Association between genetically proxied PPARG activation and early onset Alzheimer's disease: A drug target Mendelian randomization study

Abstract

Background:

Early-onset Alzheimer's disease (EOAD), which has an onset before the age of 65, is overshadowed by the more prevalent late-onset Alzheimer's disease (LOAD). Nevertheless, rather than being merely a form of LOAD that occurs at a prematurely defined younger age, EOAD differs from LOAD in multiple ways. Given these disparities, understanding the potential treatment options for EOAD becomes crucial.

Objective:

We aim to assess anti-diabetic drugs’ potential utility in treating EOAD and LOAD from a novel perspective via drug-targeted Mendelian randomization (MR) analysis.

Methods:

Through Summary-data-based MR (SMR) and inverse-variance-weighted MR (IVW-MR) analysis, we assessed the associations between anti-diabetic drug targets (including DPP-4 inhibitor, Thiazolidinedione, GLP1R agonist, Sulfonylureas, SLC5A2 inhibitor, and Insulin/Insulin analog) and AD outcomes (EOAD and LOAD). We utilized two types of genetic instruments to represent the exposure to anti-diabetic drugs: eQTLs of genes encoding drug target proteins and genetic variants within or near genes encoding these target proteins associated with HbA1c from genome-wide association studies.

Results:

SMR analysis showed that enhanced PPARG gene expression in the blood was a protective factor for EOAD (OR = 0.733, 95%CI = 0.548–0.979, p = 0.035). Additionally, an IVW-MR association was found between HbA1c mediated by PPARG and EOAD (OR = 0.295, 95%CI = 0.092–0.949, p = 0.041).

Conclusions:

This study suggests that Thiazolidinedione therapy could help suppress the development of EOAD, supporting further exploration of PPARG-targeted anti-diabetic drug development.

Keywords

Alzheimer's disease anti-diabetic drugs early onset Alzheimer's disease Mendelian randomization analysis thiazolidinedione

Introduction

Alzheimer's disease (AD) is mainly categorized into two major types: early-onset AD (EOAD) and late-onset AD (LOAD). EOAD makes up 5% to 10% of all AD cases, and people diagnosed with it usually fall within the age range of 30 to 65 years old.¹ In contrast, LOAD, the most common subtype of AD, impacts individuals aged 65 years and older. Rather than being merely a form of LOAD at an arbitrarily defined younger age threshold, EOAD differs from LOAD in multiple ways. Firstly, over 90% of AD patients are affected by sporadic LOAD, whereas more than 60% of EOAD cases involve multiple family members with AD.¹ Secondly, EOAD diagnosis requires a more comprehensive evaluation than LOAD.² Thirdly, dementia risk factors like reduced cardio - and cognitive fitness significantly impact EOAD.² Also, the younger the age of dementia onset, the more significant the potential consequences of traumatic brain injury can be for EOAD.³ Moreover, there are unique psychosocial issues related to early-onset dementia.^4–6 Finally, compared with LOAD patients, EOAD patients typically have fewer overall comorbidities, such as diabetes, obesity, and circulatory disorders.⁷ Unfortunately, there are currently very few authorized agents or medications for the treatment of AD and even fewer that can alleviate symptoms. Regardless, no clear evidence exists that these drugs have neuroprotective benefits.⁸ Consequently, there is an urgent need for research into new disease mechanisms that could lead to improved therapeutic outcomes.

Multiple studies have established type II diabetes mellitus (T2DM) as a risk factor for AD.^9–12 As early as 50 years ago, signs suggesting a causal connection between impaired brain metabolism and the development of dementia were already noted.^13,14 Brain imaging techniques and biomarker analysis can further identify the root causes of cognitive symptoms. Glucose (fluorodeoxyglucose) positron emission tomography (PET) scans are frequently employed to assess brain function (metabolism), and such measurements often reveal abnormalities in individuals with EOAD.¹⁵ Furthermore, in diabetes, high glucose and insulin levels drive insulin desensitization. Intriguingly, insulin desensitization has also been observed in the brains of AD patients, even among those without diabetes.¹⁶ This finding aligns with emerging research that has associated AD with common phenomena in T2DM, including insulin resistance, malfunctioning insulin signaling, neuroinflammation, advanced glycosylation end products (AGEs), and metabolic dysfunction.¹⁷ This association has led to AD being referred to as “type 3 diabetes”.¹⁸

Based on their original intention of enhancing insulin signaling and regulating glucose metabolism, anti-diabetic drugs have been highlighted as repurposing candidates for AD.¹⁹ Numerous randomized clinical trials (RCTs) have been carried out to explore the disease-modifying potential of these drugs in AD patients. However, thus far, the evidence gathered remains inconclusive.²⁰ When seeking to establish a correlation between various phenotypes and diseases, one analytical tool is Mendelian randomization (MR).²¹ This method can predict adverse drug reactions and presents opportunities for drug repurposing.²² When it comes to drug targets, MR can effectively reflect the effects of drug use.²³

While prior Mendelian MR studies have investigated antidiabetic drugs and AD,^24,25 these studies have treated AD as a homogeneous phenotype, failing to distinguish EOAD from LOAD subtypes. Another study has explored associations between antidiabetic drug targets and LOAD, demonstrating that GLP1R agonism may reduce LOAD risk.²⁶ However, this study has primarily focused on LOAD, overlooking EOAD – a distinct subtype with unique genetic foundations. In this study, we conducted a drug-target MR study to fill the existing knowledge gap and expand our understanding of anti-diabetic medication treatment for AD. We employed single nucleotide polymorphisms (SNPs) within or near the target as instrumental variables(IVs) to delve into the causal associations between anti-diabetic medications (Dipeptidyl peptidase 4 (DPP-4) inhibitors, Thiazolidinediones, Glucagon-like peptide 1 receptor (GLP-1R) agonists, Sulfonylurea, Sodium-glucose cotransporter 2 (SLC5A2) inhibitors, and Insulin/Insulin analog) and AD outcomes, encompassing EOAD and LOAD. The results of this study set the stage for a more comprehensive exploration of anti-diabetic drugs in AD treatment, aiming to uncover novel potential therapeutic strategies.

