Optimized post-GWAS analysis identifies therapeutic targets for essential thrombocythemia and polycythemia vera Authors: Xiaoli Li, Wenbin An, Xin Wang, Zixi Yin, Yangyang Gao, Yang Lan, Luyang Zhang, Beibei Zhao, Lixian Chang, Min Ruan, Li Zhang, Yao Zou, Wenyu Yang, Xiaofan Zhu Published: 2026 ABSTRACT BackgroundLimited insight into the biology of essential thrombocythemia (ET) and polycythemia vera (PV) has constrained therapeutic development. We therefore aimed to prioritize genetically supported molecular determinants underlying disease susceptibility and hematologic phenotypes, and to distinguish shared and subtype-specific regulation.MethodsWe implemented an optimized post-GWAS framework combining omics-wide Bayesian colocalization with Mendelian randomization (MR) to prioritize genetically supported molecular determinants of ET and PV. Colocalization-guided MR analyses were used to infer causal relationships, with additional analyses incorporating other phenotypic traits to support biological interpretation.ResultsHere, we show that higher genetically predicted levels of PARP1, BRAP, ERP29, PPP1CC, and RPN1 are consistently associated with risk of both ET and PV, whereas MYB is specifically associated with ET. Several biomarkers influenced blood cell counts, with BRAP, ERP29, and PPP1CC increasing platelet, red blood cell, and white blood cell counts, while MYB increased platelet production but reduced red and white blood cells. Additionally, MYB promoted BRAP, ERP29, PPP1CC, and EPO signaling while suppressing THPO, whereas MYB gene effects showed opposite patterns on these pathways, blood cell counts, and ET risk. BRAP, PPP1CC, and ERP29 exerted concordant regulatory effects on MPL, RPN1, THPO, and PARP1, except for EPO, which they negatively regulated. Finally, hypermethylation at cg17916418 and cg23511909 suppressed RPN1 expression and increased ET and PV risk, with RPN1 acting as a key mediator.ConclusionsThis study provides a genetics-guided map of shared and subtype-specific molecular features in ET and PV, highlighting potential therapeutic pathways and molecular signals distinguishing the two diseases. Introduction The classical BCR-ABL1-negative myeloproliferative neoplasms (MPNs) include polycythemia vera (PV), essential thrombocythemia (ET), and primary myelofibrosis (PMF) and are primarily driven by mutations in JAK2. Mutations in CALR and MPL also occur in ET and PMF1,2. In the early or chronic phases of MPNs, elevated counts of mature myeloid cells, especially platelets, are commonly observed, with JAK2 mutations present in the majority of cases. Notably, approximately 10–15% of ET and PMF patients lack all three canonical driver mutations and are therefore categorized as “triple-negative”3. The clinical distinction among classical MPN subtypes can be challenging due to overlapping phenotypic and molecular features. Among BCR-ABL1-negative MPNs, ET and PV represent two closely related disease entities with substantial overlap in clinical presentation and molecular features1. Both diseases frequently harbor JAK2 mutations but diverge in their dominant hematologic phenotypes, namely thrombocytosis in ET and erythrocytosis in PV4,5. In early-stage, atypical, or treated cases, the clinical distinction between ET and PV can be challenging, and misclassification remains a recognized issue in routine practice6. In addition, ET and PV together account for the majority of diagnosed BCR-ABL1-negative MPN cases, providing sufficient case numbers and balanced representation for robust genetic and molecular analyses. However, the molecular mechanisms underlying the shared and divergent clinical manifestations and molecular features of ET and PV remain incompletely understood7. Despite such close clinical and molecular overlap and the incomplete understanding of the underlying mechanisms, advances in sequencing technologies have led to the identification of additional recurrently mutated genes in ET and PV; however, the core driver mutations remain largely unchanged. Moreover, the newly discovered mutations are relatively rare and do not directly affect hematopoiesis8. Importantly, genomic mutations do not always translate into changes in circulating protein levels. Circulating proteins, which mediate key biological processes and may represent potential therapeutic targets, have gained attention as druggable biomarkers in hematologic malignancies9–11. However, to date, no drugs have been specifically developed to target circulating proteins in ET and PV patients. Emerging large-scale genome-wide association study (GWAS) datasets are ushering in a post-GWAS era enhanced by advanced bioinformatics techniques12,13. Traditionally, drug target identification has primarily relied on proteome-wide two-sample Mendelian randomization (MR), followed by confirmation using Bayesian colocalization analysis14–16. However, the number of cis-protein quantitative trait loci (cis-pQTLs) varies across different proteins17, complicating the selection of instrumental variables (IVs) for MR. Establishing a consistent standard across thousands of proteins is challenging, as setting thresholds too high may result in overlooking important proteins. In addition, MR requires a series of sensitivity analyses, making the process of obtaining accurate outcomes tedious. Systematic MR screening results also need multiple testing corrections, and conducting MR first may increase the risk of Type II errors. In principle, Bayesian colocalization analysis can provide direct evidence as to whether two traits are driven by the same causal variant. When the colocalization result is positive, it typically supports a non-zero effect observed in MR. However, the reverse is not necessarily true18. Even when MR suggests a causal relationship, the results may still be influenced by multiple causal variants, complex linkage disequilibrium (LD) structures, or biases in the instrumental variables, which can lead to discrepancies with colocalization findings. Moreover, conventional Bayesian colocalization assesses only one variant–trait pair at a time19, making it operationally inefficient when scaled to omics-wide validation, limiting its suitability for identifying multidimensional causal mechanisms. To address these challenges, we adopt an optimized post-GWAS analytical framework. Specifically, we first perform omics-wide Bayesian colocalization analysis as a primary screening step. Because cis-acting genetic instruments do not require clumping or P-value filtering, a broader set of biomarkers can be evaluated. Biomarkers sharing causal variants with disease traits are then prioritized for downstream MR analysis. Colocalization offers evidence of shared genetic architecture19, whereas MR assesses causality under specific assumptions20. Thus, performing MR after colocalization is more logical and yields stronger causal inference. The primary objective of this study was to systematically identify genetically supported molecular determinants underlying ET and PV, clarify biological mechanisms contributing to disease susceptibility and hematologic phenotypes, and delineate shared and subtype-specific molecular regulation between