空气污染与阿尔茨海默病因果关联的两样本孟德尔随机化研究
1.
2.
Two-sample Mendelian randomization study on the causal association between air pollution and Alzheimer′s disease
1.
2.
通讯作者: 姚俊岩,博士,主任医师;电子信箱:sunshineyao@163.com。
编委: 邢宇洋
收稿日期: 2024-06-10 接受日期: 2024-10-02 网络出版日期: 2025-01-17
| 基金资助: |
|
Corresponding authors: YAO Junyan, E-mail:sunshineyao@163.com.
Received: 2024-06-10 Accepted: 2024-10-02 Online: 2025-01-17
目的·利用两样本孟德尔随机化(Mendelian randomization,MR)的方法探究空气污染与阿尔茨海默病(Alzheimer′s disease,AD)发病风险之间的因果关系。方法·基于全基因组关联研究(genome-wide association study,GWAS)的数据,采用两样本MR分析评估空气污染与AD发病风险的因果关系。以空气污染指标包括细颗粒物(particulate matter 2.5,PM2.5)、粗颗粒物(particulate matter 2.5-10,PM2.5-10)、可吸入颗粒物(particulate matter 10,PM10)、二氧化氮以及氮氧化物为暴露因素,从英国生物银行(UK Biobank)数据库中获得其汇总数据。PM2.5的数据集包括暴露人群423 796例,包含9 851 867个单核苷酸多态性(single nucleotide polymorphisms,SNPs)的关联分析;PM2.5-10的数据集包括暴露人群423 796例,包含9 851 867个SNPs的关联分析;PM10的数据集包括暴露人群455 314例,包含9 851 867个SNPs的关联分析;二氧化氮的数据集包括暴露人群456 380例,包含9 851 867个SNPs的关联分析;氮氧化物的数据集包括暴露人群456 380例,包含9 851 867个SNPs的关联分析。以AD为结局因素,从国际阿尔茨海默病基因组学项目(International Genomics of Alzheimer′s Project,IGAP)中获得AD的数据。AD的数据集包括患者25 580例和对照人群48 466例,包含7 067 513个SNPs的关联分析。以与AD显著相关的SNPs作为工具变量,以逆方差加权(inverse variance weighted,IVW)法为主要分析方法,选择加权中位数法、MR-Egger回归、基于众数的简单估计和基于众数的加权估计4种方法进行质量控制,并通过异质性检验、基因多效性检验和敏感性分析来评估研究结果的可靠性。结果·异质性检验(IVW法和MR-Egger回归)显示,空气污染指标与AD的SNP之间不存在异质性(均P>0.05)。基因多效性检验(MR-Egger回归)显示,未检测到多效性(P>0.05)。敏感性分析显示,PM2.5的研究结果稳定。IVW法的分析结果发现,在欧洲人群中PM2.5(P<0.001)与AD之间存在统计学关联,而PM2.5-10(P=0.664)、PM10(P=0.664)、二氧化氮(P=0.284)、氮氧化物(P=0.567)这4种因素与AD之间不存在统计学关联。结论·PM2.5暴露与AD发病风险之间存在显著的因果关系,PM2.5的暴露会增加AD的发病风险,但尚未发现PM2.5-10、PM10、二氧化氮和氮氧化物暴露导致AD发病风险增加的证据。
关键词:
Objective ·To explore the causal relationship between air pollution and the risk of Alzheimer′s disease (AD) by using two-sample Mendelian randomization (MR). Methods ·Based on the data from the genome-wide association study (GWAS), a two-sample MR analysis was conducted to evaluate the causal relationship between air pollution and the risk of AD. Air pollution indicators, including particulate matter 2.5 (PM2.5), particulate matter 2.5-10 (PM2.5-10), particulate matter 10 (PM10), nitrogen dioxide and nitrogen oxides, were used as exposure factors, and summarized data were aggregated from the UK Biobank database. The PM2.5 dataset included 423 796 cases, with correlation analysis conducted on 9 851 867 single nucleotide polymorphisms (SNPs); the PM2.5-10 dataset