Antidepressant use and Parkinson's disease risk: Overall protective effect with drug‐specific variation

ABSTRACT

Background

Antidepressants can cause extrapyramidal side effects, but their impact on Parkinson's disease (PD) risk remains unclear.

Aims

We aimed to investigate the association between genetic liability to antidepressant use and PD.

Methods

We examined the causal effect of antidepressant use on PD using Mendelian randomisation (MR) analysis. We further performed multivariable MR (MVMR) analysis to determine the independent effect of genetic liability to antidepressant use on PD by accounting for the effect of major depressive disorder (MDD) on PD. The United States Food and Drug Administration Adverse Event Reporting System (FAERS) database (2004 Q1–2024 Q2, 53.4 million event‐level records) was mined to assess associations between 13 antidepressants and PD. Colocalisation analyses identified shared genetic variants between genetic liability to antidepressant use and PD.

Results

MR analyses demonstrated that genetic liability to antidepressant use was inversely associated with PD (odds ratio [OR]: 0.74, p‐false discovery rate (FDR) = 0.016), whereas MDD itself showed no causal association (OR: 1.01, p‐FDR = 0.934). MVMR confirmed that the inverse association between genetic liability to antidepressant use and PD was independent of genetic liability to MDD (OR: 0.69, p‐FDR = 0.018). Colocalisation highlighted LHFPL3, KMT2E and SRPK2 at the chromosome 7q22.3 locus as shared genetic components. Furthermore, real‐world FAERS data revealed that although most antidepressants (10/13) showed no significant disproportionate reporting for PD, clomipramine and mirtazapine were associated with increased reporting odds, whereas duloxetine exhibited a significant inverse reporting signal.

Conclusions

Genetic and observational evidence together support an inverse association between antidepressant use and PD. Given varying individual drug profiles, agents such as duloxetine warrant further prospective evaluation.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Depression is common among patients with Parkinson's disease (PD), but it is unclear whether antidepressant use influences PD risk or whether observed associations are driven by depression itself.

WHAT THIS STUDY ADDS

• Using genetic and real‐world data, we found that genetic liability to antidepressant use is associated with lower odds of PD, whereas genetic liability to major depressive disorder itself is not. Most antidepressants showed no clear association with PD in real‐world reporting data. However, drug‐specific differences were observed, with duloxetine showing an inverse reporting signal and a few other drugs showing increased PD reporting.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE, OR POLICY

• These findings suggest that antidepressant use may partly account for previously reported depression‐PD associations. They emphasise distinguishing drug effects from disease processes and considering antidepressants at the individual drug level in future research.

INTRODUCTION

Major depressive disorder (MDD) is characterised by persistent depressed mood and diminished interest or pleasure in daily activities, ¹ imposing an increasingly heavy global burden. ² The impact of depression extends far beyond mental health, as it is a risk factor for various chronic diseases, including cardiovascular diseases, dementia and type 2 diabetes, ³ , ⁴ among others. ⁵ , ⁶ Antidepressants are the first‐line treatment for depression. In addition to MDD, antidepressants are commonly prescribed for anxiety, eating and sleep disorders, pain syndromes (e.g., fibromyalgia and migraine) and gastrointestinal irritation. Although patterns vary geographically, population‐level exposure data indicate increased antidepressant use among older adults. For example, in the USA, exposure to antidepressants increases from 3.4% among individuals aged 12–19 years to 19.1% among those aged 60 and over. ⁷

Parkinson's disease (PD) is a progressive neurodegenerative disorder primarily affecting movement, with cardinal symptoms including tremor, rigidity, bradykinesia and postural instability. Approximately 1% of the global population over the age of 60 is affected by PD or PD‐like conditions such as parkinsonism, with prevalence rates increasing with age. ⁸ Understanding the genetic and environmental risk factors associated with PD is crucial for its effective management. Previous retrospective and prospective studies have examined the relationship between depression and PD, ⁹ , ¹⁰ but their findings have been inconsistent. This relationship is further complicated by a growing trend in prescribing antidepressants both for PD‐related depression ¹¹ and for gastrointestinal symptoms that often precede the development of motor symptoms of PD and may be considered prodromal symptoms of PD. ¹²

