Genetic and personality interactions shape impulse control disorders in Parkinson's disease

Abstract

Background

In Parkinson's disease (PD), predictive factors of Impulse Control Disorders (ICDs) remain insufficiently understood. Personality and polygenic risk scores (PRS) have both been independently associated with ICDs.

Objectives

Our objective was to study the interaction effects between personality dimensions and PRS on the odds of developing ICDs.

Methods

This case-control study included PD patients with and without ICDs (n = 250 patients). Personality was assessed using the Temperament and Character Inventory-Revised, and PRS were computed from genotyping data across 18 traits linked to psychiatric/behavioral phenotypes. Logistic regression models were used to evaluate main and interaction effects between personality dimensions and PRS on ICD odds.

Results

Higher scores in Self-Directedness, attention deficit hyperactivity disorder (ADHD)-PRS, and major depressive disorder (MDD)-PRS were associated with increased odds of ICD (odds ratios (ORs) between 1.31 and 1.54). Several personality*PRS interactions also significantly modulated the odds of developing ICDs. Notably, Novelty Seeking*MDD PRS and Harm Avoidance*MDD PRS interactions amplified the odds of ICD (ORs = 1.34 and 1.30, respectively), while the Harm Avoidance*ADHD PRS interaction attenuated it (OR = 0.74). Self-Directedness interactions showed both odds amplification (with PD-PRS [OR = 1.34] and MDD-PRS [OR = 1.43]) and slight attenuation (with ADHD-PRS [OR = 0.73]).

Conclusions

This study provides novel evidence that the odds of having ICD are modulated by cumulative and interactive effects of personality and genes in PD. Self-Directedness, ADHD and MDD PRS were individually associated with ICDs, and their combinations yielded complex, non-additive effects—either amplifying or attenuating odds. These findings highlight the importance of integrating both psychological and genetic profiles in personalized risk assessment for ICDs.

Plain language summary

Interactions between genetics and personality can be associated with a higher risk of Impulse Control Disorders development in Parkinson's Disease

In Parkinson's disease (PD), treatment by dopamine agonists can lead to the development of adverse effects such as Impulse Control Disorders (ICDs), whom causes are yet poorly understood. It is so important to look for predictive factors of ICDs to limit the use of dopamine agonists in patients at risk. Currently, personality and genetics have both been independently associated with ICDs. Hence, our research team studied the interactions between personality and genetics on the risk of developing ICDs, by comparing two groups of patients with PD with and without ICDs. Personality dimensions were assessed using a self-questionnaire, the Temperament and Character Inventory-Revised (TCI-R), and polygenic risk scores (PRS) linked to psychiatric or behavioral phenotypes were computed from genotyping data. Analyses (logistic regression models) were done to evaluate the interaction effects between personality dimensions (TCI-R scores) and PRS on ICD risk. It was showed that higher scores in Self-Directedness personality, in attention deficit hyperactivity disorder (ADHD) PRS, and in major depressive disorder (MDD) PRS were associated with increased risk of developing ICD. Moreover, several interactions between personality and PRS significantly modulated this risk. Notably, the interaction between higher scores in Novelty Seeking personality and MDD PRS, and the interaction between Harm Avoidance personality and MDD PRS amplified the risk of ICD; while the interaction between Harm Avoidance and ADHD PRS attenuated this risk. Self-Directedness personality also showed both risk amplification and slight attenuation depending on genetical interactions. Therefore, it is now important to consider both psychological and genetic profiles in personalized risk assessment for ICDs in PD. Indeed, the personality of Self-Directedness, and the PRS of ADHD and MDD were associated with a higher ICD risk, and their combinations could either amplify or attenuate this risk.

Introduction

In the context of personalized medicine, increasing number of studies are being conducted on predictive factors to avoid the occurrence of adverse effects following dopaminergic treatments in Parkinson's disease (PD), such as Impulse Control Disorders (ICDs). ICDs are a common adverse effect of dopamine agonists, for which certain risk factors are known ¹ such as younger age of onset, male gender, longer disease duration, greater disease severity, presence of depression or anxiety, and personal or familial history of substance addiction. Despite those, we need other reliable predictive factors to identify high-risk patients.