Methods

Study design

In the current study, we performed SMR and IVW-MR based on expression quantitative trait loci studies (eQTLs) of blood and publicly available summary-level data from genome-wide association studies (GWAS) for HbA1c, respectively, to estimate these genetic variants on EOAD and LOAD. Three key MR model assumptions are crucial for valid causal estimation: (1) A strong instrumental variables (IVs)-target genes association (relevance) is needed. IVs should closely relate to target genes to effectively proxy the exposure; (2) IVs must be independent of confounders (exchangeability). That is, no confounding factors should affect IVs, enabling unbiased estimation; (3) IVs should have no direct effect on AD risk except via drug targets (exclusion restriction), ensuring the IV-drug targets solely mediate AD risk effect for stronger causal inference. A summary of the study design is shown in Figure 1.

Figure 1.

Flowchart of study design.

Selection of genetic instruments

We employed the DrugBank and ChEMBL databases to discover genes that encode the target proteins of these anti-diabetic medications.^27,28 Initially, seven classes of anti-diabetic drugs were recognized, namely Metformin, DPP-4 inhibitor, Thiazolidinedione, GLP-1R agonist, Sulfonylurea, SLC5A2 inhibitor, and Insulin/Insulin analog.²⁹ However, because the gene targets of Metformin varied between the two databases and the molecular mechanisms underlying Metformin's physiological effects remain mostly unclear,³⁰ Metformin was excluded from subsequent analyses. As shown in Table 1.

Table 1.

Target genes of anti-diabetic drugs from DrugBank and ChEMBL databases.

Drug Class	Encoding genes of target proteins		Gene location (GRCh37)	Included in analysis
Drug Class	DrugBank	ChEMBL	Gene location (GRCh37)	Included in analysis
Metformin	PRKAB1	58 encoding genes	NA	No
Metformin	ETFDH	GPD2	NA	No
DPP-4 inhibitor	DPP4	DPP4	chr2: 162848755 .. 162930904	Yes
Thiazolidinedione	PPARG	PPARG	chr3: 12328867..12475855	Yes
GLP1R agonist	GLP1R	GLP1R	chr6: 39016557..39059079	Yes
Sulfonyluresa	KCNJ11	KCNJ11	chr11: 17386719..17410878	Yes
Sulfonyluresa	ABCC8	ABCC8	chr11: 17414045..17498441	Yes
SLC5A2 inhibitor	SLC5A2	SLC5A2	chr16: 31494323 .. 31502181	Yes
Insulin and its analogue	INSR	INSR	chr19: 7112266..7294425	Yes

The eQTLs summary-level data were obtained from the eQTLGen Consortium, the details of which are presented in Supplemental Table 1. We identified common (MAF > 1%) eQTLs significantly (p < 5.0 × 10⁻⁸) associated with the expression of DPP-4, PPARG, GLP-1R, KCNJ11, SLC5A2 and INSR in blood. The expression of ABCC8 was excluded due to the absence of eQTLs in blood or other tissues at a significant level. This study used only cis-eQTLs (eQTLs inside 1 Mb of an encoded gene) to generate genetic instruments. As shown in Table 2.

Table 2.

Information of genetic instruments.

	Genetic instruments
Exposure	Genetic variants (MAF > 1%, p < 5e-8) associated with mRNA expression levels (eQTLs) in blood of eQTLGen	Genetic variants (MAF > 1%) in low linkage disequilibrium (r² < 0.10), associated with LDL cholesterol (p < 5e-8), located within ±100kb windows from lipids-related genes region
DPP-4	top cis-SNP: rs13015258	NA (excluded due to <3 instruments)
PPARG	top cis-SNP: rs1699346	rs4493414, rs112496224, rs4135286, rs1822534, rs28595399, rs2920502, rs568984015, rs2921186, rs709159, rs66603943, rs17671592
GLP1R	top cis-SNP: rs9283907	rs10305518 ((excluded due to <3 instruments)
KCNJ11/ ABCC8	top cis-SNP: rs2074310	rs757110 (excluded due to <3 instruments)
SLC5A2	top cis-SNP: rs6565236	rs45625038, rs12932429, rs8050500, rs28675289, rs28692853
INSR	top cis-SNP: rs9283907	NA (excluded due to <3 instruments)

We selected common SNPs (MAF > 1%) from GWAS summary data on glycated hemoglobin (HbA1c) levels from the UK Biobank (n = 389,889). Genetic variants were selected within the target gene (±100 kb windows) of each drug associated with HbA1C at a genome-wide significance level (p < 5.0 × 10⁻⁸) to proxy exposure to anti-diabetic drugs. To minimize linkage disequilibrium (LD), we applied LD clumping with a threshold of r²< 0.10, retaining only nearly independent SNPs. Subsequently, to ensure robust causal inference, we excluded targets with <3 independent SNPs (after LD clumping, r²<0.10), following the 2018 BMJ guide, which advises ≥3 instruments to enable pleiotropy detection and minimize weak tool bias.³¹ Valid SNPs were identified for PPARG and SLC5A2, as detailed in Table 2.