Antidepressants are among the most widely prescribed medications worldwide, indicated for MDD as well as anxiety, chronic pain and certain gastrointestinal disorders. However, many antidepressants, particularly tricyclic antidepressants (TCAs) and selective serotonin reuptake inhibitors (SSRIs), are associated with extrapyramidal side effects, including tremor and drug‐induced parkinsonism. ¹ These drug‐induced symptoms can closely resemble PD, raising concerns that antidepressant use may contribute to the development of idiopathic PD.

The relationship between antidepressants and PD is complex. Observational studies have yielded inconsistent results: some have suggested that antidepressant use could increase PD risk, ¹³ whereas others have found no independent effect once depression itself is accounted for. ¹⁴ , ¹⁵ Adding to this complexity, depression often appears in the prodromal phase of PD, leading to the possibility of reverse causation. ⁹ Recent pharmacoepidemiological and clinical evidence further highlights this intricate interplay. Medium‐to‐high‐frequency use of TCAs and serotonin‐noradrenaline reuptake inhibitors has been linked to an increased incidence of PD, whereas low‐frequency use showed no such association. ¹⁶ Conversely, some pharmacologic agents with antidepressant properties may confer beneficial effects in PD. Pramipexole, a dopamine agonist, exhibits antidepressant efficacy comparable to that of conventional antidepressants in patients with PD and depression, while significantly improving motor symptoms. ¹⁷ Glucagon‐like peptide‐1 receptor agonists, primarily used for diabetes, have also demonstrated antidepressant effects in patients with PD. ¹⁸ Similarly, monoamine oxidase‐B inhibition has been associated with antidepressant effects and potential therapeutic benefits in PD. ¹⁹ Additionally, certain antidepressants have been linked to improved survival; mianserin use, for example, has been associated with reduced 8‐year mortality in patients with PD. ²⁰

We hypothesised that, although antidepressants have been reported to cause transient drug‐induced parkinsonism, they may not increase the risk of developing true neurodegenerative PD and might even be inversely associated with PD risk. Mendelian randomisation (MR) can help address these uncertainties by minimising confounding and reverse causation using genetic variants as instrumental variables. ²¹ In this study, we applied MR and multivariable MR (MVMR) to disentangle the causal effects of antidepressant use and MDD on PD risk, complemented by analyses of data from the United States Food and Drug Administration Adverse Event Reporting System (FAERS). Clarifying these associations could provide critical insights into the long‐term neurological impact of antidepressants and inform future research on therapeutic strategies for patients with comorbid neuropsychiatric conditions.

METHODS

GWAS summary datasets

Figure 1 describes the study design. Publicly available summary statistics of the genome‐wide association study (GWAS) were used for the MR and MVMR analyses in this study. The GWAS dataset of MDD (170 756 cases and 329 443 controls, n = 500 199) was derived from the Psychiatric Genomics Consortium. ²² The GWAS dataset for antidepressant use was obtained from the FinnGen project R11, including 131 176 cases and 104 642 controls (n = 235 818). ²³ The PD dataset was sourced from the International Parkinson's Disease Genomics Consortium. ²⁴ Detailed data characteristics are provided in Supporting Information S1: table S1. All participants in the datasets were of European ancestry. Ethical approvals were obtained in all original studies.

FIGURE 1.

