The causal relationship between neuroimaging and Parkinson’s disease: A two-sample Mendelian randomization study

Abstract

Parkinson’s disease (PD) is a neurodegenerative disorder whose etiology and progression remain not fully understood, posing significant challenges for both diagnosis and treatment. This study leverages the 2-sample Mendelian Randomization (TSMR) approach to systematically examine the possible causal connections involving neuroimaging characteristics and the risk of developing PD. Making use of genetic variations from extensive genome-wide association studies (GWAS) as instrumental factors, this analysis focuses on neuroimaging traits related to cortical and subcortical gray matter volumes. The primary findings reveal that specific changes in neuroimaging traits, notably in regions such as the right ventral diencephalon and the thalamus, have a strong causal connection with the risk of PD. These associations were robust across multiple MR methods, including MR-Egger, Inverse-Variance Weighted (IVW), and Weighted Median approaches, supplemented by extensive sensitivity analyses to guarantee the accuracy of the findings. Our research’s insights into the structural and functional brain changes provide a deeper understanding of PD’s etiology, suggesting that certain neuroimaging features may not only precede but also potentially drive the disease process. This underscores the possibility of these neuroimaging traits serving as early biomarkers or targets for preventive measures against PD. Despite its strengths in causal inference and the use of extensive GWAS datasets, limitations include reliance on data primarily from European populations and inherent assumptions of Mendelian Randomization.

1. Introduction

Parkinson’s Disease (PD) is a common neurodegenerative disorder distinguished by hallmark clinical features including bradykinesia, muscular rigidity, resting tremor, and loss of balance and coordination.^[1,2] PD not just considerably deteriorates the quality of life for individuals, but also places a considerable load on public health systems. It is estimated that millions worldwide are affected by PD, and with the increasing aging population, this number is expected to rise.^[3,4] Despite extensive studies on its clinical manifestations, the precise etiology and pathogenesis of PD remain largely elusive, limiting the development of effective preventive and therapeutic strategies.

Recent advancements in neuroimaging technologies have offered new perspectives for PD research.^[5] Techniques such as Positron Emission Tomography (PET) and Magnetic Resonance Imaging (MRI) allow noninvasive observation of changes in brain structure and function.^[6] These tools are instrumental not just in the early diagnosis of PD, but also in providing valuable insights into its neuropathological underpinnings.^[7] For instance, MRI studies have revealed a reduction in gray matter density and a decline in white matter integrity in PD patients,^[8] while PET scans have detected alterations in the function of dopaminergic neurons.^[9] However, despite a wealth of neuroimaging studies elucidating brain changes associated with PD, these findings predominantly stem from observational studies, and the causal relationships remain unclear. Specifically, it is still uncertain whether the neuroimaging characteristics are causes of PD, consequences of the disease, or part of a complex interaction involving both.

Traditional observational studies have played a crucial role in revealing associations between diseases and potential exposure factors, yet their inherent limitations, such as confounding factors and potential reverse causation, often constrain accurate causal inference. To address these concerns, Mendelian Randomization (MR), using genetic variations as instrumental factors, has gained widespread use in epidemiological research. By leveraging genetic variants assigned at conception—which theoretically are free from the confounding typically seen in observational studies—MR offers more reliable causal inferences.^[10–12]

Under the MR framework, Two-sample Mendelian Randomization further enhances the flexibility and applicability of this approach. This method allows researchers to use genetic variants and disease outcome data from different samples, effectively increasing the available sample size and statistical power. It has been successfully applied to explore the causal relationships in complex diseases like cardiovascular diseases, metabolic disorders, and neurodegenerative diseases.^[13,14] Nevertheless, the causal relationship between neuroimaging features and PD remains underexplored, particularly through systematic assessment using MR techniques.

