Predicting Collisioninduced Dissociation Spectra

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

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 those different molecular geometry,[19] so has the potential to give ions. DFT is not routinely used to explain CID product ion sufficiently accurate estimates of bond elongation for this mass spectra, however, because the amount of computational application, especially as the absolute values are not required. resource required, both in terms of time and computer AM1 calculations run in seconds rather than the hours specification and in the computational chemistry expertise required for DFT; for example, geometry optimisation of required, limits its accessibility to the mass spectrometrist. indole takes 5 s by AM1 but more than 1.5 h by DFT The time taken to calculate the geometry of a single, drug-like (B3LYP 6.31G).[26] If AM1 were found to give adequate molecule can be anything from minutes to days depending on estimates of bond lengths, this would extend the potential the size and flexibility of the molecule. for this application of quantum computational chemistry to Calculating low-energy geometries and electronic struc- mass spectral interpretation as both the speed and the lack tures with semi-empirical methods is considerably faster than of requirement for specialist computational resource offer with DFT, and can be undertaken on any desktop PC of the possibility of routine desktop use by non-expert users. reasonable specification using widely available commercial and academic software. Semi-empirical methods were [12,13] championed by Dewar in the 1950s to 1970s, when EXPERIMENTAL computers were still severely limited by processor speed and memory creating a real need for an approach which Chemicals would allow computational chemistry calculations to be undertaken on realistic time scales. HPLC grade methanol and acetonitrile were supplied by Semi-empirical methods are used to calculate heats of Rathburn (Walkerburn, UK). HPLC grade water was obtained formation, geometry, dipole moment and ionisation energy from VWR International Ltd (West Chester, PA, USA). Formic [14]

1128 as well as chemical reactivity. They give similar results to acid (99% +) was supplied by Biosolve (Valkenswaard, DFT for calculating bond dissociation energies but they tend The Netherlands).