Study design and analytical framework. (A) Two‐sample MR was used to estimate the causal effects of antidepressant use on PD risk using SNPs as instrumental variables. MVMR was used to estimate the independent effects of genetic liability to antidepressant use and MDD on PD. Colocalisation analysis was used to identify shared genetic variants between antidepressant use and PD. (B) FAERS database analysis workflow. The FAERS database was analysed to assess drug‐specific PD associations. CI, confidence interval; FAERS, Food and Drug Administration Adverse Event Reporting System; MDD, major depressive disorder; MR, Mendelian randomisation; MVMR, multivariable Mendelian randomisation; PD, Parkinson's disease; Q1, the first quarter; Q2, the second quarter; ROR, reporting odds ratio; SNPs, single nucleotide polymorphisms.

MR analysis

The MR analyses were conducted using three models in TwoSampleMR, ²¹ adhering to the reporting guidelines outlined in the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) for MR studies. The inverse‐variance weighted (IVW) model was selected as the main method, whereas the MR‐Egger and weighted median (WM) models were used as complementary analytic techniques. The intercept from the MR‐Egger regression was employed to evaluate the average directional pleiotropy. Heterogeneity was assessed using both I ² statistics and Cochran's Q test (with I ² > 0.25 and p < 0.05 indicating significant heterogeneity). ²⁵ Significant associations were determined using false discovery rate (FDR)‐adjusted p values from the IVW model. Single‐nucleotide polymorphisms (SNPs) with genome‐wide significance (p < 5 × 10⁻⁸) in the exposure datasets were used to derive independent instrumental variables (r ² < 1 × 10⁻³ within a 10 Mb window). Crucially, depression often presents as a prodromal symptom of PD, raising concerns of reverse causation. To address this, we performed the MR Steiger directionality test. This test computes the variance explained by the instrumental variables in both the exposure and the outcome. Any SNP explaining more variance in PD than in antidepressant use, indicating inconsistency in the inferred causal direction, was excluded from the final analyses to ensure the robustness of the causal inference.

MVMR analysis

To estimate the direct effects of each exposure on PD, the MVMR analysis was conducted using the TwoSampleMR package. For each exposure, instrumental variables were extracted, combined into a set of genetic instruments, and then entered into the analytic pipeline.

Mining of adverse event signals

We extracted adverse event reports from the FAERS quarterly ASCII files (2004 Q1–2024 Q2), encompassing a total of 21 558 936 raw reports. Thirteen candidate antidepressants were predefined, including amitriptyline, citalopram, clomipramine, duloxetine, escitalopram, esketamine, fluoxetine, fluvoxamine, mirtazapine, paroxetine, sertraline, trazodone and venlafaxine. Adverse events in FAERS are coded using preferred terms (PTs) from the Medical Dictionary for Regulatory Activities (MedDRA). Because MedDRA is updated biannually (March and September), with each update potentially involving revisions to PT hierarchies and corresponding System Organ Classes (SOCs), we recoded all adverse event terms in the database to the latest MedDRA version (v27.0) to ensure terminological consistency. For the analysis of parkinsonism, we then strictly restricted the data to the single PT ‘Parkinson's disease’, thereby maximising phenotypic specificity while minimising confounding from drug‐induced parkinsonism. Data cleaning adhered to FDA‐recommended deduplication protocols as follows: records were sorted by case identification number (CASEID), FDA receipt date (FDA_DT) and Primary Identification Number (PRIMARYID), retaining the record with the latest FDA_DT for duplicate CASEIDs, or the highest PRIMARYID if dates were identical. This process removed 3 611 216 redundant reports, yielding 17 947 720 unique case‐level records, corresponding to 53 352 754 event‐level records. Disproportionality analyses were conducted using reporting odds ratios (RORs). A positive signal (indicating increased reporting) was deemed statistically significant if the lower bound of the 95% confidence interval (CI) exceeded 1 and the case count was ≥ 3. Conversely, an inverse signal was defined as statistically significant if the upper bound of the 95% CI was < 1 and the case count was ≥ 3. ²⁶ The data extracted for this study encompassed over 17.9 million deduplicated case reports and 53.4 million associated adverse event records.

