A multilevel genomic approach to uncover causal connections between CPFE and lung cancer subtypes: A two-sample Mendelian randomization study.

In summary, our study provides a multilevel genomic view of how CPFE-associated genes might drive oncogenesis in 3 main lung cancer subtypes. By combining transcriptomic profiling with robust Mendelian randomization validation, we identified pivotal genes such as CPPED1 and CD300LF that appear to mediate tumor-promoting or tumor-suppressive effects in CPFE. These findings offer a foundation for future large-scale validation and mechanistic interrogation, with the ultimate goal of refining both the diagnostic assessment and targeted treatment of CPFE patients at high-risk for lung cancer.

Combined pulmonary fibrosis and emphysema (CPFE) is a distinct pulmonary disorder characterized by the coexistence of upper-lobe predominant emphysema and lower-lobe diffuse pulmonary fibrosis. [ 1 , 2 ] Initially described in case reports, CPFE was formally recognized as an independent radiological entity in 2005 when Cottin et al systematically analyzed 61 patients. [ 1 – 3 ] Importantly, CPFE not only presents distinct pulmonary pathology but also confers a substantially increased risk of lung cancer compared to isolated idiopathic pulmonary fibrosis (IPF) or chronic obstructive pulmonary disease (COPD). [ 2 , 4 – 9 ] Squamous cell carcinoma (SCC) and adenocarcinoma (ADC) are the predominant histological subtypes of CPFE-associated lung cancer, with SCC frequently arising in fibrotic regions of the lower lung lobes and often displaying keratinizing features. [ 5 , 6 ] However, the causal relationships between CPFE-associated genes and specific lung cancer subtypes remain largely unexplored, and key driver genes have yet to be fully elucidated. Mendelian randomization (MR), a genetic variation-based approach for causal inference, has been employed to evaluate the potential causal effects of CPFE-associated genes on lung cancer development. This study integrates transcriptomic and genomic data to systematically assess the causal relationships between CPFE-related differentially expressed genes (DEGs) and the 3 major lung cancer subtypes: SCC, ADC, and small cell lung cancer (SCLC). Additionally, 3 independent validation steps were performed to ensure the robustness of MR findings through external consistency verification. Our results not only provide molecular insights into the link between CPFE and lung cancer but also offer a theoretical foundation for early intervention and precision therapy in high-risk patients.

Conceptualization: Xinhua Chai, Huanhuan Zhao, Yongxia Bao. Data curation: Chengcheng Yang, Ping Chen. Formal analysis: Chengcheng Yang, Ping Chen. Investigation: Chengcheng Yang, Ping Chen, Qiuyuan Yang, Siyu Wu, Zefeng Wang. Methodology: Xinhua Chai, Huanhuan Zhao, Yongxia Bao. Resources: Qiuyuan Yang, Siyu Wu, Zefeng Wang. Software: Qiuyuan Yang, Siyu Wu, Zefeng Wang. Supervision: Xinhua Chai, Huanhuan Zhao, Yongxia Bao. Writing – original draft: Chengcheng Yang, Ping Chen. Writing – review & editing: Xinhua Chai, Qiuyuan Yang, Siyu Wu, Huanhuan Zhao, Zefeng Wang, Yongxia Bao.