It has been reported that certain Single Nucleotide Polymorphisms (SNPs) in various genes involved in neurotransmission were implicated in the emergence, progression, or clinical phenotype of PD, as well as in the response to different treatments. Notably, associations between ICDs and specific genomic variants have been identified in SNPs within the opioid receptor gene OPRM1 in PD,^2,3 within the dopamine transporter gene DAT1, ² and within the following genes (PRKAG2, MEFV, and PRKCE). ⁴

Moreover, studies have shown that specific personality dimensions derived from the Temperament and Character Inventory (TCI) or its revised version (TCI-R) are correlated with behavioral symptoms in PD, such as anxio-depressive states and impulsivity.^5,6 Indeed, impulsivity (UPPS-P scale) has been correlated with higher Novelty Seeking scores in PD patients with motor fluctuations, ⁵ as well as in de novo (never-medicated) PD patients. ⁶ Impulsivity related to enthusiasm or contrariety (“positive urgency” and “urgency” subscales of the UPPS-P) has also been associated with lower Self-Directedness scores in PD patients. ⁵ Novelty Seeking was also significantly higher in PD patients with gambling disorders, ⁵ which are strongly associated with impulsivity. ⁷

Hence, it appears that both genetic background and personality dimensions may influence the emergence of ICDs in PD following the introduction of dopamine agonists.

Furthermore, the bio-psycho-social model behind the TCI suggests that certain personality dimensions -the temperaments, namely Novelty Seeking, Harm Avoidance, Reward Dependence and Persistence- are associated with specific neurotransmitters (dopamine, serotonin, noradrenalin, and glutamate, respectively),^8,9 and are genetically influenced. Associations have been found between Novelty Seeking scores and a SNP within the dopaminergic receptor D4 gene in healthy subjects, ¹⁰ Harm Avoidance scores and a SNP within a serotoninergic transporter gene in both healthy and PD subjects, ¹¹ and Reward Dependence scores and a SNP within a noradrenergic transporter gene in healthy subjects. ¹² In the TCI-R, additional associations have also been identified: Novelty Seeking and Harm Avoidance scores were linked to the variable number tandem repeat polymorphism of the serotoninergic transporter, Reward Dependence scores were associated with a SNP in the brain-derived neurotrophic factor gene, and Persistence scores related to a SNP in the catechol-O-methyltransferase gene. Nonetheless, the genetic bases underlying these personality dimensions remain poorly studied.

Since both personality and ICDs are related to neurotransmitter systems and genetic background, we suggest that ICDs may be influenced by interactions between genetic predisposition and personality dimensions. As such, personality dimensions combined with polygenic risk scores (PRS) could represent a valuable factor to consider in clinical practice to limit the use of dopamine agonist therapy in PD patients at higher odds of developing ICDs. In this study, we evaluated the relationship between ICDs, specific PRS, and personality dimensions assessed using the TCI-R in a French European population of PD patients treated by dopamine agonists with and without ICDs. Our main objective was to study the interaction effects between TCI-R personality dimensions and PRS on the increased odds of developing ICDs.

Materials and methods

Study design

This study is an ancillary analysis based on data collected from the BADGE-PD study, a multicentric case-control candidate genetic association study in PD patients with or without ICDs (clinicalTrial.gov n°NCT02319395). This study was sponsored by the AP-HP, conducted between 2011 and 2015 in twelve expert centers of the French clinical research network for PD (NS-PARK/FCRINnetwork), and approved by the CPP Paris VI Ethics Committee. All patients provided written informed consent. The study was conducted in accordance with Good Clinical Practice, local regulations, and data collection was compliant with GDPR guidelines.

Population

PD patients with or without ICDs were included. The inclusion and exclusion criteria and data collection were defined in a previous publication. ² Mainly, patients with PD according to United Kingdom Parkinson's Disease Society Brain Brank criteria, age>30, with ICDs and related disorders (cases) or without ICDs and related disorders (controls) were included. ICDs or related behaviors of cases were related to the initiation of dopaminergic therapy (dopamine agonists and/or levodopa). Patients with ICDs or related behaviors prior to the initiation of the dopaminergic therapy were excluded. Exclusion criteria for both groups were atypical parkinsonism and treatment with neuroleptics, ICDs and related disorders before dopamine replacement therapy initiation. Controls were matched for sex, age at inclusion and at PD diagnosis within a 5-year interval.