Colocalisation analysis

We performed Bayesian colocalisation analysis using the R package coloc. ²⁷ This method evaluates whether a pair of datasets shares common genetic variants at specific genomic loci. The analysis calculates posterior probabilities (PP.H⁴) to determine if both traits are likely to share the same causal genetic variation, with PP.H⁴ > 0.80 considered strong evidence for colocalisation. We tested colocalisation signals shared between the traits of antidepressant use and PD.

RESULTS

MR analysis

We first investigated the overall causal effects of MDD and antidepressant use on PD risk using two‐sample MR analyses. To rule out reverse causation, we applied the MR Steiger directionality test prior to the final causal estimates. We found that genetic predisposition to MDD was not causally associated with the risk of PD (odds ratio [OR]: 1.01, 95% CI 0.78–1.31, p‐FDR = 0.934), with all instrumental variables supporting the hypothesised causal direction (Supporting Information S1: table S2). For the effects of antidepressant use on PD, the Steiger test identified only one SNP (rs61944361) exhibiting inconsistency with the hypothesised causal direction, which was subsequently excluded to prevent reverse causation bias (Supporting Information S1: table S3). Using this refined, Steiger‐validated set of instrumental variables (all F‐statistics > 10), the updated IVW analysis confirmed that genetic liability to antidepressant use was inversely associated with PD (OR: 0.74, 95% CI 0.60–0.92, p = 0.008, p‐FDR = 0.016) (figure 2). Sensitivity analyses yielded directionally consistent estimates across MR methods (table 1). Importantly, formal pleiotropy testing via the MR‐Egger regression provided no evidence of directional horizontal pleiotropy. The intercept p values for both exposures were nonsignificant (for MDD on PD: intercept = 0.02, p = 0.265; for antidepressants on PD: intercept < −0.01, p = 0.969), thus ruling out widespread pleiotropic confounding. To assess instrument heterogeneity, we performed Cochran's Q test. For the primary IVW analyses, mild to moderate heterogeneity was observed among the genetic instruments for both MDD (I ² = 0.30, p = 0.027) and antidepressant use (I ² = 0.41, p = 0.007). Given this observed heterogeneity, all primary IVW estimates were appropriately derived using a multiplicative random‐effects model to ensure conservative and robust variance estimation.

FIGURE 2.

Causal effects of antidepressant use and MDD on PD. The forest plot shows the two‐sample MR estimates of the causal effects of genetic liability to antidepressant use and MDD on PD obtained using the IVW method. Dots represent ORs, and the horizontal lines represent 95% CIs. CI, confidence interval; FDR, false discovery rate; IVW, inverse‐variance weighted; MDD, major depressive disorder; MR, Mendelian randomisation; OR, odds ratio; PD, Parkinson's disease.

TABLE 1.

Causal associations between MDD, antidepressant use and PD

Exposure	Outcome	Method	N_IV	β (SE)	OR (95% CI)	p‐FDR
MDD	PD	IVW	47	0.01 (0.13)	1.01 (0.78–1.31)	0.934
MDD	PD	WM	47	−0.17 (0.17)	0.85 (0.61–1.17)	0.325
MDD	PD	MR‐Egger	47	−0.75 (0.68)	0.47 (0.12–1.81)	0.560
Antidepressant use	PD	IVW	34	−0.30 (0.11)	0.74 (0.60–0.92)	0.016
Antidepressant use	PD	WM	34	−0.27 (0.13)	0.76 (0.60–0.97)	0.060
Antidepressant use	PD	MR‐Egger	34	−0.27 (0.70)	0.77 (0.19–3.04)	0.706

Abbreviations: CI, confidence interval; FDR, false discovery rate; IVW, inverse‐variance weighted; MDD, major depressive disorder; MR, Mendelian randomisation; N_IV, number of instrumental variables; OR, odds ratio; PD, Parkinson's disease; SE, standard error; WM, weighted median.