Outcome sources

GWAS summary data related to AD outcomes were obtained from the FinnGen consortium. For EOAD, the sample size was 201,078 individuals, and for LOAD, it was 207,746. Given that PPARG is a key target of anti-diabetic drugs and plays a vital role in metabolic regulation, we conducted positive control studies to verify the reliability of our IVs. We deliberately selected two positive control outcomes closely linked to PPARG's known functions. Positive Control Outcome 1: waist circumference, sourced from the UK Biobank (n = 407,661) and the GIANT consortium (n = 127,469), is a well-recognized indicator of adiposity. PPARG is known to influence fat distribution and adipocyte function, and an association between PPARG-related gene targets and waist circumference in our MR analysis would validate the proper functioning of our IVs. Positive Control Outcome 2: HDL cholesterol, sourced from the UK Biobank (n = 315,133) and the Within Family GWAS consortium (n = 37,120), is another metabolic trait affected by PPARG. PPARG activation can increase HDL cholesterol levels through various biological pathways. Demonstrating an MR association between PPARG gene targets and HDL cholesterol serves as a crucial check for the reliability of our IVs. More details of these GWAS datasets can be found in Supplemental Table 1.

Statistical analyses

Primary MR analysis. Effect estimates were generated using the SMR method to investigate the association between expression of genes encoding drug target proteins and outcome of interest using GWAS summary data and eQTL studies.³² We utilized the fixed-effects inverse-variance weighted (IVW) method to integrate effect estimates from genetic variations linked to LDL cholesterol levels. Allele harmonization and analysis were performed using the SMR software (version 1.31) and the TwoSampleMR package in R software (version 4.4.1).

Sensitivity analysis. The SMR method employed the heterogeneity in dependent instruments (HEIDI) test to determine whether the gene expression-outcome connection was due to a linkage situation in the SMR software (version 1.31).³² Heterogeneity was signified by p < 0.01, indicating that a single SNP may correlate with the expression of numerous genes, leading to horizontal pleiotropy.

The IVW-MR approach assessed heterogeneity in the Cochran Q test, with p < 0.05 signifying evidence of heterogeneity. MR-Egger regression and MR-PRESSO analysis were applied to evaluate the possible horizontal pleiotropy of SNPs as IVs. In MR Egger regression, the intercept term indicates directional horizontal pleiotropy, with p < 0.05 suggesting evidence of horizontal pleiotropy. All analyses were performed with the R program (4.2.3).

MR may yield false-positive associations between exposures and outcomes because of genetic confounding. Specifically, a genetic variant influences the outcome through a non-glucose metabolism pathway strongly correlated with the actual causal variant. We additionally carried out colocalization analysis, which is meticulously detailed as a sensitivity analysis within the MR context.³³ Based on these five hypotheses,³⁴ a colocalization analysis was conducted: Hypothesis 0 (H0): Neither trait has a causal genetic instrument; Hypothesis 1 (H1): There is one causal genetic instrument for trait 1; Hypothesis 2 (H2): One causal genetic instrument for trait 2 exists; Hypothesis 3 (H3): Both traits have a causal variant in the region, yet these are distinct; Hypothesis 4 (H4): One shared causal genetic instrument for both traits exists. The outputs of interest are the posterior probability of different causal variants (H3) and shared causal variants (H4). H4/(H3 + H4) represents the probability that the exposure and the outcome share a causal variant when hypotheses H0 – H2 are excluded (i.e., when at least one causal variant exists for the outcome), as described in previous studies.^35,36 Colocalization analyses were performed using the coloc package, assuming uniform prior probabilities (π0 = π1 = π2 = π3 = π4 = 0.2) for the five hypotheses, as recommended by the method's standard implementation.

Results

Selection of instrumental variables

For SMR analysis, Cis-eQTLs were identified from eQTLGen for drugs target gene DPP-4, PPARG, GLP1R, KCNJ11, SLC5A2, and INSR, respectively, and the most significant cis-eQTL SNP was selected as a genetic instrument for the target gene of each drug (Table 2 and Supplemental Table 2).

For IVW-MR analysis, we extracted genetic variants from the target gene region and applied LD clumping with a threshold of r² < 0.10 to select nearly independent variants as genetic instruments. Initially, 11, 1, 1, and 5 SNPs within or near PPARG, GLP1R, KCNJ11/ABCC8, and SLC5A2 genes were identified from HbA1c GWAS summary data (Table 2 and Supplemental Table 3). Targets with <3 IVs (GLP1R, KCNJ11/ABCC8) were excluded to ensure robust statistical power, as MR guidelines recommend ≥3 independent instruments to detect pleiotropy and minimize weak tool bias.³¹ No valid SNPs were identified for DPP-4 inhibitor or INSR/insulin analogs at genome-wide significance (Table 2 and Supplemental Table 3). Finally, the validated genetic instruments included 11 SNPs for PPARG (proxying Thiazolidinedione) and 5 SNPs for SLC5A2 (proxying SLC5A2 inhibitor), as shown in Table 2.