included 423 796 cases, with correlation analysis conducted on 9 851 867 SNPs; the PM10 dataset included 455 314 cases, with correlation analysis conducted on 9 851 867 SNPs; the nitrogen dioxide dataset included 456 380 cases, with correlation analysis conducted on 9 851 867 SNPs; the nitrogen oxides dataset included 456 380 cases, with correlation analysis conducted on 9 851 867 SNPs. AD was used as the outcome factor, and data were obtained from the International Genomics of Alzheimer′s Project (IGAP). The AD dataset included 25 580 cases and 48 466 controls, with correlation analysis of 7 067 513 SNPs. SNPs significantly associated with AD were used as instrumental variables. The main analysis was conducted by using the inverse variance weighted (IVW) method, and four methods including weighted median, MR-Egger regression, mode-based simple estimation and mode-based weighted estimation were used for quality control. Heterogeneity testing, gene pleiotropy testing and sensitivity analysis were conducted to assess the reliability of the study results. Results ·Heterogeneity testing indicated no evidence of heterogeneity among SNPs associated with air pollution indicators and AD (both IVW and MR-Egger results, P>0.05). Gene pleiotropy testing did not detect any pleiotropic effects (MR-Egger results, P>0.05). Sensitivity analysis confirmed the stability of the PM2.5 results. IVW analysis revealed a statistically significant association between PM2.5 and AD in European populations (P<0.001), while no statistically significant associations were observed between PM2.5-10 (P=0.664), PM10 (P=0.664), nitrogen dioxide (P=0.284), nitrogen oxides (P=0.567) and AD. Conclusion ·There is a significant causal relationship between PM2.5 exposure and the risk of AD, with PM2.5 exposure increasing the incidence of AD. However, no evidence has been found to suggest that PM2.5-10, PM10, nitrogen dioxide or nitrogen oxides cause an increased risk of AD.
Keywords:
本文引用格式
张迎迎, 张俊瑶, 宋际伟, 王声杰, 姚俊岩.
ZHANG Yingying, ZHANG Junyao, SONG Jiwei, WANG Shengjie, YAO Junyan.
随着工业的快速发展,空气污染问题对世界的危害逐年增加。世界卫生组织的一项科学声明指出,全球每年约有420万人因暴露于环境(室外)污染而死亡[3]。空气污染物主要为颗粒物质[细颗粒物(particulate matter 2.5,PM2.5)、粗颗粒物(particulate matter 2.5-10,PM2.5-10)、可吸入颗粒物(particulate matter 10,PM10)等]、气体(一氧化碳、二氧化氮等)和有机化合物(烃类等)的组合[4];其中,PM10对人体的危害最大。现有研究[3,5]指出,PM2.5与呼吸系统、心血管系统疾病的发生密切相关。既往观察性研究[5]显示,PM2.5与AD、帕金森病(Parkinson′s disease,PD)等中枢神经系统疾病间存在一定的相关性。同时,流行病学的研究[6]也提示,长期暴露于PM2.5环境中的人群的认知功能显著下降,且罹患AD的风险增加。此外,PM2.5还能引发系统性炎症反应,而这些炎症因子可通过血脑屏障迁移至中枢神经系统,诱发神经炎症[1];其还会引发氧化应激反应,这种应激状态可进一步导致神经元损伤,并促进认知功能减退[7]。然而,传统的观察性研究常会受到样本量、混杂因素、反向因果关系等的影响,流行病学研究也可能受到数据收集和分析方法的限制,这使得空气污染与AD之间的因果关系尚未得到充分证实,仍需开展更进一步的分析。
