Causal relationship between diabetes mellitus, glycemic traits and Parkinson’s disease: a multivariable mendelian randomization analysis

Diabetes mellitus (DM) and Parkinson’s disease (PD) are disorders associated with aging, and their prevalence is increasing worldwide. In the past few decades, the global number of adult diabetes patients has increased from 108 million in 1980 to 422 million in 2014 [1]. At the same time, the age-standardized prevalence of diabetes in men increased from 4.3 to 9.0% and in women from 5.0–7.9% [1]. It is estimated that by 2045, the number of diabetes patients will increase to 783 million [2]. Parkinson’s disease (PD) is also a rapidly developing neurodegenerative disease, with a global average prevalence of 1–2‰ [3]. With the exacerbation of population aging, the burden of PD will become even heavier [4]. According to statistics, from 1990 to 2016, the incidence, disability burden, and mortality related to Parkinson’s disease have more than doubled. Furthermore, a global survey of neurological diseases shows that PD may be the fastest-growing neurological disease globally [5]. In recent years, the role of DM in neurodegeneration has grown special interest not only as a contributing factor to disease onset but also as a modifying factor of motor and nonmotor symptoms [6].

Some epidemiological studies suggest an association between diabetes and PD, but the results are not entirely consistent with some positive correlation studies [7]. A recent meta-analysis included 15 cohort studies (including over 86,000 PD cases and nearly 30 million participants), reporting a 27% increased risk of PD in patients with diabetes [8]. An earlier meta-analysis included 7 cohort studies (including 1,761,632 patients) and found that the risk of PD in patients with diabetes also increased by 38% [9]. It is worth noting that the results of a few case-control studies suggest that diabetes may reduce the risk of PD [10, 11]. This difference may be attributed to heterogeneity, confounding factors, and biases between studies (such as inclusion and recall biases) [10]. Therefore, the causal relationship between diabetes and PD is still controversial.

Factors such as ethical and moral constraints, methodological confounding, and reverse causality contribute to the lack of high-quality randomized controlled trial (RCT) data in observational studies. However, Mendelian randomization (MR) provides a promising alternative. MR, which conceptually resembles a randomized controlled trial, is based on the principle of random allocation of genetic variations during meiosis. This random allocation makes genetic variations independent of many factors influencing observational studies. To investigate the causal relationship between genetic liability to diabetes and glycemic traits with Age at onset (AAO), risk of PD, and progression (UPDRS3/MMSE/MOCA), we conducted univariable Mendelian randomization (UVMR). UVMR allows us to examine the potential causal effects of genetic variations on these outcomes. Considering the everyday use of clinical anti-diabetic medications in diabetes management, we implemented multivariable Mendelian randomization (MVMR) to account for biases induced by the concomitant use of anti-diabetic drugs. This approach allows us to control for the potential confounding effects of these medications on the observed associations.

Univariate conventional MR analysis showed a correlation between the genetic prediction of T1DM and a reduced risk of PD (IVW OR = 0.9708; 95% CI: 0.9466, 0.9956; P = 0.0214) (see Figs. 2 and 3; Table 2, S3). The estimated associations from MR Egger and Weighted median analyses were consistent with the observed associations in the primary study, but the confidence intervals were often imprecise. It is worth noting that these sensitivity methods have lower statistical power than IVW because they rely on more stringent assumptions; thus, their results are expected to provide weaker statistical evidence but cannot offer effect sizes. There is no statistical evidence for an impact of T2DM on the risk of PD (IVW OR = 1.0292; 95% CI: 0.9714, 1.0905; P = 0.3284). Furthermore, there is no statistical evidence to suggest an association between diabetes, glycemic traits, and other phenotypes of PD.