Data collection

Patients were primarily classified using the Ardouin Scale of Behavior in PD. For cases, ICDs were defined as the presence of at least one item score≥3, or at least two items scored≥2, on the Ardouin scale domains associated with ICDs or related behaviors: pathological gambling, shopping, eating, hypersexuality, hypercreativity, hobbyism, risk-taking behaviors, and punding. Controls (PD patients without ICDs) had no current or past history of ICDs, defined as all hyperdopaminergic items<2 on the Ardouin scale and no more than two items scoring 1. Subjects with intermediate scores were not included in the analysis. In addition, controls must have had PD for at least 5 years and a daily dose of dopamine agonists>300 mg (Levodopa equivalent daily dose (LEDD)) for at least 12 months, to minimize biases related to limited drug exposure.

LEDD was calculated for each patient following recommendations from Tomlinson et al. (2010). ¹³

Personality was assessed using the French version of the TCI-R, a validated 240-item self-questionnaire. ¹⁴ It evaluates seven personality dimensions, consisting of four temperaments - Novelty Seeking (NS), Harm Avoidance (HA), Reward Dependence (RD), and Persistence (P) -, and three characters - Self-Directedness (SD), Cooperativeness (C), and Self-Transcendence (ST). ⁸ Each item is rated on a 5-point Likert scale, and scores for each dimension are calculated as the sum of the relevant items. The questionnaire also includes five validity control questions to verify the attentiveness of respondents.

DNA samples were genotyped using the NeuroX array containing 487,687 sites. ¹⁵ After quality control, genotypes were imputed using the TopMed server ¹⁶ with an imputation R² threshold of 0.3. PRS were calculated from imputed dosages using PLINK, applying the published effect alleles and weights provided in the corresponding PGS Catalog scoring files (https://www.pgscatalog.org/). Eighteen PRS of interest were selected for analysis: these were related to dependency behaviors, addictions, impulsivity, or psychological phenotypes. Supplementary Table S1 presents the 18 selected PRS, their corresponding scoring methods and their respective coverage.

TCI-R: missing values treatment

In cases of missing responses in the TCI-R, data were handled as follows:

1. Patients presenting more than 20% of missing responses in the whole TCI-R (i.e., more than 48 unanswered questions) were excluded from the analyses.

2. Patients presenting more than 20% of missing responses within a single personality dimension were also excluded from the analyses: that is more than 7 missing responses for NS, 6 for HA, 6 for RD, 7 for P, 8 for SD, 7 for C, and 5 for ST.

3. For patients with fewer than 20% of missing responses in a given dimension, unanswered items were imputed using the mean score of the individual's answered items within that same dimension.

4. The five validity control questions were also reviewed. Patients who provided at least one incorrect response were excluded from the analyses.

This imputation approach is based on the method previously defined and validated for the standard TCI questionnaire in earlier studies on PD.^5,17–19

Statistical analyses

Descriptive statistics were computed separately for the two groups (cases: PD patients with ICDs, and controls: PD patients without ICDs), using means and standard deviations for quantitative variables, and frequencies and percentages for qualitative variables.

Group comparisons were conducted for each variable using appropriate statistical tests: two-sample t-tests or Mann–Whitney U tests for continuous variables (depending on data distribution evaluated through Shapiro tests), and chi-squared tests for categorical variables.

As a preliminary step to validate the assumptions of the main models, pairwise Spearman correlations were calculated to explore relationships: among the 7 TCI-R personality dimensions, among the 18 PRS, and between all 126 possible personality-PRS pairs (7 × 18 combinations).

To address the main objective -investigating the joint contribution of TCI-R and PRS to ICD status-, logistic regression analyses were performed with diagnostic group (ICD vs. control) as a binary outcome. Each model included one TCI-R personality dimension, one PRS, and their interaction term. This resulted in 126 separate logistic models (7 personality dimensions × 18 PRS). All predictor variables were standardized (mean = 0, standard deviation = 1) prior to analysis to facilitate comparability and interpretability. For each model, we reported the log-odds coefficients, standard errors, and p-values for: the personality dimension (main effect adjusted for the corresponding PRS), the PRS (main effect adjusted for the corresponding personality dimension), and their interaction term. A positive interaction coefficient (log-odds) indicates a greater likelihood of belonging to the ICD group. Corresponding odds ratios (ORs) and 95% confidence intervals (95%CI) were obtained by exponentiating the coefficients. We also reported the combined OR for the full model expression (i.e., OR = exp(β₁+β₂+β₃)), reflecting the total effect of both main effects and their interaction. The combined OR corresponds to the risk for an individual with a simultaneous increase of 1 standard deviation in both PRS and personality dimension score. Moreover, some supplementary analyses were performed as sensitivity analyses: the same logistic regression models were done using covariates (age, sex, disease duration and total LEDD) to adjust our analyses. Going further, correlations were done between significant variables and LEDD at the occurrence of the first ICD in PD patients with ICDs.

