Transcriptome-Proteome Decoupling in Smoking-Induced Cardiovascular Disease: A Multi-Omics Mendelian Randomization Study

Objective: Smoking is a major modifiable cardiovascular risk factor. We aimed to dissect the causal relationships between multiple smoking phenotypes and cardiovascular diseases (CVDs), and critically evaluate whether smoking-induced molecular alterations propagate linearly from DNA methylation to gene expression and circulating proteins. Methods: We performed bidirectional two-sample Mendelian randomization (MR) for 11 smoking behaviors and 12 CVD outcomes, and applied multivariable MR (MVMR) to estimate direct effects independent of cardiometabolic confounders. To investigate molecular mechanisms, we integrated summary-data-based MR (SMR) with blood cis- expression quantitative trait locus (eQTL) data and performed Bayesian colocalization between smoking-associated DNA methylation (mQTL), gene expression, and plasma/cerebrospinal fluid (CSF) protein QTL (pQTL). Finally, we systematically interrogated the cross-platform coverage of candidate proteins across major proteomic datasets. Results: Genetically proxied "ever smoked" and "cigarettes per day" significantly increased the risk of CAD, PAD, and HF, with direct effects persisting independent of cardiometabolic confounders. eQTL-SMR identified 640 genes shared across CVD subtypes. However, this cascade was not linear. While 11 CpG-gene pairs showed robust mQTL-eQTL colocalization, the classical smoking marker AHRR showed a complete absence of colocalization (PP.H3 = 0.94–1.00), indicating that epigenetic imprints reflect exposure rather than causal mediators. A second layer of systemic decoupling was observed between transcription and the proteome. Among the seven candidate genes with robust mQTL-eQTL colocalization, six were entirely absent from the SomaLogic 7k official menu. For the sole covered gene, SPTLC2, eQTL-pQTL colocalization was negative in both plasma (PP.H1 = 0.65) and CSF (minimum P = 4.96×10⁻⁴). In parallel, pQTL MR independently identified 18 subtype-specific causal proteins (e.g., ABO for DVT/PE, CTF1/VPS29 for CAD) that were not direct downstream effectors of smoking. Conclusion: Smoking behaviours exert heterogeneous effects on CVDs through shared upstream transcriptional programs with subtype-specific downstream protein effectors. However, this process is characterized by multi-layer decoupling—driven by the systematic exclusion of intracellular regulatory proteins from current proteomic platforms. Our findings redefine the molecular architecture of smoking-induced CVD and highlight the necessity of distinguishing molecular enrichment from shared causal continuity in multi-omic MR studies.