Introduction

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder and the leading cause of dementia worldwide1. With aging populations, its prevalence is rising rapidly, yet disease-modifying treatments remain elusive. Although hallmark pathologies—amyloid-β (Aβ) plaques and neurofibrillary tangles—have been well documented2,3,4, the molecular mechanisms that initiate and drive AD remain poorly understood, particularly in sporadic late-onset cases, which account for the vast majority of cases. Large-scale genome-wide association studies (GWAS) have identified more than 40 risk loci for AD5,6. However, the majority of these variants are located in noncoding regions5, limiting their direct interpretability and functional translation. As a result, the current challenge in AD genetics has shifted from locus discovery to understanding how genetic variation disrupts regulatory mechanisms and downstream molecular pathways that ultimately lead to neurodegeneration.

To bridge this gap, molecular quantitative trait loci (QTL) analyses have emerged as a powerful approach. These analyses link genetic variation to intermediate molecular phenotypes, such as DNA methylation (mQTL)7,8, gene expression (eQTL)9,10, and protein abundance (pQTL)11, providing insights into the regulatory architecture between genetic variation and disease. Methylation QTLs capture the epigenetic modulation of gene activity; eQTLs reveal transcriptional consequences; and pQTLs reflect terminal functional outputs. By integrating these datasets with Mendelian randomization (MR), it is possible to infer causal relationships between molecular traits and AD risk, moving beyond correlation to mechanistic inference.

Despite this potential, prior studies remain constrained by several limitations. Most have analyzed only a single molecular layer—either mQTL, eQTL12,13,14,15, or pQTL—without integrating evidence across levels16,17,18,19. Furthermore, tissue specificity has been insufficiently addressed. While brain tissue is etiologically relevant, molecular data from blood and plasma are more accessible and may serve as peripheral biomarkers. However, the extent to which peripheral molecular changes reflect brain-based processes remains unclear. Finally, there has been no systematic effort to integrate mQTL, eQTL, and pQTL data to identify consistent, multi-layer regulatory pathways driving AD risk.

In this study, we address these gaps by conducting a comprehensive multi-omics analysis that integrates DNA methylation, gene expression, and protein abundance data across both central (brain, CSF) and peripheral blood. We leveraged genome-wide association studies (GWAS) in conjunction with summary-data-based Mendelian randomization (SMR) to infer potential causal relationships between molecular traits and AD risk. To distinguish pleiotropic effects from true mediation, we applied the HEIDI (Heterogeneity in Dependent Instruments) test, which helps determine whether observed associations are due to a shared causal variant or linkage disequilibrium. We further employed colocalization analysis to evaluate whether QTL signals and AD risk loci are driven by the same underlying variant. To examine the translatability of peripheral markers, we conducted tissue concordance analysis—assessing correlation of effect sizes across blood and brain/CSF—aiming to determine whether peripheral molecular signals reflect central nervous system biology. Finally, we validated key findings in an independent population-based cohort (FinnGen) (Fig. 1).

Fig. 1
figure 1

Study Design. We filtered molecular quantitative trait loci (QTL) data as the exposure factors. GWAS summary statistics for Alzheimer’s disease (AD) outcomes were sourced from the GWAS Catalog and the FinnGen study. Summary-data-based Mendelian randomization (SMR) analysis was conducted across three biological layers: methylation, gene expression, and protein abundance. We identified significant risk genes via colocalization analysis and integrated multiple omics results to identify risk genes at various levels, validated by the FinnGen study. No sample overlap occurred between the exposure and outcome populations. SMR, summary-data-based mendelian randomization; QTL, quantitative trait loci; AD, alzheimer’s disease; SNP, single nucleotide polymorphisms; PP.H4, posterior probability of H4; p, significance threshold; FDR, false positive rate; HEIDI, the Heterogeneity in Dependent Instruments.

Full size image

Methods

Quantitative trait loci datasets

Peripheral blood-derived methylation QTL (mQTL) data were obtained from McRae et al.20. This dataset includes peripheral blood samples from two European-ancestry cohorts: the Brisbane Systems Genetics Study (BSGS; n = 614) and the Lothian Birth Cohort (LBC; n = 1,366). DNA methylation levels were assessed using Illumina HumanMethylation450 arrays. For our analysis, we used mQTL summary statistics from a meta-analysis combining both BSGS and LBC. This meta-analysis retained only methylation probes with at least one significant cis-mQTL (P < 5 × 10⁻⁸) and single nucleotide polymorphisms (SNPs) located within 2 Mb of each probe21. Brain mQTL data were obtained from Qi et al.22. These data were derived from a meta-analysis of the Religious Orders Study and Memory and Aging Project (ROSMAP)23, Hannon et al.24, and Jaffe et al.25.