Primary analysis

In Figure 2 and Supplemental Table 2, SMR analysis revealed that genetically predicted enhanced PPARG gene expression in blood was a protective factor for EOAD (OR = 0.733, 95% CI: 0.548–0.979, p = 0.035). This association aligns with PPARG's known metabolic and neuroprotective roles.^37,38 In Figure 3 and Supplemental Table 4, IVW-MR analysis demonstrated a significant protective association between PPARG-mediated HbA1c levels. It reduced EOAD risk (OR = 0.295, 95% CI: 0.092–0.949, p = 0.041), consistent with the SMR findings.

Figure 2.

Forest plot of SMR analysis examining the association between drug – related genes (DPPIV, PPARG, GLP1R, KCNJ11, SLC5A2, INSR) and AD subtypes (early-onset AD, late-onset AD). Each row represents a gene-AD subtype pair. The dot indicates the OR, the horizontal line denotes the 95% CI, and the p-value for each analysis is provided. AD: Alzheimer's disease; SMR: summary-data-based Mendelian randomization; OR: odds ratio; 95%CI: 95% confidence interval.

Figure 3.

Forest plot of IVW-MR analysis for the association between genes (PPARG, SLC5A2) and AD subtypes (early-onset AD, late-onset AD). Each row shows a gene-AD subtype combination with its corresponding OR (dot), 95% CI (horizontal line), and p-value. AD: Alzheimer's disease; IVW-MR: inverse-variance weighted Mendelian randomization; OR: odds ratio; 95%CI: 95% confidence interval.

The positive control analyses demonstrated significant associations. PPARG genetic instruments (eQTLs) showed a strong correlation with HDL cholesterol as per UK Biobank data and waist circumference, also from UK Biobank data (refer to Supplemental Table 5). Similarly, when using HbA1c levels GWAS – proposed IVs, drug-exposed PPARG was associated with HDL cholesterol as indicated by the Within Family GWAS consortium and waist circumference from UK Biobank data (as detailed in Supplemental Table 6). These associations validate the role of these instruments in adiposity regulation, which is a well-established, PPARG-mediated mechanism.

Sensitivity analysis

For SMR analysis in Supplemental Table 2, the HEIDI test suggested that none of the observed associations were attributable to a linkage (p > 0.01). For IVW-MR sensitivity analysis in Supplemental Table 4, Cochran's Q, MR-Egger regression and MR-PRESSO global tests detected no heterogeneity and horizontal pleiotropy.

For colocalization analysis, conditional on the presence of a causal variant for the outcome, the posterior probability of colocalization between eQTL PPARG and EOAD in the gene region was 78.2% (Supplemental Table 7). For HbA1c and EOAD in the PPARG gene region, it was 98.6% (Supplemental Table 8).

Discussion

We conducted drug target MR analyses on AD outcomes using GWAS data from the FinnGen Consortium. Our study investigated the causal relationship of six common antidiabetic drug targets – DPP-4 inhibitor, Thiazolidinedione, GLP1R agonist, Sulfonylurea, SLC5A2 inhibitor, and Insulin/Insulin analogue – on EOAD and LOAD. Notably, Thiazolidinedione emerged as a robust protective factor for EOAD (SMR: OR = 0.733, p = 0.035; IVW-MR: OR = 0.295, p = 0.041), with no effect on LOAD. This EOAD-specific benefit aligns with PPARG's multifunctional role in insulin sensitization, Aβ clearance, and anti-inflammation. Specifically, PPARG activation reduces IRS-1 serine phosphorylation to restore AKT/GSK3β insulin signalling³⁷ and upregulates insulin-degrading enzyme (IDE) to enhance Aβ metabolism,³⁸ mechanisms directly targeting EOAD's Aβ-overproduction pathology. The PPARG-EOAD association was supported by convergent genetic evidence from SMR and IVW-MR analyses, which both demonstrated consistent protective effects across independent genetic instruments.

Previous retrospective studies have indicated that EOAD and LOAD are different. This difference is particularly evident in their genetic foundations, crucial in differentiating the two subtypes. Three specific gene variants, namely those in the Amyloid Precursor Protein (APP), Presenilin-1 (PSEN1), and Presenilin-2 (PSEN2) genes, are known to trigger EOAD.³⁹ APP gene variants account for 10–15% of EOAD cases.^1,40 A significant aspect is that most APP mutations occur near the γ-secretase cleavage site. This proximity is associated with elevated levels of Aβ_42.^1,41 PSEN1 encodes Presenilin-1, the proteolytic subunit of γ-secretase, and mutations in PSEN1 enhance the activity of γ-secretase.³⁹ This enhancement results in an increased production of Aβ₄₂,⁴² ultimately leading to the most severe forms of AD.¹ Shifting the focus to PSEN2, its variants can also elevate γ-secretase activity and boost the production of Aβ₄₂.³⁹ Nevertheless, in contrast to PSEN1 variants, PSEN2 variants are less common and are linked to a less aggressive disease progression.⁴³ For many years, research in this field was predominantly guided by the amyloid hypothesis.⁴⁴ Initially, this hypothesis emphasized Aβ aggregation as the root cause of AD. Over time, however, the amyloid hypothesis has been refined. Current understanding now identifies soluble Aβ oligomers, rather than Aβ plaques, as the primary toxic amyloid species.⁴⁵ So far, among the various forms of amyloid, small oligomers composed of the aggregation-prone Aβ₄₂ have been identified as the most harmful.⁴⁶ Even as the amyloid hypothesis develops, the search for effective therapies remains important and imperative.

