Assessment of quantitative structural property relationships for prediction of pharmaceutical sorption during biological wastewater treatment

In this study, we critically examined the available data related to pharmaceutical (PhAC) sorption in biological treatment processes. Using these data, we developed and assessed single and polyparameter quantitative structural activity models to better understand the role of sorption in PhAC attenuation. In contrast to other studies, our analysis suggests that values of the sorption coefficient (KD) are poorly correlated to single parameter models employing log KOW or the apparent partition coefficient (i.e., KOW corrected to the experimental pH). Results from the development of polyparameter models suggest that the range of functional moieties typically incorporated in PhAC molecules offers a diverse set of interactions between PhAC and sludge surface (e.g., hydrogen bonding, electrostatic interactions, and hydrophobic interactions). Of particular importance is the role of dissociation and resulting charge(s) of a PhAC in solution. Results demonstrate that when developing predictive models it is advantageous to separate PhACs based upon the charge of the dominant species at the experimental pH. Yet, use a single model for PhACs which are negatively charged and uncharged may have practical utility. Performance of the polyparameter models, however, was found to plateau with a pred-R2 between 0.50 and 0.60, even when six statistically relevant predictors are included. This outcome suggests that effective predictive models for PhAC sorption cannot include solely PhAC descriptors, rather they must incorporate critical properties related to the sorbent (i.e., mixed liquor) surface.

Figure optionsDownload as PowerPoint slideHighlights
► QSAR models are developed for PhAC sorption during biological wastewater treatment.
► Models employing log KOW or log D alone generally offer insufficient predictive capability.
► Performance of best-fit polyparameter models plateau between 0.45 and 0.65 pred-R2.
► Predictive capability is greater when the fraction of uncharged species >0.85.
► Results suggest biosolids characteristics should be considered with sorption data.