Blood expression QTL (eQTL) data were obtained from the eQTLGen Consortium26. This large-scale meta-analysis assessed gene expression in blood samples from 31,684 individuals, identifying approximately 11 million SNP-gene associations. We included 88% of the identified cis-eGenes in our analysis. For brain tissues, we used eQTL data from 13 regions in the Genotype-Tissue Expression (GTEx) project, version 8 (GTEx v8)27. This dataset, released in 2020 by the National Human Genome Research Institute (NHGRI) and the NIH, contains gene expression data across 54 tissue types from nearly 1000 donor.

Plasma protein QTL (pQTL) data were derived from a study by Ferkingstad et al.11, which involved 35,559 Icelandic participants and identified 18,084 variants associated with plasma protein levels. Cerebrospinal fluid (CSF) pQTL data were obtained from the National Institute on Aging Alzheimer’s Disease Data Storage Site (NIAGADS) at the University of Pennsylvania (Accession Number: NG00102.v1; https://dss.niagads.org/datasets/ng00102/). This dataset includes multi-tissue proteomic data curated and quality-controlled by the Cruchaga Lab at Washington University in St. Louis. Proteomic profiling of 869 proteins was performed on 770 CSF samples from 1,157 subjects using the SomaLogic SomaScan 1.3K platform.

GWAS summary statistics of Alzheimer’s disease

GWAS summary statistics for AD were primarily sourced from the study by Bellenguez et al.28, which analyzed data from 39,106 clinically diagnosed AD cases, 46,828 proxy cases (based on parental history), and 401,577 controls, all of European ancestry. Genotyping was conducted using genome-wide arrays from Affymetrix and Illumina platforms. For replication, we used the R10 release of the FinnGen study29, which included 15,617 AD cases and 396,564 controls.

Summary-data-based Mendelian randomization (SMR) analysis

We conducted SMR analysis using the SMR software (v1.3.1)30 to test whether genetic effects on AD are mediated through gene expression or other molecular traits. We focused on SNPs located within ± 1,000 kb of each target gene and applied a significance threshold of P ≤ 5 × 10⁻⁸. To minimize potential biases related to population stratification or dataset heterogeneity, SNPs with allele frequency differences greater than 0.2 across datasets were excluded. To correct for multiple testing, we applied the Benjamini–Hochberg procedure to control the false discovery rate (FDR) at 0.05. SNPs with p-FDR < 0.05 were considered statistically significant. To distinguish pleiotropy from linkage, we applied the Heterogeneity in Dependent Instruments (HEIDI) test30. SNPs with p-HEIDI < 0.01 were excluded as potential linkage artifacts21. Visualization of the SMR results was performed using the “forestploter” R package (v1.1.1) in R 4.3.

Colocalization analysis

Colocalization analysis were conducted using the “coloc” R package (v5.2.3)31. This method tests whether a shared causal variant influences both a molecular trait (e.g., gene expression, protein level, or methylation) and AD risk within a specific genomic region. We included SNPs meeting the criteria of p-FDR < 0.05 and p-HEIDI > 0.01. Colocalization evaluates four hypotheses: H0 (no association), H1 (only one trait associated), H3 (both traits associated, but with different causal variants), and H4 (both traits associated with the same causal variant). Colocalization was applied to pQTL-GWAS, eQTL-GWAS, and mQTL-GWAS datasets. Based on prior studies, we used region windows of ± 1,000 kb for pQTL-GWAS32 and eQTL-GWAS33, and ± 500 kb for mQTL-GWAS34. A posterior probability for H4 (PP.H4) > 0.5 was considered strong evidence of colocalization between the molecular trait and AD risk35,36,37,38.

Pearson correlation

Following prior studies39,40, we performed Pearson correlation analysis to assess whether risk genes identified in blood were representative of those in brain and CSF. We used effect estimates from SMR analysis and selected genes with p < 0.005 in both blood and brain (or CSF) datasets.

