房颤与认知障碍的因果关系:一项孟德尔随机化研究
Causal relationship between atrial fibrillation and cognitive impairment: a Mendelian randomization study
通讯作者: 修建成,电子信箱:xiujch@163.com王月刚,电子信箱:wyg06@fimmu.com。
编委: 徐敏
收稿日期: 2023-09-19 接受日期: 2023-11-03 网络出版日期: 2023-11-28
基金资助: |
|
Corresponding authors: XIU Jiancheng, E-mail:xiujch@163.comWANG Yuegang, E-mail:wyg06@fimmu.com.
Received: 2023-09-19 Accepted: 2023-11-03 Online: 2023-11-28
目的·探讨心房颤动(房颤)与认知障碍之间的因果关系。方法·采用两样本孟德尔随机化(Mendelian randomization,MR)分析方法,利用房颤的大规模全基因组关联研究(genome-wide association study,GWAS)汇总数据集,提取与房颤强相关的单核苷酸多态性(single nucleotide polymorphism,SNP)作为工具变量。基于公开的认知功能障碍的GWAS数据,分别提取SNPs与阿尔茨海默病性痴呆、帕金森病痴呆、血管性痴呆、路易体痴呆、额颞叶痴呆、未定义的痴呆、总体认知功能评估等的关联程度。采用逆方差加权法(inverse variance weighted,IVW)进行主要分析,Cochran′s Q检验、MR-Egger回归、留一法(leave-one-out)进行敏感性分析。为了验证结果的稳健性,使用不同GWAS数据进行重复分析及荟萃分析。结果·初次分析从一项涉及多达1 030 836名个体的全基因组关联研究荟萃分析中提取了101个SNPs作为工具变量,IVW结果未发现房颤与认知障碍的因果联系[痴呆:OR=1.032(95%CI 0.973~1.094),P=0.290;帕金森病痴呆:OR=1.004(95%CI 0.780~1.291),P=0.977;血管性痴呆:OR=1.123(95%CI 0.969~1.301),P=0.125;未定义的痴呆:OR=1.013(95%CI 0.910~1.129),P=0.807]。重复分析从FinnGen网站的房颤GWAS数据提取了27个SNPs作为工具变量,IVW结果与初次分析一致[认知功能:OR=0.999(95%CI 0.982~1.016),P=0.874;阿尔茨海默病性痴呆:OR=0.977(95%CI 0.943~1.012),P=0.193;路易体痴呆:OR=1.014(95%CI 0.898~1.145),P=0.826;额颞叶痴呆:OR=0.996(95%CI 0.745~1.333),P=0.980]。2次孟德尔随机化分析及荟萃分析均表明遗传预测的房颤与不同类型痴呆及总体认知功能评估均无相关证据。MR-Egger回归提示不存在水平多效性,留一法逐个剔除SNP后发现结果稳定。结论·未发现房颤与认知障碍之间的因果关系证据。在观察性研究中观察到的关联可部分归因于共同的生物学或共患病等混杂因素。
关键词:
Objective ·To investigate the causal relationship between atrial fibrillation (AF) and cognitive impairment. Methods ·A two-sample Mendelian randomization (TSMR) analysis was used to assess the potential causality of AF on cognitive dysfunction. Single nucleotide polymorphisms (SNPs) strongly associated with AF were extracted as instrumental variables by using a dataset of a large-scale genome-wide association study (GWAS) on AF. The associations of SNPs with Alzheimer′s disease dementia, Parkinson′s disease dementia, vascular dementia, Lewy body dementia, frontotemporal dementia, undefined dementia, and overall cognitive function assessment were extracted separately from publicly available GWAS data on cognitive dysfunction. The inverse variance-weighted (IVW) method was used for the main analysis, and sensitivity analyses were conducted by using Cochran′s Q test, MR-Egger regression, and leave-one-out method. To verify the robustness of the results, replicate analyses and meta-analyses were performed