Development and Validation of Multiple Linear Regression Models for Predicting Chronic Zinc Toxicity to Freshwater Microalgae

Abstract

Multiple linear regression (MLR) models were developed for predicting chronic zinc toxicity to a freshwater microalga, Chlorella sp., using three toxicity-modifying factors (TMFs): pH, hardness, and dissolved organic carbon (DOC). The interactive effects between pH and hardness and between pH and DOC were also included. Models were developed at three different effect concentration (EC) levels: EC10, EC20, and EC50. Models were independently validated using six different zinc-spiked Australian natural waters with a range of water chemistries. Stepwise regression found hardness to be an influential TMF in model scenarios and was retained in all final models, while pH, DOC, and interactive terms had variable influence and were only retained in some models. Autovalidation and residual analysis of all models indicated that models generally predicted toxicity and that there was little bias based on individual TMFs. The MLR models, at all effect levels, performed poorly when predicting toxicity in the zinc-spiked natural waters during independent validation, with models consistently overpredicting toxicity. This overprediction may be from another unaccounted for TMF that may be present across all natural waters. Alternatively, this consistent overprediction questions the underlying assumption that models developed from synthetic laboratory test waters can be directly applied to natural water samples. Further research into the suitability of applying synthetic laboratory water-based models to a greater range of natural waters is needed. Environ Toxicol Chem 2023;42:2630-2641. © 2023 The Authors. Environmental Toxicology and Chemistry published by Wiley Periodicals LLC on behalf of SETAC.

Keywords: Metal toxicity; Multiple linear regression; Toxicity modifying factors; Toxicity prediction modeling.