This research attempts to close this information gap by methodically assessing the potential causal relationships between neuroimaging characteristics and PD risk using the 2-sample Mendelian Randomization approach. Specifically, we will employ genetic variants associated with neuroimaging traits from publicly available Genome-Wide Association Studies (GWAS) data to evaluate their causal influence on PD risk. By utilizing various MR analysis methods, including MR-Egger regression, Inverse-Variance Weighted (IVW), and Weighted Median, we strive to guarantee the validity of our findings and minimize potential pleiotropic biases. Additionally, sensitivity analyses will further validate the reliability of these causal associations, ensuring the scientifical validity and effectiveness of our findings.

Understanding the causal relationship between neuroimaging characteristics and PD could not only help elucidate the etiological nature of PD but also inform the development of preventative and therapeutic strategies. For example, if certain neuroimaging traits are identified as causal risk factors for PD, interventions taking aim at these characteristics may be essential in preventing the onset of the disease.^[15] Moreover, the causality in neuroimaging traits could offer novel biomarkers for early detection and personalized treatment of PD, thus improving patient prognosis and quality of life.^[16]

Although MR offers significant advantages for causal inference, its application also faces challenges and limitations. For instance, the choice of instrumental variables is critical and must satisfy core assumptions including relevance, independence, and exclusion criteria.^[17] Furthermore, potential associations involving genetic variants and confounding factors, along with the presence of multiple genetic effects, could impact the accuracy of MR analyses. Therefore, in this study, we will adhere strictly to the best practices of MR analysis, employ various methods for result validation, and conduct sensitivity analyses to evaluate potential biases, ensuring the reliability and robustness of our conclusions.

In summary, while neuroimaging has shown considerable potential in PD research, its causal relationship with PD has not been definitively determined. By employing the 2-sample Mendelian Randomization method, the purpose of this research is to systematically evaluate the causal effect of neuroimaging traits on PD risk, providing fresh understanding into the mechanisms of PD and informing optimized prevention and treatment strategies. As genomic and neuroimaging data continue to accumulate and integrate, MR-based causal inference studies will play a more important part in elucidating the etiology of PD and other complex diseases.

2. Materials and methods

2.1. Study design

This study employs a 2-sample Mendelian Randomization (MR) approach, utilizing single nucleotide polymorphisms (SNPs) derived from publicly available data of a previous GWAS study on neuroimaging traits.^[18] The regions associated with neuroimaging traits were considered as exposures, with Parkinson’s disease acting as the outcome. To minimize the influence of potential confounders, we applied the core Mendelian Randomization (MR) assumptions: Relevance: Genetic variations and exposure are significantly correlated; Independence: the instrumental variable (IV) is not connected to any confounding factors; Exclusivity: the exposure is the only pathway through which the genetic variants affect the result of interest. Furthermore, to rule out horizontal pleiotropy and other confounding effects, multiple sensitivity analyses including MR-Egger regression, weighted median, and MR-PRESSO were performed. These approaches ensured that any residual confounding from genetic variants was minimized and that the causal estimates remained robust. Figure 1 provides a diagrammatic illustration of the Two-sample Mendelian Randomization (TSMR).^[19]

Figure 1.

Two-sample Mendelian Randomization (TSMR) schematic.

Three main presumptions form the basis of the study: The first assumption is that genetic variance and exposure are significantly correlated; the second is that there are no confounding variables linked to the instrumental variable (IV); and the third is that the IV is only indirectly related to Parkinson’s disease (outcome) through the neuroimaging traits (exposure), not directly.

2.2. Data acquisition

The exposure data for this study comprises summarized statistics of 83 gray matter volumes in both cortical and subcortical regions (33 cortical regions per hemisphere + 8 subcortical areas per hemisphere + brainstem),^[18] based on a European population. All information is available at https://gwas.mrcieu.ac.uk/. The GWAS data relating to Parkinson’s Disease (PD) are derived from public datasets available in the Finnish database (https://r11.finngen.fi/pheno/G6_PARKINSON), which includes large case-control samples from a pan-European population. See Table 1.

