Introduction

Impulse control disorders (ICD) can be triggered by dopamine replacement therapy (DRT) used in Parkinson’s disease (PD) patients, especially with dopamine agonists (Weintraub et al. 2010). Converging evidence suggest that PD itself does not confer an increased risk for development of ICD or related behaviour symptoms in the absence of treatment, reinforcing the reported association between PD medications and ICDs (Weintraub et al. 2013). However, not every PD patients under dopaminergic therapy develop ICDs that seem to occur only in a specific subset of patients that could be more at risk. Thus, even if DRT appears to be the main risk factor for the occurrence of ICDs in PD patients, non-pharmacological risk factors for ICDs have also been identified in that population, such as early age at disease onset, male gender, personal or family history of alcohol abuse or pathological gambling, novelty seeking personality, cognitive or psychiatric disorders, and sleep disorders (Fantini et al. 2015; Vitale et al. 2011; Voon and Fox 2007; Voon et al. 2011b). A specific genetic profile may also be associated with a higher risk to develop ICDs in PD (Kraemmer et al. 2016). Lately, neuroimaging and preclinical studies have helped improve our understanding of the mechanisms underlying the development of ICDs in PD.

Clinical aspects

Clinical description

ICDs are an increasingly recognized non-motor complication in PD and are described as behaviours in which a person fails to resist the drive to behave repetitively, excessively and compulsively in ways that could be dangerous for him-/herself or other people and could interfere in major areas of life functioning. In regard to drug addiction, ICDs have been defined as “behavioural” addictions and include compulsive eating, gambling, buying and sexual behaviour (Voon and Fox 2007; Weintraub et al. 2015). Other impulsive–compulsive behaviours (ICBs) have been reported in PD, as consequences of DRT. One of them is dopamine dysregulation syndrome (DDS) (Giovannoni et al. 2000) characterized by compulsive overuse of DRT particularly with l-dopa and short-acting dopamine agonists. Several other ICBs have also been related in PD such as punding (stereotyped, repetitive, purposeless behaviours) (Evans et al. 2004), hobbyism (Weintraub et al. 2013) (similar but higher level of repetitive behaviours: internet use, reading, art work, work on projects). Walk-abouts (excessive aimless wandering) (Giovannoni et al. 2000; Weintraub et al. 2013) and hoarding (acquisition and failure to discard a large number of items with little or no objective value) (O’Sullivan et al. 2010) are also reported ICBs. Enhanced creativity has also been described in PD patients under dopaminergic therapy but it is still debated whether creativity is or not related to ICBs (Canesi et al. 2016; Faust-Socher et al. 2014). Yet, artistic production and creative thinking were reported not to be influenced by DRT in PD (Canesi et al. 2016), and verbal and visual creativity in PD were found to be unrelated to ICDs (Faust-Socher et al. 2014).

These behavioural disorders may be present with various degrees of severity, but may lead to major functional and quality of life impairment (Phu et al. 2014), particularly for the caregiver (Leroi et al. 2012). The excessive nature of these behaviours is not always easy to define, depending on premorbid behaviours of patients and cultural differences; thus it is generally admitted that these behaviours become a disorder when they become harmful for the patients or his relatives, or interfere with social relationships (Weintraub et al. 2015).

Evaluation tools for ICDs in PD

There are few validated instruments for the evaluation of ICDs and related disorders in PD. Yet these tools are of clinical relevance, as patients may often not report such behaviours to their neurologists either because they are not aware of the association with antiparkinsonian drugs or because of embarrassment. Thus screening and rating instruments are very important for the diagnosis of these disorders, to monitor changes in symptoms over time and to assess the efficacy of future therapeutic strategies.

DSM V

Diagnostic and statistical manual of Mental Disorders 5 (DSM-5) has a new chapter on disruptive, impulse-control and conduct disorders (American Psychiatric Association 2013). This chapter includes oppositional defiant disorder, conduct disorder, disruptive behaviour, impulse-control disorders, intermittent explosive disorder, pyromania, and kleptomania. The ICD-11 working group on obsessive–compulsive and related disorders was asked to review the approach of the DSM-V to these conditions with a particular emphasis.

Modified version of MIDI (Minnesota Impulsive Disorders Interview) (Christenson et al. 1994)

The MIDI is a semi-structured interview used to assess the degree of impulsivity related to compulsive behaviour. It comprises 36 items and different modules for each of the ICDs: compulsive buying (CB), pathological gambling (PG), Hypersexuality (HS), compulsive eating (CE) and other impulsive disorders not or unusually reported to occur in PD (pyromania, kleptomania, trichotillomania, pathological skin picking) (Bonfanti and Gatto 2010). Each module is composed by a screening general question to which the subject has to answer affirmatively (1 point) or negatively (0 point). In case of affirmative answer to the screening question, the interviewers asks a series of additional questions reflecting DSM-IV criteria [(1) urges to perform the behaviour, (2) efforts to resist the urges, (3) feelings of tension prior to performing the behaviour, (4) feeling of release after performing the behaviour, and (5) distress and impairment associated with the behaviour], and also disorder-specific questions to assess clinically relevant details. The MIDI has been widely used in clinical and research settings but neither the internal consistency of the items or the inter-reliability of the diagnoses made according to this tool have been demonstrated. The validity of the MIDI has been reported in a sample of patients with various psychiatric disorders, but not in PD (Grant et al. 2005). Moreover, the MIDI does not cover punding, DDS or hobbyism.

