Convergent Fragmentation: A Shared Ion Vocabulary Limits Structure−Spectrum Fidelity

Tandem mass spectrometry (MS/MS) library searching and molecular networking assume that molecules with similar structures give similar spectra. We examined what limits this structure−spectrum fidelity (SSF) using high-resolution electrospray ionization MS/MS spectra of authentic standards in METLIN, acquired from even-electron [M + H]+ and [M − H]− ions by collision-induced dissociation at 0, 10, 20, and 40 eV. Across 4,000 compound pairs spanning the range of structural similarity, structural and spectral similarity were only moderately correlated (R2 = 0.28). Two fragmentation processes weaken this relationship in opposite directions. First, fragments that form only after hydrogen atoms move between the separating pieces (hydrogen redistribution) can make the spectra of closely related molecules diverge. Second, structurally unrelated molecules often fragment to the same small product ions. Across 958,450 METLIN 960K compounds, we found 56 such recurrent ion formulas in five chemical families: aromatic/allylic, iminium/ammonium, saturated alkyl and acylium cations, and small heteroatom-containing anions. At least one occurs in 80.7% of library compounds, every family becomes more prevalent with collision energy, and all 56 were independently detected in MassBank of North America. In pairwise comparisons, shared aromatic/allylic cations raised spectral similarity only among the few pairs sharing many of them. Removing the recurrent cations raised the structure−spectrum correlation from ρ = 0.410 to 0.468, but only aromatic/allylic cations carried less structural information than ordinary fragments of equal prevalence; iminium/ammonium ions did not. The 56 shared fragment ions can inflate spectral similarity between structurally different molecules, creating a major obstacle to inferring structural relationships from tandem mass spectra, particularly in molecular networking.