Table 1.

Characteristics of the GWAS in terms of exposure and outcome.

Exposure	Consortium	Total population	Cases/controls	Ethnicity
Neuroimaging features	The UK Biobank	36,778	NA	European
Outcome				European
PD	The FinnGen R11 project	453,733	5150/448583	European

2.3. Selection of instrumental variables

In order to find SNPs that are strongly linked to each exposure, we used a strict genome-wide significance criterion of P < 5 × 10⁻⁵ for the selection of instrumental variables (IVs). This ensures the relevance of genetic variants to the exposures studied. To maintain the independence among SNPs and avoid bias due to linkage disequilibrium (LD), we establish an r² threshold of 0.001 for windows exceeding 10,000 kB. Additionally, we assessed the strength of the IVs using the proportion of variance explained (R²) and the F-statistic, ensuring that selected IVs meet the robustness required for MR analysis (F > 10 is considered strong). We further ensured the eligibility of selected IVs as they are not influenced by confounders and exclusively affect the outcome through the exposure pathway.

2.4. Mendelian randomization analysis

A 2-sample Mendelian randomization method was used in this investigation. The effect estimates are based on extracting the regression coefficients and their standard errors of the instrumental variables on neuroimaging features (exposure) and PD (outcome). If only 1 instrumental variable is available, the Wald ratio method is used to estimate the MR effect. The primary method for estimating the causal effect of neuroimaging traits on PD is the standard Inverse-Variance Weighted (IVW) method, augmented by MR-Egger and Weighted Median methods. IVW method consolidates the effect estimates of each instrumental variable through weighting to determine the total causal influence. MR-Egger regression assesses horizontal pleiotropy via the intercept term and provides an adjusted estimation of the causal influence. The Weighted Median method, assuming at least 50% of the instrumental variables are valid, offers a robust causal effect estimate. Additionally, many sensitivity studies were carried out to guarantee the accuracy of the findings: The heterogeneity across instrumental factors was evaluated using Cochran’s Q test; if considerable heterogeneity is found, further research into its causes is required. In order to make sure that the instrumental factors are not influencing the result in other ways, the MR-Egger regression’s intercept is used to test for horizontal pleiotropy.

3. Ethics statement

Ethical committees have examined the linked research, and the GWAS data utilized in this analysis came from publically accessible sources. Therefore, no further ethical clearance was needed for this research.

4. Statistical analysis

For all statistical studies, R software (version 4.2.0 or above) was used, primarily utilizing the “TwoSampleMR” package. Specific steps include data cleaning, matching instrumental variables, effect estimation, and sensitivity analysis. Two-sided P < .05 was established as the cutoff point for statistical significance.

5. Results

5.1. Impact of neuroimaging traits on PD

This study initially assessed the causal impact of neuroimaging traits on Parkinson’s Disease (PD) (Fig. 2). The Inverse-Variance Weighted (IVW) analysis indicated that exposures to neuroimaging traits in 27 regions have a causal effect on PD, with 25 demonstrating a positive causal relationship, and 2 (Left pars triangularis volume and Left parahippocampal volume) showing a negative causal relationship (Supplementary Table S1). Moreover, we identified the 4 regions with the most significant differences: Right ventral diencephalon (OR = 1.0008, 95%CI: 1.0005–1.0012; P = 1.465E-06); Right thalamus proper volume (OR = 1.0005, 95%CI: 1.0003–1.0008; P = 3.768E-06); Left ventral diencephalon (DC) volume (OR = 1.0005, 95%CI: 1.0003–1.0010; P = 1.3918E-5); and Left thalamus proper volume (OR = 1.0003, 95%CI: 1.0001–1.0006; P = 3.1338E-5) (Fig. 3). Furthermore, results related to these 4 regions from MR-Egger regression and Weighted Median methods were in accordance with those from IVW. Cochran’s Q tests and MR-Egger regression revealed no heterogeneity or horizontal pleiotropy (Supplementary Table S2), and leave-one-out tests confirmed the reliability of the findings. To further minimize the influence of potential confounders, we conducted sensitivity analyses using MR-Egger intercept, Cochran’s Q test, and MR-PRESSO global test. The results, summarized in Supplementary Table S3, revealed no evidence of directional pleiotropy (MR-Egger intercept P > .05), no significant heterogeneity (Cochran’s Q test P > .05), and no global pleiotropy detected by MR-PRESSO, indicating that the observed causal effects were unlikely to be biased by confounding from genetic variants.