QUIP (screening questionnaire for impulsive–compulsive disorders in Parkinson’s Disease) (Weintraub et al. 2009), and a rating scale version of this instrument (QUIP-RS) (Weintraub et al. 2012)

The QUIP is a screening questionnaire developed and validated to assess ICDs and related disorders specifically in PD with a good discriminating validity. It provides a dichotomous choice (yes/no) as a response to various questions.

While the QUIP is valid as a screening instrument for ICDs it does not allow an assessment of the severity of these disorders, which is necessary to establish diagnosis cutoff points and for the follow-up of these behaviours. Therefore, the QUIP rating scale (QUIP-RS) has been developed for that purpose and appears to be a valid and reliable rating scale for ICDs in PD. It uses a five-point Likert scale that requires individuals to rate the severity of each symptoms depending on its frequency. Cutoff points for the diagnosis for all four ICDs (but not for DDS) have been proposed with a sensitivity and specificity > 80%. Both of these instruments can be self- or rater-administered and have been translated and validated in numerous languages.

The SCOPA-PC (Visser et al. 2007)

This is a validated screening instrument for Psychiatric complications in PD, containing seven questions related to symptoms that have been occurring in the past month. It includes questions about hallucinations, illusions, paranoid ideation, altered dream phenomena and confusion, but also one question about sexual preoccupation and compulsive shopping or gambling. Together with the QUIP questionnaire, the SCOPA-PC are until now the only validated tools for screening ICDs in PD.

The Ardouin scale of behaviour in Parkinson’s disease (Rieu et al. 2015)

This ASBPD is a new instrument consisting of a semi-structured standardized interview designed to assess neuropsychiatric features in PD, with a particular focus on hypo-and hyperdopaminergic mood and behaviours. This scale includes 21 items grouped into three parts: hypodopaminergic disorders (part I), non-motor fluctuations (part II), and hyperdopaminergic behaviours (part III). Part I successively evaluates depressive mood, anxiety, irritability and aggressiveness, hyperemotionality, and apathy. Part II assesses the psychological state associated with the motor symptoms in the OFF and ON states in fluctuating patients. Part III evaluates behavioural disorders mostly induced by dopaminergic treatment, including hypomanic mood, psychotic symptoms, nocturnal hyperactivity, diurnal somnolence, risk-taking behaviour, excess motivation, and some items referring to ICDs (eating behaviour, compulsive shopping, pathological gambling, hypersexuality) and related behaviors (creativity, hobbyism, punding, dopaminergic addiction). The timeframe of the assessment is the preceding month. Each item is rated on a five-point scale (4 = severe disorder; 3 = marked disorder; 2 = moderate disorder; 1 = mild disorder; 0 = absence of disorder), by taking into account the severity and frequency of the disorder, as well as its impact. This scale allows for a quantitative measurement of a large range of behavioural disorders, including all ICDs and related behaviours, and is thus a valuable tool for follow-up.

Risk factors for ICDs in PD

Dopaminergic therapy

The main risk factor for developing ICDs in PD is the use of DRT. Indeed, an association between the use of dopaminergic drugs, in particular dopamine agonists (DA), and the occurrence of ICDs and ICBs in PD has been consistently reported, since the first published case reports. Indeed, at least one ICD has been reported in 17.7% of PD patients treated with both levodopa and DA, in 14% of Patients under DA only, 7.2% for those treated with levodopa only, and 1.7% in PD without DA or levodopa (Weintraub et al. 2010). Many studies subsequently confirmed the predominant role of dopaminergic agents in the emergence of ICDs in PD, with the strongest association reported for DA versus levodopa (Isaias et al. 2008; Lee et al. 2010; Marechal et al. 2015; Weintraub et al. 2006). However, it remains debated whether there is (Lee et al. 2010; Marechal et al. 2015; Weintraub et al. 2006) or not (Avanzi et al. 2006; Singh et al. 2007; Voon et al. 2006) a dose effect relationship in the ICD and related behaviours associated with dopaminergic agents. There could exist specific mechanisms depending on different categories of ICDs as eating compulsive disorders and punding could be associated with higher levodopa dose, whereas buying and gambling could be associated with higher DA dose (Lee et al. 2010). Comparative studies of clinical occurrence of ICDs on long-acting versus short-acting DA are lacking, but it has been reported a relatively low rate of ICDs with long acting or transdermal DA in preliminary observational studies (Garcia-Ruiz et al. 2014; Rizos et al. 2016). Longer duration of dopaminergic agents may also increase ICD risk (Bastiaens et al. 2013; Giladi et al. 2007). The reversibility of ICDs and related behaviours after withdrawal of dopamine agonist has been well documented in several studies (Bostwick et al. 2009; Dodd et al. 2005; Klos et al. 2005; Mamikonyan et al. 2008; Weintraub et al. 2006).