An identical modeling strategy was applied to assess interactions within each domain -i.e., personality dimension*personality dimension, and PRS*PRS- to evaluate the independence of each TCI-R dimension and each PRS.

To investigate interaction effects between personality and genetic predisposition, we also conducted a cluster-based analysis using standardized TCI-R and PRS variables. Hierarchical clustering was performed separately on the personality dimensions and the PRS using Euclidean distance and Ward's minimum variance linkage method (ward.D2). The optimal number of clusters in each domain was determined using the Bayesian Information Criterion, which indicated a two-cluster solution for both personality and PRS domains. Participants were assigned to one of two personality clusters and one of two PRS clusters based on the dendrogram splits. To assess whether the combination of personality and PRS clusters predicted ICD group membership, we fitted a binary logistic regression model including an interaction term: group∼pers_cluster*prs_cluster, where pers_cluster and prs_cluster are categorical variables denoting cluster membership. To avoid the bias to capture an effect associated to genetic ethnicity, comparisons of ancestry principal component analyses (aPCA) were done between both PRS clusters.

Analyses were conducted using RStudio4.3. A significance threshold of p < 0.05 was applied. Given that this is an exploratory study, we chose not to perform multiple comparison corrections in order to detect clues that will need to be confirmed in hypothesis-driven independent studies.

Results

Population description

Figure 1 presents the flowchart of the 250 PD patients included: n = 139 ICD patients and n = 111 control patients (without ICDs). A complete description of the population is provided in Table 1.

Figure 1.

Flowchart of PD patients included in the analysis (n = 250). 323 PD patients initially screened, and n = 250 included. Among the 27 patients excluded for invalid responses on the quality control items, 70% had ICDs. Of the 94 patients with missing data imputed using the mean substitution method, 57% had ICDs.

Table 1.

Description of the population and comparisons between groups (n = 250).

