QSTR and interspecies-QSTR modelling for aquatic toxicity data gap filling of cationic polymers

Abstract

Polymers are extensively used in several fields representing a growing multi-billion dollar industry covering several thousands of materials and billions of kilos used globally every year. The widespread use of cationic polymers may significantly discharge such compounds into the aquatic environment, which may cause potential toxic effects on aquatic organisms. The amount of publicly available, high-quality environmental toxicity data for industrial polymers such as cationic polyquaterniums is low. We have developed here individual quantitative structure–toxicity relationship (QSTR) models for toxicity prediction against fish and algae. These models against fish and algae showed optimistic statistical quality in terms of several internal and external quality and validation metrics such as determination coefficient R2 (0.703 and 0.676), cross-validated leave-one-out Q2 (0.638 and 0.516) and predictive R2pred or Q2ext (0.776 and 0.703) for fish (Ntrain = 72, Ntest = 23) and algae (Ntrain = 40, Ntest = 14) toxicity datasets, respectively. The study revealed that higher charge density increases the toxicity against both the response endpoints. However, a higher percentage of oligomers with a molecular mass of lower than 1000 Daltons results in a decreased toxicity towards both the studied endpoints. Similarly, primary amines in the molecular building block result in a reduction in the toxicity against the algal species. However, acceptable individual QSTR models against D. magna could not be generated with the limited feature information obtained from the United States Environmental Protection Agency and additional data provided by Environmental Climate Change Canada (ECCC). Therefore, we have also proposed interspecies quantitative structure–toxicity relationship (i-QSTR) models among three species (D. magna, fish and algae species) to bridge the toxicity data gap for cationic polymers. The mechanistic interpretation of i-QSTR models revealed several important characteristic features of polymers along with the experimental response of one species which are also helpful for toxicity prediction of other species, ultimately helping to reduce the experimental testing for toxicity prediction. Finally, the proposed QSTR and i-QSTR models can be helpful to compute the toxicity of polymers in the early stages of screening for regulatory purposes and data gap filling for new or untested polymers falling within the applicability domain of the models.