We conducted a series of sensitivity tests to assess the accuracy of the optimistic estimates. These tests included Cochran’s Q-test, MR-Egger intercept, leave-one-out analysis, and funnel plot. The results of Cochran’s Q-test indicated no heterogeneity (P = 0.3968), suggesting that the studies included in our calculation were consistent. Additionally, the MR-Egger intercept test (P = 0.9704) did not detect potential horizontal pleiotropy, further supporting the reliability of our findings. Furthermore, the leave-one-out analysis results indicated that the causal effect was not driven by a single instrumental variable, suggesting that the observed association was robust. The symmetrical funnel plot also stated the results’ reliability, suggesting minimal publication bias. We conducted directionality checks using Steiger’s analysis to validate our findings further. These checks did not indicate a violation of the observed causal relationship, strengthening the evidence for our significant associations. Moreover, we performed reverse MR analysis to assess the influence of reverse causality on our results. The analysis showed that the results were unlikely to be influenced by reverse causality (IVW OR = 0.9347; 95% CI: 0.8657, 1.0092; P = 0.0844), providing additional support for the robustness of our findings(see Table S4).

In the context of MVMR, we evaluated the genetic risk of T1DM in combination with anti-diabetic drugs (see Table 3, 5). After adjusting for phenotypes related to anti-diabetic medications, such as drugs used in diabetes (IVW OR = 0.9812; 95% CI: 0.9324, 1.0325; P = 0.4740), diabetes, insulin treatment (IVW OR = 0.9822; 95% CI: 0.9463, 1.0194; P = 0.3380), and Metformin (IVW OR = 1.0000; 95% CI: 0.9825, 1.0178; P = 0.9930), the correlation between T1DM and PD risk was no longer significant. This suggests that the observed association between T1DM and PD risk may be confounded by the use of anti-diabetic drugs. The estimated associations from MR Egger and Weighted median analyses consistently aligned with the associations observed in IVW. Moreover, Cochran’s Q-test and MR-Egger intercept test did not reveal potential heterogeneity and pleiotropy, further supporting the robustness of our findings.

In this analysis, we have demonstrated the potential protective effect of T1DM on PD risk. Our MVMR analysis suggests that this observed causal relationship may be driven by drug-related features of specific anti-diabetic medications. We thoroughly examined the data using various sensitivity methods in the MR analysis and found no significant pleiotropy or heterogeneity. Moreover, no evidence supports a causal relationship between genetically predicted T2DM and PD. To delve deeper into the topic, we further analyzed the causal relationship between glycemic traits and PD. However, the results of this analysis do not support a causal relationship between the two.

There has been a long-standing controversy regarding the association between DM and PD. Epidemiological evidence suggests an association between DM and PD, but the results are inconsistent, ranging from significant negative correlations to significant positive correlations [35‐40]. Biological evidence demonstrates that both conditions are characterized by abnormal protein accumulation, lysosomal and mitochondrial dysfunction, and chronic systemic inflammation [41, 42]. Moreover, hypoinsulinemia in T1DM patients or insulin resistance (IR) in T2D patients leads to hyperglycemia, exposing neurons to increased metabolic stress, neuronal dysfunction, and death, thereby directly contributing to the development of PD [43]. Furthermore, several anti-diabetic drugs have been shown to have anti-PD effects, such as DPP-4 inhibitors and GLP-1 receptor agonists [44‐46]. However, these studies often have relatively small sample sizes, which may introduce confounding, selection bias, and reverse causality, further limiting the interpretability of the results [47]. Additionally, case-control studies do not adequately address the temporal relationship between diabetes and PD since they rely on retrospective data and often fail to specify the time window for exposure assessment. Although large-scale prospective studies hold promise in overcoming these limitations, conducting such research requires significant human, financial, and time resources.

Although clinical trials have various limitations, early identification of risk factors for PD is crucial. Early intervention targeting relevant risk factors is currently the most effective approach to delay or prevent the onset of PD [48]. However, there is currently no effective cure once PD occurs. Compared to traditional epidemiology, MR analysis reveals the causal relationship between DM and PD cost-effectively, reducing confounding biases in epidemiological studies, including reverse causation [49, 50]. Three Mendelian randomization studies have recently been reported, investigating the causal inference of DM on PD in different populations. Chohan et al.‘s MR study on the European population reveals that genetically predicted T2DM leads to an increased risk and faster progression of PD, particularly in motor impairment [51]. Park et al.‘s MR study based on the Korean (East Asian) population suggests no evidence of a causal association between T2DM and PD. The authors explain this seemingly contradictory result as being due to a small sample size and ethnic differences [51, 52]. Additionally, Senkevich et al.‘s MR study on the European population suggests a potential protective association between genetically predicted T1DM and the risk and progression of PD, possibly driven by latent pleiotropy [53].

