Molecular Descriptors and Formula-Level Signatures Associated with Dissolved Organic Matter Chemodiversity across Freshwater Trophic Gradients

Abstract We characterized dissolved organic matter (DOM) chemodiversity across freshwater trophic gradients by combining Fourier transform ion cyclotron resonance mass spectrometry with interpretable machine learning. Chemodiversity was quantified from the richness and evenness of detected molecular formulas using a chemodiversity index (CDI) in 48 lake and river samples spanning oligotrophic to middle-eutrophic conditions. Overall, CDI was negatively correlated with the trophic state index (TSI) (r = −0.700, P < 0.001), with a pronounced decline in lakes and no detectable relationship in rivers. A random forest model linked 13 molecular descriptors of 2840 recurrently detected formulas to their formula-level Spearman correlations with CDI. Shapley additive explanations analysis highlighted correlated molecular-property dimensions associated with saturation, aromaticity, and oxidation state. In VK diagrams, CDI-positive formulas were enriched in lignin-like, tannin-like, and condensed aromatic regions, whereas CDI-negative formulas were concentrated in lipid-like and aliphatic regions. Representative 20 high-CDI formulas were predominantly lignin-like; all but one met compositional criteria for carboxyl-rich alicyclic molecules. In 24 independent samples, the summed relative peak intensity of these 20 formulas declined significantly with increasing TSI, whereas the corresponding association for the 20 low-CDI formulas was not significant. These findings identify molecular signatures of DOM chemodiversity and show that its relationship with trophic state differs between lakes and rivers within the sampled trophic range.