Variables		ICD group (n = 139)			PD controls group (n = 111)			p.values (ICD vs control)
		mean	±	sd	mean	±	sd
Sex (women; men)		55;84			35;76			0.24
Age (years)		59.5	±	8.2	62.7	±	7.8	0.002*
Disease duration (years)		9.8	±	5.3	10.4	±	5.7	0.67
Total LEDD (mg/day)		930.8	±	582.9	1043.1	±	446.0	0.01*
LEDD levodopa (mg/day)		671.1	±	579.6	595.3	±	406.4	0.64
LEDD dopamine agonists (mg/day)		215.5	±	195.1	389.7	±	200.0	3 × 10−16*
LEDD others (mg/day)		44.2	±	49.7	58.1	±	51.9	0.03*
MMSE		28.1	±	1.9	28.3	±	1.9	0.29
MDS-UPDRS part I		2.3	±	2.0	1.1	±	1.4	4 × 10−07*
MDS-UPDRS part II		10.9	±	4.8	9.7	±	4.4	0.05
MDS-UPDRS part III		15.8	±	8.9	16.4	±	9.3	0.67
MDS-UPDRS part IV		4.8	±	3.3	3.8	±	3.2	0.02*
Temperament and Character Inventory (TCI)
Novelty Seeking (NS)		102.9	±	12.1	102.6	±	13.5	0.69
NS1		30.6	±	4.6	30.5	±	5.0	0.92
NS2		28.8	±	4.5	30.0	±	4.5	0.04*
NS3		23.7	±	4.0	22.7	±	4.3	0.07
NS4		19.7	±	3.9	19.4	±	3.9	0.49
Harm Avoidance (HA)		93.4	±	12.1	92.8	±	12.1	0.56
HA1		30.8	±	4.7	31.1	±	4.7	0.64
HA2		19.6	±	4.6	18.5	±	4.1	0.04*
HA3		18.6	±	3.3	19.4	±	3.4	0.12
HA4		24.4	±	4.2	23.8	±	4.0	0.22
Reward Dependence (RD)		96.4	±	10.2	98.0	±	10.3	0.11
RD1		28.8	±	4.0	28.2	±	3.6	0.18
RD2		32.8	±	4.7	33.1	±	4.5	0.64
RD3		18.2	±	3.7	19.2	±	3.0	0.01*
RD4		16.5	±	4.1	17.5	±	4.0	0.08
Persistence (P)		111.5	±	16.1	111.2	±	16.8	0.88
P1		31.3	±	4.6	31.5	±	4.2	0.70
P2		26.2	±	4.4	26.5	±	4.6	0.62
P3		28.9	±	6.3	29.0	±	6.7	0.92
P4		25.0	±	4.4	24.1	±	4.6	0.12
Self-Directedness (SD)		112.9	±	16.8	107.3	±	15.6	0.007*
SD1		20.7	±	5.0	19.1	±	4.6	0.01*
SD2		18.3	±	3.1	18.2	±	3.5	0.62
SD3		13.2	±	3.1	12.3	±	2.6	0.02*
SD4		25.7	±	5.5	24.9	±	5.3	0.28
SD5		35.1	±	5.5	32.8	±	5.2	0.0009*
Cooperativeness (C)		107.8	±	9.9	108.0	±	11.3	0.79
C1		24.9	±	3.2	25.1	±	3.3	0.51
C2		16.3	±	3.2	15.8	±	3.0	0.19
C3		27.7	±	3.5	28.4	±	3.4	0.10
C4		15.6	±	3.1	15.4	±	3.2	0.30
C5		23.3	±	3.8	23.3	±	4.1	0.95
Self-Transcendence (ST)		69.2	±	14.1	68.4	±	13.0	0.98
ST1		28.5	±	7.3	27.1	±	7.0	0.13
ST2		20.8	±	5.5	21.7	±	5.0	0.20
ST3		19.8	±	4.1	19.6	±	3.9	0.87
Polygenic risk score (PRS)
PGS000903	PD	−3.07	±	1.54	−3.12	±	1.71	0.88
PGS001901	Alcohol dependance	−0.17	±	0.11	−0.18	±	0.10	0.85
PGS002738	Alcohol dependance	−0.01	±	0.09	−0.02	±	0.10	0.36
PGS002115	Alcohol dependance	0.00	±	0.01	0.00	±	0.01	0.66
PGS002037	Nicotine dependance	0.02	±	0.01	0.02	±	0.01	0.89
PGS003057	Nicotine dependance	0.02	±	0.12	0.03	±	0.10	0.74
PGS002098	Eating disorder	0.02	±	0.02	0.02	±	0.02	0.46
PGS001118	Drug use measurement	0.16	±	0.10	0.15	±	0.09	0.59
PGS002124	Addictive behavior	−0.01	±	0.01	−0.01	±	0.01	0.95
PGS002746	ADHD	147.93	±	141.07	131.53	±	159.73	0.40
PGS003753	ADHD	2.34	±	3.88	0.75	±	4.28	0.003*
PGS004230	Impulsivity	−0.46	±	0.11	−0.47	±	0.11	0.51
PGS000205	Risk-taking behavior	0.04	±	0.15	0.03	±	0.16	0.84
PGS002222	Self-injurious behavior	0.01	±	0.03	0.00	±	0.02	0.06
PGS004451	Anxiety disorder	−0.06	±	0.07	−0.05	±	0.07	0.35
PGS000907	Major Depressive Disorder	0.36	±	0.31	0.27	±	0.33	0.048*
PGS001016	Sensitivity / hurt feeling	0.07	±	0.23	0.05	±	0.24	0.54
PGS002342	Neuroticism	0.21	±	0.52	0.13	±	0.56	0.25

ECMP = échelle comportementale de la maladie de Parkinson; LEDD = levodopa equivalent daily dosage (Tomlinson et al. (2010)); MMSE = Mini Mental State Examination; MDS-UPDRS = Movement Disorders Society-Unified Parkinson's Disease Rating Scale; PGS = polygenic score; MDD = Major Depressive Disorder; two-sample tests according to data distribution; *p.value<0.05 in bold.

The two groups of PD patients differed significantly in age, total LEDD, dopamine agonist LEDD, and MDS-UPDRS parts I and IV scores (Table 1). PD patients with ICDs were younger, had lower total LEDD and dopamine agonist LEDD, and higher scores on MDS-UPDRS parts I and IV compared to PD controls.

