Identification of unknown contaminants in surface water : combination of analytical and computer-based approaches

Thousands of different chemicals are used in our daily life for household, industry, agriculture and medical purpose, and many of them are discharged into water bodies by direct or indirect ways. Thus, monitoring and identification of organic pollutants in aquatic ecosystem is one of the most essential concerns with respects to human health and aquatic life. Althrough liquid chromatography coupled to high resolution mass spectrometry (LC-HRMS) has made huge advancements in recent years, allowing non-target screening without initial information and reference stardards, the detection and identification of unknown compoundsfrom the large datasets generated is still challenging. To overcome this, various software-based approaches have been developed, particulalry by the metabolomics community, covering one or more steps such as peak detection and alignment, or fragmentation and retention prediction . However, an evaluation and optimization of these approaches for the nontarget screening of environmental samples is largely lacking. Thus, the main objective of this PhD thesis is evaluation and optimization of the performance of two important steps in the nontarget screening workflow to improve their application for environmental pollutants, and to demonstrate its use the prioritisation of catchment-specific organic micropollutants in rivers. In general, nontarget screening workflows start from detection of peaks and extracted their information (m/z, retention times, and intensities) as peak lists from raw data files by the peak picking software which plays an important role in deconvolution and detection of chromatographic peaks from huge LC–HRMS dataset. In order to maximum the quality and number of detected peaks the performance of one of the widely used data processing software MZmine 2 was assessed and optimized for data acquired on a LTQ Orbitrap (Chapter 2). To this end, a design of experiments (DoE) approach was used. The results show that full scan acquisition is recommended for data acquisition in nontarget screening since short cycle time (CT) significantly improves the performance of automatic peak detection. Compared to manual peak detection, MZmine 2 with the optimized parameter settings detected 75 -100% of compounds at an intensity level of 105 on both spiked and real surface water samples. In the end, an optimization workflow for MZmine 2 is provided and which can be easily applied on other LC-HRMS instruments data. While the determination of the molecular formula of an unknown peak based on accurate masses and isotope peaks is often straightforward, it is challenging to find the most likely structure for the obtained molecular formula among hundreds or thousands of possible candidates retrieved from a database search. Different software-based predictions are helpful to prioritise candidates with the most likely structure, however, their performance has rarely been evaluated. Hence, in Chapter 3, a dataset of 78 diverse known micropollutants analyzed both in ESI+ and ESI- mode was used to assess two different MS/MS fragmentation and two retention prediction approaches. For MS/MS analysis, different fragmentation technologies and energies were applied for fragmentation prediction and the rank positions the “correct” candidates were compared. Additionally, application domains of two retention prediction models were investigated according to the relatively rank of the “correct” structure and prediction accuracy of the models. To combine scores from multiple candidate selection tools, consensus score values with optimal weights were calculated to show the contribution of each approach and whether the combination could improve candidate selection. It seems unlikely that a complete identification of all peaks in sample will be ever feasible, thus identification efforts have to focus on a limited number of peaks of interest, calling for a thorough prioritization. In Chapter 4, a recently proposed approach to prioritise peaks and identify compounds solely from HRMS data was used on a set of river water samples. Frequency Scores were calculated as a combined measure for the intensity and frequency of occurrence of a peak and peak attribute information (isotopes, adducts, and homologue series) was generated automatically obtained from peak lists using the R nontarget package. Water sample extracts from 19 sites distributed over four European catchments (Saale, Danube, Sava and Emme) were analyzed using a LTQ Orbitrap instrument and the data analyzed using the outlined approach. Results show that both ubiquitous peaks in all catchment and specific peaks in each catchment occur among the top peaks ranked by Frequency Scores. About 10 to 50% of the top ranked peaks were assigned to homologue series, indicating the huge contribution of surfactants and their transformation products to the peak inventory. A number of unknown compounds could be (tentatively) identified, including several high production volume industrial chemicals and surfactants