these two closely related myeloproliferative neoplasm subtypes. Using an integrative post-GWAS analytical framework, we identify both shared and subtype-specific molecular signals associated with ET and PV, providing genetically informed insights into disease biology and highlighting potential therapeutic targets for further investigation. Methods Study design To investigate the genetic architecture of ET and PV, an optimized post-GWAS analytical framework was implemented, in which omics-wide Bayesian colocalization was used as the primary screening step to prioritize molecular traits, followed by two-sample MR analyses for causal inference. Specifically, we first conducted proteome-wide Bayesian colocalization analyses to systematically screen circulating proteins sharing causal variants with ET or PV. Proteins supported by colocalization evidence were subsequently subjected to two-sample MR to assess causal relationships. To determine whether the identified protein signals were supported at the gene expression level, transcriptome-wide Bayesian colocalization analyses were performed, followed by MR where appropriate. In addition, reverse MR analyses were conducted, treating ET and PV as exposures, to further evaluate the directionality of identified associations. To validate positive biomarker specificity and regulation in the target diseases, we included common MPN-related clinical features (i.e., abnormal blood cell counts) and four other blood cell–related disorders (secondary polycythemia, thrombocytopenia–unspecified, other white blood cell disorders, and drug-induced neutropenia) as outcomes. We then performed proteome- and transcriptome-wide Bayesian colocalization analyses for platelet (PLT), red blood cell (RBC), and white blood cell (WBC) counts, as well as for these diseases, with MR conducted when necessary. To further explore regulatory relationships and biological interpretation, we examined interactions among significant biomarkers, canonical MPN driver mutations (JAK2, CALR, MPL), and key hematopoietic regulators (EPO, EPOR, THPO) using omics-wide colocalization and MR analyses. When regulatory relationships could not be fully explained at the protein or transcript level, DNA methylation GWAS data were additionally incorporated using Bayesian colocalization and MR approaches. An overview of the study design and analytical workflow is shown in Fig. 1.Fig. 1Schematic overview of the study design.This schematic diagram outlines the key components of the study, divided into two main sections. The upper section presents detailed exposure and outcome data, including corresponding databases (deCODE, UKB-PPP, eQTLGen, GoDMC, etc.), along with sample sizes or the number of cases used for Bayesian colocalization and Mendelian Randomization (MR) analyses. It also annotates the primary objectives associated with each outcome. The lower section outlines the methodological steps involved in the analysis, including omics-wide Bayesian colocalization, two-sample MR analysis, and two-step mediation MR, along with additional sensitivity tests for two-sample MR, such as tests for heterogeneity, pleiotropy, and other sensitivity analyses. Schematic overview of the study design. This schematic diagram outlines the key components of the study, divided into two main sections. The upper section presents detailed exposure and outcome data, including corresponding databases (deCODE, UKB-PPP, eQTLGen, GoDMC, etc.), along with sample sizes or the number of cases used for Bayesian colocalization and Mendelian Randomization (MR) analyses. It also annotates the primary objectives associated with each outcome. The lower section outlines the methodological steps involved in the analysis, including omics-wide Bayesian colocalization, two-sample MR analysis, and two-step mediation MR, along with additional sensitivity tests for two-sample MR, such as tests for heterogeneity, pleiotropy, and other sensitivity analyses. GWAS summary data for diseases and blood cell counts GWAS summary statistics for diseases were all sourced from the FinnGen consortium (Round 11; https://www.finngen.fi/en/access_results, accessed June 25, 2024). The FinnGen study is a large-scale genomics initiative that has analyzed over 500,000 Finnish biobank samples and correlated genetic variations with health data to understand disease mechanisms and predispositions. The project is a collaboration between research organizations and biobanks within Finland, along with international industry partners21. The target diseases in this study were ET and PV. Other blood cell–related disorders included secondary polycythemia, thrombocytopenia-unspecified, other disorders of white blood cells, and drug-induced neutropenia. GWAS summary statistics for PLT, RBC, and WBC counts were obtained from the NHGRI–EBI GWAS Catalog22 (https://www.ebi.ac.uk/gwas/, accessed May 27, 2024). Diseases were defined using International Classification of Diseases, 10th Revision (ICD-10) codes, whereas blood cell traits were identified using GCST accession codes. Detailed information on sample sizes and phenotype codes is provided in Table 1.Table 1Overview of GWAS and molecular QTL data sources, traits analyzed, sample sizes, and linkage disequilibrium (LD) clumping parameters used in this studyData sourceCohort typeTraitCodeCasesControlsSample size (N)LD clumping criteriaR²P-valueWindow size (kb)FinnGenTarget diseasesEssential thrombocythemia (ET)D47.31171344,335345,5060.0015 × 10−810,000Polycythemia vera (PV)D451084344,463345,547Other blood cell–related disordersSecondary polycythemiaD75.1735445,490446,225---Thrombocytopenia–unspecifiedD69.62295446,531448,826---Other white blood cell disordersD721789445,490447,279---Drug-induced neutropeniaD70.822318451,415453,733---NHGRI-EBI GWAS CatalogBlood cell countsPlatelet (PLT)GCST90002357--542,827---Red blood cell count (RBC)GCST90002363--545,203---White blood cell count (WBC)GCST90002374--562,243---deCODE4907 pQTLs4899 cis-pQTLs---35,5590.15 × 10⁻⁵100UKB-PPP4907 pQTLs2937 cis-pQTLs---54,219eQTLGen Consortium19,127 eQTLs19,127 cis-eQTLs---31,6480.0015 × 10−810,000GoDMCmQTLscis-mQTLs---27,750This table summarizes all GWAS datasets and molecular QTL resources used in the study. For disease outcomes, ICD-10 codes are provided. LD clumping parameters were applied when selecting instrumental variables for Mendelian randomization analyses. Since the number of cis-pQTLs for proteins is relatively small, relatively relaxed LD clumping parameters were applied in this study to ensure that potentially positive proteins identified through Bayesian colocalization could be included in the MR analysis. For other types of exposures, stricter parameters were adopted to reduce multicollinearity arising from LD. Sample size (N) refers to the total number of individuals included in each GWAS or QTL dataset. ET essential thrombocythemia, PV polycythemia vera, PLT platelet count, RBC red blood cell count, WBC white blood cell count. Table 1Overview of GWAS and molecular QTL data sources, traits analyzed, sample sizes, and linkage disequilibrium (LD) clumping parameters used in this studyData sourceCohort typeTraitCodeCasesControlsSample