Parkinson's disease (PD) is a progressive neurodegenerative illness, characterized predominantly by motor symptoms including tremor, rigidity, and bradykinesia (Jankovic, 2008). More recently, deficits in cognition have also been described in PD patients (Dubois and Pillon, 1997). Dopaminergic medications, such as levodopa (L-3,4-dihydroxyphenylalanine) or dopamine receptor agonists, are prescribed with the main therapeutic goal of improving motor deficits in PD patients (Hornykiewicz, 1974). However, the effects of dopaminergic therapy on cognition are paradoxical, such that some aspects of cognition are improved whereas others are impaired (Gotham et al., 1988; Cools et al., 2001; MacDonald and Monchi, 2011).

Differences in endogenous dopamine levels within disparate brain regions, related to PD pathophysiology, have been proposed to account for these opposing effects of medication on cognitive functions (Cools, 2006; MacDonald and Monchi, 2011). There is a progressive loss of dopaminergic neurons in the substantia nigra (SN) resulting in dopamine depletion in the dorsal striatum (DS), the brain region that it innervates almost exclusively (Kish et al., 1988). This selective dopamine deficiency in the SN is associated with the emergence of motor impairments in PD but has also been related to deficits in DS-mediated cognitive functions such as selective attention and responding as well as task switching (Cools et al., 2001; MacDonald et al., 2011; MacDonald and Monchi, 2011). In contrast, dopaminergic neurons within the ventral tegmental area (VTA) are relatively spared in PD patients (Haber and Fudge, 1997). Subsequently, brain regions innervated by the VTA, such as the ventral striatum, prefrontal and limbic cortices, have relatively spared levels of endogenous dopamine, especially at early disease stages (Haber and Fudge, 1997). Understanding this pathophysiology, it has been contended that cognitive operations mediated by VTA-innervated brain regions, such as probabilistic reversal learning, learning associations between stimuli, as well as stimuli and responses, remain unaffected at baseline due to relatively replete dopamine levels (Cools et al., 2001; Cools, 2006; MacDonald et al., 2011; MacDonald and Monchi, 2011). The differential levels of endogenous dopamine at baseline within the DS compared to the VTA-innervated brain regions have provided a foundation to explain the paradoxical effects of dopaminergic medication on cognition (Cools, 2006; MacDonald and Monchi, 2011).

This dopamine overdose hypothesis has been proposed as an explanation for the differential effects of dopaminergic therapy on disparate cognitive functions (Gotham et al., 1988; Cools, 2006; MacDonald and Monchi, 2011). Dopaminergic medications are titrated to motor symptoms, which are improved by restoring the maximally dopamine-depleted DS (Hornykiewicz, 1974). These medication doses, however, overdose VTA-innervated brain regions, disrupting their functions (Gotham et al., 1988; Swainson et al., 2000; Cools et al., 2001, 2007; MacDonald et al., 2013b; Hiebert et al., 2014b). Accordingly, DS-mediated cognitive functions are improved whereas dopaminergic therapy has detrimental effects on functions mediated by relatively replete VTA-innervated brain regions (Cools et al., 2001; MacDonald and Monchi, 2011). In support of this, many studies have consistently shown that dopaminergic medication improves PD patient performance on various DS-mediated tasks (Cools et al., 2001, 2003; MacDonald et al., 2011; Aarts et al., 2014) but impairs performance on tasks associated with VTA-innervated brain regions (Swainson et al., 2000; Cools et al., 2001; MacDonald et al., 2011).