Notably, Thiazolidinedione emerge as a promising therapeutic strategy in this context. These agents disrupt the insulin resistance-inflammation synergy, a key pathological axis in AD,⁴⁷ providing a robust mechanistic foundation for their protective effects against the disease. Thiazolidinedione activate PPARG, driving the redistribution of visceral fat to subcutaneous tissues, reducing the secretion of pro-inflammatory adipokines.^48,49 Concurrently, PPARG-mediated IRS-1 dephosphorylation restores AKT/GSK3β signalling,⁵⁰ a pathway critical for neuronal glucose uptake and Aβ clearance. Notably, IDE upregulation by PPARG³⁸ has been shown in transgenic mice to reduce brain Aβ₄₂ levels,⁵¹ directly addressing the genetic Aβ burden in EOAD (e.g., APP/PSEN1 mutations). Reduced systemic inflammation also mitigates microglial dysfunction, enabling more efficient phagocytosis of Aβ plaques and suppressing toxic oligomer formation in the central nervous system.⁵² Further research on Thiazolidinedione has shown improvements in cognitive performance in transgenic animal models of AD marked by Aβ₄₂ overexpression.^51,53 This underlying mechanism might partially account for the protective effects of Thiazolidinedione against the development of EOAD.

In contrast to EOAD, which is characterized by an overproduction of Aβ, LOAD typically exhibits a general deficiency in Aβ clearance.⁵⁴ As individuals age, the catabolism and clearance of Aβ decline through multiple mechanisms. One such mechanism involves reducing the enzymatic activity of crucial Aβ-degrading enzymes. For example, neprilysin activity decreases,^55,56 leading to an age-related decline in Aβ clearance.⁵⁴ Another contributing trigger is a series of AD risk factors, including immune system dysregulation, mitochondrial dysfunction accompanied by oxidative stress,^57,58 metabolic abnormalities often associated with diabetes,⁵⁹ cholesterol homeostasis disruption,⁶⁰ and deregulation of cell cycle pacing mechanisms.⁶¹ All these factors can negatively impact Aβ catabolism. Clinical trials of Thiazolidinediones (including Rosiglitazone and Pioglitazone) involving people over 65 were also unpromising. For Rosiglitazone, A Phase 2 clinical trial (NCT00265148) investigating Rosiglitazone found that the drug had no statistically significant impact on outcome measures.⁶² Moreover, two Phase 3 clinical trials (NCT00428090, NCT00348309, and NCT00348140) using Rosiglitazone yielded less promising results.^63,64 Regarding Pioglitazone, neither of the two Phase 2 clinical trials (NCT00982202 and NCT00736996) showed a statistically significant association with any improvement in cognitive measures. Additionally, a Phase 3 clinical study (NCT01931566) of Pioglitazone did not succeed in delaying the development of moderate cognitive impairment.⁶⁵ These findings suggest that Thiazolidinedione may not be able to restore the age-related decline in the Aβ clearance system in LOAD.

This MR study has several notable strengths. For instance, it explores the genetically mimicked preventive effects of Thiazolidinedione on EOAD compared to other currently utilized anti-diabetic drugs. Nevertheless, this study also has certain limitations. Firstly, the genetically predicted drug effects may deviate from real-world therapeutic practices. Because genetic variants that serve as exposure instruments are present from birth and persist throughout an individual's lifetime. Our analyses, therefore, assess the long-term modulation effects of drug target proteins and cannot reflect the impacts of anti-diabetic drug exposure during particular life stages. Secondly, eQTLs in our analysis have inherent limitations, primarily due to tissue specificity and pleiotropic effects. While widely available, eQTLs often reflect gene expression regulation in non-target tissues (e.g., peripheral blood instead of relevant metabolic or neural tissues), potentially masking actual target tissue effects and introducing bias.^66,67 Additionally, genetic variants tagging eQTLs can influence multiple genes concurrently. Specifically, pleiotropy can distort the results in similar analyses.⁶⁸ Thirdly, despite using a rigorous LD threshold (r²< 0.1), the choice of threshold introduces subjectivity, and no single cutoff eliminates the trade-off between type 1 error risk and statistical power. Conversely, our strict threshold excluded these drug targets due to insufficient independent IVs, potentially limiting the comprehensiveness of our findings. Fourthly, metformin was excluded from all analyses due to its unclear mechanism of action. In addition, DPP-4 inhibitors, GLP1R agonist, Sulfonylurea and insulin/insulin analogue were not included in the IVW-MR analysis because suitable IVs were lacking. Therefore, future research is needed to re-evaluate these drug classes when more or more powerful IVs become available. Fifthly, the confidence intervals (CIs) of our key findings were relatively wide, particularly in the IVW-MR analysis (e.g., OR = 0.295, 95% CI = 0.092–0.949 for PPARG-mediated HbA1c and EOAD) reflecting estimation uncertainty and the need for a larger cohort validation. Critically, even minor deviations from MR assumptions 2 (instrument-exposure independence) or 3 (no direct effect of instruments on AD risk) could attenuate these associations toward the null.⁶⁹ For example, undetected pleiotropic effects of genetic variants on both HbA1c and AD through non-PPARG pathways might bias results. Although sensitivity analyses (HEIDI test, MR-Egger regression) did not indicate significant horizontal pleiotropy, the wide CIs underscore the potential fragility of causal inferences under relaxed MR assumptions. Finally, since we exclusively utilized GWAS data limited to the European ancestry population, the generalizability of our results is constrained to populations of European descent. Although this study has these constraints, it offers valuable insights and recommendations for future research.