Figure 2.

Two-sample MR Estimated Forest plot of the Relationship between neuroimaging features and PD.

Figure 3.

Two-sample Mendelian Randomization (TSMR) analysis of right ventral diencephalon, Right thalamus proper volume, left ventral diencephalon (DC) volume, and left thalamus proper volume in PD patients. (A) Forest plot; (B) Scatter plot; (C) Funnel plot.

Additionally, to eliminate the interference of reverse causation, reverse MR studies were conducted with PD as the exposure and neuroimaging traits as the outcomes. It was found that there is no reverse causation for Right ventral diencephalon, Right thalamus proper volume, Left ventral diencephalon (DC) volume, and Left thalamus proper volume, reinforcing the reliability and stability of the study. However, bidirectional causation was observed in the analyses using Inverse Variance Weighted for all mentioned regions (Supplementary Table S4, Figure 4), suggesting that while the primary results are reliable, complex reciprocal interactions may also exist between PD and specific neuroanatomical structures.

Figure 4.

Two-sample MR estimated forest plot of the relationship between PD and neuroimaging features.

6. Discussion

This research employs the Two-Sample Mendelian Randomization (TSMR) approach to systematically assess the causal relationship involving neuroimaging features and the risk of Parkinson’s Disease (PD). Our findings indicate that specific neuroimaging characteristics, such as decreased gray matter density and reduced white matter integrity, are significantly associated with the onset of PD. These discoveries broaden our knowledge of the causes of Parkinson’s disease and provide fresh scientific justifications for preventative and therapeutic approaches in the future.

The study identifies significant causal associations involving the risk of PD and neuroimaging characteristics in specific brain regions such as the Right ventral diencephalon, Right thalamus proper, Left ventral diencephalon (DC), and Left thalamus proper. In particular, a higher risk of Parkinson’s disease is linked to lower white matter integrity and gray matter density in certain areas. Consistent with existing cross-sectional and longitudinal observational studies,^[20,21] our research further validates, through Mendelian Randomization, the potential causal roles of these neuroimaging changes in the pathogenesis of PD.

For instance, reductions in gray matter density have been reported in multiple observational studies to be associated with cognitive impairments and motor symptoms in PD.^[22] Through MR analysis, our study further confirms the causality of this association, suggesting that changes in gray matter density could actively participate in the pathogenesis of PD, rather than merely being a result of disease progression. Additionally, as a crucial center for motor regulation, the thalamus’s volumetric changes are closely linked with PD’s motor symptoms. Our results support the hypothesis of a causal role of the thalamus in the pathological process of PD, aligning with previous studies on thalamic functional and structural alterations.^[23,24]

The causal relationships between neuroimaging features and PD risk suggest that structural and functional brain changes could play key roles in the pathology of PD. Decreases in gray matter density may reflect neuronal loss or a reduction in synaptic connections, particularly significant in the basal ganglia and cortical areas of PD patients. The basal ganglia, a vital area for motor regulation, whose dysfunction is a major cause of PD’s motor symptoms, may lead to disruptions in the basal ganglia-cortical circuit, thus inducing motor disorders.^[25]

