Research Article Received: 7 December 2013 Revised: 7 February 2014 Accepted: 13 February 2014 Published online in Wiley Online Library Rapid Commun. Mass Spectrom. 2014, 28, 1127–1143 (wileyonlinelibrary.com) DOI: 10.1002/rcm.6870 Predicting collision-induced dissociation spectra: Semi-empirical calculations as a rapid and effective tool in software-aided mass spectral interpretation Patricia Wright1*, Alexander Alex2 and Frank Pullen1 1School of Science, University of Greenwich, Medway Campus, Chatham ME4 4TB, UK 2Evenor Consulting Ltd, The New Barn, Mill Lane, Eastry CT13 0JW, UK RATIONALE: Fifteen molecules were modelled using quantum chemistry, prior to interpreting their collision-induced dissociation (CID) product ion spectra, in a ’blind trial’ to establish if calculated protonation-induced bond elongation could be used to predict which bonds cleaved during CID. Bond elongation has the potential to be used as a descriptor predicting bond cleavage. METHODS: The 15 molecules were modelled with respect to protonation-induced bond length changes using Density Functional Theory (DFT). Significant bond elongations were highlighted to flag potential bond cleavages. CID product ion spectra, obtained using positive ion electrospray ionisation (Waters Synapt G1), were interpreted to establish if observed bond cleavages correlated with calculated bond elongations. Calculations were also undertaken using AM1 (Austin Model 1) to see if this rapid approach gave similar results to the computationally demanding DFT. RESULTS: The AM1-calculated bond elongations were found to be similar to those generated by DFT. All the polarised bonds observed to cleave (n = 82) had been calculated to elongate significantly. Protonation, possibly via proton migration, on the most electronegative atom in the bond appeared to initiate cleavage, leading to a 100% success rate in predicting the bonds that broke as a result of protonation on a heteroatom. Cleavage of carbon–carbon bonds was not predicted. CONCLUSIONS: Cleavage of the polarised bonds appears to result from protonation on the more electronegative atom of the bond, inducing conformational changes leading to bond weakening. AM1-calculated bond length changes act as a descriptor for predicting bond cleavage. However, the impetus for cleavage of the unpolarised bonds may be product ion stability rather than bond weakening. Copyright © 2014 John Wiley & Sons, Ltd. Mass spectrometry (MS) is not an established rule-based compound, so that they may predict product ions structures discipline in that the MS performance of compounds, both which are not chemically feasible. This is because the in terms of quantitative sensitivity and the qualitative predictions are made on the basis of applying rules (often fragmentation behaviour, is difficult to predict even by based on electron ionisation spectra rather than CID) practitioners with many years’ experience. This makes extrapolating from databases and/or applying a ’systemic interpretation of collision-induced dissociation (CID) product bond dissociation method’ in which all possible bonds in the ion spectra time-consuming, potentially rate-limiting, and molecule are cleaved and the mass of the remaining structure sometimes subjective. In addition, novice users can find mass calculated. Although the predictions made by these spectral interpretation challenging. commercial software packages are generally useful, the There are commercial software packages available to aid method described in this manuscript of predicting spectral interpretation but, in general, these have the limitation fragmentation by calculating bond elongation is able to narrow that they over-predict the number of product ions formed. down the number of possible choices significantly and For example, four major (>5%) product ions were observed therefore enables much faster and more efficient interpretation in the product ion spectrum of protonated dofetilide, but the of spectra. Waters’ Mass Fragment software (Waters Corporation, Examples of such commercially available packages include Manchester, UK) predicted 20 possible product ions on the Mass Frontier (Thermo Scientific, San Jose, CA, USA)[2] which basis of the accurate mass data and over 100 possible product combines comparison with their database, containing over ions for the nominal mass values.[1] The reason for this over- 30 000 fragmentation schemes from the literature and in- prediction is that many of the software packages lack any house data, with the application of general fragmentation/ insight into the specificchemicalstructureofagiven rearrangement rules. MS Fragmenter (ACDLabs, Toronto, Canada)[3] predicts product ions from the imported parent structure by applying rules of fragmentation. Fragment fi [4] * Correspondence to: P. A. Wright, School of Science, University iDenti cator (FiD) takes the alternative approach of of Greenwich, Medway Campus, Chatham ME4 4TB, UK. generating all the possible fragments that correspond to the 1127 E-mail: [email protected] accurate mass of the observed ions and then ranking in order Rapid Commun. Mass Spectrom. 