Tiered classification of causal candidate genes

We integrated evidence from three molecular layers in blood (protein, expression, methylation) to prioritize causal candidate genes. Given the functional relevance of proteins, our classification required all candidate genes to show causal associations with AD at the protein level. Based on this criterion, genes were categorized into two tiers: (1) Tier 1: Genes demonstrating significant associations (p-FDR < 0.05 and p-HEIDI > 0.01) across all three omics layers (protein, expression, and methylation); (2) Tier 2: Genes showing significant protein-level associations (p-FDR < 0.05 and p-HEIDI > 0.01) plus associations at either the methylation or expression level.

Results

Causal effects of DNA methylation on AD risk

In peripheral blood, 768 CpG sites across 355 unique genes showed significant associations with AD (p < 0.05; Table S1), after excluding signals suggestive of pleiotropy or linkage (p-HEIDI < 0.01). After multiple testing correction, 39 CpG sites near 25 genes remained significant (p-FDR < 0.05) (Fig. 2 and Figure S1). Of these, 19 CpG sites linked to 13 genes showed strong evidence of colocalization with AD risk (PP.H4 > 0.5), suggesting a shared genetic variant influencing both DNA methylation and disease susceptibility (Figure S1).The strongest colocalization evidence was observed at cg19118072 near PLXNC1 (PP.H4 = 0.99). Key implicated CpG–gene pairs included ACE (cg04199256), AGRN (cg20988802, cg24730157), ATF6B (cg08188698), CA12 (cg03491459), CD33 (cg10129493), DAPK2 (cg26167930), FCRLB (cg18200075), MICA (cg08575621, cg26846810), PLXNC1 (cg19118072), QPCT (cg20647735), SERPINF2 (cg00443543, cg17514665, cg18278729), TMEM106B (cg19800032, cg23422036), and UBASH3B (cg08706575, cg20555462). Interestingly, effect directions varied across CpG sites within the same gene. For instance, methylation at cg19800032 in TMEM106B was associated with decreased AD risk (OR = 0.91, 95% CI [0.88–0.95]), whereas methylation at cg23422036 was associated with increased risk (OR = 1.06, 95% CI [1.04–1.09]). To validate these findings, we used summary statistics from the FinnGen study. Several associations, including those for SERPINF2 (cg00443543 and cg18278729), DAPK2 (cg26167930), and TMEM106B (cg23422036), were successfully replicated (Figure S2).

Fig. 2
figure 2

Circular plot of DNA methylation associations (blood) with AD risk (discovery cohort).This circular plot displays the associations between DNA methylation at specific CpG sites and AD risk. Each segment represents a CpG site, labeled with its corresponding gene name and CpG identifier. Outermost ring represents p-HEIDI value. Color gradient from light orange (0) to dark orange (1). Second ring is p value. Color gradient from light blue (0) to dark blue (1). Third ring indicates OR. Color gradient from green (0.6) to white (1) to dark blue (1.4). Innermost ring shows Posterior Probability of Hypothesis 4 (PP.H4). Color gradient from light pink (0) to dark red (1). The central area of the plot shows the chromosomal locations of the CpG sites.

Full size image

In brain tissue, we identified 316 CpG sites across 160 genes that were significantly associated with AD risk (p-FDR < 0.05) and passed the HEIDI test (p-HEIDI > 0.01), indicating likely causal relationships (Table S2). Some signals overlapped with those from blood, including TMEM106B (cg23422036; OR = 1.12, 95% CI [1.06–1.19]) and UBASH3B (cg08706575; OR = 1.05), showing consistent directions of effect. To assess whether blood-based methylation signatures reflect those in the brain, we performed Pearson correlation analysis on SMR effect estimates for CpG sites with p < 0.005 in both tissues. A strong positive correlation was observed (r = 0.70; Figure S3A), indicating that blood methylation may serve as a proxy for brain epigenetic patterns relevant to AD.

Causal effects of gene expression on AD risk

In whole blood, 125 genes met significance criteria (p-HEIDI > 0.01 and p-FDR < 0.05; Table S3). Among them, 61 genes were positively associated with AD (i.e., increased expression increased risk), and 64 were inversely associated. No significant associations were detected on chromosomes 4, 13, or 18. CASS4 showed the strongest association (OR = 2.21, 95% CI [1.72–2.83]; PP.H4 = 0.99). ACE (angiotensin-converting enzyme), which plays a key role in vascular and inflammatory pathways, was strongly associated with reduced AD risk (OR = 0.41, 95% CI [0.30–0.55]; PP.H4 = 0.99). In contrast, CD33, a gene linked to microglial activation, showed a positive association with AD risk (OR = 1.17, 95% CI [1.09–1.25]; PP.H4 = 0.95) (Fig. 3, Figure S4). Replication using FinnGen data confirmed associations for GRINA, RITA1, APH1B, ACE, and SIGLEC18P (Figure S5).

