Integration of non-target screening, statistical analyses and biossays to globally assess chemical water quality

Advancements in high-resolution mass spectrometry (HRMS)-based screening methods have enabled a shift from target to non-target analyses to detect chemicals in water samples. Non- target screening has therefore become a promising tool to evaluate the changes of chemical water quality during water treatment (Nürenberg et al., 2015). However, the wealth of data resulting from non-target screenings renders structural identification virtually impossible (Hollender et al., 2017). The aim of the exploratory research project presented here was to evaluate the use of information generated by non-target data to study water treatment, without identification of all HRMS peaks. It focused on three different levels of non-target data for water quality assessment, i.e. the “unknown feature” level, the “suspect” level, and the “trend profile” level.
A feature represents a given compound and consists of a unique combination of an accurate mass and a retention time. Without identifying the feature, information on its response – measured in instrument counts or response relative to an internal standard, presence in a homologous series, mass defect, isotopic pattern and predicted hydrophobicity presented as log octanol water partition coefficient (Kow) can be automatically extracted (Heberger, 2007; Zhang et al., 2009; Carlson et al., 2012; Sleno, 2012; Jobst et al., 2013; Nagao et al., 2014; Bade et al., 2015; Aalizadeh et al., 2016; Parry and Young, 2016; Sjerps et al., 2016; Loos and Singer, 2017). The unknown feature level refers to all this intrinsic information. The suspect level refers to potential candidates that match a feature through automated suspect screening against an in-house curated suspect list consisting of environmentally relevant compounds and predicted transformation products. Finally, the trend profile level combines the two, and reveals patterns in the data through statistical methods, with the goal to cluster both features and the effects of water treatments on water quality (Muller et al., 2011; Schollee, 2015; Schollée et al., 2016). Distinction is made between persistence, elimination and formation during treatments. The trend profile level can then be connected to responses of bioassays.

Er zijn verschillende statistische tools en workflows ontwikkeld waardoor het nu mogelijk is conclusies te trekken over de effecten van waterbehandeling(stappen) op basis van de grote hoeveelheden data die worden gegenereerd met non-target screening (hoge resolutie massaspectrometrie gecombineerd met vloeistofchromatografie) en bioassays. Deze statistische of data science tools en methoden zijn getest in twee casestudies: een pilot-scale data set uit de drinkwaterbehandeling met gedoseerde organische- microverontreinigingen (DPWE robuustheid zuiveringen) en een real-scale data set uit de innovatieve afvalwaterzuivering (H2020 AquaNES). Non-target-resultaten zijn met bioassay metingen geïntegreerd om een uitgebreider beeld van de chemische waterkwaliteit te krijgen. Daardoor wordt informatie gegenereerd, die bij alleen target screening ontbreekt: verschillen tussen monsters en behandelingstappen worden op een efficiënte manier aangetoond. De visualisatie helpt hierbij om een duidelijk beeld van complexe data te geven en vereenvoudigt de prioritering.