Since the prevalence of ICDs has not been reported to be different in untreated PD compared to general population (Weintraub et al. 2013), and since ICDs have been described in other conditions treated with dopaminergic agents such as Restless Legs syndrome (RLS) or prolactinoma (Bancos et al. 2014; Cornelius et al. 2010; Driver-Dunckley et al. 2007), Parkinson’s disease itself does not appear to confer an increased risk for development of ICDs that would only be due to DRT. However, the dopaminergic denervation observed in PD could increase the risk to develop ICDs when associated with DRT as suggested by a threefold risk to develop ICDs in PD compared to RLS in spite of both populations being treated with dopamine agonists (Grall-Bronnec et al. 2017). It could be argued that lower doses of dopamine agonists used in RLS could account for these different prevalences, but the lack of a clear dose–effect relation reported between dopamine agonists and ICDs goes against that sole hypothesis. Moreover, a different pattern of denervation has been reported in PD patients with and without ICDs (see following sections). These findings suggest that the emergence of ICDs in PD could be in fact enhanced by the particular and unique association of a dopaminergic denervation (and probably a specific pattern of denervation in subjects that could be more at risk to develop ICDs) and the administration of a chronic DRT (particularly D3/D2 agonists). Overall, even if dopaminergic replacement therapy is the most consistent risk factor for the emergence of ICDs in PD, the question of other non-pharmacological risk factors is raising more and more interest.

Non pharmacological risk factors

Demographic and clinical risk factors

Several risk factors have been identified for the occurrence of ICDs in PD, such as younger age at PD onset, being a man, being single, past or current depression, positive family history of smoking or substance abuse (Vitale et al. 2011; Voon et al. 2011b). Recently, some studies have also reported an association between ICDs and sleep disorders in PD, particularly REM sleep behaviour disorders (Fantini 2017; Fantini et al. 2015; Scullin et al. 2013). Indeed, it was showed after a multivariate analysis accounting for gender, age of onset, disease duration, PD severity, depression score, and total and agonists levodopa equivalent dose that PD patients with RBD had a 2.6-fold risk to develop ICDs (Fantini et al. 2015). RLS could also be per se an independent risk factor for the emergence of ICDs in PD (Marques et al. 2018). All these features could be interesting parameters to take into account when deciding to introduce DA in PD patients, in order to estimate their risk to develop ICDs. Moreover, both personality (including traits of negative affect and high levels of impulsivity and novelty seeking) and cognitive features (namely poor executive functioning with inhibitory control impairment) may also represent risk factors for the development of ICD in medicated PD patients (Poletti and Bonuccelli 2012a; Santangelo et al. 2013, 2017; Vitale et al. 2011).

Personality

Personality differences have been reported in PD patients at early stages of the disease and before the onset of motor symptoms, with less novelty seeking (exploratory activity in response to novel stimulation) and more harm avoidance (excessive worrying or pessimism in anticipation of future problems, passive avoidant behaviors such as fear of uncertainty and shyness of strangers) than control populations (Baig et al. 2017; Poletti and Bonuccelli 2012b). The putative role of these premorbid traits as predisposing factors for the development of ICDs in PD is still debated (Poletti and Bonuccelli 2012a). Yet personality profile could account for differences in behaviours, namely in risk taking propensity (Llewellyn 2008). While novelty seeking seems to be significantly lower in drug naïve newly diagnosed PD patients compared to healthy controls, after 12 weeks of DRT novelty seeking was significantly higher in the same patients compared to healthy controls and to the same patients before treatment (Bodi et al. 2009). Another study reported higher novelty seeking in 25 PD patients with compulsive dopaminergic drug use compared to healthy controls and to PD without this complication (Evans et al. 2005). Thus higher novelty seeking under DRT could be associated with the presence of ICDs in PD. Impulsivity is a complex personality trait distinguished from novelty seeking and defined as a tendency to enter into situations or rapidly respond to cues for potential reward without much planning or deliberation and without consideration of potential punishment or loss of reward (Zuckerman and Kuhlman 2000).

Impulsivity is associated with ICD in pathological gamblers, and could probably also play a role in the development of ICD in PD patients treated with dopamine agonists. However, the relationship between novelty seeking, impulsivity and impulsive real-life behaviours that has been reported in the general population has not been clearly confirmed in PD patients (Poletti and Bonuccelli 2012a). Indeed, in newly diagnosed drug naïve PD patients impulsivity did not influence real-life impulsive behaviors (Antonini et al. 2011), whereas a similar study conducted in PD patients under DRT reported the opposite finding with increased impulsivity in patients with real life impulsive behaviours (Isaias et al. 2008). This is probably due to the multifactorial aspects of impulsivity such as cognitive impulsivity (making quick cognitive decisions), motor impulsivity (acting without thinking) and non-planning impulsivity (lack of “looking into the future”) (Llewellyn 2008), and to a different effect of dopaminergic denervation and DRT on these different components.

It was also found that alexithymia (defined by a difficulty to identify feelings and describe them) could be considered as an independent risk factor for ICD in PD as well as in general population (Bonnaire et al. 2013; Goerlich-Dobre et al. 2014).