Fig. 3
figure 3

Circular plot illustrating gene expression associations (blood) with AD risk. The plot displays four concentric rings representing different statistical measures for each gene: outermost ring shows p-HEIDI, second ring shows p value, third ring shows OR, and innermost ring shows PP.H4. Color gradients for each measure are indicated in the legend below the plot. Each segment represents a gene, labeled with its gene symbol. The central area shows chromosomal locations of the genes. SNP, single nucleotide polymorphisms; OR, odds ratio; CI, confidence interval.

Full size image

We also examined eQTL data across 13 brain regions from GTEx v8. A total of 116 genes demonstrated significant associations with AD risk (p-HEIDI > 0.01 and p-FDR < 0.05; Table S4). Several genes exhibited consistent effects across multiple brain regions. For example, NDUFAF6 expression was positively associated with AD risk in both the cerebellar hemisphere (OR = 1.12, 95% CI [1.06–1.18]) and cerebellum (OR = 1.10, 95% CI [1.06–1.15]), with PP.H4 > 0.94. Pearson correlation analysis of SMR estimates between blood and brain confirmed strong concordance (Figure S3B–S3O). Genes such as CD33 and CTSH showed consistent pro-risk expression patterns in both blood and brain, suggesting that peripheral gene expression may reflect disease-relevant mechanisms in the central nervous system.

Regulatory mediation of gene expression by DNA methylation

To investigate the regulatory relationship between DNA methylation and gene expression, we performed an integrative SMR analysis using mQTL as exposure and eQTL as outcome, focusing on loci with prior evidence of association with AD. This approach aimed to identify methylation sites that may causally influence gene expression relevant to disease risk.

Methylation at cg04199256 and cg21657705, both located near the ACE gene, was positively associated with ACE expression. Conversely, methylation at cg10129493, located near CD33, was negatively associated with CD33 expression (Fig. 4). Given that elevated CD33 expression has been implicated in increased AD risk, this result suggests that hypermethylation at this site may act protectively by downregulating CD33—a gene known to modulate microglial activation and amyloid clearance.

Fig. 4
figure 4

Associations of genetically predicted methylation with expression (blood). SNP, single nucleotide polymorphisms; OR, odds ratio; CI, confidence interval.

Full size image

Protein-level associations with AD risk

To explore downstream effects of gene regulation, we extended the analysis to plasma protein levels. We identified 157 proteins associated with AD (p < 0.05; Table S5). After multiple testing correction, 10 proteins were negatively associated with AD risk, and 8 were positively associated (Fig. 5). Colocalization analysis confirmed strong evidence of shared genetic regulation for several proteins, including TMEM106B (PP.H4 = 0.96), CTSH (PP.H4 = 0.77), SIRPA (PP.H4 = 0.92), ACE (PP.H4 = 0.54), and CLN5 (PP.H4 = 0.92) (Fig. 6). Colocalization could not be assessed for BPNT2 due to missing data. FinnGen data supported associations for PILRA, GRN, SIGLEC9, ACE, and CD33 (Figure S6), with consistent effect directions.

Fig. 5
figure 5

Protein-level associations (blood) with Alzheimer’s Disease risk. The circular plot (left) displays four concentric rings representing different statistical measures for each protein: outermost ring shows p-HEIDI, second ring shows p value, third ring shows OR, and innermost ring shows PP.H4. Color gradients for each measure are indicated in the legend below. Each segment represents a protein, labeled with its gene name. The forest plot on the right provides detailed statistics for each protein, including gene name, SNP identifier, OR with 95% confidence interval, p value, and PP.H4. The x-axis of the forest plot represents the OR, with values less than 1 indicating decreased AD risk and values greater than 1 indicating increased risk.

Full size image
Fig. 6
figure 6

Comparison of Alzheimer’s Disease Strong Signals with pQTLs (blood). LocusCompare plots for 6 colocalized signals. Gene names are shown in the plots. (A) ACE; (B) CLN5; (C) CTSH; (D) SIRPA; (E) TMEM106B; (F) CD33. Points are colored based on linkage disequilibrium (LD) bins relative to the candidate SNP prioritized by HyPrColoc (purple diamond labeled with rsID; red, ≥ 0.8; orange, 0.6–0.8; green, 0.4–0.6; light blue, 0.2–0.4; and dark blue, < 0.2).