Learning is the cognitive operation that appears to be most commonly worsened by dopaminergic therapy. Exploring performance on various learning tasks in PD patients on relative to off medication has demonstrated therapy-related impairments in stimulus-reward reversal (Swainson et al., 2000; MacDonald et al., 2013b), sequence (Feigin et al., 2003), probabilistic classification (Jahanshahi et al., 2010), abstract figure, and list learning (MacDonald et al., 2013a). Further, stimulus-response learning is impaired in PD patients on compared to off dopaminergic therapy, but patients off medication exhibit comparable learning to healthy age-matched controls (Vo et al., 2014). In support of this, Hiebert et al. (2014a) used functional magnetic resonance imaging (fMRI) to investigate the brain regions involved in stimulus-response learning in healthy young adults. They reported that learning stimulus-response associations through feedback is mediated by the ventral striatum (Hiebert et al., 2014a), a VTA-innervated brain region (Haber and Fudge, 1997), explaining the effects of dopaminergic therapy on stimulus-response learning in PD and supporting the dopamine overdose hypothesis (Vo et al., 2014).

Overwhelmingly, investigations of the dopamine overdose hypothesis are performed in PD patients. This strategy, however, cannot resolve whether this is a main effect of dopaminergic medications or whether dopamine overdose is the result of a dopaminergic medication by PD pathology interaction. We hypothesized that dopaminergic medication impairs learning independent of PD pathology. A critical test of the dopamine overdose hypothesis consequently involves testing the effect of dopaminergic therapy on functions in healthy individuals (Vo et al., 2016). Healthy young adults provide an ideal model to investigate whether learning impairment is simply due to dopamine overdose in VTA-innervated brain regions because they have optimal endogenous dopamine levels presumably (Haber and Fudge, 1997). The use of healthy young participants eliminates a number of confounds that exist in PD such as dopamine receptor sensitization secondary to chronic exposure (Voon et al., 2009) as well as altered synaptic regulation of dopamine in the form of decreased dopamine transporter (DAT), which also occurs in PD (Harrington et al., 1996; Kordower et al., 2013). In brief, if the mechanism for the previously reported deficits in stimulus-response learning in PD (Vo et al., 2014) is the straightforward result of dopamine overdose to VTA-innervated brain regions (Hiebert et al., 2014a), we expect dopaminergic therapy-related impairments in healthy young adults that parallel effects in PD. In this respect, studying medication effects on learning in healthy individuals provides a critical test of the prevalent dopamine overdose hypothesis.

Using a modified version of the stimulus-response learning task, which previously demonstrated impairment in PD patients on dopaminergic medication (Vo et al., 2014), we compared the effects of pramipexole, a dopamine agonist, versus placebo on stimulus-response learning in healthy young adults. We predicted that participants treated with pramipexole would exhibit impaired learning relative to placebo controls.

In the present study, we demonstrated that a single dose of pramipexole, a dopamine agonist, impairs stimulus-response learning in healthy young adults. This was indicated by lower overall accuracy scores on a stimulus-response learning task in the pramipexole group relative to a matched placebo control group. We found that across blocks of this task, mean percent accuracy scores increased, and mean key-press RTs decreased for both groups, thus demonstrating that learning was achieved. The differential learning performance between the pramipexole and placebo groups was not due to pre-existing cognitive, affective, or demographic differences as no significant differences in any of these control measures were found. In addition, there were no significant differences in physiological measures of heart rate, systolic blood pressure or diastolic blood pressure between the two groups at any time point. Thus, the learning impairment by medication was not a consequence of the physiological effects of the drug. Although heart rate and blood pressure significantly decreased throughout the experiment, this occurred in both groups and we attributed this finding to increasing comfort with the testing situation and perhaps an element of general fatigue and prolonged sitting over the experimental session. This was not due to pramipexole, however, as there were no group differences in these measures.