CEN025-014 was donated by Cyclofluidic (Welwyn Garden Assignment of product ions City, UK). Sildenafil, doxazosin, ziprasidone and dofetilide Product ions of greater than approximately 8% abundance were donated by Pfizer Ltd (Sandwich, UK). All other were structurally assigned. All percentages in the assignment compounds were obtained from Sigma-Aldrich (Poole, UK). tables are relative to the most abundant product ion (this may The structures of all 15 authentic standard compounds are not be the base peak where there is a considerable amount of shown in Fig. 1. unfragmented precursor ion). All compounds were prepared at 1 mg/mL in acetonitrile/ water (between 10 and 100% acetonitrile depending on the compound solubility), then diluted with 50:50 (v/v) Criteria for prediction of bond cleavage fi methanol/water 0.1% formic acid to give a nal concentration Previous investigations into the fragmentation of protonated μ of 20 g/mL. dofetilide and four of its analogues[1] indicated that only bonds which elongated significantly as a result of protonation on a heteroatom at the site of cleavage were observed to LC/MS cleave. Therefore, only bond elongations of >0.039 Ǻ as a LC/MS data were acquired on a Synapt G1 Q-TOF result of protonation on one of the atoms to which the bond (quadrupole-time-of-flight) mass spectrometer (Waters was connected would be considered as predictive of bond Corporation) in ESI positive ion and V mode (resolution cleavage in this study. 15 000 FWHM), calibrated with sodium formate. Leucine enkephalin (MH+556.277) was infused at 5 μL/min as the reference lock mass. Samples (10 μL; 20 μg/mL)) were RESULTS AND DISCUSSION introduced via flow injection (0.5 mL/min 50:50 methanol/ water 1% formic acid; no HPLC column). Methanol was Comparison of AM1 and DFT for calculating bond length chosen as the modifier as it has a lower proton affinity than changes the other common modifier, acetonitrile, potentially enhancing Changes in the bond lengths of the 15 compounds (Fig. 1) protonation of the analyte. resulting from protonation on all heteroatoms were calculated The following instrumental conditions were applied: using both AM1 and DFT (6-31G** basis set). This represents capillary voltage 5 kV; extraction cone voltage 5 V; sampling the calculation of 4147 bond length changes by each cone voltage 35 V; transfer collision energy 5 eV; cone gas (N ) 2 computational method. Using all these data points (bond flow rate 150 L/h; desolvation gas (N ) flow rate 1800 L/h; 2 elongation, contraction and unchanged) a correlation of 0.87 source temperature 100°C; desolvation temperature 500°C; (R2 = 0.76) was found between these two methods (using the trap collision energy 25 to 35 eV (set on a compound by ’Correl’ function in Microsoft Excel 2013). Therefore, there is a compound basis to obtain a spread of product ions). Argon statistically significant correlation between the two calculation was used as the collision gas. methods. This correlation is even greater if only the significant The data acquisition settings were as follows: scan range bond length increases (>0.039 Ǻ; n = 123) were compared; the m/z 50 to 700; scan time 1 s, data centroid. correlation was increased to 0.96 (R2 = 0.88) as shown in Fig. 2. Most importantly, the predictions made as to which bonds are Computational modelling likely to cleave were the same based on data generated by either method for all 15 compounds. Considering one of the All quantum calculations were undertaken using Spartan’10 compounds as an example, for 1-methyl-2-pyrrolidinol (Wavefunction, Inc., Irvine, CA, USA). Structures were drawn (Table 1), the same bonds were calculated to elongate by in ACD Labs Chemsketch (freeware for academic or personal >0.039 Ǻ by both computational methods and these were the use), saved as .skc files and then opened in Spartan. The bonds which were observed to cleave during CID (spectrum starting geometry was obtained using molecular mechanics shown in Fig. 3 and product ion assignments in Table 2). It is MMFF minimum energy geometry optimisation. notable that considering the data in Table 2 the product ions All compounds were geometry optimised after protonation were formed from two different precursor ions, cations 1 and at all heteroatoms using both DFT 6.31G** and AM1 with the 2 (Table 1), and were not derived from a single charged a following preferences: maximum ligand distance 2.00Ǻ ; molecular species. polar area range 1000 kJ/mol; accessible area radius 1.000; AM1 is far less demanding than DFT both in terms of speed ’converge’ was selected. of calculation and in the computational resource required. All calculations were undertaken with the explicit hydrogens AM1 calculations typically took less than 30 s, whereas the (i.e. all hydrogens shown) on the molecule. All calculations DFTcalculations took between 15 min and 9 h. This means that were undertaken locally on a desktop computer of specification these bond length calculations may be undertaken routinely by the Intel® i7-3370 k CPU @ 3.50Hz, 16GB RAM, 64bit. mass spectrometrist as an aid to mass spectral interpretation without recourse to specialist computational resource. The speed of calculations also offers the potential for bond length calculations aCertain structures were shown to form internal hydrogen to be incorporated into commercial mass spectral interpretation bonds when modelled using an initial maximum ligand distance of 3.60Ǻ, the default setting. Internal hydrogen software packages to improve the accuracy of the predictions. bonding appeared to distort the bond lengths locally and Therefore, on the basis of the calculated bond lengths for fi lead to erroneous bond calculations. Therefore, hydrogen these 15 compounds, AM1 has been shown to be t-for-

bonding was eliminated from the model by reducing the purpose and only AM1-calculated bond length changes will 1129 maximum ligand distance to 2.00 Ǻ for all calculations. be reported and discussed in the rest of this manuscript.

Figure 1. The structures of the 15 compounds analysed in this study. The structures are annotated to ﬂag the most basic centre(s) in the gas* and liquid# phases. The potential sites of protonation modelled are labelled as cations C1 to C10.

Comparison of AM1 and DFT for calculating relative being the most basic. The energies did not always give the gas-phase basicities same order for gas-phase basicities when calculated by The relative gas-phase basicities were determined by AM1 and DFT. An example is shown for ziprasidone in

1130 calculating the global energy minimum for each protonated Table 3. This difference in basicity order may result from form of all the molecules, with the ion of the lowest energy the higher degree of error associated with the energies

Figure 2. Comparison of bond elongations (>0.039 Å; n = 123) ’ calculated by both AM1 and DFT using Spartan 10. Figure 3. CID product ion spectrum of protonated 1-methyl- 2-pyrrolidinol ([MH] + 102). determined by AM1; AM1 tends to only be accurate to only 3–4 kcal/mol.[22] Therefore, for ions with energy minima which differ by smaller values, the order calculated by AM1 is less reliable than that calculated by DFT. Therefore, the only parameter required to predict bond cleavages for in any further discussions in this manuscript only the DFT- these 15 compounds and therefore, in practice, only AM1 calculated relative energies (not bond lengths) will be calculations are required for predicting or rationalising referred to. In this study the relative gas-phase basicities were bond cleavage. calculated to elucidate the mechanism of bond cleavage in terms of the possibility of proton migration occurring Prediction of bond cleavage on basis of calculated bond subsequent to ionisation. Calculation of gas-phase basicities elongations was not required for the prediction of bond cleavages as [27] proton migration to the ’dissociative site’ appears to For the 15 compounds considered in this study 98 distinct [1] initiate cleavage and so knowledge of the initial site of bonds were observed to cleave in the CID product ion ionisation was not required. Bond length elongation was spectra. Of these, cleavage of 82 bonds was correctly

