A Study Of A Multi-exposure Weighted Quantile Sum Based Mediation Method And Its Application To “Atmospheric Fine Particulate Matter Constituents-Gut Microbiota-Diabetes”

Background and ObjectiveDiabetes mellitus is a common chronic non-communicable disease with many patients,high prevalence and severe disease burden.In addition to common risk factors（e.g.,age,gender,genetic factors and body mass index）,recent epidemiological studies and animal experiments have shown that atmospheric fine particulate matter with aerodynamic diameters≤2.5μm(PM_2.5)is also a significant risk factor for diabetes.Further understanding of the specific mechanisms of PM_2.5causing diabetes could help to reduce the disease burden of diabetes due to PM_2.5.The human gut microbiota is essential in nutrient metabolism,hormone secretion and immune function.Epidemiological studies have identified that PM_2.5 exposure is associated with alterations in the gut microbiota and that the gut microbiota might mediate the association between PM_2.5 exposure and diabetes.PM_2.5 is a complex mixture of constituents,including Black Carbon（BC）,ammonium,nitrates,Organic Matter（OM）,sulfates,Soil Particles（SOIL）and Sea Salt（SS）.Different constituents vary in their health effects on humans.However,in the three scenarios mentioned above(i.e.,association of PM_2.5 with diabetes,association of PM_2.5 with gut microbiota,and mediating role of gut microbiota in the association of PM_2.5 with diabetes),none of the existing studies have considered the constituent structure of PM_2.5.Understanding the role of PM_2.5 constituents in the above associations can identify important constituents of PM_2.5 and provide new perspectives for reducing the disease burden of PM_2.5-related diabetes.There is usually a strong correlation between PM_2.5 and its constituents(PM_cons),so some multi-exposure analysis methods are often used to analyze the health effects of PMcons for this scenario,including penalty algorithms such as Ridge regression and Least Absolute Shrinkage and Selection Operator（LASSO）;dimensionality reduction algorithms such as Weighted Quantile Sum（WQS）regression,Quantile G method（Quantification）;variable selection algorithms,such as Bayesian Kernel Machine Regression（BKMR）,etc.However,none of the existing mediation analysis frameworks,either the traditional Structural Equation Model or the emerging Casual Mediation Inference,contains strategies for multi-exposure mediation analysis.Even though some exploratory analytical strategies and frameworks exist in the current literature,there are no mature and methodologically validated methods for multi-exposure mediation analysis.This study addresses these gaps by conducting four sub-studies:1)exploring the association between PM_cons and diabetes;2)exploring the association between PM_consand gut microbiota;3)exploring new methods for mediator analysis in multiple exposure scenarios;and 4)exploring the mediating role of gut microbiota in the association between PM_cons and diabetes.MethodsStudy 1:Associations between PM_cons and diabetes.Based on baseline data from the China Multi-Ethnic Cohort in southwest China（CMEC）and PM_cons data from the Global Burden of Disease（GBD）study,we explored the association between PM_consand diabetes.Specifically,logistic regression was used to investigate the association between single exposure to PM_cons and diabetes（dichotomous index defined by fasting plasma glucose and glycated hemoglobin）.Multiple exposure analysis methods（e.g.,LASSO,WQS,QGC,etc.）were used to explore the association between joint exposure to PM_cons and diabetes and to identify significant components.Study 2:Associations study between PM_cons and gut microbiota.Based on data collected from participants with stool samples in the CMEC and PM_cons data from the GBD study,we explored the association between PM_cons and gut microbiota.Specifically,we used generalized propensity score weighting（GPSW）regression to investigate the association between single exposure to PM_cons and theα-diversity index of gut microbiota（Shannon index）.Multiple exposure analysis methods（e.g.,LASSO,WQS,QGC,BKMR etc.）were used to explore the association between joint exposure to PM_cons and Shannon index and to identify important components.Weighted