size (N)LD clumping criteriaR²P-valueWindow size (kb)FinnGenTarget diseasesEssential thrombocythemia (ET)D47.31171344,335345,5060.0015 × 10−810,000Polycythemia vera (PV)D451084344,463345,547Other blood cell–related disordersSecondary polycythemiaD75.1735445,490446,225---Thrombocytopenia–unspecifiedD69.62295446,531448,826---Other white blood cell disordersD721789445,490447,279---Drug-induced neutropeniaD70.822318451,415453,733---NHGRI-EBI GWAS CatalogBlood cell countsPlatelet (PLT)GCST90002357--542,827---Red blood cell count (RBC)GCST90002363--545,203---White blood cell count (WBC)GCST90002374--562,243---deCODE4907 pQTLs4899 cis-pQTLs---35,5590.15 × 10⁻⁵100UKB-PPP4907 pQTLs2937 cis-pQTLs---54,219eQTLGen Consortium19,127 eQTLs19,127 cis-eQTLs---31,6480.0015 × 10−810,000GoDMCmQTLscis-mQTLs---27,750This table summarizes all GWAS datasets and molecular QTL resources used in the study. For disease outcomes, ICD-10 codes are provided. LD clumping parameters were applied when selecting instrumental variables for Mendelian randomization analyses. Since the number of cis-pQTLs for proteins is relatively small, relatively relaxed LD clumping parameters were applied in this study to ensure that potentially positive proteins identified through Bayesian colocalization could be included in the MR analysis. For other types of exposures, stricter parameters were adopted to reduce multicollinearity arising from LD. Sample size (N) refers to the total number of individuals included in each GWAS or QTL dataset. ET essential thrombocythemia, PV polycythemia vera, PLT platelet count, RBC red blood cell count, WBC white blood cell count. Overview of GWAS and molecular QTL data sources, traits analyzed, sample sizes, and linkage disequilibrium (LD) clumping parameters used in this study This table summarizes all GWAS datasets and molecular QTL resources used in the study. For disease outcomes, ICD-10 codes are provided. LD clumping parameters were applied when selecting instrumental variables for Mendelian randomization analyses. Since the number of cis-pQTLs for proteins is relatively small, relatively relaxed LD clumping parameters were applied in this study to ensure that potentially positive proteins identified through Bayesian colocalization could be included in the MR analysis. For other types of exposures, stricter parameters were adopted to reduce multicollinearity arising from LD. Sample size (N) refers to the total number of individuals included in each GWAS or QTL dataset. ET essential thrombocythemia, PV polycythemia vera, PLT platelet count, RBC red blood cell count, WBC white blood cell count. Omics-wide Bayesian colocalization analyses Datasets and data processing GWAS summary datasets for proteome-wide protein traits were obtained from deCODE Genetics (deCODE) and the UK Biobank Pharma Proteomics Project (UKB-PPP), corresponding to the SomaScan and Olink platforms, respectively. These datasets encompass thousands of proteins measured across tens of thousands of individuals. Specifically, in 2021, deCODE published GWAS summary statistics for 4719 circulating plasma proteins (https://www.decode.com/summarydata/, accessed February, 2024), measured using 4907 aptamers (SomaScan, version 4) in 35,559 Icelandic individuals23. Throughout this study, the term “protein” refers to an aptamer targeting a specific protein. In 2023, UKB-PPP used the Olink platform, which quantifies proteins based on the binding of two distinct antibodies, to generate genetic association data for 2940 proteins in 54,219 UK Biobank participants24 (https://www.synapse.org/Synapse:syn51365301, accessed February, 2024). Transcriptome-wide data for 19,127 blood-expressed genes were obtained from the eQTLGen Consortium, based on samples from 31,684 individuals25 (http://www.eqtlgen.org, accessed April 14, 2024). In genetic studies, pQTLs and eQTLs represent genetic variants associated with protein abundance and gene expression levels, respectively. This study focused on cis-pQTLs and cis-eQTLs, defined as single-nucleotide polymorphisms (SNPs) located within ±1 Mb of the gene body of the corresponding gene. Only SNPs with a minor allele frequency (MAF) greater than 0.01 were retained for analysis. After data quality control and filtering, the final datasets comprised 4,899 proteins from deCODE, 2,937 proteins from UKB-PPP, and 19,127 genes from eQTLGen (Table 1). Large-scale batch Bayesian colocalization analysis Bayesian colocalization analysis was performed to assess whether two traits share a common causal variant. Effect alleles were aligned using the harmonise_data function from the TwoSampleMR R package. Subsequently, datasets D1 and D2 were constructed for trait 1 and trait 2, respectively, including the required parameters (i.e., type, beta, varbeta, position, N, MAF, snp, and s). Bayesian colocalization was then conducted using the coloc.abf function from the coloc R package (p1 = 1 × 10⁻⁴, p2 = 1 × 10⁻⁴, p12 = 1 × 10⁻⁵), which generated posterior probabilities for five competing hypotheses (PP.H0–PP.H4) based on approximate Bayes factors19. Evidence for colocalization was interpreted as follows: PP.H4 ≥ 0.8 indicated strong evidence for a shared causal variant; values between 0.5 and 0.8 indicated moderate evidence; and values between 0.3 and 0.5 indicated weak evidence. For large-scale batch analysis, 4,899 proteins from deCODE, 2,937 proteins from UKB-PPP, and 19,127 genes from the eQTLGen consortium were organized into dataset-specific directories. Each file, corresponding to a single protein or gene, was harmonized with its respective target trait (e.g., disease status, blood cell counts, or protein levels) and assigned as D1, which was sequentially colocalized with D2 representing the same target trait. When deCODE proteins were used as D2, the updated deCODE dataset was applied26. This workflow enabled systematic, high-throughput Bayesian colocalization analyses within individual omics layers. Two-sample MR analysis Two-sample MR analysis was performed to further evaluate the causal relevance of molecular traits identified through large-scale Bayesian colocalization analysis. MR leverages the random allocation of genetic variants at conception, analogous to randomization in randomized controlled trials, thereby reducing confounding and reverse causation. Valid causal inference in MR relies on three core instrumental variable assumptions27: (i) relevance, whereby the genetic variants are robustly associated with the exposure; (ii) independence, whereby the variants are independent of confounding factors; and (iii) exclusion restriction, whereby the variants influence the outcome exclusively through the exposure. These assumptions were assessed indirectly through instrument strength metrics and a series of sensitivity analyses, as described below. Genetic variants associated with exposures (pQTLs and eQTLs) identified in previous steps were further processed as instrumental variables (IVs) for MR analyses. For MR analyses involving plasma proteins from deCODE as either exposures or outcomes, updated summary statistics were used, expanding the effective sample size to 35,89626. Independent IVs were selected using LD