by using different GWAS data. Results ·In the initial analysis, 101 SNPs were extracted as instrumental variables from a meta-analysis of a genome-wide association study involving up to 1 030 836 individuals. The IVW analysis showed no evidence for causal associations between AF and dementia [dementia (OR=1.032; 95%CI 0.973‒1.094; P=0.290), Parkinson′s disease dementia (OR=1.004; 95%CI 0.780‒1.291; P=0.977), vascular dementia (OR=1.123; 95%CI 0.969‒1.301; P=0.125), or unspecified dementia (OR=1.013; 95%CI 0.910‒1.129; P=0.807)]. In the replication analysis, 27 SNPs were extracted as instrumental variables from the FinnGen AF GWAS data, and the IVW analysis were consistent with the initial analysis [cognitive function (OR=0.999; 95%CI 0.982‒1.016; P=0.874), Alzheimer′s disease dementia (OR=0.977; 95%CI 0.943‒1.012; P=0.193), Lewy body dementia (OR=1.014; 95%CI 0.898‒1.145; P=0.826), or frontotemporal dementia (OR=0.996; 95%CI 0.745‒1.333; P=0.980)]. Both Mendelian randomization analyses and meta-analyses showed no evidence of an association between genetically predicted AF and different types of dementia or overall cognitive function assessment. MR-Egger regression suggested no horizontal pleiotropy and leave-one-out analysis showed stable results after individually removing each SNP. Conclusion ·No evidence of a causal relationship between AF and cognitive impairment was found. The associations observed in observational studies can be partially attributed to confounding factors such as shared biology or co-morbidities.
Keywords:
本文引用格式
高雄, 张秋霞, 杨苗苗, 罗玮, 王月刚, 修建成.
GAO Xiong, ZHANG Qiuxia, YANG Miaomiao, LUO Wei, WANG Yuegang, XIU Jiancheng.
近年来,房颤与认知障碍相关性的研究逐渐得到重视。Rotterdam研究[5]表明房颤患者的认知障碍是非房颤患者的约2倍。多项大型纵向研究[6-8]表明,房颤与认知功能下降或痴呆风险的增加密切相关。在观察性研究中发现,房颤导致的认知障碍需要得到进一步研究。近年来,孟德尔随机化(Mendelian randomization,MR)方法提供了有力的研究策略:使用在全基因组关联研究中确定与暴露密切相关的遗传变异——单核苷酸多态性(single nucleotide polymorphism,SNP)作为工具变量,可以帮助克服观察性研究(特别是残留混杂因素和反向因果关系)的局限性,从而在暴露和结局风险之间做出证据更强的因果推断。
本研究通过两样本孟德尔随机化方法(two-sample Mendelian randomization,TSMR)来评估房颤与认知障碍的因果关系,以期为预防和治疗房颤相关的认知障碍提供科学依据。
1 资料与方法
1.1 研究设计
图1
图1
孟德尔随机化研究设计
Note: Assumption 1—Genetic variants were robustly associated with exposure. Assumption 2—Genetic variants were not associated with confounders. Assumption 3—Genetic variants affected the outcomes only through the exposure of interest. SNPs—single nucleotide polymorphisms. LD—linkage disequilibrium. WR—Wald ratio; WM—weighted median.
