The Mendelian Randomization Study Of Vitamin D Levels/ Vitamin D Deficiency With Primary Liver Cancer Risk

ObjectiveThis Mendelian Randomization study aimed to examine the potential causal association between VD levels/VD deficiency and primary liver cancer（PLC）,using the genetic risk score（GRS）as instrumental variable associated with VD levels or VD deficiency.Methods1.Based on a 1:1 frequency-matched case-control study design,1068newly diagnosed PLC cases,aged 30 to 79 years old,were recruited from Shunde Hospital of Southern Medical University.The 1068 healthy controls were enrolled from the Shunde Health Screening Program,as well as the 2000healthy individuals among whom the instrumental variable.2.Serum 25（OH）D concentration was detected by ELISA（Enzyme linked immunosorbent assay）.A total of night genetic variants（rs12785878,rs1790349,rs4588,rs7041,rs1155563,rs10741657,rs12794714,rs6013897 and rs2209314）potentially associated with VD were detected,including those derived from the genome-wide association studies（GWAS）and verified in Asian populations,and previously identified by our research team.The TaqMan fluorescence quantitative PCR technique was used for genotyping.3.VD deficiency was defined as 25（OH）D level<50nmol/L.In order to exclude the confounding bias caused by seasonal factors（sunshine hours）,only427 cases and 427 controls with similar sampling time（January to May）were included.The multivariate logistic regression model was used to evaluate the observational effects between VD levels/deficiency and PLC risk before and after the restrict of sampling time.4.The genetic associations with VD levels were further confirmed in 2000healthy individuals,and the unweighted GRS,unweighted core GRS,weighted GRS and weighted core GRS were calculated as the instrumental variable of VD levels.The instrumental variable for VD deficiency was constructed using the similar method.The generalized linear regression model was used to estimate the association between GRS and VD levels,while the logistic regression model for GRS and VD deficiency.5.Considering that the distribution of genetic variants was not affected by seasonal factors,in 1068 cases and 1068 controls,the multivariate logistic regression model was used to analyze the association effect between continuous GRS and PLC risk.Based on the quartile method,the nonlinear causal association between different GRS grades and PLC risk was observed.Through weak instrumental variable testing and pleiotropic screening,this study explored whether the conditions for using instrumental variables were satisfied or not.Next,the pleiotropic genetic variants were eliminated to further optimize the GRS to construct a new instrumental variable.By means of limiting the sampling time of the subjects,the causal association between the new instrumental variable and PLC risk was analyzed using the same statistical strategy.Results1.In 427 cases and 427 controls（The sampling time is from January to May）,PLC risk increased by 1%（OR=1.01,95%CI=1.00～1.03）for every1nmol/L decrease of VD levels after adjustment by the multivariate model.VD deficiency increased PLC risk by 76%（OR=1.76,95%CI=1.16～2.65）.If the sampling time of the subjects was not restricted,the analysis result of 1068 cases and 1068 controls showed that there was no statistical association between VD levels/deficiency and the PLC risk（VD levels:OR=1.11,95%CI=0.86～1.43）;VD deficiency:OR=0.99,95%CI=0.99～1.00).2.Further verification of target genetic variation in the healthy individuals showed that:rs12785878,rs1790349,rs4588,rs7041,rs1155563,rs10741657and rs12794714 were showed a statistical association with VD levels(all P_FDR<0.05).In addition,rs4588,rs7041,rs1155563,rs10741657 and rs12794714 were statistically associated with VD deficiency risk(all P_FDR<0.05).Therefore,the aforementioned genetic variants were used to construct the core GRS.The F-statistics of the unweighted GRS,unweighted core GRS,weighted GRS,and weighted core GRS at the VD level were 87.29,87.30,124.23,and 120.32,respectively.Similarly,the F-statistics for GRS of VD deficiency are also greater than 10.The statistical power of all GRS was100%,which conformed to the robust instrumental variable.3.The results of Mendelian randomization analysis showed that VD levels GRS（unweighted:P=0.534,weighted:P=0.201）and VD deficiency GRS（unweighted:P=0.993,weighted:P=0.271）,whether weighted or not,none of them were statistically related to the PLC risk.Even for core GRS constructed by retaining only positive-associated genetic variants,VD levels core GRS（unweighted core:P=0.710,weighted core:P=0.265）and VD deficiency core GRS（unweighted core:P=0.307,weighted core:P=0.083）also showed no statistical association.Besides,no matter what kinds of GRS,there was no statistical association with common confounding factors（P>0.05）,which satisfied the exclusive condition of instrumental variable.4.In the sensitivity analysis,the GRS was recalculated after excluding possible pleiotropic genetic variants（rs12785878,rs1790349,rs7041,rs12794714 and rs6013897）,and the newly GRS was not statistically associated with the PLC risk.Further analysis was performed only in cases and controls with similar sampling times,and GRS was still not statistically associated with PLC risk.ConclusionsThis Mendelian randomization study did not support a causal association between VD levels/VD deficiency and PLC risk.The association between VD and PLC risk found in case-control studies may be attributed to confounding bias or reverse causation.Further studies are needed to confirm our findings.