Table 1. Comparison of bond length changes resulting from protonation of 1-methyl-2-pyrrolidinol at the sites speciﬁed as C1 and C2, calculated using both DFT and AM1 (Spartan’10). Bond elongation calculated to be greater than 0.039 Ǻ as a result of protonation on one of the bonding atoms are highlighted in the table 1131

Table 2. The proposed product ion assignments for the ions in the CID spectrum of protonated 1-methyl-2-pyrrolidinol

Proposed ion Relative formula and intensity Experimental calculated Error (%) m/z accurate mass (ppm) Proposed structure(s) of ion Bond breaking

100 84.0817 C5H10N 84.0813 4.8 C4,O6

N5, C7 and/or N5,C3 and/or N5,C4 and/or 20 56.0506 C3H6N 56.0500 10.0 C2,C4 and/or C1,C2 and/or C1,C3

Table 3. The relative energies of the different protonated forms of ziprasidone calculated by using both AM1 and DFT. The energy values are normalised to the most stable cation

Energy difference Energy difference between most stable between most stable À À À À E (kcal mol 1) cation and others (kcal mol 1) E (kcal/mol 1) cation and others (kcal mol 1)

AM1 DFT 6.31G** 60.7 Neutral n/a À1233397 Neutral n/a 204.3 Cation 3 0 À1233643 Cation 4 0 204.3 Cation 4 0 À1233639 Cation 2 4 205.4 Cation 2 1 À1233634 Cation 3 9 223.5 Cation 1 19 À1233605 Cation 6 26 228.2 Cation 6 24 À1233608 Cation 8 36 228.9 Cation 8 25 À1233602 Cation 1 39 229.8 Cation 5 25 À1233602 Cation 5 41 267.2 Cation 7 63 À1233565 Cation 7 78

predicted on the basis of bond elongation alone. This the carbonyl group to centre the charge on C2 did not represents an overall success rate of 84%. Only carbon– result in a signiﬁcant elongation of the C2–C28 bond. This carbon bond cleavage was not predicted (n = 16). As was also the case for the other compounds which mentioned above, this may be due to the thermodynamic underwent carbon–carbon bond cleavage (cortisone, 5-(p- factors and to product ion stability, which is currently not methyl)phenylhydantoin, reserpine, trichlormethazide and considered in this approach. ziprasidone); modelling the appropriate carbocations If polarised bonds only are considered, the success rate for rather than locating the proton on a heteroatom did not predicting bond cleavage at a heteroatom is 100% (n = 82). predict the bond cleavage. Table 4 summarises the results of this ’blind trial’. It may be that bond weakening (via lengthening) The results obtained for ziprasidone (spectrum shown in around heteroatoms results from an increase in polarity Fig. 4; bond length changes in Table 5 and product ion of the bond by the addition of proton to the most assignments in Table 6) are typical for all 15 compounds. electronegative atom. This is consistent with the Bond Of the 10 bonds which were observed to cleave, 9 Activation Rule (BAR) proposed by Alcami;[7,28,29] were correctly predicted. The bond which was not the presence of the proton on the electronegative centre predicted to cleave was a carbon–carbon bond (C2–C28). pulls the bonding electrons toward the charged centre, Initial modelling was performed on cations C1 to C7. The reducing the electron density in the bonding region, with carbocation C8 (structure shown in Fig. 5) was modelled cleavage occurring if there is sufﬁcient difference in – 1132 in retrospect in order to try to rationalise cleavage of C2 electronegativity between the basic centre and the atom C28. Addition of a proton across the double bond in bonded to it.