This study obtained data from 3 CPFE patients’ fibrotic and emphysematous lung tissue samples as the experimental group from the gene expression omnibus (GEO) database. Normal lung tissue samples were also retrieved from the same platform as controls. Limma analysis was performed separately for the CPFE fibrotic versus control group and the emphysematous versus control group to identify (DEGs), which were then merged to obtain the CPFE versus control DEGs. Corresponding gene and single-nucleotide polymorphism (SNP) data were obtained from the MRC integrative epidemiology unit (IEU) OpenGWAS, and after removing confounding factors, MR analysis was conducted using GWAS data for SCC, ADC, and SCLC to identify associated genes. Three rounds of validation were performed. The first validation used cis-expression quantitative trait loci (cis-eQTL) data from the eQTLGen Consortium to match with the associated genes as exposures and lung cancer GWAS data as the outcome for MR analysis, identifying spare genes. The second validation used genotype-tissue expression (GTEx) V8 eQTL data from the GTEx Consortium to match the spare genes and performed summary Mendelian randomization (SMR) analysis with lung cancer GWAS data as the outcome to identify stronger associations between CPFE and lung cancer. The third validation involved matching protein quantitative trait loci (pQTL) data from the UK Biobank Pharma Proteomics Project (UKB-PPP) with the genes selected in the previous step, and performing MR analysis with lung cancer GWAS data to identify the final strongly associated genes (Fig. 1 ). The research content and ideas of this study. Raw microarray data were obtained from GEO ( http://www.ncbi.nlm.nih.gov/geo/ ) and comprised GSE38934 ( GPL570 ; fibrotic lesion samples: GSM952433 , GSM952435 , GSM952437 ; emphysematous lesion samples: GSM952432 , GSM952434 , GSM952436 ) [ 10 ] and normal lung tissues from GSE18842 ( GPL570 ; GSM466948 , GSM466950 , GSM466953 ). [ 11 ] Data were normalized via the robust multi-array average (RMA) procedure, ensuring background correction, logarithmic transformation, and standardization. Sample quality was validated using the arrayQualityMetrics package. Affymetrix probe IDs were mapped to GeneSymbols using the hgu133plus2.db annotation file. Nonspecific probes mapping to multiple genes were excluded, and for genes represented by multiple probes, the probe-level expressions were averaged. Limma’s linear model was applied to identify DEGs, with Bayesian adjustment for p -values. DEGs were defined by an adjusted P -value  1, and classified as upregulated or downregulated based on the positive and negative values of the fold change. In this study, SNP data related to gene expression were obtained from the IEU OpenGWAS Project for association analysis ( P -value < 5 × 10− 08 ). The GWAS clumping method (kbfilter = 10,000, r 2 filter = 0.001) was applied for linkage disequilibrium (LD) filtering based on the European population to remove SNPs in LD, followed by the removal of weak instrumental variables (using an F -statistic threshold of 10). By integrating with CPFE DEGs, SNPs corresponding to these DEGs and their related information were obtained as exposure data. For each exposure gene, we then calculated the total variance explained ( R 2 ) by its independent SNPs and converted this to Cohen f 2 . Under a 2-sided significance level of α = 0.05, and the respective GWAS sample sizes for the 3 lung cancer subtypes, statistical power was estimated using the pwr.f2.test framework. All genes achieved power >80% across SCC, ADC, and SCLC (Table S1, Supplemental Digital Content, https://links.lww.com/MD/P739 ). The FastTraitR package was used to query the association between exposure factor SNPs and lung cancer in the GWAS Catalog database, with a significance threshold of P  < 1 × 10 −05 . SNPs associated with confounders such as COPD, smoking, and cancer were then excluded. For SCC, we retrieved a GWAS dataset (ID: ieu-a-967; cases = 3275; controls = 15,038) from the IEU OpenGWAS database ( https://gwas.mrcieu.ac.uk/ ). For ADC (GCST004744; cases = 11,273; controls = 55,483) and SCLC (GCST004746; cases = 2664; controls = 21,444), outcome data were sourced from the GWAS Catalog ( https://www.ebi.ac.uk/gwas/home ). These datasets were used as outcomes in the subsequent MR analyses. MR analysis was performed using the TwoSampleMR R package (version 0.6.6; https://github.com/MRCIEU/TwoSampleMR ). We aligned effect alleles between the exposure and outcome datasets, discarding palindromic SNPs with ambiguous allele frequencies. We used the inverse variance weighted (IVW) method as the primary approach. We set the statistical significance level at P  < .05. Additionally, we used Mendelian Randomization Egger regression (MR-Egger), weighted median, simple mode, and weighted mode as supplementary methods, ensuring that the direction of the odds ratios (OR) was consistent with those from the IVW method to reflect the robustness of the results. We applied false discovery rate (FDR) correction to the IVW results, which are presented for indicative purposes. Furthermore, we removed pleiotropic effects by excluding SNPs with a pleiotropy P -value <.05, thereby further ensuring the reliability