Alertness was reduced over time, collapsed across group, from the pre-administration to the pre-test and post-test periods. Although there was not a main effect of Group on alertness scores, there was a Group × Time interaction, with lower alertness scores for participants in the pramipexole group relative to the placebo group at the pre-test and post-test time points. Sedation is a recognized side effect of pramipexole and others have demonstrated clinically significant sedation in young healthy volunteers following administration of dopaminergic medications (Micallef-Roll et al., 2001; Micallef et al., 2009). This potentially accounts for the finding that participants were above chance in accurately judging whether they had received pramipexole or placebo. The effect size for the main effect of Time was large, as suggested by η² = 0.264, which we attribute to increased comfort with the testing situation as well as some fatigue associated with completing a number of demographic, clinical, and affective questionnaires. The effect size for the Group × Time interaction was quite small by comparison (η² = 0.098) and there was no main effect of Group on alertness score. This effect on alertness was smaller than the effect of pramipexole on learning score (η² = 0.125), suggesting that alertness alone may not account for differences in learning between participants receiving pramipexole compared to those receiving placebo. Further, despite these differences across Time and to lesser extent related to pramipexole treatment, at no point in the experiment did either group rate their level of alertness as significantly somnolent (lowest mean alertness score was 45.31 on a 100-point scale). Finally, we found no significant differences in key-press RTs between the two groups during the task. Though there are likely multiple influences on RT, including possibly enhanced stimulus-reward value in the pramipexole group (Hickey et al., 2010), if alertness was significantly compromised due to pramipexole, equivalent RTs across groups would be unexpected. Taken together, we interpret that pramipexole-related learning impairment was likely not entirely attributable to reduced alertness.

The current findings are consistent with the PD literature in demonstrating detrimental effects of dopaminergic medication, including pramipexole, on learning. Many studies have reported various forms of learning impairments in PD patients related to dopaminergic medication, including probabilistic associative (Jahanshahi et al., 2010), list (MacDonald et al., 2013a), and stimulus-reward reversal learning (Swainson et al., 2000; Cools et al., 2001, 2007; MacDonald et al., 2013b). Vo et al. (2014) reported that stimulus-response learning was impaired in PD patients tested on their usual dopaminergic medications (i.e., levocarb and/or dopamine agonists as prescribed by their treating neurologist), compared to when they were tested off medication in a within-subject design. Our results are consistent with this study in demonstrating that pramipexole, a dopamine agonist, impaired learning in a slightly modified version of this same stimulus-response learning paradigm.

As we have done here, investigations of the effects of dopaminergic therapy on cognition have also recently been conducted in healthy controls, to understand the main effect that these medications have on cognition independent of PD pathology (Vo et al., 2016). Our study contributes to this small but growing literature that can determine whether cognitive results are attributable in a straightforward manner to dopaminergic therapy or whether they occur as an interaction between medication and PD pathophysiology. The use of healthy young adults, in particular, provides an ideal control model for exploring the effects of dopaminergic medication on cognition. This strategy is advantageous because it avoids the significant variability in typical PD patient groups related to wide age ranges, large differences in disease severity, medication doses and types (Kalia and Lang, 2015). These studies also can rule out the possibilities that these effects occur only (a) secondary to dopamine receptor sensitization, through chronic exposure to dopaminergic therapy, or (b) due to the fact that PD-associated as well as aging-related reductions in DAT levels, which clears and regulates dopamine at the synapse, predisposing to dopamine overdose (Harrington et al., 1996; Voon et al., 2009; Kordower et al., 2013; Kalia and Lang, 2015). Further, it is important to understand cognitive effects of dopaminergic therapy independent of PD because these medications are used in other conditions such as restless leg syndrome (Liu et al., 2016), in some cases of dystonia (Jankovic, 2013), and are being considered as therapy for depression (Cusin et al., 2013). Finally, if these cognitive effects are main results of dopaminergic therapy, this should alert clinicians to the possibility of cognitive improvements and impairments related to dopaminergic medications at any stage of PD.

Previous studies involving healthy individuals have evaluated the effects of dopamine agonists on learning. These investigations have revealed medication-related impairment in associative (Breitenstein et al., 2006), probabilistic reversal (Mehta et al., 2001), and reinforcement learning (Pizzagalli et al., 2008; Santesso et al., 2009) in healthy adults. Results from the present study are entirely in line with these findings. However, the current results contribute the novel observation that pramipexole impairs stimulus-response learning in healthy young adults, which has not previously been investigated to our knowledge.