Cognition

In the same line, whether executive dysfunction is or not a risk factor for the development of ICDs in PD remains a matter of debate. Several neuropsychological studies have demonstrated that PD patients with ICDs show an altered cognitive profile compared to patients without ICDs, including impaired cognitive flexibility and planning capability as well as poor feedback process (Santangelo et al. 2009; Vitale et al. 2011). Thus it has been reported that ICDs in PD patients are associated with impaired working memory and executive functions as well as dysfunctions in tasks of visuospatial planning and set-shifting (Biundo et al. 2015; Djamshidian et al. 2010; Santangelo et al. 2009; Voon et al. 2010). When looking more precisely into different subtypes of ICDs, a specific association between pathological gambling and dysfunction of both set-shifting and cognitive flexibility has been described (Vitale et al. 2011). Compared to patients with pathological gambling, those with hypersexuality showed more severe impairments of inhibitory control and verbal learning, while patients with compulsive eating showed an intermediate pattern (Vitale et al. 2011). As suggested by the authors, these differences of cognitive impairment among subtypes of ICDs may reflect differential involvement of the neural substrates devoted to natural rewards (such as sex and feeding) or learned rewards (such as money) (Vitale et al. 2011). However, other studies showed either no difference for cognitive functions between PD patients with and without ICDs, or better performances for Mini Mental Status and tasks of episodic memory, verbal fluency and attention, suggesting that pathological gambling is associated with preserved cognitive functioning in PD patients (Rossi et al. 2010; Siri et al. 2010).

A longitudinal cohort study investigated cognitive functions in PD patients with and without ICDs in order to identify possible cognitive predictors of behavioural outcome. ICDs in PD patients were not related to greater cognitive impairment or dysexecutive dysfunction but rather to a relatively lower cognitive decline over time (Siri et al. 2015). These findings are not necessarily conflicting and the authors suggest that impaired executive functions observed in PD patients with ICD could actually be due to a reversible drug-induced impairment on selective frontal lobe tasks in predisposed individuals (Williams-Gray et al. 2009). The occurrence of ICDs in patients treated with DA for disorders sparing cognitive functions such as RLS strengthens that hypothesis (Cilia and van Eimeren 2011; Voon et al. 2011a). Thus, it appears that impaired top-down inhibitory control of behaviour from prefrontal cortical areas associated with DRT could predispose PD patients to develop ICDs, (Bentivoglio et al. 2013; Cilia and van Eimeren 2011; Santangelo et al. 2013; Voon et al. 2011a) but ICDs per se in PD patients do not seem to be associated with a decline in executive functions overtime (Siri et al. 2015).

Genetic risk factors

The risk to develop ICDs in PD might be enhanced by several demographic and clinical risk factors that we exposed earlier. Yet, genetic polymorphisms in PD could also play a role in the development of ICDs. Thus genetic studies have linked the risk to develop ICDs with polymorphism of several genes such as genes encoding for dopamine receptors (DRD3 and not DRD2 conversely to general population) and serotonin receptors (5HT2AR) (Cormier et al. 2013). However, no difference was found regarding allelic variations for genes involved in the control of dopamine homeostasis (COMT, DAT) when comparing PD patients with and without ICDs (Lee et al. 2009; Vallelunga et al. 2012). These discrepancies could be due to the small size of the studied samples and to the lack of strong defining criteria for ICDs in PD, and underline the need for further larger studies in that field. Recently a longitudinal cohort study including 276 de novo PD patients aimed to estimate ICD heritability and to predict ICD by a candidate genetic multivariable panel in patients with PD (Kraemmer et al. 2016). It showed that ICD behaviours had a substantial heritability in early PD and that a genetic panel including 13-SNP (Single nucleotide polymorphisms) in 12 candidate genes significantly increased ICD predictability, leading to a predictive model reaching clinically relevant accuracy. These data suggest the existence of a premorbid genetically determined neurobiological risk factor for ICDs in PD, but need to be confirmed by additional studies to determine the respective weight of pharmacological and genetic risk factors for the development of these behaviors in PD.

Pathophysiology

Anatomopathological studies

There is no available anatomopathological study in the field of ICDs in PD, and further research will be needed in this area in order to improve our understanding of these disorders.

Imaging studies

Neuroimaging has been largely used to understand the mechanisms underlying ICDs in PD and to identify the structures involved in the reward system as well as its abnormalities elicited by exogenous stimulation in PD (Aracil-Bolaños and Strafella 2016). All data from brain metabolism, functional and morphometric imaging studies converge to show an overactivity of the meso-cortico-limbic pathway in PD patients with ICD, involving cortical but also subcortical areas which are critical in the reward system, such as orbitofrontal cortex (OFC), amygdala, insula, anterior cingular cortex (ACC), and ventral striatum (Balarajah and Cavanna 2013; Keitz et al. 2008).