clumping, with clumping criteria specified in Table 1 according to data type and analytical context. All candidate IVs were evaluated using F-statistics and the proportion of variance explained (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${R}^{2}$$\end{document}R2), and only variants with F-statistics greater than 10 were retained for downstream analyses (Supplementary Data 1). To minimize potential pleiotropic effects, all IVs were further queried against the IEU Open GWAS database (https://gwas.mrcieu.ac.uk/) to confirm their exclusive association with the exposure of interest. When two or more SNPs were available for a given exposure–outcome pair, causal estimates were derived using the inverse variance weighted (IVW) method. When only a single SNP was available, the Wald ratio method was applied. To assess the robustness of MR findings, a series of sensitivity analyses were conducted, including heterogeneity tests, horizontal pleiotropy tests, RadialMR, MR-PRESSO, and leave-One-Out analyses. Bonferroni correction was applied to account for multiple testing in MR analyses. For associations supported by more than four IVs, consistency in the direction of effect across IVW, MR-Egger, weighted median, simple mode, and weighted mode methods was additionally required. DNA methylation data processing CpG site identifiers (CpG IDs) denote specific genomic loci at which DNA methylation (DNAm) levels are quantified. CpG IDs corresponding to the target gene were extracted using the minfi R package. Summary statistics for DNAm quantitative trait loci (mQTLs) were obtained from the Genetics of DNA Methylation Consortium (GoDMC), which comprises data from 27,750 European individuals and includes 420,509 DNAm sites measured in whole blood28 (http://mqtldb.godmc.org.uk/downloads, accessed August 6, 2024). For downstream analyses, we focused on cis-mQTLs, defined as genetic variants located within ±1 Mb of the corresponding DNAm site. Cis-mQTLs associated with the target gene were subjected to large-scale Bayesian colocalization analysis and MR, following the same analytical procedures and criteria applied in transcriptome-wide gene expression analyses. In addition, a two-step mediation MR framework was implemented to estimate the total effect, direct effect, indirect effect, and proportion mediated29. Our research primarily set RPN1 methylation (cg17916418, cg23511909) as exposures, RPN1 protein as the mediator, and target diseases (i.e., ET, PV) as outcomes to confirm whether RPN1 methylation plays a role through its protein. Effect estimates (β) and corresponding standard errors (SE) obtained from MR analyses, including exposure–outcome, exposure–mediator, and mediator–outcome models, were used to calculate mediation parameters and the corresponding P values. Statistical software and computational environment Statistical analyses were performed using R (version 4.4.2). The primary R packages used in this study included TwoSampleMR (v0.6.14) for two-sample MR, coloc (v6.0.0) for Bayesian colocalization analyses, MR-PRESSO (v1.0) for the detection and correction of horizontal pleiotropy, and RadialMR (v1.1) for radial MR analyses and outlier detection. Genotype data processing and LD pruning were conducted using PLINK (v1.9.0), with Phase 3 European-ancestry samples from the 1000 Genomes Project used as the reference panel. All analyses were conducted using publicly available GWAS summary statistics, accessed in accordance with the ethical approvals and data governance policies of the original studies and consortia. Ethical approval and informed consent were obtained in all original studies. The datasets used in this study are comprised of publicly available data, and no additional ethical approval was necessary for the secondary use of these data. All study populations were of European ancestry. Results Biomarkers identified as positive in ET and PV After screening with a PP.H4.abf ≥ 0.3 using a proteome-wide Bayesian colocalization analysis (Supplementary Data 2), the deCODE and UKB-PPP studies identified 27 and 26 proteins in ET (Fig. 2a), 21 and 23 proteins in PV (Fig. 2c), respectively. The overlap in identified proteins between the two datasets was 7 for ET and 6 for PV. Subsequent MR analyses revealed that PARP1, BRAP, ERP29, PPP1CC, and RPN1 collectively play roles in ET and PV. Here, we primarily focus on proteins with a PP.H4.abf exceeding 0.8, as they show a significant association with these target diseases. For ET patients (Bonferroni-corrected P < 0.0009, Fig. 2b and Supplementary Data 3), ERP29 (OR = 507.62, 95% confidence interval [CI] = 91.02–2830.88), PPP1CC (OR = 506.28, 95% CI = 90.81–2822.54), MYB (OR = 486.25, 95% CI = 46.17–5121.24), BRAP (OR = 150.08, 95% CI = 49.46–455.38), PARP1 (deCODE, OR = 13.88, 95% CI = 4.82–39.94; UKB-PPP, OR = 3.27, 95% CI = 2.22–4.81), and ATF6B (OR = 1.81, 95% CI = 1.55–2.12) showed positive genetic associations with ET risk. For PV (Bonferroni-corrected P < 0.001, Fig. 2d and Supplementary Data 4), BRAP (OR = 57.40, 95% CI = 17.42–189.20) and PARP1 (deCODE, OR = 19.75, 95% CI = 7.20–54.19; UKB-PPP, OR = 3.60, 95% CI = 2.37–5.45) were positively associated with disease risk, whereas RPN1 (OR = 0.31, 95% CI = 0.22–0.42) exhibited a negative effect.Fig. 2Positive identification of biomarkers in essential thrombocythemia (ET) and polycythemia vera (PV).Venn diagrams represent the results from a proteome-wide Bayesian colocalization analysis based on deCODE and UKB-PPP data, using a PP.H4.abf threshold of ≥ 0.3. Forest plots display the MR results, with biomarkers derived from the inverse variance weighted method (IV ≥ 2) and the Wald ratio method (IV = 1). All biomarkers in the forest plots are statistically significant after applying Bonferroni correction and through sensitivity analyses, including tests for heterogeneity, pleiotropy, RadialMR, MR-PRESSO, and Leave-One-Out analyses. Bayesian colocalization results, along with PP.H4.abf values, are displayed in the forest plots. The 95% confidence interval (CI) of the odds ratio (OR) is shown as the corresponding log value. Biomarkers with PP.H4.abf ≥ 0.8 are highlighted in bold and represented by orange squares, while those with PP.H4.abf ≤ 0.5 are shown in gray text. a and b correspond to ET. a highlights MYB, which meets the PP.H4.abf ≥ 0.8 threshold. b includes MYB from eQTLGen, along with its MR results and PP.H4.abf value. c and d) correspond to PV. In (d), PDE5Aa and PDE5Ab correspond to entries 16805 and 5256 in the deCODE dataset, respectively. Positive identification of biomarkers in essential thrombocythemia (ET) and polycythemia vera (PV). Venn diagrams represent the results from a proteome-wide Bayesian colocalization analysis based on deCODE and UKB-PPP data, using a PP.H4.abf threshold of ≥ 0.3. Forest plots display the MR results, with biomarkers derived from the inverse variance weighted method (IV ≥ 2) and the Wald ratio method (IV = 1). All biomarkers in the forest plots are statistically significant after applying Bonferroni correction and through sensitivity analyses, including tests for heterogeneity, pleiotropy, RadialMR, MR-PRESSO, and Leave-One-Out analyses. Bayesian