1 资料与方法
1.1 数据来源
从英国生物银行(UK Biobank)数据库中获取空气污染物包括PM2.5、PM2.5-10、PM10、二氧化氮、氮氧化物等的全基因组关联研究(genome-wide association study,GWAS)数据,下载网址为
表1 空气污染与AD因果关联研究的样本量、SNP数量及人群信息
Tab 1
| Exposure/outcome | Dataset | Sample size/n | Number of SNPs/n | Population | Publication year |
|---|---|---|---|---|---|
| PM2.5 | ukb-b-10 817 | 423 796 participants | 9 851 867 | European | 2018 |
| PM2.5-10 | ukb-b-12 963 | 423 796 participants | 9 851 867 | European | 2018 |
| PM10 | ukb-b-589 | 455 314 participants | 9 851 867 | European | 2018 |
| Nitrogen dioxide | ukb-b-2 618 | 456 380 participants | 9 851 867 | European | 2018 |
| Nitrogen oxides | ukb-b-12 417 | 456 380 participants | 9 851 867 | European | 2018 |
| AD | NG00053 | 25 580 cases and 48 466 controls | 7 055 881 (stage 1) and 11 632 (stage 2) | European | 2013 |
1.2 研究设计
本研究以PM2.5、PM2.5-10、PM10、二氧化氮和氮氧化物为空气污染的暴露因素,以AD为结局因素,以与AD显著相关的SNPs作为工具变量,采用两样本MR方法对上述5种空气污染指标与AD发病风险之间的因果关系进行分析。
两样本MR分析的设计原理见图1。其中该分析需满足3个基本假设:① 关联性假设,即工具变量必须与暴露因素具有显著的统计学关联,以确保工具变量能够有效地反映暴露因素的变化。② 独立性假设,工具变量需要与混杂因素之间彼此独立,即工具变量对结局因素的影响均应通过其对暴露因素的作用来体现,而不存在其他直接影响途径。③ 排他性假设,工具变量仅能通过暴露因素对结局因素产生作用,即不存在多效性[11]。同时,SNPs的筛选条件包括:① 设置P<5×10-6,筛选出与空气污染各指标具有全基因组意义和相关性的SNPs。② 去除连锁不平衡的影响,设置参数R2=0.001、位点距离10 000 kb,并剔除R2 >0.001的SNPs。③ 去除弱工具变量,计算SNPs的F值;筛选F>10的SNPs纳入MR研究[12]。
图1
图1
两样本MR分析的研究设计
Note: LD—linkage disequilibrium; IVW—inverse variance weighted; WM—weighted median.
Fig 1
Study design of the two-sample MR analysis
1.3 数据分析
1.3.1 效应估计
1.3.2 敏感性分析、异质性检验及基因多效性检验
2 结果
2.1 MR分析
IVW法的分析结果显示,PM2.5与AD发病风险之间存在显著的因果关系,且PM2.5可增加AD的发病风险(OR=2.302,95% CI 1.421~3.729,P<0.001);WM的分析结果(OR=2.186,95% CI 1.073~4.453,P=0.031)亦支持上述结论;而其他3种分析方法得出的结果并不存在统计学意义。由于在该5种方法中,IVW法最为准确且其获得的分析结果最具判断价值,因此我们认为PM2.5与AD的发病风险相关。
图2
图2
SNP对PM2.5 和AD效应的散点图分析
Note: Each black point on the plot represents an individual SNP, with its position on the exposure (horizontal axis) and the outcome (vertical axis). Error bars indicate the standard error (SE) for each SNP.
Fig 2
Scatter plot analysis of SNP effects on PM2.5 and AD
图3
图3
PM2.5 对AD效应的森林图分析
Note: The red and black bars represent the causal estimates of PM2.5 levels on the risk of AD.
