Erle Austdal Aabrekk, Lisbet Sørensen, and Anders Lervik
Department of Chemistry, Norwegian University of Science and Technology (NTNU), Trondheim, Norway
Email: erleaaa@stud.ntnu.no
Chemicals are everywhere, several of which are problematic to the environment or human health. Understanding how these chemicals are distributed is essential for risk assessment. Public databases list up to 250 million chemicals, but few of them have associated information of environmental properties (0.03 % and 0.01 % of chemicals in PubChem have physicochemical properties and toxicological properties, respectively) (1). The need for methods that can estimate these properties for a wide range of chemicals is therefore important.
The octanol-water partitioning coefficient (Log KOW) is the most common way of expressing the lipophilicity of a compound, indirectly serving as a measure of the bioaccumulation potential of neutral organic compounds. Log KOW has also been shown to have a strong correlation to the sediment-water partitioning coefficient and is therefore an overall useful proxy to predicting how contaminants in the environment will distribute (2). Log KOW values, either experimental or modelled, are widely available, and as such lend themselves to development and optimization of prediction tools (3).
Gas chromatography (GC) has great potential for non-target analysis of complex mixtures (4). GC retention times are primarily determined by analyte properties, including boiling point, size, and polarity – all determined by chemical structure. Two-dimensional GC (GC×GC) retention has shown promise towards inferring environmental behaviour of analytes (5). Coupling with high-resolution mass spectrometry (HRMS) provides ultimate separation in addition to potential for unknowns’ identification through coupling of mass spectral signatures with structural properties. If direct structural annotation is not possible, our hypothesis is that the spectral fingerprint can provide diagnostic information towards prediction of (environmental) partitioning properties.
In the current work, the aim is to develop a universal prediction tool for Log KOW for neutral water-soluble chemicals based on their GC×GC retention times and/or their HRMS signatures. Both linear regression models and a k nearest neighbour machine learning model are tested. The models are first implemented for individual chemical groups (PCBs, PBDEs, PAHs…) and later combined for preparation of a universal model. Improvements in prediction strength by use of experimental EI MS spectra are tested, and finally recommendations for use of either GC×GC, GC-HRMS or GC×GC-HRMS for non-target analysis of environmental samples are made based on the combination of increased prediction power and investment-benefit considerations for laboratories.
References
1 Zushi Y. Direct prediction of Physicochemical Properties and Toxicities of Chemicals from Analytical Descriptors by GC-MS. Analytical Chemistry. June 28th 2022; Volume (94): Pages 9149-9157.
2 Dewulf J, Langenhove HV, Graré S. Sediment/water and octanol/water equilibrium partitioning of volatile organic compounds: temperature dependence in the 2-25°C range. Water research. July 1st 1999; Volume (33): Pages 2424-2436.
3 Amézqueta S, Subirats X, Fuguet E, Rosés M, and Ràfols C. Chapter 6 – Octanol-Water Partitioning Constant. I: Poole CF, red. Liquid-Phase Extraction. Barcelona: Elsevier; 2020. 183-208.
4 Wilson MB, Barnes BB, Boswell PG. What experimental factors influence the accuracy of retention projections in gas chromatography-mass spectrometry? Journal of Chromatography A. December 19th 2014; Volume (1373): Pages 179-189.
5 Nabi D, Gros J, Dimitriou-Christidis P, and Arey JS. Mapping Environmental Partitioning Properties of Nonpolar Complex Mixtures by Use of GC×GC. Environmental Science & Technology. June 17th 2014; Volume (48): Pages 6814-6826.