colocalization results, along with PP.H4.abf values, are displayed in the forest plots. The 95% confidence interval (CI) of the odds ratio (OR) is shown as the corresponding log value. Biomarkers with PP.H4.abf ≥ 0.8 are highlighted in bold and represented by orange squares, while those with PP.H4.abf ≤ 0.5 are shown in gray text. a and b correspond to ET. a highlights MYB, which meets the PP.H4.abf ≥ 0.8 threshold. b includes MYB from eQTLGen, along with its MR results and PP.H4.abf value. c and d) correspond to PV. In (d), PDE5Aa and PDE5Ab correspond to entries 16805 and 5256 in the deCODE dataset, respectively. Transcriptome-wide colocalization using eQTLGen consortium data across three cohorts showed that MYB gene level was negatively associated only with ET. Further MR showed effects opposite to those of its protein (OR = 0.02, 95% CI = 0.00–0.08, P = 4.99 × 10−⁷, Fig. 2b and Supplementary Data 3). To assess potential reverse causality, ET and PV were additionally modeled as exposures in reverse MR analyses. Apart from a negative association between PV and KLK14 (Supplementary Fig. 3a, b), no evidence was found to support disease-driven effects on the remaining proteins (Supplementary Data 5, 6). In addition, the results of the Leave-One-Out analyses for ET and PV are illustrated in Supplementary Figs. 1 and 2. Effects of positive biomarkers on blood cell counts and other blood cell–related disorders To determine whether the positive biomarkers identified in ET and PV also influence the key clinical hallmark of these diseases (i.e., dysregulated blood cell counts), we screened potential candidates using proteome-wide and transcriptome-wide Bayesian colocalization with PLT, RBC, and WBC counts, requiring a PP.H4.abf value exceeding 0.3. These biomarkers not only surpassed this threshold but were also verified in the previous step to be involved in ET and PV. As a result, 14, 7, and 7 proteins were detected for PLT, RBC, and WBC, respectively (Fig. 3a, c, e and Supplementary Data 7). Notably, MYB gene appeared in all three blood cell counts. MR analysis was performed to confirm their effects (PLT: Bonferroni-corrected P < 0.003; RBC: Bonferroni-corrected P < 0.006; WBC: Bonferroni-corrected P < 0.006). Specifically, the MYB protein upregulated PLT counts (β = 3.27, 95% CI = 3.16–3.39, Fig. 3b and Supplementary Data 8), while it downregulated RBC (β = −4.91, 95% CI = −5.03 to −4.80, Fig. 3d and Supplementary Data 9) and WBC counts (β = −1.25, 95% CI = −1.37 to −1.14, Fig. 3f and Supplementary Data 10). However, MYB gene had opposite effects compared to its protein on PLT, RBC, and WBC traits. Furthermore, the proteins BRAP, ERP29, and PPP1CC demonstrated positive effects across three cohorts (Fig. 3b, d, f and Supplementary Data 8–10). Additionally, the results of the Leave-One-Out analyses for PLT, RBC, and WBC are illustrated in Supplementary Figs. 4 and 5.Fig. 3Impact of significant biomarkers on platelet (PLT), red blood cell (RBC), and white blood cell (WBC) counts.Venn diagrams illustrate the overlap among three analytical components: yellow represents proteome-wide Bayesian colocalization analyses with blood cell counts (PP.H4.abf ≥ 0.3); blue denotes positive biomarkers identified in ET and PV; and light green represents transcriptome-wide Bayesian colocalization analyses with blood cell counts (PP.H4.abf ≥ 0.8). The intersection between yellow and blue indicates positive proteins that potentially share a causal variant with blood cell counts, whereas the intersection between blue and light green indicates positive genes with shared causal variants. Forest plots display the MR results, with biomarkers derived from the inverse variance weighted method (IV ≥ 2) and the Wald ratio method (IV = 1). All biomarkers in the forest plots are statistically significant after applying Bonferroni correction and through sensitivity analyses, including tests for heterogeneity, pleiotropy, RadialMR, MR-PRESSO, and Leave-One-Out analyses. Bayesian colocalization results, along with PP.H4.abf values, are displayed in the forest plots. The 95% confidence interval (CI) for β is presented. In addition, the MYB gene and protein are highlighted in bold, with orange squares indicating PP.H4.abf ≥ 0.8, and gray text representing PP.H4.abf <0.8. a and b depict PLT; c and d depict RBC; and e and f depict WBC. In (a), PDE5A from deCODE corresponds to entry 16805. In (f), PDE5Aa and PDE5Ab correspond to entries 16805 and 5256 in the deCODE dataset, respectively. Impact of significant biomarkers on platelet (PLT), red blood cell (RBC), and white blood cell (WBC) counts. Venn diagrams illustrate the overlap among three analytical components: yellow represents proteome-wide Bayesian colocalization analyses with blood cell counts (PP.H4.abf ≥ 0.3); blue denotes positive biomarkers identified in ET and PV; and light green represents transcriptome-wide Bayesian colocalization analyses with blood cell counts (PP.H4.abf ≥ 0.8). The intersection between yellow and blue indicates positive proteins that potentially share a causal variant with blood cell counts, whereas the intersection between blue and light green indicates positive genes with shared causal variants. Forest plots display the MR results, with biomarkers derived from the inverse variance weighted method (IV ≥ 2) and the Wald ratio method (IV = 1). All biomarkers in the forest plots are statistically significant after applying Bonferroni correction and through sensitivity analyses, including tests for heterogeneity, pleiotropy, RadialMR, MR-PRESSO, and Leave-One-Out analyses. Bayesian colocalization results, along with PP.H4.abf values, are displayed in the forest plots. The 95% confidence interval (CI) for β is presented. In addition, the MYB gene and protein are highlighted in bold, with orange squares indicating PP.H4.abf ≥ 0.8, and gray text representing PP.H4.abf <0.8. a and b depict PLT; c and d depict RBC; and e and f depict WBC. In (a), PDE5A from deCODE corresponds to entry 16805. In (f), PDE5Aa and PDE5Ab correspond to entries 16805 and 5256 in the deCODE dataset, respectively. To exclude biomarkers that are only involved in ET and PV, we performed proteome and transcriptome-wide Bayesian colocalization analyses for other blood cell–related disorders (i.e., secondary polycythemia, thrombocytopenia-unspecified, other white blood cell disorders, and drug-induced neutropenia). Among the biomarkers identified in ET, PV, and MPN, HSPA1A (primarily implicated in PV with a PP.H4.abf of 0.79, Fig. 2d) was also found to be associated with other white blood cell disorders (PP.H4.abf of 0.57; Supplementary Data 11). All other positive biomarkers showed no association with other blood cell–related disorders. Causality among significant positive biomarkers Our analyses indicated that the MYB protein (Bonferroni-corrected P < 0.01, Fig. 4a and Supplementary Data 12,13) was significantly associated with increased levels of EPO (β = 5.67, 95% CI = 5.20–6.15), ERP29 (β = 1.17, 95% CI = 0.70–1.65), PPP1CC (β = 1.02, 95% CI = 0.55–1.49), and BRAP (β = 0.93, 95% CI = 0.47–1.40), while showing an inverse association with THPO (β = −1.15, 95% CI = −1.62 to −0.68). In