Regarding the main TCI-R personality dimensions, only Self-Directedness scores differed significantly between groups, with higher scores observed in patients with ICDs compared to controls (Table 1). Some significant differences were also noted in certain TCI-R subdimensions, which will not be discussed here. Among the PRS examined, only two showed significant group differences: the PRS for major depressive disorder (MDD) (PGS000907) and for attention deficit hyperactivity disorder (ADHD) (PGS003753) were both higher in the ICD group (p = 0.048, 95%CI = [0.0008; 0.16], and p = 0.003, 95%CI = [0.56; 2.6], respectively) (Table 1).

Correlation analyses between TCI-R personality dimensions and PRS variables (Supplementary Figure S2) revealed modest intercorrelations among personality dimensions, varying from 0.45 to 0.68, indicating moderate internal consistency. In contrast, PRS correlations were more heterogeneous, ranging from 0 to 0.5. These findings support treating personality and PRS as largely independent predictor domains.

For the main analyses, we fitted a series of logistic regression models to examine interactions between each pair of personality dimensions and PRS (Figure 2, Table 2, and Supplementary Table S3). Several significant interactions emerged, some with robust effect sizes.

Figure 2.

Heat map of odds ratio from logistic regression models examining the joint effect of personality dimensions and PRS on group status. Each cell represents ORs from a logistic regression model (personality dimensions*PRS to predict the group status). Color reflects the magnitude and direction of the interaction effect (blue = negative effect; red = positive effect = increased odds of ICD). Asterisks and bold fonts indicate significant OR (p.value<0.05).

Table 2.

Odd ratios and respective 95% C.I. for the main and interaction effects for pairs of personality-polygenic variables: significant results.

Focusing on the main effects of personality dimensions or PRS alone, we found that the Self-Directedness personality dimension and the PRS for ADHD (PGS003753) and MDD (PGS000907) had significant effects independently. Higher scores in Self-Directedness or in the ADHD or MDD PRS were associated with an increased odds of having ICDs, with ORs ranging from 1.31 to 1.54. This corresponds to a 31% to 54%-fold increased odds of developing ICDs for each one standard deviation increase in the respective variables.

Regarding the combined and interaction effects, we observed the following (Supplementary Figure S4):

• -The interaction between Novelty Seeking (NS) and the MDD PRS (PGS000907) significantly increased the odds of developing ICDs (interaction OR = 1.34), with a global OR of 1.8 — indicating 80% increased odds of having ICDs for each one standard deviation increase in both NS and PGS000907.

• -The interaction between Harm Avoidance (HA) and the ADHD PRS (PGS003753) significantly attenuated the odds of developing ICDs (interaction OR = 0.74), effectively “neutralizing” the odds of having ICD (deriving from ADHD PRS), as the global OR was not significant.

• -The interaction between Harm Avoidance (HA) and the MDD PRS (PGS000907) significantly increased the odds of developing ICDs (interaction OR = 1.30), with a global OR of 1.82 — indicating 82% increased odds of having ICD for each one standard deviation increase in both variables.

• -The interaction between Self-Directedness (SD) and the PD PRS (PGS000903) significantly increased ICD odds (interaction OR = 1.34), with a global OR of 2.07 — indicating 107% increased odds of ICD for each one standard deviation increase in both variables.

• -The interaction between Self-Directedness (SD) and the ADHD PRS (PGS003753) significantly attenuated the odds of ICDs (interaction OR = 0.73) without fully canceling it: the global OR was 1.69, indicating 69% increased odds of developing ICD for each one standard deviation increase in both variables.

• -The interaction between Self-Directedness (SD) and the MDD PRS (PGS000907) significantly increased ICD odds (interaction OR = 1.43), with a global OR of 2.78 — corresponding to 178% increased odds of developing ICD for each one standard deviation increase in both variables.

Adjustment of these analyses with the following covariates (age, sex, disease duration and total LEDD) showed similar results that are presented in Supplementary Table S5.

In PD patients with ICDs, correlations were done between LEDD at the occurrence of the first ICD and the previous significant variables (NS, HA, SD, PGS000907, PGS003753 and PGS000903). No correlations were significant (Supplementary Figure S6).

Concerning the logistic regression models testing interactions within the TCI-R dimensions, no significant interaction effects were observed between pairs of personality dimensions (Supplementary Figure S7). Similarly, we examined all PRS*PRS interaction terms. While a small number of significant interactions were identified, given the large number of comparisons, some are likely false positives statistically (Supplementary Figure S8). Nonetheless, several of these interactions showed clinically meaningful effects, with predicted probability shifts of 20–50%, highlighting the potential for non-additive polygenic mechanisms in complex phenotypes such as PD.