Molecular imaging studies

Molecular imaging in PD patients, especially treated with DA, has largely focused on dopamine and its receptors, the autoregulatory mechanisms and the inhibitory inputs from cortical structures. Thus striatal (Cilia et al. 2011; O’Sullivan et al. 2011; Politis et al. 2013; Steeves et al. 2009) and extrastriatal (Ray et al. 2012) molecular imaging studies have demonstrated the presence of an “hyperdopaminergic state” in the brain of PD patients with ICDs, when exposed to reward stimuli. An increased ventral striatal dopamine release in response to rewards, but also a low ventral striatal D2/D3 receptor availability have been described in [11C] raclopride PET studies, suggesting a sensitization of ventral striatal circuits to reward and a blunted transmission of reward signals (reward deficiency syndrome), leading to increased reward seeking (Blum et al. 1996; O’Sullivan et al. 2011; Steeves et al. 2009). The low DAT expression reported in ventral striatum could account either for a more pronounced dopaminergic denervation, a functional downregulation (compensating for lower dopamine availability) or a premorbid susceptibility trait (Cilia et al. 2010; Vriend et al. 2014). A lack of homeostatic control over striatal DA release, causing increased sensitivity to reward/propensity for impulsivity, was also suggested by low midbrain dopamine autoreceptor in a [11C] FLB-457 and PET study (Ray et al. 2012). From all these studies, it appears that DA could enhance the risk to develop ICDs in PD by preventing pauses in dopamine transmission and thus impairing the processing of negative reinforcement, together with an augmentation of the encoding of positive reinforcement due to a tonic occupation of postsynaptic receptors (Ray and Strafella 2013).

Cortical perfusion studies in ICD

Regional cerebral blood flow studies complete dopaminergic assessment and allow the understanding of networks and systems involved in impulsive behaviors in PD. Tonic stimulation of dopamine receptors could desensitize the dopaminergic reward system by preventing decreased in dopaminergic transmission that usually occurs with negative feedback. Indeed, a study using functional MRI showed impaired cortical and striatal responses to feedback when on dopamine agonists in PD patients with pathological gambling suggesting an increased learning from reward and decreased learning from loss, promoting the repetition of inadequate behaviours (van Eimeren et al. 2009; Voon et al. 2011b). The authors propose that ICDs may be caused by an impaired capacity of the OFC to guide behaviour when facing negative consequences (van Eimeren et al. 2009). However, ICDs are a multidimensional concept including a cognitive component (that could lead to impulsive choices) a motor component (that could lead to impulsive actions), and these different components could be differentially sensitive to dopaminergic treatments. A study investigating regional cerebral blood flow modifications in PD patients with and without DA medication during tasks assessing impulsivity showed that DA may influence mainly the neural network underlying impulsive choices (the medial prefrontal cortex and posterior cingulate) but not the network mediating impulsive action (lateral prefrontal cortex) (Antonelli et al. 2014). Resting-state fMRI allows the understanding of functional connectivity between different cortical regions without the bias that could be due to difference in task paradigms during functional imaging studies. Resting state brain networks were found to be different in PD patients with ICD compared to those without ICD (Tessitore et al. 2017). Indeed, ICDs were associated with a specific pattern of activation characterized by a reduced functional connectivity within the central executive network (prefrontal cortex and inferior parietal cortex mediating external attentionally driven executive functions, judgment and decision making) and an increased connectivity within the salience network (anterior insula, anterior cingulate cortex and ventral striatum mediating affective, reward processing and interoceptive awareness) and sensorimotor network (posterior cingulate, medial prefrontal and temporal cortex mediating ruminations, mind-wandering, emotional processing and cognitive social functions), and these modifications were correlated with the severity of ICDs in PD (Tessitore et al. 2017).

Morphometric studies

Parkinson’s disease patients with ICDs have an increased cortical thickness in limbic regions when compared to PD patients without ICD, regardless of the disease stage, cognitive deficits or daily levodopa equivalent dose. These corticometric changes may play a role in the lack of inhibition of compulsive behaviours as suggested by the positive correlation between these differences in cortical thickness and the severity of ICDs in this study. The presence of such structural abnormalities may be induced by a ventral striatum “hyperdopaminergic state” leading to an aberrant cortical plasticity within ACC and OFC, and hereby to cortical thickening (Tessitore et al. 2016). Interestingly, the same cortical thickening has been observed in the inferior frontal sulcus in PD patients with dyskinesias, in line with the hypothesis of a similar mechanism in limbic or motor circuits leading to ICDs and dyskinesias in PD (Cerasa et al. 2013).

Preclinical aspects

Clinical studies have provided valuable insight into the clinical and socio-demographic risk factors of ICD, in addition to the identification of dysfunctional networks from neuroimaging studies (see previous sections). Furthermore, numerous clinical studies consistently identified D3/D2 agonists as a pharmacological class of dopamine replacement therapy (DRT) strongly associated with the emergence of ICD (Isaias et al. 2008; Weintraub et al. 2010). Unlike dyskinesia that inexorably arises following chronic l-dopa intake, the fact that only a portion of treated PD patients develop ICD indicates that pre-existing individual vulnerability and/or specific disease-related features play a significant role. However, the pathophysiological mechanisms leading to the development of ICD in only a portion of PD patients remain to be fully elucidated.