The dopamine overdose hypothesis explains the paradoxical effects of dopaminergic therapy on cognition in PD patients as owing to the varying endogenous levels of dopamine in the SN-innervated DS compared to VTA-innervated brain regions. A large literature links learning to VTA-innervated brain regions such as the ventral striatum, hippocampus, and medio-frontal and medio-temporal regions (Schendan et al., 2003; Rudy et al., 2007; Denayer et al., 2008; Talpos et al., 2008; Cavanagh et al., 2010; Guo et al., 2011; Hiebert et al., 2014a; Lungu et al., 2014; Mattfeld and Stark, 2015) and not surprisingly in the PD literature learning is the function vastly, most commonly impaired by dopaminergic therapy. Indeed, with the stimulus-response learning paradigm used here, (a) PD patients were impaired by dopaminergic medication (Vo et al., 2014) and (b) feedback-based learning was shown to be mediated by ventral striatum using fMRI in healthy controls (Hiebert et al., 2014a), strongly supporting this dopamine overdose hypothesis.

In the small literature investigating the effects of dopaminergic therapy in healthy controls, learning is also by far the function most frequently found to be impaired. In healthy young participants, however, all brain regions are dopamine-replete and therefore the straightforward dopamine overdose hypothesis predicts that all cognitive functions that implicate the neurotransmitter dopamine, not just those mediated by VTA-innervated brain regions, should actually worsen with dopaminergic therapy. There should be no specific predilection or vulnerability for learning functions in healthy controls. For example, cognitive functions such as task-set switching (Cools et al., 2001), integration of various influences involved in response selection (MacDonald et al., 2011), set-shifting and conflict monitoring (Monchi et al., 2006; Ko et al., 2009), as well as spatial attention (Christian et al., 2006), Stroop interference (Vernaleken et al., 2007), and spatial working memory (Sawamoto et al., 2008) depend upon dopamine and should all be worsened in the on condition according to the dopamine overdose account. These examples do not exist and indeed there are a very small number of investigations in healthy controls that in fact revealed improvements related to exogenous dopamine administration in short-term spatial memory (Mehta et al., 2001), working memory (Murphy et al., 2016), response inhibition (Roesch-Ely et al., 2005), and novel word learning (Knecht et al., 2004; Shellshear et al., 2015). At this point, it is possible that the effects of dopaminergic therapy on these other cognitive functions have not been investigated. It is equally possible that these investigations have not revealed detrimental effects in the form of between-group or between-condition differences and therefore do not appear in the publication record. Future studies should directly contrast, within the same participants, the effects of dopaminergic therapy on cognitive functions known to be mediated by brain regions innervated by SN and those innervated by VTA, to investigate any particular vulnerabilities. These studies should also include neuroimaging measures to fully substantiate interpretations.

In summary, we have found that a single dose of the dopamine agonist pramipexole impairs stimulus-response learning in healthy young adults who have intact dopamine levels and dopamine regulation functions. This confirms that the detrimental cognitive effects of dopaminergic therapy are a main effect of these drugs and do not arise only as an interaction between these medications and PD pathology. The dopamine overdose hypothesis is supported here. This has implications for use of pramipexole in other conditions such as restless leg, dystonia, or for potential future use of this medication in depression. Further, this alerts clinicians to the potential of detrimental effects of dopaminergic therapy, here pramipexole, on learning in PD patients at all stages of disease, independent of the nature or extent of their PD pathology. The serious motor and cognitive symptoms in PD related to DS dopamine deficiency obviously warrant dopaminergic therapy. However, these results caution that the goal of therapy should be to strike a better balance between these benefits and potential side effects based on individual patient priorities and symptomatology.

HG participated in data acquisition and interpretation, as well as writing the first draft and revising the manuscript. AV, KS, and PM were involved in study design and conceptualization, obtaining materials, data analysis and interpretation, and revising the manuscript.

This research was supported by a Canada Research Chair (CRC) Tier 2 in Cognitive Neuroscience and Neuroimaging to PM (Grant: 950-230372), a Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery Grant to PM (Grant: 6621), an Internal Research Fund Grant from Lawson Research Institute to PM, and an Ontario Graduate Scholarship to AV.