Hierarchical clustering separated the personality dimensions and PRS into two latent groups each (Supplementary Table S9 for description of variables within each cluster). Regarding the personality clusters, patients in cluster 2 exhibited higher scores across all TCI-R dimensions, with statistically significant differences in HA, RD, P, SD, and C. For the PRS clusters, patients in cluster 2 had higher scores in most studied PRS: “Alcohol dependence” (PGS001901, PGS002738), “Nicotine dependence” (PGS002037, PGS003057), “Eating disorder” (PGS002098), “Addictive behavior” (PGS002124), “ADHD” (PGS002746, PGS003753), “Impulsivity” (PGS004230), “Risk-taking behavior” (PGS000205), “Self-injurious disorder” (PGS002222), “Anxiety disorder” (PGS004451), “MDD” (PGS000907), “Sensitivity” (PGS001016), and “Neuroticism” (PGS002342). The cluster-based interaction analysis revealed a statistically significant interaction term (β = −1.34, SE = 0.66, p = 0.04), indicating that the effect of personality cluster on increased odds of developing ICD was moderated by PRS cluster membership (Figure 3). Specifically, PD patients with high scores in the PRS cluster 2 exhibited a higher probability of ICDs (i.e., above average regardless of the personality cluster). Conversely, for those with low scores in these PRS (PRS cluster 1), personality dimension scores took over, playing a much more important role in predicting ICD increased odds (much higher for personality cluster 2, and well below average for personality cluster 1). Comparisons of aPCA between both PRS clusters were not significant, confirming these PRS did not measure an ethnicity effect (Supplementary Table S10).

Figure 3.

Predicted probability of ICD group membership by patient clusters-based combination. The barplot displays predicted probabilities of being in the ICD group for each combination of personality*PRS cluster, derived from the fitted model (Pers1xPRS1 = 0.42, 95%CI = [0.30–0.54]; Pers1xPRS2 = 0.60, 95%CI = [0.51–0.68]; Pers2xPRS1 = 0.71, 95%CI = [0.49–0.87]; Pers2xPRS2 = 0.58, 95%CI = [0.42–0.72]). Pers1 = personality cluster1; Pers2 = personality cluster2; PRS1 = PRS cluster1; PRS2 = PRS cluster2.

Discussion

This study reveals that the odds of developing ICD in PD patients are modulated by cumulative and interactive effects of personality and genes evaluated through PRS. Indeed, this study shows that modest effect of some ICD risk factors (such as the Self-Directedness dimension, and the “ADHD” and “MDD” PRS) could increase in combination with other factors. For instance, the interaction between Self-Directedness personality dimension and the MDD PRS increased the odds of having ICD of 178%. Hence, combinations of certain personality dimensions and PRS domains either amplified or, conversely, attenuated these odds in a non-linear and significant manner.

First, some demographic differences were observed between PD patients with and without ICDs. As expected, patients with ICDs were younger ¹ and had lower total LEDD, mainly due to lower doses of dopamine agonists, likely reflecting a reduction in dopamine agonists due to the occurrence of this adverse effect.

Concerning personality, on the one hand, PD patients with ICDs scored higher on the Self-Directedness dimension. This result was quite surprising as previous studies had reported an inverse correlation between Self-Directedness and impulsivity traits in PD, ⁵ and in gambling disorders subjects. ⁷ This personality dimension could therefore represent a specific consequence of ICDs, rather than a predisposition of “basic” impulsivity, since the characters evaluated by the TCI-R can change over time. ⁸ It is conceivable that patients with ICDs would develop greater Self-Directedness (reflected in higher scores in responsibility, resourcefulness, and congruent second nature sub-dimensions) during the course of the disease, as an adaptive response to manage the behavioral challenges associated with ICDs, ¹⁹ since personality was here evaluated in patients already presenting ICDs. On the other hand, no difference was observed in Novelty Seeking scores between PD patients with or without ICDs, although Novelty Seeking temperament is related to impulsivity in the literature.^5,7,20–22 This could be explained by the fact that Novelty Seeking temperament more specifically measures an impulsive trait rather than impulsive behavior induced by medication. Furthermore, PD can alter personality dimensions in various ways, ¹⁹ and the literature has shown that PD patients generally have lower Novelty Seeking scores, ²³ likely related to the depletion of the dopaminergic system.