In this regard, experimental models of PD can be useful to investigate the pathophysiology of ICD by determining how identified risk factors individually contribute and how they interact with DRT and dopaminergic neurodegeneration to promote the development of ICD. Indeed, disentangling the respective contributions of DRT, disease-specific patterns of neurodegeneration and pre-existing vulnerability is virtually impossible in clinical studies. These pre-clinical models will also be valuable to decipher the neural substrates and molecular mechanisms potentially involved, even though there is so far no preclinical model with face and construct validity [discussed in (Cenci et al. 2015)]. Developing experimental models to investigate the pathophysiology of ICD is associated with a number of technical and methodological issues and proves to be more challenging than investigating motor outcomes such as motor benefits of DRT or l-dopa induced dyskinesia. Although unilateral models (e.g.; unilateral 6-OHDA lesion) are well suited for investigating motor response and motor side-effects of DRT due to a marked dopaminergic asymmetry, they are unsuitable to investigate non-motor deficits and non-motor side-effects of DRT. Consequently, models with bilateral nigrostriatal denervation are required. Since the bilateral injection of 6-OHDA in the medial forebrain bundle or substantia nigra is associated with substantial lethality and animal welfare issues (weight loss, adipsia, aphagia) other options must be considered. Currently available strategies include bilateral striatal injections (Baunez and Robbins 1999; Rokosik and Napier 2012) or repeated intracerebroventricular administration of 6-OHDA (Engeln et al. 2013a; Quiroga-Varela et al. 2017), as well as bilateral viral-mediated overexpression of alpha-synuclein in the substantia nigra using adeno-associated viral vectors (AAV) (Engeln et al. 2013b, 2016a; Loiodice et al. 2017). Among these different lesioning strategies, AAV-mediated overexpression of alpha-synuclein provides multiple advantages over acute 6-OHDA models, including progressive motor impairments and neurodegeneration that have been described with good reproducibility by several groups (Bourdenx et al. 2015; Decressac et al. 2012a, b; Gombash et al. 2013). In addition, this model enables to recapitulate the accumulation of alpha-synuclein, one of the cytopathological hallmarks of PD, as well as synaptic dysfunction in surviving neurons consecutive to alpha-synuclein accumulation (Lundblad et al. 2012).

Behavioural effects of dopaminergic lesion and/or DA relevant to ICD

Clinical studies suggest that the occurrence of ICD involve multiple interactions between D2/D3 agonists, individual susceptibility, and the disease process, with a strong contribution of the drugs (see previous sections). Even though there is so far no experimental model that faithfully recapitulates ICD occurring in PD (Cenci et al. 2015), several behavioural features potentially relevant to the pathophysiology of ICD have been investigated in healthy animals and in pre-clinical models of PD. These include the assessment of the effects of DRT and/or dopaminergic loss on reward processing, risk-taking, impulsivity, gambling-like tasks, as well as perseverative/compulsive behaviours.

The rewarding properties of several DRT have been investigated in normal and preclinical rodent models of PD. Among classes of DRT, D3 and D2 rather than D1 agonists display rewarding properties (Zengin-Toktas et al. 2013). DRT incriminated in ICDs such as pramipexole display significant rewarding properties (Engeln et al. 2013a; Riddle et al. 2012) that are enhanced by the dopaminergic lesion when assessed using conditioned placed preference (Riddle et al. 2012). Increased rewarding properties of apomorphine and levodopa were also demonstrated in lesioned animals compared with sham rats (Campbell et al. 2014; Engeln et al. 2013b). The pattern of dopaminergic lesion may also influence the rewarding properties of DRT since lesion of the posterior, but not anterior ventral tegmental area, increase the rewarding properties of pramipexole and bromocriptine (Ouachikh et al. 2013, 2014).

Impulsivity being a multifaceted construct, several dimensions of impulsivity can be individually assessed using dedicated instrumental conditioning tasks. Regarding impulsive choice, it was first demonstrated that pramipexole (0.1–0.32 mg/kg) increased delay discounting in healthy rats, but at the expense of increased response latency and omissions (Koffarnus et al. 2011; Madden et al. 2010). Although such finding may be interpreted as increased impulsive choice at first glance, it should be noted that pramipexole decreased the choice for the large reinforcer even when no delay was imposed and highest pramipexole doses tested shifted preference towards indifference (Koffarnus et al. 2011; Madden et al. 2010). Surprisingly, both D2/D3 agonists and antagonists were shown to induce similar effects as shown with a decreased choice of the large reinforcer regardless of the delay (Koffarnus et al. 2011). Interestingly, further insight into the role of nigrostriatal degeneration was provided by the demonstration that striatal infusions of 6-OHDA increased delay discounting in rats without affecting the number of omissions during trials (Tedford et al. 2015). Whether the dopaminergic lesion combined with DA may lead to an additive effect on delay discounting has yet to be assessed.

Other dimensions of impulsivity have been assessed in the context of PD and ICD. Using a differential reinforcement of low rate of responding and fixed consecutive number schedules of reinforcement to, respectively, assess waiting and action inhibition/motor impulsivity, viral-mediated lesion of the nigrostriatal pathway led to increased impulsivity in both tasks (Engeln et al. 2016a). Acute challenge with pramipexole (0.3 or 1 mg/kg) increased waiting impulsivity and drastically decreased action inhibition. However, the effects of pramipexole on waiting impulsivity were more pronounced in high impulsive lesioned rats indicating a possible interaction between the pattern of dopaminergic denervation, DRT, and impulsivity trait. Furthermore, this study demonstrated that despite being associated with a reduced neurodegeneration following alpha-synuclein overexpression, a premorbid impulsivity trait exacerbates the effects of pramipexole on waiting and motor impulsivity (Engeln et al. 2016a).