MVMR analysis

To disentangle the independent effects of these intertwined clinical factors, we conducted an MVMR analysis. This approach estimates the direct effect of each exposure on PD while adjusting for the genetic liability to the other. Within this joint framework, the conditional F‐statistics for MDD and antidepressant use were 9.73 and 7.87, respectively. These values reflect the expected attenuation in instrument strength when partitioning genetic variance between highly correlated exposures. Crucially, the MVMR results revealed that genetic liability to MDD is not independently associated with the risk of PD (OR: 1.16, 95% CI 0.80–1.67, p‐FDR = 0.435). In contrast, genetic liability to antidepressant use was independently associated with lower odds of PD (OR: 0.69, 95% CI 0.52–0.91, p‐FDR = 0.018) (figure 3).

FIGURE 3.

Independent causal effects of antidepressant use and MDD on PD from MVMR analysis. Forest plot of the MVMR estimates for genetic liability to antidepressant use and MDD in relation to PD. Points indicate ORs and horizontal lines indicate 95% CIs. CI, confidence interval; FDR, false discovery rate; MDD, major depressive disorder; MVMR, multivariable Mendelian randomisation; OR, odds ratio; PD, Parkinson's disease.

Mining of adverse event signals

Adverse event data were extracted from the FAERS database, covering the period from the first quarter of 2004 to the second quarter of 2024. The initial dataset contained 21 558 936 raw reports. The final dataset included adverse event reports in which one of the 13 target drugs was recorded as the primary suspect (n = 292 636 reports; n = 1 195 184 adverse events) (figure 4 and Supporting Information S1: table S4). Most antidepressants (10 out of 13) showed no significant disproportionate reporting for PD. Two drugs were associated with increased reporting odds for PD: clomipramine (ROR: 3.93, 95% CI 1.77–8.76, p‐FDR = 0.006) and mirtazapine (ROR: 1.84, 95% CI 1.33–2.54, p‐FDR = 0.003). In contrast, one drug, duloxetine, exhibited a significant inverse reporting signal for PD (ROR: 0.59, 95% CI 0.44–0.79, p‐FDR = 0.003) (figure 5).

FIGURE 4.

Flowchart of data selection and processing from the FAERS database. Data were obtained from FAERS quarterly ASCII files spanning 2004 Q1–2024 Q2. Raw reports (n = 21 558 936) were deduplicated using the FDA‐recommended algorithm based on the raw data. Records were sorted by CASEID, FDA_DT (FDA receipt date) and PRIMARYID; for duplicate CASEID entries, the record with the latest FDA_DT was retained; if FDA_DT was identical, the record with the highest PRIMARYID was kept, yielding a total of 17 947 720 unique case‐level records. Subsequently, irrelevant reports (non‐target drugs or incomplete data) were excluded, followed by the exclusion of irrelevant adverse events. The final dataset included adverse event reports in which one of the 13 target drugs was recorded as the primary suspect. ASCII, American Standard Code for Information Interchange; CASEID, case identification number; FAERS, Food and Drug Administration Adverse Event Reporting System; FDA, Food and Drug Administration; PRIMARYID, Primary Identification Number.

FIGURE 5.

Parkinson's disease reporting signals associated with antidepressants in FAERS. Forest plot of RORs and 95% CIs for PD reports associated with individual antidepressants. (A) number of PD reports for which the target antidepressant was recorded as the primary suspect; (B) number of reports of other adverse events for which the target antidepressant was recorded as the primary suspect; (C) number of PD reports for which a drug other than the target antidepressant was recorded as the primary suspect; (D) number of reports of other adverse events for which a drug other than the target antidepressant was recorded as the primary suspect. CI, confidence interval; FAERS, Food and Drug Administration Adverse Event Reporting System; FDR, false discovery rate; PD, Parkinson's disease; ROR, reporting odds ratio.

Colocalisation analysis

The colocalisation analysis of the genes shared by antidepressant use and PD revealed a common genetic region on chromosome 7q22.3, with a PP.H⁴.abf value of 0.86 (Supporting Information S1: table S5). This genomic region contains three genes, including LHFPL3, KMT2E and SRPK2 (figure 6).

