Fig. 4: Molecular-level characterization of complex PM2.5 extracts.
High-throughput characterization of PM2.5 molecules was performed by combining information from MS1 and MS2 HRMS data. The integration of multiple cheminformatics approaches is illustrated here for molecular features detected in the WSOC extracts by LC-HRMS with ESI+ (see also Fig. S6). a Molecular formulae were assigned to a proportion of features (45% in this example, see Fig. S5) allowing color-coded visualization in van Krevelen diagrams, and in b molecular networks (GNPS). In the network, each feature is shown as a node linked to other nodes by edges indicating the degree of similarity among deconvoluted MS2 spectra (minimum cosine score = 0.65). In this example, 10,051 molecular features are clustered into 1064 molecular families having inferred structural analogy. c Zoom-in on a molecular family cluster of 15 nitrogen-containing benzenoids. The first-candidate structures from re-ranked in silico predictions (NAP/MetFrag) are shown in blue. The highlighted node (red outline) shows two putative annotations for the same molecular formula (C9H10N2), i.e., 5,6-dimethylbenzimidazole (red structure) from the GNPS library match (Fig. S18), and 1-indanonehydrazone (blue structure) as the top in silico first-candidate ranked by the network consensus (Fig. S20). The structure of 5,6-dimethylbenzimidazole was also predicted in silico, but ranked as the sixth candidate.
