QSAR models for oxidation of organic micropollutants in water based on ozone and hydroxyl radical rate constants and their chemical classification

Abstract

Ozonation is an oxidation process for the removal of organic micropollutants (OMPs) from water and the chemical reaction is governed by second-order kinetics. An advanced oxidation process (AOP), wherein the hydroxyl radicals (OH radicals) are generated, is more effective in removing a wider range of OMPs from water than direct ozonation. Second-order rate constants ($k_{\text{OH}}$ and $k_{{\text{O}}_3}$) are good indices to estimate the oxidation efficiency, where higher rate constants indicate more rapid oxidation. In this study, quantitative structure activity relationships (QSAR) models for O₃ and AOP processes were developed, and rate constants, $k_{\text{OH}}$ and $k_{{\text{O}}_3}$, were predicted based on target compound properties. The $k_{{\text{O}}_3}$ and $k_{\text{OH}}$ values ranged from 5 * 10⁻⁴ to 10⁵ M⁻¹s⁻¹ and 0.04 to 18 * (10⁹) M⁻¹ s⁻¹, respectively. Several molecular descriptors which potentially influence O₃ and OH radical oxidation were identified and studied. The QSAR-defining descriptors were double bond equivalence (DBE), ionisation potential (IP), electron-affinity (EA) and weakly-polar component of solvent accessible surface area (WPSA), and the chemical and statistical significance of these descriptors was discussed. Multiple linear regression was used to build the QSAR models, resulting in high goodness-of-fit, r² (>0.75). The models were validated by internal and external validation along with residual plots.

Introduction

Organic micropollutants (OMPs) in water are of a great concern in drinking water and the environment. Studies have shown their negative impacts on different strata of living organisms such as plants, sea-life, and animals (Ericson et al., 2010; Oaks et al., 2004; Qin et al., 2011). Regulatory agencies (U.S. EPA, EU directive) have taken measures to identify the problems associated with OMPs and have included several OMPs in their directive frameworks (Richardson and Ternes, 2005). Pharmaceuticals and personal care products which constitute many of the OMPs are either highly hydrophobic or metabolized to a more hydrophilic molecule than the parent OMP and hence pass through the water treatment plant. The hydrophobic OMPs are adsorbed onto biosolids in wastewater treatment and can eventually be removed whereas the hydrophilic OMPs and their metabolites are highly resistant to advanced/conventional water treatment methods (Klavarioti et al., 2009).

Ozonation is used for disinfection and oxidation or a combination of both (Camel and Bermond, 1998). Ozone is highly unstable in water and undergoes either direct ozonation with OMPs or decomposes into OH radicals (advanced oxidation process (AOP)) which are strong oxidants in water (Staehelin and Hoigne, 1985). An ozone molecule can act as a dipole, electrophilic or nucleophilic agent while reacting with organic solvents and thereby undergoes cycloaddition due to dipolar structure; electrophilic reactions with electron dense compounds such as aromatics and alkenes; and nucleophilic reactions (rare) with compounds having low electron dense molecular sites. In the case of OMPs in water, the electrophilic and cycloaddition reactions are more common. Aromatics with electron-donor groups (alcohols, amines, etc.) have high electron-density on carbon atoms in the ortho–para positions compared to meta position and hence are highly reactive with ozone at these positions (Langlais et al., 1991). The (difference in) energies of the molecular orbitals, ionisation potential, electron-affinity, hydrophobic/hydrophilic surface area also influence ozonation (Karelson et al., 1996; Sudhakaran et al., 2012). During the ozonation process, the rate of the reaction is indicated by a second-order rate constant (direct ozone: $k_{{\text{O}}_3}$, AOP: $k_{\text{OH}}$), which is a constant for a given reaction at a particular temperature. The $k_{\text{OH}}$ values (10⁹ M⁻¹ s⁻¹) are much higher than the corresponding $k_{{\text{O}}_3}$ values (∼10 M⁻¹ s⁻¹) which indicate that OH radical mediated reactions are faster reactions since they are radical-based. The ozonation of OMPs is a second-order type of reaction, i.e., first order with respect to ozone and first order with respect to an OMP. There is plenty of scientific literature focusing on the ozone/hydroxyl radical rate constants which are also compiled in the kinetic data base Radiation Chemistry Data Center of the Notre Dame Radiation Laboratory, available at http://kinetics.nist.gov/solution/. The decomposition of dissolved ozone is highly affected by pH, ozone concentration, and the concentration of various scavengers (von Gunten, 2003). Since the ozonation process is governed by the structural attributes of a molecule, the ozone/hydroxyl radical rate constant (property) of the OMPs can be associated with their structure. The association of the chemical structure to the activity/property (rate constants) leads to a Quantitative Structure Activity Relationship (QSAR) model.