FIGURE 6.

Shared genetic region on chromosome 7q22.3 between antidepressant use and PD. Regional association plots for antidepressant use and PD on chromosome 7q22.3. The plots show one shared region containing the three genes LHFPL3, KMT2E and SRPK2. The −log₁₀(p) values for SNPs are shown with lead SNPs highlighted. Colours indicate linkage disequilibrium (r ²) with the lead SNP. Recombination rates (cM/Mb) are shown on the right y‐axis and annotated genes are displayed below. The upper plot represents antidepressant use and the lower plot represents PD. cM/Mb, centimorgans per megabase; PD, Parkinson's disease; SNPs, single nucleotide polymorphisms.

DISCUSSION

Main findings

This study provides new genetic evidence that genetic liability to antidepressant use is inversely associated with PD, whereas MDD itself does not appear to have a causal effect. Using MR and MVMR, we found that the previously reported association between MDD and PD may be partly attributable to antidepressant use rather than MDD itself.

The effects of TCAs and SSRIs on parkinsonism present a complex picture. In the UK Biobank anxiety cohort, SSRI use was consistently associated with an increased incidence of PD across all frequency groups (multivariate‐adjusted ORs [95% CIs]: 1.584 [1.315–1.893] for 1–4 times, 1.956 [1.572–2.408] for 5–10 times, and 1.728 [1.475–2.021] for > 10 times; all p < 0.001). ¹⁶ Conversely, fluoxetine demonstrated therapeutic potential in a rat model of PD by restoring impaired hypercapnic ventilatory response, suggesting that it may alleviate respiratory distress associated with parkinsonism. ²⁸ TCAs such as desipramine, when combined with levodopa (L‐3,4‐dihydroxyphenylalanine, L‐DOPA), enhanced motor behaviour in a rat model of PD, likely via increased contralateral rotations and elevated dopamine and noradrenaline levels. ²⁹ These findings highlight the potentially context‐dependent effects of SSRIs and TCAs on parkinsonism, warranting further investigation.

Our genetic analyses and FAERS data indicate that most antidepressants are not strongly linked to PD risk. Clomipramine and mirtazapine show increased PD reporting, whereas duloxetine is associated with an inverse reporting signal (ROR 0.59), suggesting drug‐specific effects. Preclinical and limited clinical data suggest that clomipramine may reduce L‐DOPA‐induced dyskinesia and provide neuroprotective benefits, ³⁰ , ³¹ whereas mirtazapine may improve depressive symptoms in patients with PD, ³² , ³³ although both drugs require careful monitoring due to potential adverse effects. Overall, FAERS and available literature do not support disproportionate PD reporting for most antidepressants.

In FAERS, duloxetine was the only antidepressant that showed a significant inverse reporting signal for PD. Duloxetine has been associated with at least partial improvement in mood and motor symptoms in patients with PD, ³⁴ including those with comorbid depression. ³⁵ Duloxetine appears to improve motor and non‐motor symptoms through leucine‐rich pentatricopeptide repeat‐containing protein‐related mechanisms. ³⁵ , ³⁶ Duloxetine acts by blocking the reuptake of both serotonin and noradrenaline and has a low potential to cause extrapyramidal symptoms due to its low affinity for D2 receptors. Duloxetine has been proposed as a promising treatment option for depression in patients with PD. ³⁷ Although these findings should not be interpreted as evidence of disease prevention, they support further prospective evaluation of duloxetine and other antidepressants with favourable safety profiles.