of the analysis. To ensure the robustness of our Mendelian randomization results, we conducted a series of sensitivity analyses. For each gene-phenotype pair (exposure-outcome), we assessed heterogeneity using Cochran’s Q test, evaluated horizontal pleiotropy via the MR-Egger intercept test, and applied the Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO) global test to detect and correct potential outlier SNPs. We then employed the Steiger directionality test to confirm that the instruments explain more variance in the exposure than in the outcome, thereby validating the causal direction. With these preparations, the 3 core assumptions of the Mendelian randomization design were realized: the genetic instrument must be strongly associated with the exposure factor; the genetic instrument affects the outcome solely through the exposure factor and not through other pathways; the genetic instrument is not associated with confounding factors related to the exposure-outcome relationship. Cis-eQTL summary statistics and SNP frequency data were obtained from the eQTLGen Consortium. The datasets included “Full cis-eQTL summary statistics” and “Allele frequencies based on 26,609 eQTLGen samples (excluding FHS).” Datasets were merged, and key parameters (effect allele frequency, β values, and standard errors) were calculated. SNPs were filtered to retain only those with P  < 5 × 10 −8 and with LD clumping ( r 2  = 0.001, kb = 10,000) based on the 1000 Genomes Project (European population). SNPs with F-statistics < 20 were excluded. The filtered cis-eQTL data were matched with the associated genes identified from the MR analysis as exposures, and lung cancer GWAS data were used as outcomes for MR analysis. The IVW method was used to estimate the causal effects, retaining only genes that were statistically significant ( P   .05). These genes were designated as spare genes for downstream analyses. GTEx V8 eQTL data were obtained from the GTEx Consortium, including only SNPs with P  < 1 × 10 −5 . Spare gene-related cis-eQTL summary statistics were converted into binary expression summary data (BESD) format for SMR analysis. The European 1000 Genomes Project data were used as the LD reference panel. The heterogeneity in dependent instruments (HEIDI) test was conducted to distinguish between horizontal pleiotropy and linkage effects. Lung cancer GWAS summary statistics were integrated with the eQTL data and analyzed using SMR software (v1.3.1). Genes were selected based on an SMR P -value threshold ( P -SMR  .05. The European pQTL data from the UKB-PPP website were downloaded, decompressed, and merged. The data were then formatted using the “BSgenome.Hsapiens.NCBI.GRCh38,” “NPlocs.Hsapiens.dbSNP155.GRCh38,” and “MungeSumstats” packages to meet the requirements for further analysis. The data were filtered by selecting SNPs with P  < 1 × 10 −5 for significance. Additionally, to reduce the impact of LD, the clumping method ( r ² = 0.1, kb = 10,000) was applied for LD filtering based on the European population, and weak instrumental variables with an F -statistic < 10 were excluded. The filtered pQTL data were then matched with the genes obtained from the SMR analysis as exposures, with lung cancer GWAS data as the outcome for MR analysis. Genes that were significant under the IVW method ( P   .05) were retained, resulting in the final strongly associated genes.

Comparisons between fibrotic lesions and normal controls identified 809 DEGs (253 upregulated and 556 downregulated) (Table S2, Supplemental Digital Content, https://links.lww.com/MD/P739 ), as illustrated by the heatmaps and volcano plots (Fig. 2 ). Heatmaps of DEGs between CPFE emphysema (A)/fibrotic (C) tissues and the control group, indicating significant transcriptional alterations. Volcano plots of DEGs between CPFE emphysema (B)/fibrotic (D) tissues and the control group provide a clear visualization of the DEGs. In the volcano plots, red points represent upregulated genes, blue points represent downregulated genes, and gray points indicate nonsignificant genes. These plots Heatmaps and volcano plots of DEGs between CPFE emphysema/fibrotic tissues and the control group. CPFE = combined pulmonary fibrosis and emphysema, DEGs = differentially expressed genes. Comparisons between emphysematous lesions and controls yielded 13 DEGs (9 upregulated and 4 downregulated) (Table S3, Supplemental Digital Content, https://links.lww.com/MD/P739 ), all of which overlapped completely with the fibrotic lesion-derived DEGs, including the regulation direction. Thus, a final set of 809 CPFE-related DEGs was established. Quality control analyses detected no significant outliers or batch effects (see Supplemental Materials 1 and 2, Supplemental Digital Content, https://links.lww.com/MD/P740 for the control-analysis reports). A total of 809 CPFE-related genes were queried in the IEU OpenGWAS database to identify corresponding eQTLs (Table S4, Supplemental Digital Content, https://links.lww.com/MD/P739 ). After removing 15 SNPs (Table S5, Supplemental Digital Content, https://links.lww.com/MD/P739 ) flagged as potential confounders for chronic obstructive pulmonary disease, smoking, or cancer, the remaining SNPs were used for MR analysis to assess the causal