correlation analysis was used to assess the association between single exposure to PM_cons and the abundance of gut microbiota.Study 3:A developing research of Weighted Quantile Sum-Causal Mediation Analysis method（WQS-CMA）.The WQS-CMA was developed,considering both exposure-mediator and exposure-outcome associations,i.e.,double-supervised downscaling of multiple exposures using mediation and outcome.Then the existing causal mediation analysis method was used for analysis.Specifically,Maximum Likelihood Estimation（MLE）was used to complete the parameter estimation of WQS-CMA.Simulation studies were conducted to determine the statistical performance of WQS-CMA under different scenarios compared with existing strategies.Study 4:The mediating effect of gut microbiota in the association between PM_cons and diabetes.Specifically,traditional regression methods were used to assess the associations between PM_cons single exposure-Shannon index,PM_cons single exposure-diabetes,and gut microbiota-diabetes.The traditional causal mediation analysis method was used to explore the mediating role of the Shannon index in the association between PM_cons single exposure and diabetes.WQS-CMA was used to investigate the mediating effect of the Shannon index in the association between PM_cons joint exposure and diabetes and estimate the mediating effect of PM_consmixture and the weight of each constituent.ResultsStudy 1:Per-SD increases in the 3-year average concentrations of PM_2.5（odds ratio[OR]1.08,95%CI 1.01–1.15）,black carbon（BC;1.07,1.01–1.15）,ammonium（1.07,1.00–1.14）,nitrate（1.08,1.01–1.16）,OM（1.09,1.02–1.16）,and SOIL（1.09,1.02–1.17）were positively associated with diabetes.The associations were stronger in those≥65 years.Joint exposure to PM_cons was positively associated with diabetes（1.04,1.01–1.07）.The estimated weight of OM was the largest among PM_cons.Study 2:An interquartile range increase of 3-year average BC,ammonium,nitrate,OM,sulfate,and SOIL were negatively associated with Shannon index with mean difference（95%CI）being-0.144（-0.208,-0.080）,-0.141（-0.205,-0.078）,-0.126（-0.184,-0.068）,-0.117（-0.172,-0.062）,-0.153（-0.221,-0.085）,and-0.153（-0.222,-0.085）.BKMR indicated joint exposure to PM_cons was associated with decreased Shannon index,and BC had the largest posterior inclusion probability（0.578）.Weighted correlation analyses indicated PM_cons were associated with decreased Bacteroidetes(r=-0.204,P<0.001 for PM_2.5)and increased Proteobacteria(r=0.273,P<0.001 for PM_2.5).Study 3:This study proposed and validated WQS-CMA.The simulation showed that WQS-CMA performed well and accurately estimated the weights of multiple exposures when the mediator and outcome were continuous or,at most,one dichotomous.For example,the mean absolute error of the WQS-CMA weight estimates was 0.013 when both the mediator and outcome were continuous variables.The mean absolute error of the existing WQS downscaling strategy for exposure and outcome was 0.142.The new method was not precise when the mediator and outcome were all dichotomous.Study 4:Gut microbiota mediated the association of PM_cons with diabetes.Specifically,the Shannon index mediates the association between BC,ammonium,nitrate,OM,PM_2.5,sulfate,SOIL and diabetes,with proportions mediated of 8.95%,9.38%,9.83%,10.36%,9.96%,9.43%and 9.39%,respectively.WQS-CMA revealed that the highest weights in the mixture were OM（27.4%）.The Shannon index could mediate the association between joint PM_cons exposure and diabetes with a proportion mediated of 10.17%.This study found that exposure to some constituents may elevate the risk of diabetes by decreasing Elusimicrobium,Prevotella and increasing Proteobacteria,Flavonifractor.ConclusionLong-term exposure to PM_cons is associated with an increased risk of diabetes,with OM being the most important constituent.Long-term exposure to PM_cons was associated with reduced gut microbiota diversity and altered abundance,with BC being the most important constituent.Gut microbiota mediates the association between long-term exposure to PM_cons and diabetes,with OM being the most important constituent.The WQS-CMA proposed in this study performed well when both mediator and outcome were continuous,or at most one dichotomous,and bad when both mediator and outcome were dichotomous.