Several biological pathways may underlie the observed inverse association between genetic liability to antidepressant use and PD. Antidepressants are frequently prescribed for gastrointestinal disorders such as inflammatory bowel disease and irritable bowel syndrome. ³⁸ , ³⁹ By alleviating prodromal gastrointestinal dysfunction, antidepressants could theoretically delay α‐synuclein aggregation and its propagation to the brain. ⁴⁰ In addition, some agents, including monoamine oxidase‐B inhibitors, possess both antidepressant and neuroprotective properties, ¹⁹ while pramipexole has antidepressant efficacy comparable to that of conventional antidepressants and also improves motor symptoms in patients with PD. ¹⁷

Body composition may represent another potential mediating pathway. Certain antidepressants are associated with weight gain, and recent genetic studies have demonstrated causal relationships and shared genetic architectures between body fat composition and PD. ⁴¹  Higher waist‐to‐weight index, conicity index and body shape index have been associated with increased PD prevalence, ⁴² whereas patients with established PD commonly have lower body weight and visceral fat. ⁴³ Furthermore, inadequate protein intake has been linked to altered fat mass and sarcopenia in patients with PD. ⁴⁴ These observations suggest that antidepressant‐related alterations in body composition may influence PD susceptibility, although the direction and magnitude of this relationship may differ across disease stages.

Our colocalisation analysis further identified a shared genetic region on chromosome 7q22.3 containing LHFPL3, KMT2E and SRPK2. These genes are implicated in neuronal function and antidepressant response. Some of these genes have reported roles in α‐synuclein regulation, ⁴⁵ duloxetine response, ⁴⁶ tau phosphorylation ⁴⁷ and gut–brain interactions. ⁴⁸ Notably, LHFPL3 has previously been associated with antidepressant treatment response in MDD through the SNP rs4352778. ⁴⁹ These findings highlight possible molecular links between antidepressant use and the observed inverse association with PD.

The integration of MR and FAERS data is an important strength of this study. Although FAERS provides valuable real‐world pharmacovigilance data capable of detecting early safety and inverse reporting signals, it is inherently susceptible to confounding by indication, underreporting and reverse causation. Complex real‐world prescribing patterns may further complicate the interpretation of spontaneous reporting data. A recent nationwide study reported frequent off‐label use of antidepressants among paediatric patients, with prescription prevalence varying substantially by age, indication and dose. ⁵⁰ Such heterogeneous usage patterns across indications and age groups can substantially confound reporting patterns in FAERS. In contrast, MR uses genetic variants as instrumental variables to minimise environmental confounding and reverse causation. In the present study, FAERS identified drug‐specific reporting patterns, and our MR and MVMR analyses independently supported an inverse association between genetic liability to antidepressant use and PD. Similar MVMR approaches have demonstrated that antidepressant use may account for an apparent causal effect of MDD on type 2 diabetes. ⁵¹ These approaches provide complementary evidence, increasing confidence in the robustness of our observations.

Limitations

Several limitations should be acknowledged. First, although MR substantially reduces confounding, residual pleiotropy cannot be completely excluded. Mild‐to‐moderate heterogeneity was observed among the genetic instruments, but this was accommodated using random‐effects modelling without evidence of directional pleiotropy, consistent with previous MR studies of complex polygenic traits. Second, the conditional F‐statistics in the MVMR analyses were modest after accounting for the shared genetic architecture of MDD and antidepressant use. This reduction is expected when partitioning genetic variance between closely related phenotypes. Although the independent selection of instrumental variables and the stability of the estimates argue against substantial weak‐instrument bias, such bias cannot be entirely excluded. Third, all GWAS datasets were restricted to individuals of European ancestry, limiting the generalisability of our findings to other populations. Fourth, because large GWAS datasets stratified by antidepressant class are currently unavailable, our MR analyses evaluated antidepressant use as a single aggregated phenotype. Consequently, genetic liability reflects a lifetime propensity for antidepressant use rather than exposure to specific drugs or doses and cannot capture the compound‐specific heterogeneity observed in the FAERS analyses, including short‐term drug‐induced parkinsonism. Finally, FAERS is based on spontaneous reports and is therefore susceptible to underreporting, reporting bias and confounding by indication. To maximise specificity, we restricted the analysis to the preferred term ‘Parkinson's disease’ rather than the broader Standardised MedDRA Query, thereby reducing misclassification of drug‐induced parkinsonism at the expense of sensitivity. In addition, incomplete indication data prevented adjustment for psychiatric comorbidities. Nevertheless, these limitations underscore the complementary value of the MR analyses, which reduce indication‐related confounding through the use of genetic instruments.