contrast, The MYB gene exhibited opposite effects compared to its protein on the aforementioned proteins (Fig. 4a and Supplementary Data 12, 13). BRAP, PPP1CC, and ERP29, which share the same causal variant (rs10774624; Supplementary Fig. 6a–c), demonstrated consistent effects on MPL, RPN1, THPO, PARP1, and EPO (Fig. 4b and Supplementary Data 12, 13), with the exception of a negative control on EPO. Furthermore, the CALR gene, associated with the ATF6B protein (Supplementary Fig. 6d), and the JAK2 protein, associated with the MPL protein (Supplementary Fig. 6e), shared the same causal variant.Fig. 4Causal relationships among notable positive biomarkers and regulatory mechanisms involving MYB.Forest plots summarize the Mendelian Randomization (MR) results, derived using the inverse variance weighted method (IV ≥ 2) and the Wald ratio method (IV = 1), along with Bayesian colocalization findings (PP.H4.abf). All results are statistically significant after applying Bonferroni correction (except for panel e, line 2, which corresponds to PV as the outcome), and after undergoing sensitivity analyses, including tests for heterogeneity, pleiotropy, RadialMR, MR-PRESSO, and Leave-One-Out analyses. Bayesian colocalization results, along with PP.H4.abf values, are displayed in the forest plots. The 95% confidence interval (CI) for β or the odds ratio (OR) is presented. a and b illustrate causal relationships among selected positive biomarkers. c–f depict how MYB gene expression regulates MYB protein abundance, and how hypermethylation at cg19623877 is associated with ET. a highlights the regulatory effects of MYB protein (from deCODE) and MYB gene expression (from eQTLGen) on EPO, ERP29, PPP1CC, BRAP, and THPO. b outlines the influence of BRAP, ERP29, and PPP1CC on MPL, RPN1, THPO, PARP1, and EPO protein levels. c illustrates the genetically supported regulation of MYB protein abundance by MYB gene expression. d demonstrates the effects of rs9376092 on MYB gene expression and MYB protein levels. e shows the methylation of cg19623877 in the MYB gene as the exposure, with ET and PV as outcomes. f depicts the cg19623877 methylation on both MYB gene regulation and MYB protein expression. In (a) and (b), gray text indicates PP.H4.abf <0.8. In (c), (e), and (f), gray text indicates PP.H4.abf <0.5. Additionally, MYB is italicized when referring to the gene and presented in regular font when referring to the protein. Causal relationships among notable positive biomarkers and regulatory mechanisms involving MYB. Forest plots summarize the Mendelian Randomization (MR) results, derived using the inverse variance weighted method (IV ≥ 2) and the Wald ratio method (IV = 1), along with Bayesian colocalization findings (PP.H4.abf). All results are statistically significant after applying Bonferroni correction (except for panel e, line 2, which corresponds to PV as the outcome), and after undergoing sensitivity analyses, including tests for heterogeneity, pleiotropy, RadialMR, MR-PRESSO, and Leave-One-Out analyses. Bayesian colocalization results, along with PP.H4.abf values, are displayed in the forest plots. The 95% confidence interval (CI) for β or the odds ratio (OR) is presented. a and b illustrate causal relationships among selected positive biomarkers. c–f depict how MYB gene expression regulates MYB protein abundance, and how hypermethylation at cg19623877 is associated with ET. a highlights the regulatory effects of MYB protein (from deCODE) and MYB gene expression (from eQTLGen) on EPO, ERP29, PPP1CC, BRAP, and THPO. b outlines the influence of BRAP, ERP29, and PPP1CC on MPL, RPN1, THPO, PARP1, and EPO protein levels. c illustrates the genetically supported regulation of MYB protein abundance by MYB gene expression. d demonstrates the effects of rs9376092 on MYB gene expression and MYB protein levels. e shows the methylation of cg19623877 in the MYB gene as the exposure, with ET and PV as outcomes. f depicts the cg19623877 methylation on both MYB gene regulation and MYB protein expression. In (a) and (b), gray text indicates PP.H4.abf <0.8. In (c), (e), and (f), gray text indicates PP.H4.abf <0.5. Additionally, MYB is italicized when referring to the gene and presented in regular font when referring to the protein. Relationship between the MYB gene and its encoded protein Transcriptome-wide Bayesian colocalization analyses revealed a strong association between the MYB gene and its protein (Supplementary Data 12). Further MR framework confirmed that the association between the two is negative (β = −0.64, 95% CI = −0.98 to −0.30, P = 1.94 × 10⁻⁴, Fig. 4c and Supplementary Data 14). Additionally, rs9376092, which has been shown to reduce MYB expression in other studies30, was identified as a cis-eQTL and cis-pQTL for the MYB gene and protein. It was negatively linked to the gene but positively linked to the protein (Fig. 4d). The MYB methylation batch Bayesian colocalization analysis revealed that cg19623877 shares variation (Supplementary Data 15) with ET, but not with PV. Subsequent MR confirmed that cg19623877 is negatively related (Fig. 4e and Supplementary Data 14) to ET. However, colocalization between cg19623877 and MYB gene expression and protein levels was relatively weak (Fig. 4f and Supplementary Data 15), indicating limited evidence for a direct shared causal variant. Hypermethylation of cg17916418 and cg23511909 enhances ET and PV by downregulating RPN1 protein Our analyses showed that RPN1 protein levels were inversely associated with ET and PV risk, whereas BRAP, ERP29, and PPP1CC, which were consistently upregulated across three cohorts, were positively associated with RPN1 protein abundance. These findings suggest the presence of additional regulatory mechanisms influencing RPN1 expression. Given that DNA methylation is commonly regarded as a transcriptionally repressive epigenetic modification, we next examined CpG sites at the RPN1 locus. A total of 22 CpG sites were annotated, of which seven had available data suitable for Bayesian colocalization analyses. Among these, cg17916418 and cg23511909 were identified as sharing putative causal variants with RPN1 protein levels in both ET and PV (Supplementary Data 16). Subsequent MR showed that (Bonferroni-corrected P < 0.025, Fig. 5a and Supplementary Data 17) cg17916418 (β = −0.38, 95% CI = −0.42 to −0.34) and cg23511909 (β = −1.17, 95% CI = −1.29 to −1.04) suppress RPN1 protein while both promoted ET and PV (Fig. 5b and Supplementary Data 17). We then performed mediation MR to confirm that RPN1, as a mediator for cg17916418 and cg23511909, participated in ET and PV, and all mediated proportions were above 96% (Fig. 5c).Fig. 5Hypermethylation of cg17916418 and cg23511909 promotes ET and PV by downregulating RPN1 protein expression.Forest plots summarize the Mendelian Randomization (MR) results, derived using the inverse variance weighted method (IV ≥ 2) and the Wald ratio method (IV = 1), along with Bayesian colocalization findings (PP.H4.abf). Results with PP.H4.abf <0.8 are displayed in gray text. All MR results are statistically significant after applying Bonferroni correction and undergoing sensitivity analyses, including tests for