Fig 1
Overview of the current Mendelian randomization (MR) study
进行MR分析时,工具变量的选择需满足3个核心假设:① 工具变量必须与暴露因素存在强关联(关联性假设)。② 工具变量必须独立于结局的其他混杂因素(独立性假设)。③ 工具变量只能通过影响暴露因素影响结局发生,而不能通过其他途径对结局产生作用(排他性假设)[11]。
1.2 数据来源
表1、表2展示了全基因组关联研究(genome-wide association study,GWAS)数据来源详细信息。初次分析中房颤相关的遗传变异来自一项最近发表的研究[12]。这项研究是一项包含60 620例房颤患者和970 216例非房颤对照的6个研究队列组成的群体,98.6%是欧洲血统,检测了34 740 186个基因变异与房颤之间的关联。认知障碍的GWAS数据来自一项芬兰人群的研究。重复分析中房颤的遗传关联数据来自芬兰联盟,包含22 068例患者和116 926名对照。认知障碍的遗传关联数据来自4项认知障碍相关的全基因组关联研究[13-16]。有关芬兰联盟参与者、基因平台和统计分析方案的详细信息,可访问FinnGen网站(
表1 孟德尔分析中暴露因素的GWAS信息
Tab 1
Analysis | Exposure | ID | Sample size/n | Race | Year | PMID |
---|---|---|---|---|---|---|
First analysis | Atrial fibrillation | ebi-a-GCST006414 | 1 030 836 | European | 2018 | 30061737 |
Duplicate analysis | Atrial fibrillation | finn-b-I9 | 138 994 | European | 2021 | NA |
表2 孟德尔分析中结局因素的GWAS信息
Tab 2
Analysis | Outcome | ID | Sample size/n | Race | Year | PMID |
---|---|---|---|---|---|---|
First analysis | Dementia | finn-b-F5_DEMENTIA | 216 771 | European | 2021 | NA |
First analysis | Dementia due to Parkinson′s disease | finn-b-PD_DEMENTIA | 216 895 | European | 2021 | NA |
First analysis | Vascular dementia | finn-b-F5_VASCDEM | 212 389 | European | 2021 | NA |
First analysis | Undefined dementia | finn-b-F5_Dementia_U | 215 511 | European | 2021 | NA |
Duplicate analysis | Cognitive performance | ebi-a-GCST006572 | 257 841 | European | 2018 | 30038396 |
Duplicate analysis | Alzheimer′s disease | ebi-a-GCST90012877 | 472 868 | European | 2021 | 33589840 |
Duplicate analysis | Dementia with Lewy body | ebi-a-GCST90001390 | 6 618 | European | 2021 | 33589841 |
Duplicate analysis | Rontotemporal dementia | ieu-b-43 | 3 024 | European | 2010 | 20154673 |
1.3 工具变量的选择
首先,选择与房颤显著相关的SNPs(P<5×10-8,R2<0.001,遗传距离=10 000 kb),这满足第一个MR假设(关联性假设)。然后,通过在线网站PhenoScanner(
1.4 MR分析、荟萃分析及敏感性分析
所有MR分析均使用R软件(version 4.1.2 with packages)中的“TwoSample MR”“MRPRESSO”包进行,并使用“Meta”程序包进行荟萃分析。在这项研究中,逆方差加权法(inverse variance weighting,IVW)被作为主要方法来评价房颤与认知障碍的因果关系。使用Wald比值法(Wald ratio,WR)、加权中位数法(weighted median,WM)、MR-Egger回归分析进行补充。为了验证结果的稳健性,我们使用初次分析及重复分析结果进行荟萃分析以确定最终结果。使用Cochran′s Q统计量检验判断工具变量之间是否存在异质性[18]。采用MR-Egger回归评估潜在的多效性。此外,我们进行了留一法检验(leave-one-out)逐个剔除SNP来计算剩余SNPs的效应,用于检验单个SNP对结果稳定性的影响[19]。