In conclusion, this study provides evidence that Thiazolidinedione might have a preventive effect against EOAD. Further clinical trials are required to understand better the probable molecular pathways that link Thiazolidinedione with EOAD. In addition, it is essential to optimize the rational application of anti-diabetic medications.

Supplemental Material

sj-docx-1-alr-10.1177_25424823251359579 - Supplemental material for Association between genetically proxied PPARG activation and early onset Alzheimer's disease: A drug target Mendelian randomization study

Supplemental material, sj-docx-1-alr-10.1177_25424823251359579 for Association between genetically proxied PPARG activation and early onset Alzheimer's disease: A drug target Mendelian randomization study by Lingyu Zhang, Yi Wu, Qianqian Jin and Chunfang Wang in Journal of Alzheimer's Disease Reports

Supplemental Material

sj-xlsx-2-alr-10.1177_25424823251359579 - Supplemental material for Association between genetically proxied PPARG activation and early onset Alzheimer's disease: A drug target Mendelian randomization study

Supplemental material, sj-xlsx-2-alr-10.1177_25424823251359579 for Association between genetically proxied PPARG activation and early onset Alzheimer's disease: A drug target Mendelian randomization study by Lingyu Zhang, Yi Wu, Qianqian Jin and Chunfang Wang in Journal of Alzheimer's Disease Reports

Footnotes

Acknowledgements

The authors have no acknowledgments to report.

ORCID iD

Lingyu Zhang

Author contributions

Lingyu Zhang: Funding acquisition; Software; Supervision; Visualization; Writing – original draft; Writing – review & editing.

Yi Wu: Software.

Qianqian Jin: Funding acquisition.

Chunfang Wang: Supervision; Writing – review & editing.

Funding

This study was supported by the Provincial Doctoral Funding Program at Shanxi Medical University (Grant No. SD2109), Doctoral Funding Program at Shanxi Medical University (Grant No. XD2143), the Innovation and Technology Projects for Higher Education Institutions in Shanxi Province (Grant No.202122L127), and the Fundamental Research Program of Shanxi Province (202203021221191).

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data availability statement

The GWAS Summary statistics used in this study were publicly accessed from the IEU OpenGWAS project (https://gwas.mrcieu.ac.uk/), FinnGen (https://www.finngen.fi/en), GWAS Catalog (https://www.ebi.ac.uk/gwas/home), the eQTLGen Consortium (https://eqtlgen.org/) and the GTEx Consortium ().

Supplemental material

Supplemental material for this article is available online.

References

Bekris

Bird

, et al. Genetics of Alzheimer disease. J Geriatr Psychiatry Neurol 2010; 23: 213–227.

Eriksson

Fereshtehnejad

Falahati

, et al. Differences in routine clinical practice between early and late onset Alzheimer's disease: data from the Swedish Dementia Registry (SveDem). J Alzheimers Dis 2014; 41: 411–419.

Mendez

Paholpak

Lin

, et al. Prevalence of traumatic brain injury in early versus Late-Onset Alzheimer's disease. J Alzheimers Dis 2015; 47: 985–993.

Clemerson

Walsh

Isaac

. Towards living well with young onset dementia: an exploration of coping from the perspective of those diagnosed. Dementia (London) 2014; 13: 451–466.

Kaiser

Melrose

Liu

, et al. Neuropsychological and neuroimaging markers in early versus late-onset Alzheimer's disease. Am J Alzheimers Dis Other Demen 2012; 27: 520–529.

Ducharme

Kergoat

Antoine

, et al. The unique experience of spouses in early-onset dementia. Am J Alzheimers Dis Other Demen 2013; 28: 634–641.

Gerritsen

Bakker

Verhey

, et al. Prevalence of comorbidity in patients with Young-Onset Alzheimer disease compared with Late-Onset: a comparative cohort study. J Am Med Dir Assoc 2016; 17: 318–323.

Mangialasche

Solomon

Winblad

, et al. Alzheimer's disease: clinical trials and drug development. Lancet Neurol 2010; 9: 702–716.

Luchsinger

Tang

Shea

, et al. Hyperinsulinemia and risk of Alzheimer disease. Neurology 2004; 63: 1187–1192.

10.

Craft

. Insulin resistance and Alzheimer's disease pathogenesis: potential mechanisms and implications for treatment. Curr Alzheimer Res 2007; 4: 147–152.

11.

Haan

. Therapy Insight: type 2 diabetes mellitus and the risk of late-onset Alzheimer's disease. Nat Clin Pract Neurol 2006; 2: 159–166.

12.

Ohara

Doi

Ninomiya

, et al. Glucose tolerance status and risk of dementia in the community: the Hisayama study. Neurology 2011; 77: 1126–1134.

13.

Hoyer

Oesterreich

Wagner

. Glucose metabolism as the site of the primary abnormality in early-onset dementia of Alzheimer type? J Neurol 1988; 235: 143–148.

14.

Hoyer

Oesterreich

. Blood supply and metabolism of the brain in substantial traumatic brain damage. Nervenarzt 1971; 42: 180–185.

15.