The decline in white matter integrity could affect inter-regional information transfer and the coordination of neural networks. Myelin in white matter is crucial for the conduction of nerve fibers^[26]; damage to white matter integrity could make neuronal signal transmission less effective, thereby impacting motor control and other cognitive functions. This mechanism aligns with the commonly observed symptoms of motor slowness and instability in balance in PD patients.^[27]

Furthermore, brain regions such as the ventral diencephalon and thalamus play significant roles in regulating the dopamine system.^[28] The loss of dopaminergic neurons in the substantia nigra, which results in lower dopamine levels, is one of the main pathogenic characteristics of Parkinson’s disease.^[29,30] Structural changes in the ventral diencephalon and thalamus could further exacerbate the dysfunction of the dopamine system, forming a vicious cycle that promotes disease progression.

This study has several strengths. Firstly, by using genetic variations as instrumental factors and the Two-Sample Mendelian Randomization technique, it effectively avoids common confounders and reverse causality issues present In conventional observational research, thereby enhancing the credibility of causal inferences. Secondly, the research utilizes large-scale GWAS data, covering extensive neuroimaging features and PD cases, ensuring the statistical power and broad applicability of the study results. Moreover, multiple MR analysis methods (including weighted median, IVW, and MR-Egger regression) and multiple sensitivity analyses further confirm the robustness and consistency of the findings, enhancing the reliability of the conclusions.

Despite these significant findings, the research has several drawbacks. Firstly, the majority of the data originates from European populations which may limit the generalizability of the results. Since our analyses were based primarily on GWAS datasets from European populations, the findings may not be fully generalizable to other ancestries; future studies incorporating more diverse cohorts are needed to confirm the universality of these associations. Furthermore, although MR suggests a causal link between neuroimaging traits and PD risk, it does not directly reveal disease-specific pathways, and further mechanistic studies are warranted. Genetic backgrounds and neuroimaging features may vary across different races and populations, and to confirm the conclusions’ universality, future research should include a more varied population. Secondly, although Mendelian Randomization has advantages in causal inference, it relies on the validity of the instrumental variables. If the chosen genetic variants do not fully satisfy the 3 MR assumptions (relevance, independence, and exclusion), it could lead to bias. Although this study attempts to minimize this risk through various methods and sensitivity analyses, the results should still be interpreted with caution. Thirdly, the age of onset of PD may act as a potential source of bias in genetic associations with neuroimaging traits. Although GWAS summary statistics used in this study were adjusted for age, and MR leverages genetic variants that are fixed at conception (thus not influenced by disease progression or age), residual confounding related to onset age cannot be fully excluded. Future stratified analyses of early- vs late-onset PD may help clarify whether age at onset differentially impacts these causal relationships.

Moreover, the connection is the main subject of this investigation involving neuroimaging features and PD risk, without delving into their associations with PD’s clinical symptoms and disease progression. Future research could further explore the impact of neuroimaging changes on PD’s clinical manifestations and prognosis, to fully understand their role in PD.

In conclusion, this study through Two-Sample Mendelian Randomization confirms significant causal relationships between specific neuroimaging features and the risk of Parkinson’s Disease. These discoveries not only advance our knowledge of the processes behind Parkinson’s disease (PD), but they also provide fresh ideas for preventative and therapeutic approaches in the future. Nevertheless, further research is required to confirm and broaden these findings due to the study’s limitations. As multidisciplinary research advances, the combination of neuroimaging and genetics will provide a solid foundation for the precision medicine of PD.

Acknowledgments

We would like to acknowledge the dedicated efforts of all staff members involved in implementing the intervention and evaluation components of this study. Their contributions were invaluable to the successful completion of this work.

Author contributions

Data curation: Yuanru Wang, Shili Zhou.

Formal analysis: Yuanru Wang, Shili Zhou.

Supervision: Yuanru Wang.

Validation: Yuanru Wang.

Writing – original draft: Yuanru Wang, Shili Zhou.

Writing – review & editing: Shili Zhou.