Enhancing Molecular Characterization of Dissolved Organic Matter by Integrative Direct Infusion and Liquid Chromatography Nontargeted Workflows

Abstract

Dissolved organic matter (DOM) in aquatic systems is a highly heterogeneous mixture of water-soluble organic compounds, acting as a major carbon reservoir driving biogeochemical cycles. Understanding DOM molecular composition is thus of vital interest for the health assessment of aquatic ecosystems, yet its characterization poses challenges due to its complex and dynamic chemical profile. Here, we performed a comprehensive chemical analysis of DOM from highly urbanized river and seawater sources and compared it to drinking water. Extensive analyses by nontargeted direct infusion (DI) and liquid chromatography (LC) high-resolution mass spectrometry (HRMS) through Orbitrap were integrated with novel computational workflows to allow molecular- and structural-level characterization of DOM. Across all water samples, over 7000 molecular formulas were calculated using both methods (∼4200 in DI and ∼3600 in LC). While the DI approach was limited to molecular formula calculation, the downstream data processing of MS2 spectral information combining library matching and in silico predictions enabled a comprehensive structural-level characterization of 16% of the molecular space detected by LC-HRMS across all water samples. Both analytical methods proved complementary, covering a broad chemical space that includes more highly polar compounds with DI and more less polar ones with LC. The innovative integration of diverse analytical techniques and computational workflow introduces a robust and largely available framework in the field, providing a widely applicable approach that significantly contributes to understanding the complex molecular composition of DOM.

Keywords: LC-orbitrap; dissolved organic matter; environmental water; molecular fingerprinting.

Figure 1

Experimental workflow. Samples from three different sources, i.e., seawater, river water, and drinking water, were filtrated, extracted, and processed by two HRMS acquisition methods (DI and LC-DIA). Raw data analysis included preprocessing, molecular formula calculations, and, for LC data, computational workflows of molecular networking and in silico structure prediction.

Figure 2

Molecular formulas distribution among samples and methods. Venn diagrams showing the formulas that were unique to each sample in direct infusion (a) and liquid chromatography in negative (b) and positive (c) ionization modes. The percentage of formulas considering all three samples is reported in blue, the percentage in the parentheses represents the % of formulas unique in that single sample type. (d) Venn diagram and (e) van Krevelen plot showing the overlap of the whole set of formulas assigned by DI and LC (DI in yellow, LC-ESI- in blue, and LC-ESI+ in red; green triangles represent the common compounds in the two methods DI and LC).

Figure 3

Molecular networks and in silico structural annotations. Molecular networks were built in GNPS based on MS2 spectra similarity. In this example, 17 796 features were clustered into 1789 molecular families (a). Zoom-in on a molecular family including five aromatic compounds (b). Each feature is represented as a node and colored by its relative abundance among the different samples, i.e., river water (orange), seawater (blue), and drinking water (light blue). Edges linking each node are colored based on the modified cosine score (>0.65) representing the spectral similarity between each feature (nodes). The molecular formulas and the associated first candidate structures from in silico predictions reranked by the MetFrag/NAP workflow are shown for some of the features in the example subnetwork clusters (b1, b2). The bottom cluster (b2) includes a node with putative annotation in the GNPS library (3,4-dihydroxybenzoate, m/z 163.019, RT 1.81 min). In this example, the first candidate structure from NAP is an isomer of the level 2 annotated compound (2,5-dihydroxybenzoate), while the 3,4-dihydroxybenzoate is predicted and ranked as third candidate.

Figure 4

Examples of in silico predicted candidate structures. Molecular formulas in green represent natural compounds or metabolites, while formulas in brown and underlined can be attributed to environmental contaminants, according to the information provided for each structure by NAP databases.

Figure 5

(a) Van Krevelen diagrams showing assigned peaks in seawater, drinking water, and river water samples by DI-HRMS (orange), LC-(ESI−)HRMS (blue), and LC-(ESI+)-HRMS (red). (b) Corresponding Kendrick diagrams of data described in van Krevelen areas of (a).