Infrared ion spectroscopy (IRIS) integrates MS with vibrational spectroscopy for the structural characterization of gas-phase ions [[1], [2], [3], [4], [5]] and is currently evolving from a tool in fundamental gas-phase ion chemistry [[6], [7], [8], [9]] to a routine method for molecular identification in analytical chemistry [10]. This evolution in scope has been enabled by adapting high-end MS platforms for the application of IRIS [[11], [12], [13]], providing high sensitivity and facile integration with liquid-chromatography [14], which are required for the analysis of complex (bio)analytical samples. In this development, it is clear that analytical samples containing isomeric species pose specific challenges to identification through IRIS [15], as mass selection alone is not sufficient to isolate the species of interest prior to spectroscopic investigation. To obtain isomer-selective IR spectra from a mixture, IRIS has been applied to species separated by liquid-chromatography MS (LC-MS) [14,[16], [17], [18], [19]]. Depending on the application, separation by ion mobility [[20], [21], [22]] may be preferred over – or applied in addition to – separation by LC, for instance when MALDI ionization is used instead of ESI, when isomers are formed upon protonation in the source, or when conformational isomers must be separated. Moreover, an ion mobility measurement aids in the structural characterization of ions in MS, especially since the collision cross section (CCS) is a computable molecular property [[23], [24], [25], [26]].

Various of implementations of the integration of IMS-MS with IR spectroscopy have been reported over the past years, greatly enhancing the structural information that can be obtained from CCS and m/z values alone. Combination of drift-tube IMS and a tunable free-electron laser has been used to address the structure of large biomolecular assemblies such as the serine-octamer [27] and the native structure of proteins during their transfer to the gas phase [28]; in small molecules, these methods have been employed for a detailed characterization of protonation tautomers of aminobenzoic acid derivatives [29,30]. Field-asymmetric IMS (FAIMS) has been used in combination with IRIS to discriminate analytically challenging saccharide isomers [31,32]. The combination of IMS with cryogenic IRIS is emerging as a bioanalytical tool for glycan analysis, where IMS not only separates precursor glycans [33], but multistage IMS (IMSⁿ) is used to separate and IR-fingerprint isomeric fragment ions [34].
Instrumentation used for combined ion mobility/ion spectroscopy experiments comprises home-built and commercial IMS-MS platforms, adapted to allow for optical access of a laser beam to the ion cloud [29,31,32,[35], [36], [37], [38], [39], [40]]. IMS-MS combined with IRIS has been demonstrated on instruments that employ drift tube IMS (DTIMS) [29], FAIMS [31,32] or travelling-wave IMS (TWIMS), including structures for lossless ion manipulation (SLIM) designs, which enable cyclic and multistage IMS approaches providing high-resolution separation [34,35]. Photodissociation is performed using tunable IR lasers, either ‘on the fly’ in ion guides [29], e.g. in QTOF instruments, or in various types of ion traps [31,32], which can also be cooled to cryogenic temperature to facilitate messenger tagging spectroscopy [35].

Notably, mass analysers capable of multistage MSⁿ, such as FT-ICR and quadrupole ion trap mass spectrometers, have only sparsely been integrated into IMS-IRIS experiments [31]. In contrast, these platforms are frequently used for IRIS [2,5,11,13,[41], [42], [43], [44], [45], [46], [47], [48], [49]], as their ion storage capability allows one to easily adjust the duration of laser irradiation; moreover, their ability to also obtain IRIS spectra of MSⁿ product ions makes these platforms very valuable in mechanistic ion chemistry studies [9,[50], [51], [52], [53], [54], [55], [56], [57], [58], [59]]. In addition to their ultra-high mass resolution, FT-ICR instruments have the advantage that virtually no collisional cooling occurs, which makes these platforms ideal for ion spectroscopy approaches based on IR multiple-photon dissociation (IRMPD)...

Owing to its high sensitivity and resolution, mass spectrometry (MS) has become indispensable in the detection and identification of molecular species in complex mixtures, e.g., in metabolomics and many other small-molecule applications1. Modern MS instruments detect and resolve thousands of molecular compounds in a sample in a matter of just seconds. However, most features detected in untargeted MS analyses are not identified in terms of their full molecular structure. While the resolving power is usually sufficient to assign a unique chemical formula to the detected ions, the mass value alone gives little information on the arrangement of the atoms within the molecule. Consequently, definitive identification of the molecular structure remains challenging as it requires structural and stereoisomers corresponding to the same mass-to-charge ratio (m/z) to be distinguished.