There is ongoing controversy regarding the relationship between DM and PD; given the complex association and significant clinical implications between the two, it is imperative to robustly replicate this association in larger GWAS study cohorts and explore potential underlying mechanisms. Consistent with the findings of Senkevich et al., our results confirm the causal relationship between T1DM and reduced risk of PD, and we further discovered that the use of anti-diabetic medications may mediate this causal relationship. Some traditional epidemiological approaches have also reported a lower risk of PD incidence in DM patients [37, 38, 54]. It has been reported that long-term use of anti-diabetic medications such as GLP-2 receptor agonists and DPP1 inhibitors may potentially reduce the risk of PD [45]. In recent years, an increasing body of research evidence supports the potential of anti-diabetic medications in reducing the risk of PD [55]. Using commonly used anti-diabetic drugs targeting the insulin signaling pathway has induced neuroprotective effects in preclinical studies and clinical trials. A longitudinal study of 5,528 veterans with T2DM showed that treatment with metformin for more than four years can reduce the risk of AD and PD [56]. The neuroprotective effect of metformin is mediated through the regulation of AMP-activated protein kinase (AMPK) activity, which modulates several critical cellular processes such as autophagy, cell growth, and mitochondrial function, as well as inhibiting microglial activation and inflammation [57‐60]. Some studies have explored the neuroprotective potential of intranasal insulin. Preclinical data indicate that intranasal delivery of recombinant human insulin can reach deep brain structures, including the hippocampus and nigrostriatal pathway [61]. The study by Novak et al. showed that intranasal short-acting (regular) insulin treatment improved motor performance and function compared to placebo, resulting in lower disability scores (HY scale) and improved UPDRS motor scores compared to placebo [46].

Furthermore, other drugs, such as glucagon-like peptide 1 (GLP-1) agonists, can provide neuroprotection. Liraglutide and lixisenatide, both GLP-1 analogs, have been shown to induce neuroprotection in PD animal models [62]. These drugs can cross the blood-brain barrier (BBB), enhance hippocampal neurogenesis, and increase brain-derived neurotrophic factor (BDNF) expression, promoting neuroprotection in AD and PD [63, 64].

Our study highlights the potential protective effect of genetic prediction of T1DM on PD, suggesting that anti-diabetic drugs may play a crucial role in reducing PD risk. However, the exact mechanism underlying this protective effect remains unclear. Therefore, it is necessary to gather further direct evidence to validate our findings and develop effective PD prevention and management strategies.

We want to acknowledge certain limitations in our study. Firstly, it is essential to note that the associations observed through MR analysis do not provide information about temporal patterns but rather reflect lifelong effects on specific risk factors. Secondly, the sample size used for analyzing PD progression (UPDRS3/MMSE/MOCA) is relatively small, which may reduce the analytical power and potentially lead to false-negative results. Conducting larger-scale MR analyses will be essential to ensure the robustness of our findings. Additionally, it should be considered that genetic variations associated with T1DM may be correlated with multiple factors, which could represent alternative pathways through which these genetic variations influence PD. This potential horizontal pleiotropy should be taken into account when interpreting our results. Lastly, it is worth mentioning that our study primarily focuses on individuals of European ancestry. Further research is needed to determine whether our findings can be generalized to other ethnicities.

In summary, our study discovered a direct causal relationship between genetic predictions of T1DM and a decreased risk of PD in individuals of European ancestry. Moreover, there is indirect evidence indicating that anti-diabetic drugs may mediate the protective effect of T1DM against PD. However, further research is needed to fully understand the mechanisms by which anti-diabetic drugs exert their anti-PD effects and to identify potential therapeutic targets.