Full size image

In cerebrospinal fluid (CSF), we identified six with strong evidence of a causal relationship with AD risk (p-FDR < 0.05 and p-HEIDI > 0.01; Table S6). CD33 (OR = 1.22, 95% CI [1.10–1.37]; PP.H4 = 0.59) and SIRPA (OR = 1.25, 95% CI [1.13–1.38]; PP.H4 = 0.95) were both significantly associated with increased AD risk in CSF, findings that mirrored those observed in plasma. To determine whether peripheral protein associations reflect central nervous system biology, we performed Pearson correlation analysis between effect estimates in plasma and CSF. This analysis aimed to assess the translatability of plasma-based protein markers to brain-relevant pathways. The results revealed a strong positive correlation between plasma and CSF protein effects (Figure S3P), suggesting that systemic protein levels can serve as informative proxies for CNS-related AD processes.

Multi-omics integration and prioritization of AD risk genes

To prioritize biologically meaningful AD risk genes, we integrated evidence across three molecular layers—DNA methylation, gene expression, and protein abundance—based on both SMR and colocalization results. A tiered classification system was developed to rank candidate genes by the strength and consistency of their multi-omics associations with AD (Fig. 7). Tier 1 genes demonstrated significant associations (p-FDR < 0.05 and p-HEIDI > 0.01) and strong colocalization evidence (PP.H4 > 0.5) across all three omics layers. Tier 2 genes met the same criteria in two of the three layers. Using this framework, we identified two Tier 1 genes. ACE, a key regulator of vascular and inflammatory pathways, showing protective effects. CD33, an innate immune receptor involved in microglial function, consistently associated with increased AD risk across all layers.

Fig. 7
figure 7

Manhattan plot for associations (blood) between gene molecular features and Alzheimer’s Disease. Manhattan plot for protein abundance (A), gene methylation (B) and expression (C). Genes with significant signals in protein abundance levels were labeled.

Full size image

In addition, five genes were categorized as Tier 2, including TMEM106B, SIRPA, CLN5, CTSH, and SERPINF2, each supported by two convergent omics datasets (Table 1). All prioritized genes showed strong colocalization evidence in at least one molecular layer, reinforcing their candidacy as causal genes for AD. Many of these genes—particularly ACE, CD33, and TMEM106B—are expressed both peripherally and in brain tissue, and play roles in immune regulation, protein homeostasis, and neuroinflammation. Their multi-layered genetic support and biological relevance make them promising targets for further functional validation and potential therapeutic development.

Table 1 Classified risk genes: based on colocalization results and multi-omics analysis.
Full size table

Discussion

In this study, we conducted a comprehensive multi-omics analysis to investigate potential causal pathways linking DNA methylation, gene expression, and protein abundance to AD risk. Prior studies have highlighted the role of genetically driven epigenetic variation in AD using either mQTL or eQTL datasets to identify candidate genes associated with the disease12,41. However, the integration of methylation, transcriptomic, and proteomic data across both peripheral and central tissues has remained limited. By combining SMR analysis, tissue concordance testing, and replication in an independent population-based cohort (FinnGen), our study addresses this gap and provides a more layered understanding of the regulatory architecture.

We identified 39 CpG sites in blood and 316 in brain significantly associated with AD, 125 AD-associated genes in blood, 116 in brain, and 24 AD-linked proteins (18 in plasma, 6 in CSF). Strong cross-tissue correlation and replicated associations (e.g., CD33, TMEM106B, SERPINF2) further supported the robustness of our results. We explored regulatory mediation by linking mQTLs to eQTLs, identifying functionally relevant methylation-expression axes. For example, methylation at cg04199256 and cg21657705 near ACE was associated with increased gene expression and reduced AD risk, consistent with ACE’s potential protective role. Conversely, methylation at cg10129493 near CD33 suppressed expression. We developed a tiered prioritization strategy by integrating SMR and colocalization results. Two genes, ACE and CD33, were supported by consistent signals across all three omics layers (Tier 1). Five additional genes (TMEM106B, CTSH, SIRPA, CLN5, SERPINF2) were supported by two molecular levels (Tier 2)(Table 1).