Table 4. Overall summary of the effectiveness of using calculated bond elongations to predict bond cleavages during CID fragmentation

Number of bond cleavages

Observed to cleave Type of bond Correctly but not predicted Percentage accuracy cleavage not Compound predicted by bond elongation of prediction correctly predicted

1-Methyl-2-pyrrolidinol 4 0 100% n/a Sulphride 3 0 100% n/a Ziprasidone 9 1 90% C-C bond Ephredine 3 0 100% n/a Doxazosin 9 0 100% n/a CEN024-014 3 0 100% n/a Trichlormethazide 10 1 90% C-C bond Reserpine 9 2 82% C-C bond 5-(p-Methylphenyl)-5-phenylhydantoin 2 2 50% C-C bond 1,1-Dimethyl biguanide 5 0 100% n/a Amlodipine 8 0 100% n/a Cortisone 2 6 25% C-C bond Desipramine 3 0 100% n/a Sildenaﬁl 4 0 100% n/a Trimethaprim 6 0 100% n/a Total 80 12 87% n/a

formed in the source.[31] Komaromi et al. observed that N-acetyl-O-methoxyproline exhibits two distinct fragmentation pathways indicative of the coexistence of several protonated forms.[32] Komaromi used appropriate DFT calculations to support these observations. As the carbon–carbon bonds are not (or are less) polarised, addition of the proton may have a limited effect. In particular, there is less ’incentive’ for the proton to remain associated with a particular carbon within an unpolarised bond and it may move along the molecule, possibly via hydride shifts. Therefore, formation of a carbocation has less effect on the polarity of an individual carbon–carbon bond. The carbon–carbon bond cleavage may occur via an alternative mechanism to protonation-induced bond weakening. Figure 4. CID product ion spectrum of protonated ziprasidone A study of the fragmentation of sulphur–sulphur bond ([MH] + 413). containing heterocycles suggests that cleavage of sulphur–sulphur bonds was driven by the stability of the product ion formed.[33] Carbon–carbon bond cleavage may be analogous to this, especially as all the product The product ions of ziprasidone were formed from seven ions formed via carbon–carbon bond cleavage in different [M+H]+ precursor ions, cations 1 to 7, and were thus this investigation (i.e. for cortisone, 5-(p-methyl) not derived from a single species. This was observed to be phenylhydantoin and ziprasidone) showed increased true of all 15 compounds in that their product ions were conjugation and/or increase in planar geometry relative to derived from several protonated precursors, and further the precursor [M+H]+ ions. These proposed product ions are exemplified by the data for 1-methyl-2-pyrrolidinol shown shown in Table 6 for ziprasidone and Table 7 for the other in Table 2. This is consistent with previous studies which compounds. highlighted that the precursor ions appear to be a mixture This bond elongation approach over-predicted bond of ions which are protonated on a number of different basic cleavage by 33% (i.e. 32 bonds were predicted to cleave but sites across the molecule.[1,10,11] Other groups have also were not observed to do so). This is a significantly improved reported that precursor ions appear to be a mixture of ions over-prediction rate compared with many commercial protonated at different positions. Two isobaric ions observed packages. In one example, Waters Mass Fragment was in an MS/MS spectrum could only be assigned if they are observed to over predict by 400% based on accurate mass derived from precursor ions protonated at different sites, data and by over 2000% for nominal mass data.[1] Basing [30] giving rise to different product ions. Kaufmann reported fragmentation predictions on bond lengthening has the 1133 that a mixture of singly charged difloxacin species were advantage that the predictions are entirely in silico, based on

Table 5. Changes in bond length in ziprasidone resulting from protonation of ziprasidone at the sites speciﬁed as C1 to C8, calculated using AM1 (Spartan’10). Bond elongation calculated to be greater than 0.039 Ǻ as a result of protonation on one of the bonding atoms are highlighted in the table

the inherent properties of the molecule itself and will give the that cleavage of multiple bonds to the same atom was not same predictions for nominal mass data as for accurate mass favoured. In addition, there was a tendency for the most data. extended bond to cleave preferentially, but there were From the summary shown in Table 8 it appears that there were sufﬁcient exceptions for this not to be considered as a ’rule’. certain classes of bond which were prone to over-prediction: • Protonation of nitrogens (n = 6; or 19% of the incorrectly predicted bond cleavages) within a conjugated system • Over half (56%; n = 18) of the bonds incorrectly predicted to was predicted to initiate cleavage, but did not do so. This cleave were to an atom to which one of the other bonds was may be due to stabilisation by delocalisation of the charge observed to cleave. The bonds that did cleave were elongated across the conjugated system, resulting in the charge not to a signiﬁcantly greater extent (30–600%) than the bonds to being associated with a single centre. Because the charge

1134 the same atom which did not cleave in eight of the 18 cases is delocalised, the proton will have less effect on the (i.e. 25% of the total over-predicted cleavages). This suggests polarity, and hence the strength, of any single bond.