effects of lung cancer subtypes (Table S6, Supplemental Digital Content, https://links.lww.com/MD/P739 ). We observed significant associations between several CPFE-related genes and SCC, ADC, and SCLC. Full Mendelian randomization effect estimates for all genes are provided in Table S7, Supplemental Digital Content, https://links.lww.com/MD/P739 , and detailed SNP information with instrument strength metrics are provided in Table S8, Supplemental Digital Content, https://links.lww.com/MD/P739 . Nine genes met the threshold for SCC: 5 risk-enhancing ( ADGRE1, CA4, C1orf162, GZMH, GIMAP6 ; OR > 1) and 4 protective ( ASF1A, CPPED1, CDK2AP2, FADS1 ; OR < 1) (Figs. 3 and 6 A). Seven genes showed significant links with ADC: KLRF1 and GSTO1 increased risk, whereas CYP1B1, FAM167A, CD33, CEBPA and notably FOXO4 (strongest effect) were protective (Figs. 4 and 6 B). Ten genes correlated with SCLC: 7 conferred risk ( AKAP10, CD160, CPNE2, CD300LF, FAM167A, FGFBP2, IER3 ) and 3 were protective ( ARRDC2, DOK4, H2BC12 ) (Figs. 5 and 6 C). The bidirectional behavior of FAM167A – protective in ADC yet deleterious in SCLC – underscores tissue-specific functionality. Forest plot of MR analysis results for SCC. MR = Mendelian randomization, SCC = squamous cell carcinoma. Forest plot of MR analysis results for ADC. MR = Mendelian randomization, ADC = adenocarcinoma. Forest plot of MR analysis results for SCLC. MR = Mendelian randomization, SCLC = small cell lung cancer. Volcano plots of gene effect sizes versus significance for SCC (A), ADC (B), and SCLC (C). ADC = adenocarcinoma, SCC = squamous cell carcinoma, SCLC = small cell lung cancer. To ensure the validity of our Mendelian randomization estimates, we conducted a comprehensive sensitivity analysis for each gene–outcome pair (Table S9, Supplemental Digital Content, https://links.lww.com/MD/P739 ). Cochran Q test yielded Q statistics from 0.009 to 19.67 (df = 1–11) with all P  > .05, indicating no significant heterogeneity among the SNP instruments. MR-Egger intercepts were uniformly close to zero (estimates between −0.021 and 0.079; SEs 0.014–0.105) and nonsignificant (all P  > .05), demonstrating absence of unbalanced horizontal pleiotropy. MR-PRESSO global tests – performed for gene–outcome pairs with ≥ 3 SNPs – detected no significant outliers (all global test P  > .05); those with fewer than 3 instruments were inapplicable for this test. Steiger directionality tests were highly significant and confirmed that genetic variants explained more variance in gene expression than in the cancer outcomes. These results support the robustness of our primary MR findings and substantiate a causal effect of CPFE-related gene expression on lung cancer subtypes. All associated genes were sequentially subjected to 3 rigorous validations, and a summary of all validation stages is provided in Table 1 (Table S10, Supplemental Digital Content, https://links.lww.com/MD/P739 for full dataset.) Results from 3 independent validations. CI = confidence interval, eQTL = expression quantitative trait loci, MR = Mendelian randomization, OR = odds ratios, pQTL = protein quantitative trait loci, SMR = summary Mendelian randomization. P <.05, – indicates no result obtained. MR analysis was performed by matching the aforementioned associated genes with cis-eQTL data from the eQTLGen Consortium. The spare genes were identified, and the results showed: SCC: ASF1A , CPPED1 , and FADS1 all exhibited significant associations in the IVW analysis, with OR all <1, suggesting a potential protective effect, consistent with the MR analysis. ADC: FAM167A , CEBPA , and GSTO1 were significantly associated with adenocarcinoma risk, with FAM167A and CEBPA showing OR values 1, aligning with the MR analysis direction. SCLC: FAM167A, CD300LF, ARRDC2, CPNE2, DOK4, and IER3 all reached significant levels. ARRDC2 and DOK4 exhibited protective effects, while the other genes increased the risk, consistent with the MR analysis results. FAM167A remained protective in adenocarcinoma (OR  1), underscoring its tissue- or pathology-specific functional differences. SMR analysis was conducted on the above candidate genes using GTEx V8 eQTL data from the GTEx Consortium, and the results showed: In SCC, only ASF1A passed the SMR analysis successfully ( P -SMR = .0095, P -HEIDI = .5327; OR = 0.7016, 95% CI: 0.5368–0.9169), suggesting a more reliable causal association with SCC, and it may have a protective effect. In ADC, FAM167A , CEBPA , and GSTO1 did not maintain statistical significance in the SMR analysis, indicating that no genes passed the rigorous test in the second validation phase for this subtype. In SCLC, FAM167A ( P -SMR = .0014, P -HEIDI = .5601; OR = 1.1567, 95% CI: 1.0581–1.2644) and CD300LF ( P -SMR = .0116, P -HEIDI = .9557; OR = 2.3324, 95% CI: 1.2087–4.5008) continued to show significant associations, further supporting their potential pathogenic or susceptibility roles in the pathogenesis of SCLC. By performing MR analysis with the pQTL data from the UKB-PPP, genes strongly associated with lung cancer subtypes were identified. The results show that CPPED1 is strongly associated with SCC, with an OR value of 1, which is consistent with previous results, further confirming its high-risk role in SCLC.