Tandem mass spectrometry (MS/MS = MS2) is commonly used to advance structure annotation beyond the chemical formula. Collision-induced dissociation (CID) of a precursor ion selected in the first MS stage produces a structurally diagnostic fragmentation pattern in the second MS stage. To delineate a molecular structure for the precursor ion, this fragmentation pattern is compared against the entries in MS/MS spectral libraries2. Many application-specific libraries exist2,3,4, e.g. for metabolites, agrochemicals, toxicological substances, drug compounds, etc. However, even taken together, MS/MS reference libraries cover only a minute fraction of chemical space, estimated as perhaps 1% or so3,5,6,7. For de novo identifications beyond these ‘known unknowns’, in silico strategies to identify structures have been developed. High-level quantum-chemical computation of MS/MS spectra8 is developing, but far too costly to screen large numbers of candidate structures. Much faster methods originally relied mostly on rule-based9,10 and combinatorial9,11 fragmentation approaches7, while more recently, these heuristic models are being updated with strategies involving elements of machine learning12,13,14,15,16.

The combinatorial approach (Fig. 1) to predict MS/MS spectra is still widely in use and underlies some of the more recent machine-learning strategies13,14. A large compound database is screened for entries with the accurate mass (MS1) of the unknown query molecule. For each hit, a quick rule-based algorithm first determines possible small neutral losses (H2O, NH3, etc.), which are common in MS/MS and often occur early on in the CID breakdown cascade. A list of possible fragment m/z values is then generated in silico by breaking each bond in the molecule (excluding X-H bonds). The resulting combinatorial fragment m/z’s are fed back into the in silico fragmenter for a second (and third) round of fragmentation, forming so-called fragmentation trees (not to be confused with experimental MSⁿ spectral trees). Each type of chemical bond represents a preset bond dissociation energy, so that the probability of generating a specific m/z fragment—and hence its relative intensity in the MS/MS spectrum—can be quantified from the summed bond dissociation energies along the fragmentation tree ending up at that m/z. The resulting in silico MS/MS spectrum is then matched against the query MS/MS spectrum...

...A caveat involves the neglect of possible structural rearrangement occurring upon bond dissociation7,10. Rearrangement constitutes the replacement of broken bonds with new ones. In the in silico fragmenter, rearrangement would thus lead to a different set of combinatorial fragments being generated upon a next round of fragmentation. Neglecting rearrangement, as indicated by the question marks in Fig. 1, may thus lead to fragment ion m/z values in the in silico MS/MS spectrum that are incorrect...

... Commonly used in silico fragmenters include MetFrag, which generates MS/MS spectra for the METLIN (https://metlin.scripps.edu) database17,18 and Competitive Fragmentation Modeling (CFM-ID)9,15 which provides spectra for the Human Metabolome Database (HMDB) (https://hmdb.ca)19,20,21,22,23. mzCloud implements heuristic approaches by combining general fragmentation rules with fragmentation mechanisms published and (partly) relies on manual evaluation24. mzCloud (https://www.mzcloud.org) also uses ab initio and density functional theory (DFT) quantum-chemical calculations, assigning the lowest-energy isomer found as the fragment ion structure and thus ignoring the kinetic aspects of the CID reaction.

The question that arises is whether the annotated structures of the MS/MS fragment ions (that are often included in in silico MS/MS libraries) are indeed correct. Even though these structural annotations are not used directly in MS/MS spectral analysis, the in silico MS/MS spectra that are compared with the experiment are derived from the combinatorial algorithm that involves these structures. An estimate of the reliability of these structures is, therefore, an indirect measure for the reliability of predicted MS/MS spectra. Incorrect structures are expected to degrade the quality of the predicted MS/MS spectrum, especially in the lower mass range, where ions occur that are formed upon cleavage of more than one bond. Deviations are, therefore also expected to be more severe at higher-energy CID settings...