Table 6. The proposed product ion assignments for the ions in the CID spectrum of protonated ziprasidone

Proposed ion formula and Relative Experimental calculated Error Proposed structure(s) intensity m/z accurate mass (ppm) of ion Bond breaking

5% 220.0924 C11H14N3S7 N13,C12 220.0908

100% 194.0331 C10H9ClNO 22 N13,C12 194.0373

< 23% 177.0487 C9H9N2S 1 N13,C25 N13, 177.0486 C14 N16, C15 N16,C26

20% 166.0427 C9H9ClN 2 N13,C12 C2, 166.0424 C28 C2,N3

25% 159.0678 C10H9NO 4 N13,C12 C8, 159.0684 Cl19

S19,C20 N18,

8% 131.0738 C9H9N2 C17 N16, 131.0735 C25 N16, C13 N13, C14 N13,C25

• Although some sulphonamide cleavage was observed there was a tendency to over-predict the cleavage of all the bonds within sulphonamide groups (n = 4 or 13%). This may be because sulphonamide bonds are ﬂexible[34] and able to absorb conformational change, and are also able

Figure 5. Structure of ziprasidone carbocation modelled by to delocalise the charge across the sulphonamide group 1135 AM1 for which delta bond length data are reported in Table 5. such that it is not strongly associated with any single atom.

Table 7. The proposed assignments of product ions resulting from carbon-carbon bond cleavage for trichlormethazide, reserpine, 5-(p-methyl)-phenylhyndantoin and cortisone

Proposed ion formula and Relative Experimental calculated Error Proposed structure(s) intensity m/z accurate mass (ppm) of ion Bond cleaved Compound

40% 448.1190 C23H30NO8 4 C10,C12 oyih 04Jh ie os Ltd. Sons, & Wiley John 2014 © Copyright 448.1971 C24,C23

10% 336.1573 C18H24O6 8 C47,C5 336.1600 C5,C7 C26,N25 C9,C10

20% 236.1268 C H NO 8 C22,C21 ai omn asSpectrom. Mass Commun. Rapid 13 18 3 236.1287 N13,C14 O32,C33 .Wih,A lxadF Pullen F. and Alex A. Wright, P.

50% 174.0933 C11 H11 NO 8 C10,C12 174.0919 C24,C23 2014 (Continues) , 28 1127 , – 1143 rdcigCDsetawt eieprclcalculations semi-empirical with spectra CID Predicting ai omn asSpectrom. Mass Commun. Rapid

Table 7. (Continued)

Proposed ion formula and Relative Experimental calculated Error Proposed structure(s) intensity m/z accurate mass (ppm) of ion Bond cleaved Compound 2014 , 28 1127 , 8% 239.1200 C15 H15 N2O7 N13,C8 239.1184 C8,N9 – 1143 oyih 04Jh ie os Ltd. Sons, & Wiley John 2014 © Copyright

100% 196.1108 C14 H14N10 C12,C10 196.1126 N13,C8

8% 104.0505 C7H6N5 C12,C14 104.0500 C12,C10 N13,C8

10% 183.9639 C4H6Cl2N2S5 S13,O15 wileyonlinelibrary.com/journal/rcm 183.9629 S13,O14 C16,C17 C9,C8

(Continues) 1137 1138

wileyonlinelibrary.com/journal/rcm Table 7. (Continued)

Proposed ion formula and Relative Experimental calculated Error Proposed structure(s) intensity m/z accurate mass (ppm) of ion Bond cleaved Compound

15% 258.1617 C17H22O2 1 C7,C5 258.1620 C28,C5 oyih 04Jh ie os Ltd. Sons, & Wiley John 2014 © Copyright 10% 241.1597 C17H21 O2 C7,C8 241.1592 C28,C5 C2,O1

100% 163.1119 C11 H15 O2 C23,C25 163.1123 C9,C8 C5,C28

30% 145.1022 C11 H13 3 C23,C25 145.1017 C9,C11 C17,O18 ai omn asSpectrom. Mass Commun. Rapid

25% 121.0660 C8H9O5 C23,C21 121.0653 C13,C14 .Wih,A lxadF Pullen F. and Alex A. Wright, P.

