak et al., 2012, Muslimovic et al., 2007). For example, Muslimovic et al. (2007) found that sequence learning in a subgroup of non-medicated early stage PD patients did not differ from that of controls, while PD patients on medication performed worse than controls. In addition, Feigin et al. (2003) observed that medication reduced motor sequence learning in an explicit reaching task. These findings suggest that the use of dopaminergic medication might negatively affect the learning of movement sequences.

Kwak et al., 2010, Kwak et al., 2012 suggested that these inconsistent findings regarding the role of medication could be due to the differential effects of dopaminergic medication on early versus later phases of sequence learning. In both studies, PD patients ON and OFF their medication, as well as healthy control participants, practiced key-press sequences. Kwak et al. (2010) observed that the early phase of sequence learning was impaired in PD patients ON their medication compared to PD patients OFF their medication and healthy controls. Kwak et al. (2012) subsequently demonstrated that both healthy controls and PD patients OFF their medication, but not those ON their medication, showed activity in the ventral striatum during early sequence learning. Whereas the early phase of sequence learning is assumed to rely on the ventral and anterior striatum (i.e., associative striatum), later phases of sequence learning show a shift to reliance on the dorsal and posterior striatum (i.e., sensorimotor striatum; e.g., Doyon et al., 2003; Lehéricy et al. 2005; Miyachi et al., 2002).

In early to moderate PD, the degeneration of DA-producing cells in the substantia nigra, projecting to the dorsal striatum, is more pronounced than cell loss in ventral tegmental area which projects to the ventral striatum. Consequently, the DA-deficiency is greater in the dorsal compared to the ventral striatum (for a review, see MacDonald and Monchi, 2011). While the use of dopaminergic medication (DA precursors, DA receptor agonists, or inhibitors of DA metabolism) enhances motor functions that strongly rely on the dorsal striatum – and therewith alleviates the classical motor symptoms associated with PD – it has been reported to at the same time hinder other motor functions (e.g., Kwak et al., 2010, Kwak et al., 2012) as well as specific cognitive functions (e.g., Cools et al., 2001; Duthoo et al., 2013) that rely on the relatively intact ventral striatum and prefrontal cortex. This discrepancy has been explained by the overdose hypothesis, which postulates that the use of dopaminergic medication restores DA levels within the depleted dorsal striatum, but overdoses the relatively spared areas (e.g., Cools, 2006; Cools et al., 2001; Cools and D’Esposito, 2011; Vaillancourt et al., 2013).

So far, most studies on sequence learning in PD have used relatively simple responses in terms of movement requirements (i.e., key-presses or vocal responses; see Ruitenberg et al., 2015). This provides little information with respect to how response selection and motor control unfold over time. An exception to this is the work by Ghilardi and colleagues (Ghilardi et al., 2003, Ghilardi et al., 2007), who distinguished between onset time and movement time during sequencing performance in PD. Participants in their studies performed an explicit sequence learning task in which they moved a cursor on a digitizing tablet back and forth between a centered starting point and one of eight radially organized targets. They were instructed to learn a fixed sequence by trial-and-error discovery and to reach each target in synchrony with a tone, so that movements had to be initiated before stimulus presentation (i.e., timed sequencing task). Moreover, they were instructed to complete their movements to a target without making corrections towards another target during a trial. Onset time was defined as the time between stimulus presentation and movement onset, and movement time was defined as the time between movement onset and the end point of the movement. Using this task, Ghilardi et al. (2003) compared sequence learning performance in PD patients who performed the task OFF medication to performance of healthy control participants. In both PD patients and controls sequence learning was evident as a reduction in mean onset time across the trials, but this reduction occurred later and less rapidly in patients. There were no differences in movement time. In another study, Ghilardi et al. (2007) directly studied effects of dopaminergic medication by comparing sequencing performance in patients ON versus OFF their medication. Results showed that patients' movement times were significantly shorter ON compared to OFF medication, but medication status did not affect onset times or the rate of sequence learning as reflected in the reduction of onset and movement times across the trials. Although these studies provide additional insight on sequencing performance in PD by investigating both onset and movement times rather than simple reaction times, participants’ movements were restricted as no corrections were allowed during performance, and this may obscure the dynamics of ongoing sequential action.

