AbstractOver the past forty years concerns over the presence of pharmaceuticals in the environment have grown considerably. Some pharmaceuticals can be effectively biodegraded in wastewater treatment plants but others can be sorbed onto sludges that are often subsequently used as fertilisers or disposed of to landfill.
This work aimed to understand how a given pharmaceutical will be distributed between the aqueous and solid phases (characterised by the sorbed:dissolved partition coefficient, 퐾푑) within a treatment plant, which is important to be able to make accurate risk assessments. An official guideline test to measure the partitioning of a pharmaceutical in sewage sludge is available, but it is time consuming and fastidious. As activated sewage sludge is a complex matrix, commercially available solid-phase extraction (SPE) cartridges with different chemistries were used to characterise the pharmaceutical-sludge binding processes. As part of this work a new solid-phase extraction screening method has been developed to provide rapid measurements of 퐾푑 and its performance was evaluated against partition coefficients obtained with the official guideline method with a correlation coefficient of 0.93 and a 푟2 of 0.94. In addition, this rapid method allowed the measurement of partition coefficients for pharmaceuticals for which values were not available in the literature and these have been used to further validate new predictive models.
Predictive models based on the octanol-water partition coefficient have been developed to calculate partitioning properties of compounds in soil, and these have been extended for application to sewage sludge. These models are optimised mainly for neutral organic chemicals, and only a few consider ionic substances. The work described in this thesis compared the performance of these soil-based models for a range of pharmaceuticals, including ionisable compounds, and assessed their application for the binding of ionic and non-ionic pharmaceuticals in sewage sludge. It also explored other predictive models based on molecular descriptors obtained from chemical structure. These models provided improved predictions over previous models based solely on the octanol-water partition coefficient. In this thesis, partial least squares (PLS) and Bayesian artificial neural network (ANN) models were evaluated for their accuracy in predicting the partition coefficient for neutral, basic, acidic and zwitterionic pharmaceuticals. Literature values were used to develop the models based on a range of molecular descriptors, and their predictive ability was assessed on an external test set of compounds excluded from the model-building process. The performance of the linear PLS and non-linear ANN models were discussed, and their predictive performance and interpretability were compared.
Attempts to apply the method for rapid measurement of the sorption of pharmaceuticals to soils were also made to investigate potential read-across from one environmental matrix to another but the two matrices were too dissimilar to achieve this.
|Date of Award||Jan 2015|
|Supervisor||Graham Mills (Supervisor), David Whitley (Supervisor), Richard Greenwood (Supervisor), David Whitley (Supervisor) & Graham Mills (Supervisor)|