30% 105.0708 C8H9 4 C23,C21 105.0708 C14,C13 2014 C17,O18 , 28 1127 ,

15% 93.0700 C7H9 5 C23,C21 –

1143 93.0704 C14,C13 C17,O18 Predicting CID spectra with semi-empirical calculations

Table 8. Summary of over prediction of bond cleavage on the basis of proton-induced bond elongation

Number of bonds predicted to cleave but did not break Compound (i.e. over-predicted) Type of bond over predicted

1-Methyl-2-pyrrolidinol 0 n/a Sulphride 5 Centred around sulphur (n = 4) ; N18-C21: other bond to N18 broke in preference (i.e. N18,C19)(n = 1) Ziprasidone 1 N3-C4: other bond to N3 broke in preference (i.e. N3-C2 elongates twice as much) (n = 1) Ephredine 0 n/a Doxazosin 3 C-O (aliphatic; n = 3): other bond to same oxygen breaks in preference CEN024-014 4 C26-N25 and C31-N25: other bond to N25 broke in preference (C24-N25) (n = 2) C18-N19: other bond to N19 broke in preference (C20-N19 elongates twice as much) (n = 1) C5-N4: other bond to N4 broke in preference (C3-N4 elongates half as much) (n = 1) Trichlormethazide 0 n/a Reserpine 6 N25-C10: other bond to N25 broke in preference (C24-N25)(n = 2); C-O (aliphatic; n = 6), other bond to same oxygen broke in preference 5-(p-Methylphenyl)-5-phenylhydantoin 2 N13-C12 other bond to N13 broke in preference (C8-N13 elongates by 50% more ) (n = 1) 1,1-Dimethyl biguanide 1 C4-N5 other bond to C4 broke in preference (C4-N2 elongates by 30% more much) (n = 1) Amlodipine 0 n/a Cortisone 0 n/a Desipramine 2 N15 to aromatic carbons (n = 2); two bonds need be broken to generate leaving group Sildenaﬁl 8 N4-C5 other bond to 029 broke in preference (N4-S2 elongates by 600% more) (n = 1); C28-O29 other bond to 029 broke in preference (C30-O29 elongates by 100% more) (n = 1); Bonds to N23 (n = 3)* ; Bonds to N20 (n = 2)* ; N14,C14 (n = 1)* *All in extended congugated systems Trimethaprim 0 n/a

Product ion intensity Overall, no correlation was observed between the extent fl of bond lengthening and the intensity of the product ion. A summary of factors that may in uence product ion abundance In a previous study using bond length changes to predict is shown in Table 9. There appears to be no correlation between the CID fragmentation of protonated dofetilide, there was the basicity of the molecule in water (pKa) and formation of the a quantitative relationship between the extent of bond fi major product ion; for only ve of the 15 compounds was the elongation and product ion intensity.[1] However, this major product ion formed by protonation at the most basic larger study shows that although product ion intensity fi centre in solution. For ve compounds, none of the product ions may be predictable on the basis of bond lengthening for resulted from protonation at the most basic centre in solution. certain compounds, it is not valid to apply this approach fi For all these ve compounds, however, the most basic atom in indiscriminately. Other research groups have successfully the gas phase was part of a conjugated system which could applied DFT to predict ion intensities for peptides[35] and delocalise the charge. This both stabilises the precursor ion, quinazolines[36] by calculating product ion energies. Thus, hence reducing the propensity for fragmentation, and means it appears that bond weakening may dictate which that the charge is not associated with any particular bond. The polarised bonds cleave, but it may the relative product results of this study to date indicate that the charge has to be ion stability which determines the relative intensity of centredononeoftheatomsinthebondtoinitiatecleavage. the product ions formed as a result of these bond Potentially, the number of bonds cleaved to form an ion cleavages. may reflect in its relative abundance in that more energy is There was no obvious correlation between the pKa of the required to cleave multiple bonds. However, the data in molecule and the appearance of spectra in terms of the Table 9 show that the intensity of the product ions does not number and abundance of the ions. For example, appear to depend on the number of bonds broken during its trichlormethazide, which has no basic centre, produces six formation. For only six of the 15 compounds was the major product ions, four of which are major (greater than 30%).

product ion formed by single bond cleavage; the other nine Amlodipine contains a primary amino group of pKa 9.5 and 1139 resulted from multiple bond cleavages. gives a product ion spectrum containing eight ions, four of

which are major. Similarly, there was no correlation with the gas-phase basicities of the protonation sites and the type or intensity of product ion.