In this study, we employed a comprehensive approach that integrated transcriptomic data and multilevel MR analyses to investigate the potential causal relationships between CPFE and the 3 major lung cancer subtypes – SCC, ADC, and SCLC. We identified 809 DEGs associated with CPFE and, through multiple validation steps, confirmed that some key genes have causal relationships with specific lung cancer subtypes. Because CPFE is a distinct pulmonary disease, it has a close association with lung cancer, particularly SCC. Numerous clinical studies have reported a markedly increased incidence of SCC in patients with CPFE and poorer prognosis. [ 5 , 6 , 8 ] Further MR analysis in our study revealed that certain genes, such as ASF1A , CPPED1 , and FADS1, exhibit potential protective roles in SCC. Notably, CPPED1 consistently showed a stable causal protective effect across multiple validation steps, suggesting an important tumor-suppressive function in the development of SCC in CPFE patients. Previous studies indicate that aberrant CPPED1 expression is closely related to the progression of various malignancies. CPPED1 is a newly identified serine/threonine protein phosphatase that specifically dephosphorylates key kinases in the phosphatidylinositol 3-kinase-protein kinase B (PI3K-AKT) pathway (e.g., AKT1 and PAK4), implying that it may exert its tumor-suppressive effect by inhibiting PI3K-AKT pathway activity. [ 12 ] Earlier research on laryngeal SCC found that downregulation of CPPED1 activates the PI3K-AKT signaling pathway, thereby promoting tumor cell migration and invasion while correlating with poor patient prognosis. [ 13 ] In light of these findings, we propose that decreased CPPED1 in CPFE patients might remove inhibitory control on the PI3K-AKT pathway and thus raise the risk of developing SCC. ASF1A , a member of the H3-H4 histone chaperone family, is involved in chromatin assembly, DNA replication, and damage repair, all of which are vital biological processes. Certain studies have shown that high ASF1A expression is closely associated with tumor progression. [ 14 , 15 ] Other research, however, has suggested potential heterogeneity regarding the impact of ASF1A expression on tumor aggressiveness. For instance, in lung adenocarcinoma, ASF1A does not exhibit notable aberrant expression nor is it significantly related to patient survival, whereas its homolog ASF1B is markedly elevated and strongly linked to poor prognosis. [ 16 ] Considering ASF1A ’s protective role in SCC observed in our study, we propose that ASF1A might perform distinct biological functions, or even act as a tumor suppressor, depending on the tumor type or subtype. Specifically, in the development of SCC under CPFE conditions, ASF1A expression may help maintain normal chromatin stability, preserve genomic integrity, and inhibit carcinogenesis. Our study also revealed that the FAM167A gene displays a potential risk association in both ADC and SCLC. FAM167A encodes a novel disordered protein, DIORA-1, which belongs to a highly conserved but functionally unclear family. [ 17 ] Recent research has shown that genetic variations in the FAM167A-BLK locus are significantly associated with various autoimmune diseases (such as rheumatoid arthritis, systemic lupus erythematosus, and Sjögren’s syndrome). [ 18 ] Intriguingly, FAM167A expression is highest in the lung, notably in alveolar macrophages and bronchial epithelium, and it exhibits a distinctive endosomal localization pattern, [ 17 ] suggesting a role in maintaining pulmonary immune homeostasis and regulating immune responses. Given the pronounced pulmonary inflammation and immune imbalance inherent in CPFE, the genetic risk posed by FAM167A might disrupt immune homeostasis within lung tissue, thereby triggering or promoting lung cancer development and progression. Moreover, we observed that the CD300LF gene exhibited a consistently strong risk association with SCLC, indicating that CD300LF may play a significant oncogenic role in the development of SCLC among CPFE patients. Previous studies have shown that CD300LF belongs to the CD300 family of immunoinhibitory receptors, is primarily expressed on the surface of immune cells, and is involved in various immunoregulatory functions, [ 19 ] correlating with persistent inflammation, immune imbalance, and the tumor immune microenvironment. [ 19 – 21 ] From a molecular perspective, CD300LF has been proven to recognize external phospholipids on cell membranes – particularly in mast cells, where it acts in tandem