Graff-Radford

Yong

Apostolova

, et al. New insights into atypical Alzheimer's disease in the era of biomarkers. Lancet Neurol 2021; 20: 222–234.

16.

Talbot

Wang

Kazi

, et al. Demonstrated brain insulin resistance in Alzheimer's disease patients is associated with IGF-1 resistance, IRS-1 dysregulation, and cognitive decline. J Clin Invest 2012; 122: 1316–1338.

17.

Michailidis

Moraitou

Tata

, et al. Alzheimer's disease as type 3 diabetes: common pathophysiological mechanisms between Alzheimer's disease and type 2 diabetes. Int J Mol Sci 2022; 23: 2687.

18.

de la Monte

Wands

. Alzheimer's disease is type 3 diabetes-evidence reviewed. J Diabetes Sci Technol 2008; 2: 1101–1113.

19.

Boccardi

Murasecco

Mecocci

. Diabetes drugs in the fight against Alzheimer's disease. Ageing Res Rev 2019; 54: 100936.

20.

Nowell

Blunt

Gupta

, et al. Antidiabetic agents as a novel treatment for Alzheimer's and Parkinson's disease. Ageing Res Rev 2023; 89: 101979.

21.

Sekula

Del

GMF

Pattaro

, et al. Mendelian randomization as an approach to assess causality using observational data. J Am Soc Nephrol 2016; 27: 3253–3265.

22.

Walker

Davey

Davies

, et al. Mendelian randomization: a novel approach for the prediction of adverse drug events and drug repurposing opportunities. Int J Epidemiol 2017; 46: 2078–2089.

23.

Schmidt

Finan

Gordillo-Maranon

, et al. Genetic drug target validation using Mendelian randomisation. Nat Commun 2020; 11: 3255.

24.

Xie

Liu

Huang

, et al. Effects of antidiabetic agents on platelet characteristics with implications in Alzheimer's disease: Mendelian randomization and colocalization study. Heliyon 2024; 10: e30909.

25.

Meng

Wang

, et al. Causal association evaluation of diabetes with Alzheimer's disease and genetic analysis of antidiabetic drugs against Alzheimer's disease. Cell Biosci 2022; 12: 28.

26.

Tang

Wang

Jiang

, et al. Genetic variation in targets of antidiabetic drugs and Alzheimer disease risk: a Mendelian randomization study. Neurology 2022; 99: e650–e659.

27.

Wishart

Feunang

Guo

, et al. Drugbank 5.0: a major update to the DrugBank database for 2018. Nucleic Acids Res 2018; 46: D1074–D1082.

28.

Gaulton

Hersey

Nowotka

, et al. The ChEMBL database in 2017. Nucleic Acids Res 2017; 45: D945–D954.

29.

Chaudhury

Duvoor

Reddy

, et al. Clinical review of antidiabetic drugs: implications for type 2 diabetes mellitus management. Front Endocrinol (Lausanne) 2017; 8: 6.

30.

Rena

Hardie

Pearson

. The mechanisms of action of metformin. Diabetologia 2017; 60: 1577–1585.

31.

Davies

Holmes

Davey Smith

. Reading Mendelian randomisation studies: a guide, glossary, and checklist for clinicians. Br Med J 2018; 362: k601.

32.

Zhu

Zhang

, et al. Integration of summary data from GWAS and eQTL studies predicts complex trait gene targets. Nat Genet 2016; 48: 481–487.

33.

Zuber

Grinberg

Gill

, et al. Combining evidence from Mendelian randomization and colocalization: review and comparison of approaches. Am J Hum Genet 2022; 109: 767–782.

34.

Giambartolomei

Vukcevic

Schadt

, et al. Bayesian Test for colocalisation between pairs of genetic association studies using summary statistics. PLoS Genet 2014; 10: e1004383.

35.

Chen

Zhou

, et al. Non-targeted metabolomics revealed novel links between serum metabolites and primary ovarian insufficiency: a Mendelian randomization study. Front Endocrinol (Lausanne) 2024; 15: 1307944.

36.

Zhao

Yiu

Barton

, et al. Association of Lipid-Lowering drugs with risk of psoriasis: a mendelian randomization study. JAMA Dermatol 2023; 159: 275–280.

37.

Landreth

Jiang

Mandrekar

, et al. PPARgamma agonists as therapeutics for the treatment of Alzheimer's disease. Neurotherapeutics 2008; 5: 481–489.

38.

Pan

, et al. Erythropoietin signaling in peripheral macrophages is required for systemic beta-amyloid clearance. EMBO J 2022; 41: e111038.

39.

Xiao

Liu

, et al. APP, PSEN1, and PSEN2 variants in Alzheimer’s disease: systematic re-evaluation according to ACMG guidelines. Front Aging Neurosci 2021; 13: 695808.

40.

Williamson

Goldman

Marder

. Genetic aspects of Alzheimer disease. Neurologist 2009; 15: 80–86.

41.

Esler

Wolfe

. A portrait of Alzheimer secretases–new features and familiar faces. Science 2001; 293: 1449–1454.

42.

Wolfe

. When loss is gain: reduced presenilin proteolytic function leads to increased Abeta42/Abeta40. Talking Point on the role of presenilin mutations in Alzheimer disease. EMBO Rep 2007; 8: 136–140.

43.

Cai

Kim

. Mutations in presenilin 2 and its implications in Alzheimer's disease and other dementia-associated disorders. Clin Interv Aging 2015; 10: 1163–1172.

44.