heterogeneity, pleiotropy, RadialMR, MR-PRESSO, and Leave-One-Out analyses. Bayesian colocalization results, along with PP.H4.abf values, are shown in the forest plots. The 95% confidence interval (CI) for β or the odds ratio (OR) is provided. a illustrates the effects of DNA methylation at cg17916418 and cg23511909 on RPN1 protein expression. b shows the associations of cg17916418 and cg23511909 with ET and PV risk. c depicts mediation analyses supporting a role for RPN1 protein expression in mediating the effects of cg17916418 and cg23511909 on ET and PV. Hypermethylation of cg17916418 and cg23511909 promotes ET and PV by downregulating RPN1 protein expression. Forest plots summarize the Mendelian Randomization (MR) results, derived using the inverse variance weighted method (IV ≥ 2) and the Wald ratio method (IV = 1), along with Bayesian colocalization findings (PP.H4.abf). Results with PP.H4.abf <0.8 are displayed in gray text. All MR results are statistically significant after applying Bonferroni correction and undergoing sensitivity analyses, including tests for heterogeneity, pleiotropy, RadialMR, MR-PRESSO, and Leave-One-Out analyses. Bayesian colocalization results, along with PP.H4.abf values, are shown in the forest plots. The 95% confidence interval (CI) for β or the odds ratio (OR) is provided. a illustrates the effects of DNA methylation at cg17916418 and cg23511909 on RPN1 protein expression. b shows the associations of cg17916418 and cg23511909 with ET and PV risk. c depicts mediation analyses supporting a role for RPN1 protein expression in mediating the effects of cg17916418 and cg23511909 on ET and PV. Discussion In this study, we applied a post-GWAS framework prioritizing colocalization followed by MR to systematically interrogate the molecular architecture underlying ET and PV. Using this approach, we identified MYB as a subtype-specific molecular feature, with genetically supported associations at both the gene and protein levels in ET but not in PV. Notably, our analyses revealed opposing genetic effect directions for MYB gene expression and circulating MYB protein levels with respect to ET risk, blood cell counts, and downstream molecular traits. Specifically, the ET-associated variant rs9376092, located in the intergenic region between HBS1L and MYB, was associated with reduced MYB transcript abundance while showing a positive association with circulating MYB protein levels. While reduced transcript levels are often accompanied by concordant reductions in protein abundance, accumulating evidence indicates that MYB regulation is highly context dependent31. Given that MYB encodes a transcription factor that plays a central role in hematopoietic regulation, its activity is governed by multilayered regulatory mechanisms that extend beyond transcriptional abundance alone. Several mechanisms may contribute to transcript–protein discordance in gene regulation, including post-transcriptional regulation such as microRNA-mediated repression, variation in translational efficiency, and differences in protein stability or turnover32–34. Consistent with this regulatory complexity, MYB protein has a relatively short half-life and is subject to multiple post-translational modifications, including ubiquitylation, phosphorylation, acetylation and SUMOylating, which influence both protein stability and transcriptional activity31. Although the present genetic analyses cannot directly distinguish among these mechanisms, these regulatory layers provide biologically plausible explanations for the discordance between MYB transcript and circulating protein levels observed in our study. In line with the interpretation above, several lines of prior evidence support the biological plausibility of such discordant regulation in ET30,35–37. Seminal experimental studies in murine models have demonstrated that reduced c-Myb activity compromises hematopoietic stem cell function and induces an ET-like myeloproliferative phenotype, characterized by a skewing of hematopoietic output toward the megakaryocytic lineage, marked thrombocytosis, and relative attenuation of erythroid output35. Consistent with this lineage bias, GWAS have identified rs9376092 as a germline variant associated with reduced MYB expression in normal myeloid cells and context-dependent effects across MPN subtypes, showing a preferential association with ET rather than PV30. However, subsequent experimental studies have reported concordant reductions in MYB mRNA and protein levels only within specific developmental contexts36. Notably, these effects were confined to early life and were not observed to be maintained across the full lifespan of Myb⁺/− animals, underscoring the temporal and contextual specificity of MYB regulation36. Consistent with this context-dependent regulation, similar RNA–protein discordance has also been reported in adenoid cystic carcinoma, where integrated immunohistochemistry and RNA sequencing analyses have revealed an imperfect and method-dependent relationship between MYB transcript levels and protein abundance37. Beyond this ET-specific signal, we identified a set of shared molecular biomarkers exhibiting coordinated genetic effects across both ET and PV, including PARP1, RPN1, BRAP, ERP29, and PPP1CC. These biomarkers consistently showed evidence of a shared genetic architecture underlying disease susceptibility and hematologic phenotypes, suggesting the presence of common pathogenic pathways in ET and PV. Among these shared traits, PARP1 emerges as a biologically plausible candidate given its well-established role in the DNA damage response38. Our genetic analyses indicate that increased PARP1 levels are associated with higher risks of ET and PV, suggesting that dysregulated DNA damage repair may contribute to disease pathogenesis. PARP inhibitors, such as olaparib and talazoparib, are clinically used in ovarian and breast cancers, particularly in the context of BRCA1/2-associated homologous recombination deficiency39. In response to single-strand break or double-strand break damage, PARP1 is typically activated. PARP inhibitors block single-strand break repair, and in the presence of homologous recombination deficiency, which already compromises double-strand break repair, this leads to tumor cell death through synthetic lethality40. Importantly, prior experimental studies provide supportive evidence for the relevance of PARP inhibition in MPN. For example, primary MPN samples have been shown to exhibit increased ex vivo sensitivity to PARP inhibitors41, and murine models driven by JAK2 or CALR mutations demonstrate enhanced responsiveness to PARP inhibition, particularly when combined with JAK inhibition42. In addition, a phase I clinical trial has suggested potential activity of PARP inhibitor-based combination regimens in aggressive MPN, chronic myelomonocytic leukemia, and related myeloid malignancies43. Together, these findings suggest that altered DNA damage repair may contribute to MPN biology, although the therapeutic application of PARP inhibition in MPN requires further validation. RPN1 (ribophorin I), a multifunctional protein involved in endoplasmic reticulum protein processing and proteasome