In the current study we extend the previous work described above by tracking mouse movements I) to examine the action dynamics of sequence learning in PD, and II) to examine the effect of dopaminergic medication on such learning. Freeman and colleagues (Freeman and Ambady, 2010, Freeman et al., 2011, Hehman et al., 2015) developed freely available software that allows for the study of mouse movements and kinematic corrections towards a target as they unfold over time. The analysis of such continuous hand movement trajectories can reveal hidden states in information processing that would be obscured by more traditional discrete measures such as reaction time and errors from simple key-press responses (for reviews, see Freeman et al., 2011; Song and Nakayama, 2009). In earlier studies mouse tracking has been employed to examine, among others, target selection in visual search (Song and Nakayama, 2008), social evaluation and attitudes (Wojnowicz et al., 2009), and syntactic transfer in English-speaking Spanish learners (Morett and Macwhinney, 2012; see Hehman et al., 2015 for more examples). Critically, mouse movements are indicative of participants’ tentative commitment to different action alternatives and reflect decision uncertainty during the progression of a movement.

Here, we tested PD patients both ON and OFF their medication (within-subject) on a sequence learning task that involved mouse responses. To capture the action dynamics of sequence learning, we recorded the initiation time (IT), movement time (MT), accuracy in terms of deviation from the optimal trajectory (area under the curve or AUC), and velocity of each of the participants’ mouse movements. These measures are elaborated below. We here explicitly distinguish between processes that are purely related to movement planning (i.e., preceding the start of actual movement within a trial) and those that mainly¹ reflect execution (i.e., during performance of the actual movement within a trial). Hence, IT provides an index for planning processes, whereas MT, AUC, and velocity constitute indices for execution processes.

On each trial in the present experiment, PD patients used a computer mouse to move a cursor from a start button, located at the central bottom of the screen, to one of four response boxes at the top of the screen. Clicking the start button prompted the presentation of a stimulus that indicated the correct response (see Fig. 1). Participants first performed several blocks of randomly ordered trials, allowing us to evaluate whether medication status affected general mouse movement performance (i.e., not involving sequence learning components). Participants then performed alternating blocks of sequentially presented trials (i.e., a fixed series of movements) and randomly presented trials (i.e., a randomly ordered series of movements). This allowed examination of sequence learning as well as the effect of medication status hereon, as indexed by performance differences between the sequenced and random blocks ON and OFF medication..

We hypothesized that dopaminergic medication would differently affect processes related to movement planning and execution in sequence learning. Previous studies have demonstrated that action initiation in sequence learning tasks is impaired in patients with frontal lobe lesions (Lepage and Richer, 1996). Such impairment has also been shown in healthy controls after repetitive transcranial magnetic stimulation (rTMS) of the pre-supplementary motor area (Ruitenberg et al., 2014) – known to be functionally (e.g., Picard and Strick, 2001) and anatomically (e.g., Picard and Strick, 1996; Wang et al., 2005) connected to pre-frontal areas. As dopaminergic medication has been shown to produce a DA overdose in relatively spared (pre-)frontal areas, we expected sequence learning as reflected in IT to be impaired by medication. Specifically, we anticipated observing a smaller difference on ITs between random and sequenced blocks in patients performing the task ON as compared to OFF their medication (hypothesis 1). This would also align with the studies by Kwak et al., 2010, Kwak et al., 2012 in the sense that both early stages of learning and the initiation of sequential movements can be assumed to be dominated by top-down processes (e.g., related to effortful stimulus-response translation and movement planning, respectively).

In contrast, sequence learning as reflected in the movement execution indices was expected to be enhanced by medication. Previous work showed that rTMS applied to the supplementary motor area – which has direct projections to primary motor cortex (Picard and Strick, 2001) – slowed execution of all elements within a sequence (Verwey et al., 2002). The actual execution of movements involves contributions from the primary motor cortex, which is part of a sensorimotor circuit including the sensorimotor striatum (Dayan and Cohen, 2011). As dopaminergic medication is believed to restore depleted DA levels in the areas within this circuit in PD and thus improve associated functions, we anticipated observing a larger difference on MTs, AUCs and/or velocity between random and sequenced blocks in patients performing the task ON as compared to OFF their medication (hypothesis 2). Overall, we thus predict dissociated effects of dopaminergic medication on movement planning and execution during sequence learning.