Paroni

Bisceglia

Seripa

. Understanding the amyloid hypothesis in Alzheimer's disease. J Alzheimers Dis 2019; 68: 493–510.

45.

Krafft

Jerecic

Siemers

, et al. ACU193: an immunotherapeutic poised to test the amyloid beta oligomer hypothesis of Alzheimer’s disease. Front Neurosci 2022; 16: 848215.

46.

Vadukul

Maina

Franklin

, et al. Internalisation and toxicity of amyloid-beta 1-42 are influenced by its conformation and assembly state rather than size. FEBS Lett 2020; 594: 3490–3503.

47.

Vinuesa

Pomilio

Gregosa

, et al. Inflammation and insulin resistance as risk factors and potential therapeutic targets for Alzheimer's disease. Front Neurosci 2021; 15: 653651.

48.

Zhang

Pan

Zhang

, et al. The effect of thiazolidinediones on body fat redistribution in adults: a systematic review and meta-analysis of randomized controlled trials. Obes Rev 2024; 25: e13675.

49.

Wang

Zhao

, et al. Deciphering the roles of PPARgamma in adipocytes via dynamic change of transcription complex. Front Endocrinol (Lausanne) 2018; 9: 73.

50.

Lebovitz

Banerji

. Insulin resistance and its treatment by thiazolidinediones. Recent Prog Horm Res 2001; 56: 265–294.

51.

Skerrett

Pellegrino

Casali

, et al. Combined liver X receptor/peroxisome proliferator-activated receptor gamma agonist treatment reduces amyloid beta levels and improves behavior in amyloid precursor protein/presenilin 1 mice. J Biol Chem 2015; 290: 21591–21602.

52.

Paouri

Georgopoulos

. Systemic and CNS inflammation crosstalk: implications for Alzheimer's disease. Curr Alzheimer Res 2019; 16: 559–574.

53.

Escribano

Simon

Gimeno

, et al. Rosiglitazone rescues memory impairment in Alzheimer's transgenic mice: mechanisms involving a reduced amyloid and tau pathology. Neuropsychopharmacology 2010; 35: 1593–1604.

54.

Mawuenyega

Sigurdson

Ovod

, et al. Decreased clearance of CNS beta-amyloid in Alzheimer's disease. Science 2010; 330: 1774.

55.

Hellstrom-Lindahl

Ravid

Nordberg

. Age-dependent decline of neprilysin in Alzheimer's disease and normal brain: inverse correlation with a beta levels. Neurobiol Aging 2008; 29: 210–221.

56.

Iwata

Tsubuki

Takaki

, et al. Metabolic regulation of brain Abeta by neprilysin. Science 2001; 292: 1550–1552.

57.

Moreira

Cardoso

Santos

, et al. The key role of mitochondria in Alzheimer's disease. J Alzheimers Dis 2006; 9: 101–110.

58.

Bettcher

Kramer

. Longitudinal inflammation, cognitive decline, and Alzheimer's disease: a mini-review. Clin Pharmacol Ther 2014; 96: 464–469.

59.

Moreira

. Alzheimer's disease and diabetes: an integrative view of the role of mitochondria, oxidative stress, and insulin. J Alzheimers Dis 2012; 30: S199–S215.

60.

Koudinov

Koudinova

. Cholesterol homeostasis failure as a unifying cause of synaptic degeneration. J Neurol Sci 2005; 229-230: 233–240.

61.

Arendt

Bruckner

. Linking cell-cycle dysfunction in Alzheimer's disease to a failure of synaptic plasticity. Biochim Biophys Acta 2007; 1772: 413–421.

62.

Tzimopoulou

Cunningham

Nichols

, et al. A multi-center randomized proof-of-concept clinical trial applying [(1)(8)F]FDG-PET for evaluation of metabolic therapy with rosiglitazone XR in mild to moderate Alzheimer's disease. J Alzheimers Dis 2010; 22: 1241–1256.

63.

Gold

Alderton

Zvartau-Hind

, et al. Rosiglitazone monotherapy in mild-to-moderate Alzheimer's disease: results from a randomized, double-blind, placebo-controlled phase III study. Dement Geriatr Cogn Disord 2010; 30: 131–146.

64.

Harrington

Sawchak

Chiang

, et al. Rosiglitazone does not improve cognition or global function when used as adjunctive therapy to AChE inhibitors in mild-to-moderate Alzheimer's disease: two phase 3 studies. Curr Alzheimer Res 2011; 8: 592–606.

65.

Burns

Alexander

Welsh-Bohmer

, et al. Safety and efficacy of pioglitazone for the delay of cognitive impairment in people at risk of Alzheimer's disease (TOMMORROW): a prognostic biomarker study and a phase 3, randomised, double-blind, placebo-controlled trial. Lancet Neurol 2021; 20: 537–547.

66.

Ongen

Brown

Delaneau

, et al. Estimating the causal tissues for complex traits and diseases. Nat Genet 2017; 49: 1676–1683.

67.

Human genomics. The Genotype-Tissue Expression (GTEx) pilot analysis: multitissue gene regulation in humans. Science 2015; 348: 648–660.

68.

Tian

Keller

Broman

, et al. The dissection of expression quantitative trait locus hotspots. Genetics 2016; 202: 1563–1574.

69.

Davey Smith

Hemani

. Mendelian randomization: genetic anchors for causal inference in epidemiological studies. Hum Mol Genet 2014; 23: R89–R98.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.03 MB