Implications

Our findings have several implications for future research. First, they suggest that the reported relationship between depression and PD may, at least in part, reflect antidepressant use rather than depression itself, emphasising the importance of disentangling disease effects from treatment effects in epidemiological studies. Second, the observed drug‐specific heterogeneity indicates that antidepressants should not be considered a homogeneous class when evaluating their neurological effects. In particular, duloxetine demonstrated a consistent inverse reporting signal in FAERS and therefore warrants prospective investigation to clarify whether this observation reflects a true biological association.

Future studies should replicate these findings in more diverse populations, evaluate dose–response relationships and the effects of antidepressant subclasses in large prospective cohorts and further investigate the biological mechanisms underlying the observed associations. Particular attention should be given to gut–brain interactions and their bidirectional relationships with insomnia, ⁵² as well as to neurochemical regulation and changes in body composition. ⁵³ Future studies should also examine the preservation of functional networks and regional homogeneity in the context of advanced interventions such as deep brain stimulation. ⁵⁴ , ⁵⁵ Such studies will help determine whether the observed associations reflect modifiable biological pathways relevant to PD pathogenesis.

CONCLUSION

This study provides genetic and pharmacoepidemiological evidence that genetic liability to antidepressant use is associated with lower odds of PD. These findings highlight the importance of distinguishing drug‐induced parkinsonism from true neurodegenerative PD. Furthermore, they suggest that certain antidepressants warrant further prospective investigation to clarify their potential association with PD susceptibility.

AUTHOR CONTRIBUTIONS

Fuquan Zhang conceptualised and designed the study, and supervised the project. Fuquan Zhang, Dongming Liu, and Ancha Baranova performed the genome‐wide data analyses. Fuquan Zhang, Ancha Baranova, Dongming Liu, Hongbao Cao, Chenxin Xu, and Wenxi Sun curated and analysed the FAERS pharmacovigilance data. Fuquan Zhang, Ancha Baranova, and Dongming Liu drafted the manuscript, and all authors contributed to critical revision and editing. All authors read and approved the final version of the manuscript.

FUNDING

This work was funded by the National Natural Science Foundation of China (82502317) and the China Postdoctoral Science Foundation (2023M741648, 2026M792483).

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Supporting Information S1

Biography

Ancha Baranova graduated with a PhD from Moscow State University in 1998 and with a DSci from the Russian Academy of Science in Russia. Currently, she is a professor in the School of Systems Biology at George Mason University in Virginia, USA. She is an expert in human pathophysiology and human genetics, with an emphasis on chronic liver diseases, cancer and anti‐ageing research. Her lab has discovered many novel biomarkers, the biosynthesis of melanin in human adipose, some properties of cell‐free DNA and a variety of novel functions for known biomolecules. Dr Baranova's main research has recently focused on major pathophysiological components of human ageing: systemic inflammation, insulin resistance and organ fibrosis. Her work in personalised medicine emphasises longitudinal monitoring and management of health in pre‐symptomatic individuals and augmenting the body's homoeostasis by non‐pharmacological means.

Baranova A, Cao H, Xu C, Sun W, Liu D, Zhang F. Antidepressant use and Parkinson's disease risk: Overall protective effect with drug‐specific variation. General Psychiatry. 2026;39:e70067. 10.1002/gps3.70067

DATA AVAILABILITY STATEMENT

The GWAS summary datasets are available on the website of each study. All data are available in the submitted manuscript, supplementary files, or related sources described in the manuscript.