with CD300a to detect externalized phospholipids and suppress cell activation, thereby attenuating the strength of the immune response. [ 22 ] Furthermore, recent research on brain injuries found that CD300LF deficiency exacerbates microglial inflammatory activation and neuronal damage. [ 23 ] These findings suggest that CD300LF exhibits complex functionality, potentially serving as an immunosuppressor while also indirectly contributing to tumor progression through the regulation of inflammatory pathways. Finally, in our transcriptomic analysis, we found a noteworthy association between IER3 and the development of SCLC in CPFE patients. Past research indicated that in lung adenocarcinoma, overexpression of IER3 is accompanied by enhanced phosphorylated extracellular signal-regulated kinase levels, suggesting that IER3 may promote tumor cell proliferation and survival through the extracellular signal-regulated kinase (ERK) pathway. [ 24 , 25 ] Likewise, in glioma studies, high IER3 expression correlates with an immunosuppressive microenvironment, drives tumor cell migration and invasion, and leads to poor patient prognosis. [ 26 ] Mechanistically, IER3 can bind with the PP2A-B56γ subtype of protein phosphatase, affecting sustained ERK signaling and thus facilitating carcinogenesis. [ 25 ] Furthermore, in ovarian endometriosis, elevated IER3 expression is closely linked to nuclear factor kappa-light-chain-enhancer of activated B cells (NF-κB) pathway activation, jointly contributing to disease recurrence. [ 27 ] Therefore, the aberrant IER3 expression identified in our study may promote lung cancer progression in CPFE by activating key oncogenic pathways such as ERK and NF-κB. From a clinical standpoint, CPFE patients exhibit unique pathological characteristics and multiple molecular abnormalities that predispose them to an elevated risk of lung cancer and worse outcomes; to address this, future studies should expand and diversify cohorts through multicenter, large-sample validations to confirm the robustness of candidate genes and their causal links with cancer risk, integrate complementary omics layers – such as proteomics, metabolomics, and single-cell transcriptomics – to map the comprehensive regulatory network connecting CPFE and tumorigenesis, perform targeted in vitro and in vivo functional experiments to clarify the roles of key genes like CPPED1 and CD300LF in immune modulation, cell-cycle control, and metabolic reprogramming, and ultimately develop early-screening protocols and precision therapeutic strategies tailored to high-risk CPFE subgroups defined by molecular subtyping and clinical phenotype. Our integrative MR and transcriptomic analyses used summary statistics from predominantly European ancestry cohorts and microarray data from a limited number of CPFE patients. This may limit the applicability of the DEGs to other ethnic or geographic groups because of differences in allele frequencies, environmental exposures and platform specific biases. In addition, our genetic instruments represent lifelong exposure based on cross sectional tissue sampling and do not account for variability across CPFE stages or differences in disease severity. As a result, we cannot precisely characterize the relationship between molecular burden and lung cancer risk. Although the MR framework reduces many confounding effects and reverse causation issues inherent in observational studies, residual pleiotropy remains a concern and requires validation through targeted functional experiments. Finally, heterogeneity in populations, data generation platforms and detection methods may further limit external generalizability. To address these concerns, future research should include multicenter, multi population validations with quantitative CPFE phenotyping and complementary in vitro and in vivo mechanistic studies.

We sincerely thank the MRC integrative epidemiology unit (IEU) OpenGWAS Project, the GWAS Catalog, the Genotype-Tissue Expression (GTEx) Consortium, the eQTLGen Consortium, and the UK Biobank Pharma Proteomics Project (UKB-PPP) for providing open-access genetic and transcriptomic data that were essential for conducting Mendelian randomization and validation analyses in this study. We also acknowledge the Gene Expression Omnibus (GEO) database for the microarray datasets used in the identification of differentially expressed genes. These publicly available resources made it possible to perform comprehensive multi-omics analyses and ensure the robustness of our findings.