Study To Determine And Predict The Relative Potency And Specific Relative Sensitivity Among Birds For Brominated Dioxins Analogues

Polybrominated dibenzo-p-dioxins（PBDDs）,polybrominated dibenzofurans（PBDFs）,and non-ortho polybrominated biphenyls（PBBs）are highly toxic dioxin-like environmental pollutants with effumability,high lipophilicity and environmental persistence.They have recently been detected to be widespread in various environmental and biological samples with high detection rates and levels.Recently,they were recommended to be included in the existing toxicity equivalency factor（TEF）scheme by World Health Organization.However,very few data were available concerning the toxicity of brominated dioxins in birds,thus the avian TEFs could not be determined,thereby leading to the inability to conduct ecological risk assessment of brominated dioxins for birds.Therefore,in this study,avian aryl hydrocarbon receptor 1（AHR1）-mediated luciferase report gene assay were used to determine their species-specific dioxin-like potencies（DLPs）and congener-specific interspecies relative sensitivities,which were urgently needed for ecological risk assessment in birds.In addition,binding free energy and molecular mechanical energy(E_MM)between each congener and each individual amino acid（AA）residue in AHR1-ligand binding domain（LBD）were calculated by molecular dynamics simulations and the molecular mechanics Poisson-Boltzmann surface area（MM-PBSA）method.These E_MM values could finely characterize the final conformation of species-specific AHR1-LBD for each brominated congener.On this basis,generalized linear models（GLMs）was developed using machine learning（ML）algorithms and spline approximation methods to qualitatively predict complex relationships between AHR1 conformations and DLPs or avian interspecies relative sensitivity to the brominated dioxin-like compounds（DLCs）.The models were built based on the ligand-specific and species-specific detailed conformational information of AHR1-LBDs.And they combined the advantage of high predictive accuracy of ML algorithms with the strength of better interpretability of GLM models.In addition,the regions where the key AAs determining the relative sensitivity of avian species are located in can be accurately identified using the feature selection method used in the present study.Therefore,the findings of this project provide fundamental scientific knowledge for the TEFs derivation and the ecological risk assessment in birds,a novel tool to identify potential DLCs of high priority,a novel strategy to reveal the molecular basis underlying interspecies sensitivity variations to DLCs in other vertebrates at the AA level,and a novel idea to facilitate the development of computational prediction models for supporting chemical safety assessment.The main results of this study are as follows:（1）The species-specific RePs of PBDD/Fs and PBBs in avians do not always follow the WHO-TEF of their chlorinated congeners,some brominated dioxins even have more toxicity potency.Therefore,in the risk assessments of brominated dioxins to ring-necked pheasant and Japanese quail,using WHO-TEF values of the corresponding chlorinated congeners as a reference would lead to underestimation of risk.The ReP values of different avians even exceed two orders of magnitude.Except for ReP values of some brominated dioxins,the ranking of most ReP values was similar in the three types of avians.Otherwise,these results indicated that not all brominated dioxin congeners complied with the sensitivity rank order of avian species for 2,3,7,8-tetra CDD,that is,chicken>ring-necked pheasant>Japanese quail.（2）Based on the data set of 107 E_MM features and the five key features,the predictive classifiers built using ML algorithms have high predictive accuracy（>87%）and generalization ability（>88%）.Therefore,the classification and prediction of the toxicity and sensitivity of brominated dioxins to birds can be realized.Expect for the simplified chicken-ReP classifier retained the classification accuracy rate of 10-times 10-fold CV at 87.3%,other simplified RF classifiers were higher than 90.7%,which were comparable to those of the original models（>91.3%）.However,the generalization ability of simplified classifiers is usually higher than that of original classifiers.Therefore,the performance of the simplified classifier is comparable to that of the original classifier and even has better evaluation metrics.In addition,the classification results based on the GLM models are consistent with those of the simplified RF classifiers,indicating the reliability of the GLM models.Thus the GLM model can be used instead of the simplified RF classifiers to predict the avian species-specific ReP and congener-specific interspecies sensitivity variations in birds for brominated dioxins.