2 结果
2.1 工具变量的确定
初次分析中与房颤强相关的111个SNPs可解释约4.6%的变异[12]。本研究工具变量F统计量=543,表明工具变量的强度足够。通过PhenoScanner查询,我们发现5个SNP与高血压或冠心病相关,5个SNP与教育程度或神经质相关。排除这10个SNPs后,我们使用剩余的101个SNPs作为MR分析的工具。重复分析使用了与房颤强相关的30个SNPs,剔除了3个与高血压或冠心病相关的SNPs,剩余的27个SNPs作为重复分析时的工具变量。
2.2 TSMR分析及荟萃分析结果
初次分析使用101个SNPs作为工具变量,对不同类型认知障碍相关工具变量进行MR分析汇总后,随机效应模型的IVW方法显示,遗传预测的房颤与认知功能障碍的风险无关[痴呆:OR=1.032(95%CI 0.973~1.094),P=0.290;帕金森病痴呆:OR=1.004(95%CI 0.780~1.291),P=0.977;血管性痴呆:OR=1.123(95%CI 0.969~1.301),P=0.125;未定义的痴呆:OR=1.013(95%CI 0.910~1.129),P=0.807]。加权中位数法与MR-Egger回归法得到了与IVW法相似的结果,并且结果的方向与主要的IVW法一致。重复分析时使用FinnGen房颤GWAS数据观察到了类似的趋势。重复分析使用27个SNPs作为工具变量,得到与初次分析一致的结果[认知功能:OR=0.999(95%CI 0.982~1.016),P=0.874;阿尔茨海默病性痴呆:OR=0.977(95%CI 0.943~1.012),P=0.193;路易体痴呆:OR=1.014(95%CI 0.898~1.145),P=0.826;额颞叶痴呆:OR=0.996(95%CI 0.745~1.333),P=0.980]。之后我们对不同类型认知障碍的MR估计值进行了合并,在系数比值法估计的荟萃分析中,遗传预测的不同认知障碍类型的合并OR值为1.000(95%CI 0.983~1.014)。表3显示了2次MR分析的结果。图2展示了荟萃分析结果。
表3 房颤与不同类型认知障碍之间关联的MR分析
Tab 3
Analysis | Outcome | IVW method | WM method | MR-Egger method | |||
---|---|---|---|---|---|---|---|
OR (95%CI) | P value | OR (95%CI) | P value | OR (95%CI) | P value | ||
First analysis | Dementia | 1.032 (0.973‒1.094) | 0.290 | 1.020 (0.920‒1.131) | 0.703 | 1.077 (0.962‒1.206) | 0.203 |
First analysis | Dementia due to Parkinson′s disease | 1.004 (0.780‒1.291) | 0.977 | 0.795 (0.508‒1.245) | 0.316 | 0.982 (0.603‒1.597) | 0.941 |
First analysis | Vascular dementia | 1.123 (0.969‒1.301) | 0.125 | 1.180 (0.898‒1.550) | 0.234 | 1.260 (0.947‒1.676) | 0.116 |
First analysis | Undefined dementia | 1.013 (0.910‒1.129) | 0.807 | 1.053 (0.862‒1.285) | 0.615 | 1.170 (0.951‒1.441) | 0.142 |
Duplicate analysis | Cognitive performance | 0.999 (0.982‒1.016) | 0.874 | 0.989 (0.975‒1.004) | 0.168 | 0.983 (0.946‒1.022) | 0.405 |
Duplicate analysis | Alzheimer′s disease | 0.977 (0.943‒1.012) | 0.193 | 0.976 (0.930‒1.024) | 0.324 | 0.994 (0.915‒1.079) | 0.885 |
Duplicate analysis | Dementia with Lewy body | 1.014 (0.898‒1.145) | 0.826 | 1.106 (0.932‒1.311) | 0.248 | 1.253 (0.953‒1.647) | 0.120 |
Duplicate analysis | Rontotemporal dementia | 0.996 (0.745‒1.333) | 0.980 | 0.966 (0.666‒1.399) | 0.853 | 0.874 (0.480‒1.590) | 0.672 |
图2
图2
遗传预测的房颤对不同认知障碍的荟萃分析
Note: SE—standard error; OR—odds ratio; CI—confidence interval.