regulation44, showed consistent inverse associations with ET and PV risk in our analyses. In line with these findings, prior cytogenetic evidence has linked RPN1 to myeloproliferative neoplasms. A rare reciprocal t(3;9) (q21;p24) translocation involving the RPN1 and JAK2 loci has been reported in chronic idiopathic myelofibrosis, suggesting that RPN1 resides within genomic regions recurrently involved in MPN-related chromosomal rearrangements45. Moreover, hypermethylation at two CpG sites (cg17916418 and cg23511909) was associated with reduced RPN1 protein levels and increased ET and PV risk. Taken together, these results suggest that epigenetic regulation at the RPN1 locus may contribute to disease susceptibility or progression, rather than implicating RPN1 as a direct therapeutic target. Given that hypomethylating agents are already used clinically in MPN and related myeloid neoplasms, these observations are consistent with the broader relevance of epigenetic dysregulation in these disorders, while underscoring the need for functional studies to clarify the role of RPN1-associated methylation changes. BRAP, ERP29, and PPP1CC, which share the same causal variant, showed genetically supported associations with ET and PV and were linked to increased PLT, RBC, and WBC counts. These proteins appear to be regulated by MYB and are associated with coordinated effects on downstream molecular traits, including elevated MPL, RPN1, THPO, and PARP1 levels, alongside reduced EPO production. Overall, these upstream and downstream associations suggest that these proteins exhibit close interactions and may participate in a shared regulatory pathway relevant to disease pathophysiology. Although our study provides genetic evidence implicating BRAP, ERP29, and PPP1CC in PV and ET, the precise roles of these proteins, together with RPN1, in hematopoietic regulation, clonal expansion, and inflammatory signaling remain poorly defined and require further functional validation46,47. As is well known, subnormal serum EPO levels are a minor criterion for PV48, and our findings may help explain this observation. In reverse MR analyses, treating PV as the exposure and EPO as the outcome, a negative association was observed (Supplementary Fig. 7 and Supplementary Data 18), although the magnitude of this effect appears modest. Consistent with these observations, BRAP, ERP29, and PPP1CC show coordinated genetic associations with increased RBC levels and reduced EPO production, while MYB appears to modulate these proteins and erythroid output without affecting PV disease status. This convergence of genetic effects may provide a plausible explanation for the reduced serum EPO levels observed in PV. Our study also found that MYB protein significantly increased EPO production while inhibiting THPO expression. Despite its effect on THPO, MYB was still associated with increased PLT count. Indeed, EPO49, THPO50, and blood cell counts represent distinct aspects of hematopoietic regulation: while blood cell counts (such as PLT, RBC, and WBC) are terminal cellular outputs, EPO and THPO are systemic cytokines regulated by complex feedback mechanisms. Thus, the associations between MYB protein levels and these traits do not follow a single linear pathway. The link between MYB protein and increased PLT counts is likely driven by intrinsic regulation of hematopoietic stem cell differentiation, promoting megakaryopoiesis independently of THPO signaling. Conversely, the inverse relationship between MYB and THPO expression suggests compensatory feedback mechanisms, rather than a direct causal requirement of THPO for platelet production. Similarly, the positive association between MYB and EPO expression reflects a different layer of regulation related to erythroid homeostasis and systemic signaling, with both effects operating through parallel pathways. Finally, MR captures long-term genetic effects, reflecting the cumulative impact, rather than acute physiological responses51. Additionally, an important clinical question is whether molecular features identified in chronic-phase MPNs are also relevant to leukemic transformation and the development of post-MPN secondary acute myeloid leukemia. Although our study was not designed to evaluate AML transformation risk, several candidates identified here have been implicated in AML biology more broadly. MYB is a well-established regulator of hematopoietic stem and progenitor cell self-renewal and leukemogenic programs52, while PARP1 has been associated with AML prognosis and therapeutic vulnerability53. In addition, the RPN1 locus is involved in inv (3)/t (3;3) AML, highlighting a potential intersection between MPN-associated molecular regulation and AML-related genomic architecture54. By contrast, evidence linking BRAP, ERP29, and PPP1CC to secondary AML remains limited, and their roles in leukemic transformation are currently speculative. Importantly, post-MPN AML exhibits a distinct genetic landscape characterized by complex karyotypes and high-risk somatic alterations55, which are not captured by currently available GWAS-based frameworks. Several limitations of this study should be acknowledged. First, although our colocalization and MR analyses provide genetically informed evidence supporting associations between multiple molecular traits and ET/PV risk, the study remains observational and hypothesis-generating in nature. We did not perform functional experiments and therefore cannot establish direct mechanistic causality or therapeutic efficacy for the identified biomarkers. Second, although we focused primarily on ET and PV, the integration of multiple large-scale datasets may introduce heterogeneity related to cohort composition, phenotype definitions, and technical platforms. In addition, all datasets analyzed were derived from individuals of European ancestry, which may limit the generalizability of our findings to other populations. Finally, while we aimed to improve biological interpretability by integrating proteomic, transcriptomic, and methylation layers, the causal ordering and functional hierarchy among these molecular levels remain to be resolved in disease-relevant experimental systems. Supplementary information Transparent Peer Review file Supplementary Information Description of Additional Supplementary files Supplementary Data 1 Supplementary Data 2 Supplementary Data 3 Supplementary Data 4 Supplementary Data 5 Supplementary Data 6 Supplementary Data 7 Supplementary Data 8 Supplementary Data 9 Supplementary Data 10 Supplementary Data 11 Supplementary Data 12 Supplementary Data 13 Supplementary Data 14 Supplementary Data 15 Supplementary Data 16 Supplementary Data 17 Supplementary Data 18 Transparent Peer Review file Supplementary Information Description of Additional Supplementary files Supplementary Data 1 Supplementary Data 2 Supplementary Data 3 Supplementary Data 4 Supplementary Data 5 Supplementary Data 6 Supplementary Data 7 Supplementary Data 8 Supplementary Data 9 Supplementary Data 10 Supplementary Data 11 Supplementary Data 12 Supplementary Data 13 Supplementary Data 14 Supplementary Data 15 Supplementary Data 16 Supplementary Data 17 Supplementary Data 18