Fig 3
Forest plot analysis of PM2.5 effects on AD
IVW法、MR-Egger回归、WM、基于众数的加权估计、基于众数的简单估计这5种回归模型均显示,PM2.5-10、PM10、二氧化氮和氮氧化物的P值均>0.05(表2)。继而提示,该4种空气污染指标与AD发病风险之间均不存在因果效应。
表2 5种空气污染指标与AD发病风险之间因果关系的MR分析
Tab 2
| Exposure | Outcome | SNP/n | MR method | P value | OR value | 95%CI | Pleiotropy P value | Heterogeneity P value |
|---|---|---|---|---|---|---|---|---|
| PM2.5 | AD | 48 | MR-Egger | 0.874 | 0.863 | 0.141‒5.295 | 0.278 | 0.453 |
| WM | 0.031 | 2.186 | 1.073‒4.453 | |||||
| IVW | <0.001 | 2.302 | 1.421‒3.729 | 0.444 | ||||
| Simple mode | 0.192 | 2.777 | 0.615‒12.550 | |||||
| Weighted mode | 0.291 | 2.175 | 0.524‒9.024 | |||||
| PM2.5-10 | AD | 4 | MR-Egger | 0.432 | 0.905 | 0.741‒1.106 | 0.300 | 0.967 |
| WM | 0.890 | 0.992 | 0.878‒1.119 | |||||
| IVW | 0.664 | 1.023 | 0.924‒1.131 | 0.576 | ||||
| Simple mode | 0.945 | 1.007 | 0.828‒1.225 | |||||
| Weighted mode | 0.782 | 0.980 | 0.860‒1.116 | |||||
| PM10 | AD | 4 | MR-Egger | 0.432 | 0.905 | 0.741‒1.106 | 0.300 | 0.967 |
| WM | 0.890 | 0.992 | 0.878‒1.119 | |||||
| IVW | 0.664 | 1.023 | 0.924‒1.131 | 0.576 | ||||
| Simple mode | 0.945 | 1.007 | 0.828‒1.225 | |||||
| Weighted mode | 0.782 | 0.980 | 0.860‒1.116 | |||||
| Nitrogen dioxide | AD | 84 | MR-Egger | 0.973 | 0.978 | 0.271‒3.531 | 0.705 | 0.428 |
| WM | 0.303 | 1.354 | 0.756‒2.423 | |||||
| IVW | 0.284 | 1.239 | 0.837‒1.834 | 0.457 | ||||
| Simple mode | 0.216 | 2.512 | 0.591‒10.670 | |||||
| Weighted mode | 0.294 | 2.063 | 0.538‒7.910 | |||||
| Nitrogen oxides | AD | 67 | MR-Egger | 0.682 | 0.703 | 0.131‒3.764 | 0.563 | 0.206 |
| WM | 0.892 | 0.960 | 0.536‒1.721 | |||||
| IVW | 0.567 | 1.137 | 0.733‒1.762 | 0.223 | ||||
| Simple mode | 0.444 | 0.578 | 0.144‒2.330 | |||||
| Weighted mode | 0.538 | 0.649 | 0.165‒2.549 |
2.2 敏感性分析、异质性检验及基因多效性检验
留一法敏感性分析的结果(图4)显示,在依次剔除PM2.5的每个SNP后,剩余的SNPs对结果未产生显著影响,提示PM2.5的研究结果稳定。
图4
图4
PM2.5 对AD效应的留一法敏感性分析
Note: The error bars indicate the 95%CI. The red line shows the results of the randow-effects IVW analysis.
Fig 4
Leave-one-out sensitivity analysis of PM2.5 effects on AD
异质性检验的结果显示,对于PM2.5,MR-Egger回归分析(Cochran′s Q=40.38,P=0.453)和IVW法分析(Cochran′s Q=41.60,P=0.444)均未发现异质
图5
图5
PM2.5 对AD效应的漏斗图分析
Note: Each black point on the plot represents an individual SNP, with the vertical axis displaying the effect size and the horizontal axis representing the standard error (SE). The central line indicates the overall estimated effect size. IV‒instrumental variable.