QSAR models are based on the concept that the structure of a molecule influences its properties and are an inter-discipline between chemistry/biology and statistics. There are different statistical approaches in building QSAR models. The most frequently used methods are multiple regression analysis, principal component and factor analysis, principal regression analysis, partial least squares (PLS), discriminant analysis, and neural networks. The statistical approaches are accompanied by chemical descriptors to characterize the chemical structure.

There are also 3-D QSAR methods wherein the three-dimensional structure of the molecule and its interaction with its surroundings (solvent, other molecules etc.) is studied, and comparative molecular field analysis (CoMFA) is the most commonly used approach (Trapido et al., 1995). While developing a QSAR model there are three components to focus on: dataset, molecular descriptors and statistical technique. The dataset for the QSAR should be reliable, measured in a consistent manner. The molecular descriptors should be mechanistically related to the predicted property/activity. In the case of descriptors governed by structural conformation such as molecular orbital energies, ionisation potential, etc., proper energy optimization methods must be used. In case of statistical analyses, methods that are simple, transparent, easily interpretable should be the first priority. Transparent models can be easily understood and updated. The transparency decreases with progressing from regression, to partial least squares and finally neural networks, however, the type of dataset also plays a role in choice of the statistical technique. The QSAR model should not be used to make predictions that extrapolate beyond the conditions associated with the model. Finally, validation ensures that the QSAR model can be used for prediction. For this purpose, a certain proportion of the training dataset called the test set (usually up to 50%) can be chosen. The test set should be representative of the complete dataset (Cronin and Schultz, 2003).

The applications of QSAR models are increasing. The Organization for Economic Cooperation and Development (OECD), in their environment directorate, encourage governments and industry to use QSAR models to evaluate the ecotoxicity of chemicals. The chemicals policy of the European commission (REACH: registration, evaluation and authorization of chemicals) use QSAR models to reduce experimental testing. In order to increase the acceptability of QSAR models it is necessary to follow the principles laid down in OECD for a validated QSAR model (Jaworska et al., 2005; Gramatica, 2007). In the U.S. EPA, the estimation program interface (EPI) suite includes QSAR-programs such as AOPWIN, KOWWIN, and BIOWIN. These programs help to estimate the ecotoxicity and physical/chemical properties of compounds.

QSAR models for oxidation/ozonation often use percent-removal or rate constants $\left(k_{\text{OH}},k_{{\text{O}}_3}\right)$ as the predicted variable (Sudhakaran et al., 2012; Lei and Snyder, 2007; Kusic et al., 2009; Pompe and Veber, 2001; Oberg, 2005). Percent-removal of OMPs, when used as a performance index to build a QSAR model, is limited by water quality parameters such as dissolved organic matter, alkalinity etc. whereas rate constants $\left(k_{\text{OH}},k_{{\text{O}}_3}\right)$ are more universal and are not influenced by dissolved organic matter present in water. The apparent rate constants are only affected by the speciation (positive/negative) of the OMPs due to change in pH (Sonntag and Gunten, 2012). Also, compared to the percent-removal of OMPs, the rate constants can be readily coupled with an ideal water reactor (plug-flow, continuous stirred tank).

QSAR models are also developed based on correlations between rate constants and substituent descriptor constants such as Hammett/Taft constants. In this type of study, emphasis is on compounds with a common parent structure (phenol, amine etc.) and other part of the molecules are considered as a substituent, and their corresponding Hammett/Taft (σ/σ*) constants are used and their correlations with the rate constants are studied. These constants indicate the electron-donating and -withdrawing properties of the substituents. In case of complex structures, such as pharmaceuticals, structural approximations are used to compute the Hammett/Taft constants. The structural approximation is based on the premise that inductive/resonance effects of substituents atoms are attenuated with increasing distance from the reaction center. Finally, a linear correlation equation between the rate constants and the substituent descriptors is the QSAR model (Lee and von Gunten, 2012).

In this research study, the individual physico-chemical properties of the OMPs were calculated. Then a sequential statistical approach was followed: correlation analysis, principal component analysis and finally a multi-linear regression equation with ozonation relevant descriptors. Later, the models were also validated. Correlation analysis helps to identify the significant molecular descriptors; principal component analysis helps to study the data/cluster patterns and reduce redundant variables; multi-linear regression creates a QSAR model in the form of an equation between the rate constants and the descriptors. Finally, a validation is performed to testify that the QSAR model can be used to make predictions. Also, based on the substituent parameters associated with Hammett/Taft constants, i.e., electron-withdrawing groups decrease ozonation, and electron-pumping groups enhance ozonation, in this study, the chemical classification of the OMPs based on their functional groups was carried out.