Proton migration Proton migration to a thermodynamically less favourable site ve ions in total ve ions in total ﬁ ﬁ has been proposed to be required to rationalise the formation

of spectra of certain product ions. For example, the loss of ammonia Appearance from peptidic amides required protonation on the nitrogen although the oxygen is both the most energetically favoured protonation site and the observed initial protonation – site.[37 39] Penicillin shows cleavage of the β-lactam bond after transfer of the proton from the carbonyl to the lactam nitrogen.[40,41] In addition, dibenzyl ethers,[27] the pharmaceutical compound maraviroc,[10] dialkylphosphoric acid esters[42] and thiourea/urea compounds[43] have all been reported to exhibit product ions in their mass spectra generated following proton migration from the most thermodynamically likely site. It has been proposed that the energy for proton migration At least one fteen compounds. The pKa was calculated using Marvin is obtained by the transfer of kinetic to internal energy most basic site? product ion due ﬁ to protonation at during ion molecule collisions,[21] probably in the collision cell during CID. The results of this larger scale study support the hypothesis

ions of all that proton migration from the initial site of ionisation to the [1,27] + ’dissociative’ site may be required to initiate fragmentation: solution phase and same in gas • Most basic sites As expected, the greatest effect in terms of conformational change, and hence bond length changes, occurred in the immediate area around the protonation site. In a few cases, however, protonation did result in signiﬁcant bond 3.2 Yes Yes Two major ions; seven ions in total 9.0 Yes No Only one major ion 4.1 No Yes Four major ion; six ions in total À À À elongation remote from the protonation site (shown in

basic site Table 10). None of these bond length changes gave rise to (aqueous)? pKa of most the product ions observed in the CID spectra, reinforcing the proposal that the proton needed to be adjacent to the site of cleavage for fragmentation to occur. • The spectra of four compounds (doxazosin, reserpine, 1,1-dimethyl biguanide and sildenaﬁl) did not contain any product ions derived from precursors protonated at Major ion (aqueous)?

protonated at the most basic site (Table 9). As all spectra were obtained most basic centre via ESI, the original ionisation site is likely to be the centre with the highest pKa. Reviewing the literature for evidence of gas-phase protonation during ESI, in many cases direct gas-phase ionisation is proposed because it is difﬁcult to rationalise the product ions in the spectra of certain

cleavage? compounds by protonation on the most basic site in Major ion due to single bond solution. For example, it has been proposed that

and the gas phase basicities refer to the relative stabilities (global energy minima) of the protonated species gas-phase ionisation via ion-molecule reactions plays a major role in ESI, such as by proton transfer from gaseous [46] ammonium ions to analytes of higher proton afﬁnity.[44] However, for all these four compounds the most basic site is the same in both the solution and gas phase. Therefore, direct gas-phase ionisation at the less basic, dissociative site is thermodynamically unlikely. Thus, the proton needs to move from the site of greatest basicity (i.e. the initial ionisation site) to the dissociative site to initiate fragmentation. Observations around relative abundance of product ions in the CID spectra of the [M+H] l No No 6.0 Yes No Three major ions;

ﬁ This raises the possibility that some cases of charge-remote fragmentation[45] reported in the literature may, in fact, -Methylphenyl)-5-phenylhydantoin No No p

1140 represent proton migration followed by charge-directed Trimethaprim No No 7.2 Yes No Seven major ions; nine ions in total DesipramineSildena No Yes 10.0 No Yes Two major ions; four ions in total 1,1-Dimethyl biguanideAmlodipineCortisone No No No No No 12.6 No Yes 9.5 Yes No Multiple major ions: many basic centres? Yes Four major ions; eight ions in total Reserpine5-( Yes No 7.3 Yes Yes Four major ion; eight ions in total CEN024-014Trichlormethazide No Yes Yes Yes 8.4 No Yes One major ion; four ions in total 1-Methyl-2-pyrrolidinolSulphrideZiprasidoneEphredineDoxazosin Yes No Yes Yes Yes No 8.6 Yes Yes No No Yes 8.4 7.1 9.5 7.1 Yes Yes Yes Yes Yes One major ion; two ions in total Yes Yes Yes No Two One major major ions; ion; four four ions ions in in total total One major ion; three ions Three in major total ions; Table 9. Compound (ChemAxon, Budapest, Hungary) fragmentation.