Fig 5
Funnel plot analysis of PM2.5 effects on AD
MR-Egger回归检验基因多效性发现,PM2.5(P=0.278)、PM2.5-10(P=0.300)、PM10(P=0.300)、二氧化氮(P=0.705)和氮氧化物(P=0.563)均不存在多效性(表2)。
3 讨论
本研究通过两样本MR分析发现,PM2.5与AD发病风险间存在显著的因果关系,而PM2.5-10、PM10、二氧化氮和氮氧化物与AD发病风险之间并不存在显著关联。
越来越多的研究证据[18-20]表明,PM2.5暴露可导致个体发生认知功能障碍,并可能加速其向痴呆发展的进程。流行病学研究[21-24]显示,PM2.5、氮氧化物、多环芳烃、挥发性有机物等多种空气污染物的暴露,与个体的认知功能障碍以及AD的发病率增加显著相关。WEUVE等[25]的研究显示,个体长期暴露于较高浓度的PM2.5与其认知功能快速下降存在明显关联。AILSHIRE等[26]的研究进一步指出,PM2.5暴露是50岁以上人群认知功能下降的重要危险因素。从机制上来看,PM2.5可通过肺气-血屏障进入血液循环系统,并通过肠-脑轴或嗅神经进入脑组织,进而引发机体的氧化应激和炎症反应,这一系列过程与AD发病的机制密切相关[27-28]。因此,上述研究均提示PM2.5暴露可能是AD发病的一个重要的环境诱因。
在分泌酶的作用下,淀粉样前体蛋白(amyloid precursor protein,APP)可裂解产生Aβ多肽(主要为Aβ42和Aβ40),其中Aβ42是淀粉样斑块的主要成分[32]。研究人员对长期暴露于空气污染的个体进行脑组织研究后发现,他们的海马和前额叶皮层中的Aβ42水平较高,且伴随神经炎症指标如环氧合酶2(cyclooxygenase 2,COX2)水平的升高[33]。最新的一项横断面研究[34]也显示,PM2.5暴露与脑脊液中Aβ42水平下降有关,这提示大脑中淀粉样斑块的沉积增加,进而增加了AD的发病风险。此外,PM2.5暴露还可通过影响APP的代谢过程来上调β-位点淀粉样前体蛋白裂解酶1(β-site amyloid precursor protein cleaving enzyme 1,BACE1)的表达,而BACE1的上调会加速APP裂解进而产生更多的Aβ多肽,导致淀粉样斑块沉积的进一步增加,最终加剧AD患者认知功能的恶化[35]。
近年来的研究普遍认为,tau蛋白的过度磷酸化和突触功能的损伤是Aβ发挥神经毒性不可或缺的因素[36],此外磷酸化tau蛋白水平的升高也是AD临床前阶段的重要早期生物标志物。CALDERÓN-GARCIDUEÑAS等[37-38]的前瞻性研究发现,当儿童和青年长期暴露于高于标准水平的PM2.5时,其脑脊液中的Aβ、脑源性神经营养因子(brain-derived neurotrophic factor,BDNF)水平会显著下降,而总tau蛋白和磷酸化tau蛋白水平则明显上升。这些发现均提示PM2.5暴露可能是通过影响Aβ和tau蛋白的代谢过程来参与AD的发生,从而为PM2.5暴露与AD发病风险之间的因果联系提供了证据支持。
本研究存在一定的局限性。由于目前针对亚洲人群的空气污染数据有限,与AD相关的GWAS数据不足,因此本研究仅对欧洲人群进行了MR分析,未能充分考虑不同人群对PM2.5暴露的敏感性差异;未来,我们将进一步探讨空气污染与AD发病风险在不同种族人群间的差异,从而更全面地了解空气污染对各类人群的影响机制,明确PM2.5暴露在不同遗传背景下的风险差异,为制定针对性的公共卫生政策和预防措施提供依据。
综上所述,本研究通过MR分析探讨了空气污染与AD发病风险之间的因果关系,结果发现PM2.5暴露与AD的发病风险间存在显著的因果关系,提示PM2.5暴露会增加AD的发病风险;而PM2.5-10、PM10、二氧化氮和氮氧化物对AD的发病风险并无显著影响。这一结果或将为未来深入探讨空气污染与AD的发病机制提供重要参考。
作者贡献声明
张迎迎负责研究的构思、设计和论文撰写,张俊瑶、宋际伟、王声杰负责数据分析与统计,姚俊岩负责论文构思、撰写及最终审核。所有作者均阅读并同意了最终稿件的提交。
AUTHOR's CONTRIBUTIONS
ZHANG Yingying was responsible for the conception and design of the study, as well as the writing of the paper. ZHANG Junyao, SONG Jiwei and WANG Shengjie were responsible for the data analysis and statistical analysis. YAO Junyan was responsible for paper conception, writing and final review. All the authors have read the final version of paper and consented to its submission.
利益冲突声明
所有作者声明不存在利益冲突。
COMPETING INTERESTS
All authors disclare no relevant conflict of interests.
参考文献
/
| 〈 |
|
〉 |