Among these, ACE has been extensively studied for its role in cardiovascular homeostasis via the renin-angiotensin system42,43,44,45,46,47. Previous investigations have shown conflicting evidence regarding ACE’s role in AD, with pharmacologic studies suggesting that ACE inhibition may reduce amyloid-beta accumulation and neuroinflammation, supporting a protective role48,49,50. In contrast, recent Mendelian randomization findings have suggested that genetically proxied inhibition of ACE may actually increase dementia risk51, implying a beneficial role for endogenous ACE activity in the brain. Our results reconcile this contradiction by showing that increased endogenous ACE expression—potentially driven by intragenic methylation—may indeed confer protection, emphasizing the need for cautious interpretation of ACE inhibitor trials in dementia patients. Moreover, our study is the first to identify specific methylation sites (cg21657705 and cg04199256) that influence ACE expression, offering a novel epigenetic perspective distinct from pharmacologic intervention.

Lysosomal dysfunction is a key mechanism in AD, particularly in the context of protein degradation and autophagy. Our identification of CLN5, CTSH, and TMEM106B supports prior reports linking these genes to lysosomal integrity52,53. Our study reveals opposing effects of TMEM106B methylation on AD risk, highlighting the spatial complexity of epigenetic regulation. Methylation at cg19800032 (likely promoter-proximal) was associated with decreased risk, while cg23422036 (intragenic) correlated with increased risk, consistent with the general model wherein promoter methylation suppresses and gene body methylation enhances expression54.

SIRPA, a key immune regulatory protein involved in the CD47 checkpoint axis, was also identified as an AD risk factor based on CSF proteomic data55,56. This protein modulates phagocytosis and immune tolerance, and its overexpression may impede microglial clearance of pathological substrates, echoing recent work implicating innate immune suppression in AD progression 57. Likewise, our findings on CD33 reinforce its known role in modulating microglial response to beta-amyloid. Importantly, we provide novel evidence that CD33 expression is epigenetically regulated by site-specific methylation at cg10129493, suggesting a mechanism through which AD risk could be modulated.

While several of these genes have been previously implicated in AD, our integrative approach provided new mechanistic insights. We have identified novel CpG sites (e.g., cg04199256), brain–blood concordance of expression for genes like CTSH, and protein-level validation for SIRPA and CD33, broadening current understanding of their pathogenic relevance. In addition, our results reveal new molecular targets that extend beyond Tier 1 genes (Figure S1-S6). Although not all candidate loci were supported across three layers, many showed strong and tissue-consistent associations at individual levels. These loci may still represent biologically relevant AD mechanisms and need further experimental exploration.

While therapeutic implications remain exploratory, they are supported by strong genetic evidence and provide testable hypotheses. We also recognize the ethical challenges of translating genomic insights into clinical interventions, including concerns about privacy, equity, and genetic determinism, which merit careful consideration as the field progresses.

Conclusion

In conclusion, our study provides an integrated landscape of molecular mechanisms in AD and highlights actionable regulatory nodes across methylation, transcription, and translation. This work sets the stage for translational research and drug development aimed at epigenetically and genetically modulated targets.

Strengths and limitations

This design allows us to identify convergent regulatory signatures and construct a more complete view of gene regulation in AD. Unlike previous studies focusing on single molecular layers, our approach enables the prioritization of risk genes supported by evidence across multiple omics layers. Nevertheless, several limitations should be acknowledged. First, the majority of the molecular QTL and GWAS datasets used are derived from individuals of European ancestry. While this improves internal consistency, it limits the applicability of our findings to other ancestral populations. Genetic architecture, allele frequencies, and linkage disequilibrium structures can vary significantly across populations, and future studies incorporating diverse ancestry groups are necessary to confirm the generalizability of these results. Second, while our tissue-spanning design is a strength, it also presents a challenge. Integrating multi-omics data across different tissues introduces complexity due to distinct regulatory environments. For instance, the concordance between peripheral and central tissues may be partial, and our correlation-based assessment, while informative, cannot fully resolve these differences. Longitudinal, single-cell, or matched tissue studies will be needed to further clarify these relationships. Third, the pQTL dataset used includes a limited range of proteins, restricting our ability to comprehensively investigate protein-level regulation in AD. Lastly, interpretation of PP.H4 requires caution. Low PP.H4 values may result not from a lack of shared causal variants, but from statistical limitations—such as insufficient SNP resolution, complex LD structure, or low power—particularly in regions with multiple overlapping signals.