循环炎症蛋白与胰腺癌风险的潜在因果关联的双样本孟德尔随机化研究

Potential causal associations between circulating inflammatory proteins and pancreatic cancer risk: a two-sample Mendelian randomization study

摘要:

背景与目的:炎症反应可能参与胰腺癌发生、发展，但传统观察性研究难以区分循环炎症相关蛋白与胰腺癌风险之间的潜在因果关系，以及疾病导致的继发性改变。本研究基于遗传流行病学方法，系统评估循环炎症相关蛋白与胰腺癌风险之间的潜在因果关联，并进一步筛选获得相对稳定遗传学支持的候选蛋白。 方法:基于公开全基因组关联研究（GWAS）汇总数据，采用双样本孟德尔随机化（MR）方法评估91种循环炎症相关蛋白与胰腺癌风险之间的潜在因果关联，以逆方差加权法（IVW）为主要分析方法，并结合加权中位数、MR-Egger回归及加权模式法进行稳健性分析。对符合预设筛选标准的候选蛋白进一步进行多变量孟德尔随机化（MVMR）和孟德尔随机化贝叶斯模型平均（MR-BMA）分析，以评价其条件效应及相对模型支持程度。同时进行异质性、水平多效性、工具变量强度及统计功效分析。 结果:单变量MR分析从91种循环炎症相关蛋白中筛选出5种具有名义关联且不同MR方法效应方向一致的候选蛋白，其中遗传预测的cystatin D、CXCL6和LIFR水平升高与胰腺癌风险增加相关，而IL-2和IL-24水平升高与较低的胰腺癌风险相关。进一步MVMR分析显示，仅CXCL6在MVMR-IVW、MVMR-Egger及MVMR-LASSO分析中均保持一致的正向效应。MR-BMA分析显示，CXCL6具有最高的边际纳入概率（MIP=0.569），且经验P值经Benjamini-Hochberg校正后仍具有统计学意义（P=0.005）。敏感性分析未发现明显的异质性或定向水平多效性。CXCL6的最小单个单核苷酸多态性位点（SNP）F统计量为22.67，MVMR条件F统计量为17.49，估算统计功效为70.7%，提示其工具变量强度尚可，但对中小效应的检出能力仍有限。 结论:本研究从91种循环炎症相关蛋白中筛选出5种与胰腺癌风险存在潜在关联的候选蛋白，其中CXCL6在MVMR和MR-BMA分析中获得了相对一致的遗传学支持，可作为进一步研究炎症相关通路与胰腺癌易感性关系的优先候选分子。但现有证据尚不足以确立CXCL6与胰腺癌之间的确定性因果关系，其具体生物学机制及潜在临床价值仍需通过独立人群和组织、功能实验进一步验证。

Abstract:

Background and Aims: Inflammation may contribute to the development and progression of pancreatic cancer; however, conventional observational studies have difficulty distinguishing the potential causal relationships between circulating inflammation-related proteins and pancreatic cancer risk from disease-related secondary changes. This study aimed to systematically investigate the potential causal associations between circulating inflammation-related proteins and pancreatic cancer risk using genetic epidemiological approaches and to prioritize candidate proteins with relatively consistent genetic evidence. Methods: Publicly available genome-wide association study (GWAS) summary statistics were used to perform a two-sample Mendelian randomization (MR) analysis of 91 circulating inflammation-related proteins and pancreatic cancer risk. The inverse-variance weighted (IVW) method was used as the primary analytical approach, complemented by weighted median, MR-Egger regression, and weighted mode analyses for robustness assessment. Candidate proteins meeting the prespecified screening criteria were further evaluated using multivariable Mendelian randomization (MVMR) and Mendelian randomization Bayesian model averaging (MR-BMA) to assess their conditional effects and relative model support. Heterogeneity, horizontal pleiotropy, instrument strength, and statistical power were also evaluated. Results: Univariable MR analysis identified five candidate proteins showing nominal associations with pancreatic cancer risk and consistent effect directions across MR methods. Genetically predicted higher levels of cystatin D, CXCL6, and LIFR were associated with increased pancreatic cancer risk, whereas higher levels of IL-2 and IL-24 were associated with lower risk. In subsequent MVMR analyses, CXCL6 was the only protein that consistently retained a positive association across MVMR-IVW, MVMR-Egger, and MVMR-LASSO models. MR-BMA further showed that CXCL6 had the highest marginal inclusion probability (MIP=0.569), with the empirical P value remaining significant after Benjamini-Hochberg correction (P=0.005). Sensitivity analyses revealed no substantial evidence of heterogeneity or directional horizontal pleiotropy. The minimum single-SNP F statistic and conditional F statistic for CXCL6 were 22.67 and 17.49, respectively, with an estimated statistical power of 70.7%, indicating adequate but limited power to detect modest effects. Conclusion: This study identified five circulating inflammation-related proteins potentially associated with pancreatic cancer risk among 91 candidates. CXCL6 showed relatively consistent genetic evidence across MVMR and MR-BMA analyses and may represent a priority candidate for further investigation of inflammation-related pathways in pancreatic cancer susceptibility. However, the current evidence is insufficient to establish a definitive causal relationship between CXCL6 and pancreatic cancer, and further validation in independent populations and tissue-based and functional studies is warranted.