Behavioural tasks using gambling-like schedules of reinforcement are useful to assess the effects of DA on decision-making and risk-taking behaviour in the context of PD and ICD. Studies using a probabilistic discounting task demonstrated that pramipexole promoted disadvantageous choice (e.g. maintained choice of the large reinforcer at the lowest probability of delivery) (Pes et al. 2017; Rokosik and Napier 2012). However, in the only study that investigated the combined effect of 6-OHDA induced nigrostriatal degeneration and pramipexole on probabilistic discounting, there was no effect of the lesion itself and the deleterious effect of pramipexole was similar between lesioned and sham rats (Rokosik and Napier 2012). Other studies using variable (i.e. gambling-like) versus fixed schedules of reinforcements also demonstrated that pramipexole increased choice toward the gambling-like schedule in a dose-dependent manner (Johnson et al. 2011, 2012). Using selection criteria to classify rats as high or low risk-takers based on their behaviour in a probabilistic discounting task, Holtz et al. demonstrated that following a chronic treatment with pramipexole (1.2 mg/kg/day for 28 days) in rats with intrastriatal 6-OHDA-induced dopaminergic loss, two-thirds of the animals met the criteria for high risk-taking, thereby highlighting individual differences in the response to pramipexole (Holtz et al. 2016). This study also identified that the antidepressant mirtazapine was able to reduce risk taking in lesioned rats treated with pramipexole.

Using a rodent betting task that measures choice toward certain versus uncertain outcomes, the D2/D3 agonist associated with ICD ropinirole (5 mg/kg/day) was shown to increase preference for uncertainty regardless of the baseline preference for uncertainty (Tremblay et al. 2017). In this study, 6-OHDA-induced dopamine denervation of the dorsolateral striatum did not affect choice behaviour and the effect of ropinirole in these rats was similar between sham and lesioned rats. Using a rodent slot machine task, ropinirole (5 mg/kg/day) was also demonstrated to markedly stimulate gambling behaviour, as shown with a large increase in the number of trials completed during sessions and indicating a possible compulsive-like engagement in the behaviour (Cocker et al. 2017). Perseverative, compulsive-like lever pressing was likewise assessed in a rat model of PD using the post-training signal attenuation procedure (Dardou et al. 2017). In this study, bilateral lesioning of the substantia nigra and VTA with 6-OHDA did not affect extra lever presses but when these animals were treated chronically with pramipexole (0.3 mg/kg/days for 14 days), they developed a strong compulsive lever pressing behaviour in the post-training signal attenuation procedure (Dardou et al. 2017).

Altogether, the behavioural studies concur to indicate deleterious effects of DA implicated in ICD such as pramipexole or ropinirole on various behavioural tasks involving decision-making, risk taking, impulsivity, and compulsive behaviour, whereas there was no consistent effect of the dopaminergic lesion. However, it should be noted that some of these studies were performed with doses of DRT that are likely above the therapeutic range used in patients and that some of the lesion models involve a highly restricted pattern of striatal dopaminergic loss that may only partially replicate the loss occurring in the human disease.

Cellular and molecular mechanisms potentially involved in ICD

Due to the lack of post-mortem studies on PD patients with ICD and to the absence of a preclinical model fulfilling the clinical criteria of ICD, one can only speculate about the underlying cellular and molecular mechanisms involved based on findings from heuristic preclinical models. Despite these current limitations, preclinical studies have highlighted some mechanisms susceptible to contribute to the emergence and sustained expression of these compulsive behaviours. At the cellular level, pramipexole was shown to significantly alter the firing of monoaminergic neurons (Chernoloz et al. 2009). Two days following pramipexole administration (1 mg/kg/day), the spontaneous firing of dopaminergic neurons in the ventral tegmental area and noradrenergic neurons in the locus coeruleus was significantly reduced. By 14 days of 1 mg/kg/day pramipexole administration, spontaneous firing rate of dopaminergic and noradrenergic neurons recovered, but their pattern of discharge remained altered, with an attenuated burst activity, as shown with a decreased number of bursts per minute for noradrenergic neurons and a decrease percentage of spikes occurring in bursts for dopaminergic neurons. Conversely, spontaneous firing of serotonergic neurons in the dorsal raphe was not altered after 2 days of pramipexole administration but significantly increased (together with the number of bursts per minutes) when assessed after 14 days of pramipexole (Chernoloz et al. 2009). Importantly, pramipexole also induced a desensitization of D2 autoreceptor function. Altogether, these results indicate that chronic administration of pramipexole alters the activity of monoaminergic neurons. Although normalization of spontaneous firing of dopaminergic neurons and increased firing of serotonergic neurons may underlie the positive effects of pramipexole on anhedonia and depression, the altered burst activity of DA neurons suggests that the encoding of saliency may be impaired. Subsequent work demonstrated that chronic pramipexole administration (1 mg/day for 14 days) led to increased dopamine neurotransmission in the prefrontal cortex and serotonergic neurotransmission in the hippocampus (Chernoloz et al. 2012). Because of its 100-fold selectivity for D3 over D2 receptors (Millan et al. 2002), the behavioural effects of pramipexole are intuitively attributed to its action at the D3 receptor. However, if low doses of pramipexole (0.001–0.1 mg/kg) that occupy D3 but not D2 receptors have been shown reduce responding for a conditioned reinforcer, this effect of pramipexole was not blocked by the D3 antagonist SB-277011-A (McCormick et al. 2015). Furthermore, similar experiments performed in D3 knock-out mice demonstrated an effect of pramipexole on instrumental responding at doses that produce no D2 receptor occupancy (McCormick et al. 2015). These data indicating a D3-independent action call for caution regarding the receptors involved in the behavioural effects of pramipexole and suggest that receptors other than D3 could be involved. Possible candidates include the dopamine D4 and adrenergic alpha-2 receptors (McCormick et al. 2015; Millan et al. 2002).