Regarding PRS, two genetic profiles were significantly associated with ICDs: higher scores in the PRS for “ADHD” (PGS003753), linked to impulsivity, and in the PRS for “MDD” (PGS000907), linked to depressive traits. Both conditions align well with the ICD phenotype and are known to involve dysregulation in dopaminergic pathways, and frequently co-occur in PD, supporting the hypothesis of a shared neurobiological underpinning. ²⁴ In addition, a recent study have suggested a potential genetic comorbidity between MDD and PD. ²⁵ ADHD has also been related to PD in studies showing that patients with ADHD were at higher risk of PD,^26,27 but no genetic studies have yet been conducted to support this hypothesis.

Moreover, examining genes and personality simultaneously suggested that the cumulative and interactive effects of certain personality dimensions and PRS may be synergistic or antagonistic. The most notable new finding in this study was the interaction between the Self-Directedness personality and the MDD PRS that led to 178% increased odds of having ICD. In fact, we observed that Self-Directedness, which independently increased ICDs odds, had variable interaction effects depending on the PRS. When combined with either the MDD PRS or the PD PRS highly, its interaction amplified the ICD odds. Conversely, its interaction with the ADHD PRS slightly attenuated the overall odds of having ICDs, though the global effect remained significant. The same applied to other personality dimensions. Although the Novelty Seeking temperament alone did not significantly influence ICDs odds, its interaction with the MDD PRS significantly increased the odds. In contrast, Harm Avoidance, which also had no main effect on its own, demonstrated divergent effects depending on the interacting PRS: when combined with the ADHD PRS, it appeared to neutralize the overall ICDs odds, whereas its interaction with the MDD PRS significantly increased the odds of having ICDs. These findings underscore the importance of considering both personality dimensions and genetic predispositions when evaluating an individual's risk of developing ICDs in the context of PD.

Thus, our clustering analysis suggests a potential decision-making framework: PRS profiles may serve as a primary stratification tool, with personality dimensions considered as secondary modulators depending on the patient's genetic risk profile. Indeed, PD patients who have high PRS scores in phenotypes associated with addictive behaviors, dependencies, impulsivity, or psychological issues are at higher odds of experiencing ICDs, regardless of their personality profile; while patients with low PRS scores would be at higher odds of experiencing ICDs if they score high on the following personality dimensions: Harm Avoidance, Reward Dependence, Persistence, Self-Directedness, or Cooperativeness. As observed in other diseases, decision models can be used to predict which patients are most likely to achieve durable benefit from treatment using different factors such as clinicopathological and genomic markers, ²⁸ or to assess the risk of progression of disease. ²⁹ It would therefore be interesting to develop an algorithm based on these PRS and personality profiles to predict the odds of developing ICDs. Genetic screening followed by personality assessment could be offered to PD patients before the introduction of dopamine agonists to determine their level of ICDs odds. Other already known factors (early age of onset, male gender, longer duration of the disease, etc.) should be added to this algorithm. ¹ It would also be important to consider the dose of dopamine agonists, which could modulate the effect of each factor (genetic, personality etc.), even if in this study there was no association between LEDD and PRS/personality dimensions associated with odds of having ICDs.

The main strength of our work can be seen in the collection of both genetic and psychological data which allowed us to examine a plurality of ICD risk factors simultaneously. Hence, the use of interaction studies may reduce the potential bias that can result from unmeasured risk factors and from individual risk factors. However, we are aware that not all ICD risk factors have been identified yet, and that other biological and/or clinical biomarkers need to be investigated in the future and be further included in predicting the odds of ICD occurrence. The lack of correction for multiple analyses must also be taken into account and we encourage the readers to interpret these results with caution and consider them preliminary until further independent analyses of other cohorts are conducted in order to confirm these results. Finally, the adjustment for age, sex, disease duration and LEDD seem to confirm our main results.

In summary, we simultaneously examined main effects and interactions of several PRS and personality dimensions on odds of having ICD PD. We showed main effects of SD, ADHD PRS and MDD PRS. In addition, we reported some novel gene-personality interactions. Therefore, it seems important to integrate both psychological and genetic profiles in personalized risk assessment for ICDs in PD, suggesting a stratified approach where PRS may guide initial risk categorization, and personality dimensions refine individual-level predictions.

Data availability

The data supporting the findings of this study are available on request from the corresponding author.