Fig 2
Meta-analysis of genetically predicted atrial fibrillation on different cognitive impairments
2.3 敏感性分析结果
在房颤与认知障碍的MR分析结果中,仅在总体认知评估人群间存在异质性(P<0.05),其余不同类型痴呆的Cochran′s Q统计量均未发现显著的异质性,所有类型认知障碍中MR-Egger回归分析均未发现存在显著水平多效性的证据(表4)。在逐个剔除检验中,依次剔除单个SNP后得到的估计值没有明显变化,这表明没有单个SNP对总体估计有过大的影响。
表4 房颤与不同类型认知障碍之间关联的敏感性分析
Tab 4
Analysis | Outcome | Heterogeneity | Pleiotropy | |||
---|---|---|---|---|---|---|
IVW method | MR-Egger method | P value | ||||
Cochran′s Q | P value | Cochran′s Q | P value | |||
First analysis | Dementia | 109.881 | 0.194 | 109.054 | 0.190 | 0.393 |
First analysis | Dementia due to Parkinson′s disease | 83.717 | 0.848 | 83.706 | 0.830 | 0.917 |
First analysis | Vascular dementia | 105.698 | 0.280 | 104.776 | 0.277 | 0.358 |
First analysis | Undefined dementia | 94.739 | 0.574 | 92.231 | 0.618 | 0.117 |
Duplicate analysis | Cognitive performance | 81.518 | 0.000 | 79.054 | 0.000 | 0.396 |
Duplicate analysis | Alzheimer′s disease | 29.020 | 0.180 | 28.745 | 0.152 | 0.651 |
Duplicate analysis | Dementia with Lewy body | 15.679 | 0.869 | 12.809 | 0.938 | 0.104 |
Duplicate analysis | Rontotemporal dementia | 4.755 | 0.783 | 4.512 | 0.719 | 0.637 |
3 讨论
本研究运用孟德尔随机化方法系统性评价房颤与认知障碍之间因果关系的研究。2次分析使用不同来源、多项国际队列的房颤大规模全基因组关联研究数据集,结果显示:无论是对于不同类型的痴呆还是总体认知功能评估,房颤与认知障碍均未发现因果联系的证据。
我们的研究也存在一定的局限性。首先,很难完全避免潜在水平多效性与异质性的影响,这可能导致有偏倚的因果效应估计[18]。然而这种可能性较小,因为纳入人群大多数人都是欧洲血统,同时在MR-Egger回归没有观察到多效性效应。其次,房颤可能只有在中老年开始时才会增加认知障碍的风险,并且房颤类型及房颤持续的时间也可能造成潜在痴呆的神经病理逐渐发展。而我们的研究中GWAS数据为汇总结果,由于缺乏详细的临床信息,不能进行亚组分析,因而不能确定其具体的因果联系。最后,对认知障碍进行系统筛查的工具多种多样,从综合神经心理测试到单一筛查工具,这可能导致诊断不足或过度诊断。
我们的两样本孟德尔随机化分析结果没有提供足够的证据来反映房颤与认知障碍的因果关系。在观察性研究中观察到的关联可部分归因于共同的生物学或共患病等混杂因素。
作者贡献声明
高雄负责研究的构思、设计、文献检索、论文撰写;高雄、杨苗苗负责数据提取与统计分析;张秋霞、罗玮负责文献质量评价、论文修订;修建成、王月刚负责论文的最终审核。所有作者均阅读并同意了最终稿件的提交。
AUTHOR's CONTRIBUTIONS
GAO Xiong was responsible for the conception, design, literature retrieval and paper writing of the study; GAO Xiong and YANG Miaomiao were responsible for data extraction and statistical analysis; ZHANG Qiuxia and LUO Wei were responsible for literature quality evaluation and paper revision; XIU Jiancheng and WANG Yuegang were responsible for the final review of the paper. All the authors have read and approved the submission of the final manuscript.
利益冲突声明
所有作者声明不存在利益冲突。
COMPETING INTERESTS
All authors disclose no relevant conflict of interests.
参考文献
/
〈 |
|
〉 |
![](../../../richhtml/1674-8115/richHtml_jats1_1/images/1.png)