At the molecular level, chronic administration of pramipexole was demonstrated to induce a long-lasting expression of the truncated splice variant of FosB (delta-FosB) in the striatum and nuclear accumbens both in normal and lesioned rats (Engeln et al. 2013a). Furthermore, dopamine-lesioned rats displayed increased counts of delta-FosB in the dorsomedial and ventromedial striatum compared with controls, indicating that dopamine depletion can trigger an increased post-synaptic response, as previously shown in the context of l-dopa induced dyskinesia (Andersson et al. 1999; Cenci et al. 1999; Engeln et al. 2016b). Such enhanced post-synaptic response was also recently confirmed in another study demonstrating enhanced Fos-like immunoreactivity in the orbitofrontal cortex, and striatum of dopamine-lesioned rats chronically treated with pramipexole (Dardou et al. 2017).

Interestingly, delta-FosB is a transcriptional regulator not only involved in the establishment and maintenance of maladaptive plasticity underlying psychostimulant addiction (Robison and Nestler 2011), but also in motor side-effects of dopamine replacement therapy in PD (Bastide et al. 2015). In addition to being increased by dopaminergic denervation (Doucet et al. 1996), natural rewards (e.g. food, sex) also triggers a sustained and long-lasting expression of delta-FosB in brain regions likely involved in ICD such as the nucleus accumbens, striatum, and prefrontal cortex (Pitchers et al. 2010). Behavioural traits relevant to ICD such as impulsivity are also associated with high levels of delta-FosB (Velazquez-Sanchez et al. 2014). Overall, these data suggest that additive effects of the treatment, lesion, reward, and individual traits could lead to an enhanced expression of delta-FosB that could contribute to the emergence of maladaptive plasticity underlying ICD in vulnerable PD patients. Although the molecular neuroadaptations induced by pramipexole remains largely unknown, one study identified an altered pattern of mGluR5 receptor mRNA expression following chronic pramipexole, with increased expression in the dorsal striatum and decreased expression in the nucleus accumbens, suggesting that glutamatergic dysfunction might be involved (Loiodice et al. 2017).

Conclusion

ICDs are frequent side effects of DRT and a significant concern in PD, with devastating consequences on the patients and caregivers’ lives. Management of ICDs is a real challenge in the absence of therapeutic options. On-going efforts to determine specific profiles (personality, prior history, DAT imaging) together with the development and validation of sensitive and specific evaluation tools will help clinicians to better identify patients with an increased risk to develop ICD and to monitor these adverse effects over the course of the treatment. Imaging studies have demonstrated altered dopaminergic function, reduced functional connectivity within executive networks, and cortical morphometric changes. Experimental research on the pathophysiology of ICD in PD is still an emerging field with a limited number of preclinical studies conducted so far. Yet, these works demonstrated that nigrostriatal degeneration could affect some behavioural processes potentially relevant to the pathophysiology of ICD, including several features of impulsivity (impulsive choice, waiting, and motor impulsivity) but without affecting behaviour on probabilistic/gambling-like schedules of reinforcement. Consistent with clinical findings pointing to a strong contribution of D3/D2 agonists in the emergence of ICDs, experimental studies reliably showed in lesioned or non-lesioned rodents that D2/D3 agonists induce impairments in several behavioural processes likely relevant to ICD occurring in PD patients. These notably include behaviours related to compulsive gambling such as risk-taking behaviour, preference for uncertainty, perseverative responding, and sustained drive to engage in gambling-like behaviour.

Whether interactions between dopamine denervation and DRT significantly contribute to the pathogenesis of ICD remains poorly understood so far, although features unique to PD have been identified in patients with ICD (lower DAT in the ventral striatum, increased motor fluctuations, and levodopa-induced dyskinesia). The small number of preclinical studies on PD models, different doses/duration of DRT, and different types of dopaminergic denervation used (i.e. acute intrastriatal 6-OHDA vs progressive alpha-synuclein induced neurodegeneration) does not provide conclusive evidence for or against such interactions. Results from clinical and pre-clinical studies also suggest a potential contribution of pre-existing vulnerability. Further work on this topic is needed to find relevant vulnerability endophenotypes and determine their real influence on the pathophysiology of ICD in PD. This is a critical point to understand why only a fraction of PD patients develop ICD and allow for an early identification of the most vulnerable individuals.