2014, 28, 1127–1143 Copyright © 2014 John Wiley & Sons, Ltd. P. Wright, A. Alex and F. Pullen of how likely these fragments are to be formed. EPIC to over-estimate ionisation potentials.[15,16] Semi-empirical (Elucidation of Product Ion Connectivity)[5] and MetFrag[6] methods are particularly useful for large molecules where are both ’systemic bond dissociation’ methods. DFT calculations take too long. However, the increased speed These programmes assist with mass spectral interpretation of calculation with semi-empirical methods is offset by a via different approaches, but common to them all is the lower accuracy than DFT. limitation that predictions are made on the basis of One of the most popular semi-empirical methods is Austin assumptions or extrapolations which may not be valid. This Model 1 (AM1).[17] AM1 performs well in calculating bond results in the prediction of a large number of product ions lengths, being in good agreement with experimental data which are not in practice observed in the mass spectra. For (approximately 5% error);[18] however, relative energies of software to be truly effective it needs to make predictions molecules are calculated more accurately by DFT.[19] AM1 based mainly on the properties of the molecule itself without tends to overestimate basicity, having been shown to be recourse to assumptions. Quantum chemistry offers the somewhat less reliable for calculating proton affinities.[20,21] potential to improve the accuracy of in silico product ion DFT generates more accurate heats of formation than predictions as it describes the behaviour of matter at AM1.[22] molecular, atomic and sub-atomic levels. Quantum The authors have used DFT in previous studies with the computational chemistry has been applied in mass pharmaceutical compounds fluconazole, maraviroc and spectrometry[7] for many years, often used to calculate the dofetilide to rationalise CID product ion spectra in terms of energies of the precursor ions, the product ions and any bond weakening resulting from conformational changes.[1,10,11] intermediates as a way of determining both the most likely In general, with a few exceptions,[23] lengthening a bond will routes of product ion formation and which product ions are cause it to weaken and render it more susceptible to the most energetically favourable. The approach described cleavage.[24,25] These three publications reported that in this manuscript differs from the majority of these protonation-induced elongation of bonds did correspond to previously reported studies regarding the application of the bonds that were actually observed to cleave in the tandem computational chemistry to mass spectrometry in that it mass (MS/MS) spectra. focuses on bond length changes as a result of ionisation to In order to further test the hypothesis that bond cleavage identify the bonds which are likely to cleave. during CID may be predicted by quantum computational One of the most widely applied quantum chemistry chemistry on the basis that bonds which are calculated to approaches is Density Functional Theory (DFT)[8] which elongate significantly as a result of conformational changes calculates the electronic structure of a given molecule. DFT induced by protonation cleave preferentially during CID, 15 models molecules in the gas phase and so is very well suited pharmaceutical molecules in the mass range 101 to 608 amu for determining the behaviour of ions within a mass were modelled. Major bond elongations were highlighted to spectrometer. Molecular geometries predicted by DFT are flag potential bond cleavages. The CID mass spectra were known to be accurate as they agree closely with experimental then subsequently interpreted to establish if the predicted X-ray diffraction data.[9] DFT has been used to great effect to bond cleavages had actually occurred. This represented a rationalise fragmentation based on the thermodynamic ’blind trial’ of using bond elongation as a descriptor effects that protonation has on the molecule,[10,11] by predicting bond cleavage. calculating the thermodynamically most stable protonated Bond length calculations were undertaken using both DFT species based on the global minimum energy of the three- (basis set 6.31G**) and AM1. The parameterised approach of dimensional structure, and this information has been useful AM1 is generally accepted to give good approximations for in predicting the potential cleavage sites of