Table 10. The bonds which were calculated to elongate by greater than 0.039 Å after protonation at a site other than on one of the bonding atoms

Calculated increase Site of protonation Observed to cleave Compound Bond in bond length (Ǻ) causing bond length increase during CID?

Sulphride S2-N4 0.335 N7 No S2-C5 0.040 O14 No C13-O14 0.103 O9 No Doxazosin C12-N16 0.052 O23 No Trichlormethazide S2-C5 0.070 N10 No S2-C5 0.049 N12 No S2-C5 0.490 O14 No S2-C5 0.815 O15 No Reserpine O1-C2 0.880 O49 No Sildenaﬁl S2-N4 0.043 N7 No C10-N4 0.057 O29 No S2-C11 0.092 O29 No S2-N4 0.078 O29 No N4-C5 0.067 O29 No

CONCLUSIONS • These structures were imported into a semi-empirical AM1 computational chemistry software package which was This study (15 compounds, 98 observed bond cleavages and used to calculate all the bond lengths in the neutral and over 8000 bond length calculations) has confirmed that protonated molecules. Internal hydrogen bonds may lead significant bond elongation (>0.039 Å) may be used as a to misleading results so these needed to be eliminated descriptor for cleavage of polarised bonds during CID by during modelling. flagging which bonds have been weakened as a result of • The bond lengths calculated for the protonated molecules structural changes due to protonation. This approach achieved were subtracted from the corresponding bonds in the 100% success rate in the prediction of polarised bond cleavage. neutral molecule to obtain the bond length changes. Moreover, it has been shown that the semi-empirical • Significant increases in bond length (in this case >0.039 Å), computational approach AM1 can be used to calculate these as a result of protonation on one of the atoms forming the bond length changes as it gives very similar results to those bond, may be considered predictive of polarised bond cleavage. obtained by DFT. Most studies to date applying computational chemistry to mass spectral data have used DFT, which is The behaviour of these 15 compounds is consistent with the computationally demanding both in terms of calculation time model of CID fragmentation that has been proposed in and in the computing power required. This has limited the previous publications: spread of the application by mass spectrometrists of computational chemistry to the prediction or rationalisation • Protonation caused conformational changes which of mass spectral fragmentation. The evidence that AM1 can resulted in bond length changes, which was accurately used to predict bond cleavage opens up this approach to many calculated using quantum chemistry based computational more scientists. AM1 calculations may take only seconds and software. be undertaken on a standard computer, rather than on an • Weakening of bonds is indicated by lengthening of bonds extremely high specification server which is often used for and significant bond length increases (>0.039 Å) weaken DFTcalculations. The speed of the AM1 calculations also offers the bond to such an extent that it is preferentially broken the potential for their incorporation into commercial software during CID. to improve the ’chemical sense’ of these packages and reduce • The proton had to be located on one of the atoms (the most the over-prediction of product ions. Over-prediction of bond electronegative) involved in the bond for cleavage to occur; cleavage was only 34% in this study, a significant improvement bond elongation remote from the protonation site did not over the over-prediction of product ion formation by many lead to bond cleavage. spectral interpretation software packages. • Protonation at the most basic sites (liquid and gas phase) Even without software packages tailored specifically for did not necessarily lead to bond cleavage. Therefore, for mass spectral interpretation, the procedure for using the some compounds, the proton appears to have migrated currently available computational chemistry packages as an from the primary site of protonation during ionisation to aid to predicting and/or explaining fragmentation of these a thermodynamically less stable site to initiate cleavage. 15 protonated molecules was found to be straightforward: • As the proton may migrate from the protonation site during ionisation, calculation of basicity (gas- and liquid-phase) was • Structures for each neutral molecule and the corresponding unnecessary for fragmentation predictions; the only required molecules protonated at all possible sites were generated. calculation to predict cleavage of a polarised bond was The basicity of the protonation sites did not need to be the bond length change. It is the protonation site which has

considered and so protonation at all heteroatoms needed the greatest effect on adjacent bond lengths rather than the centre 1141 to be modelled. at which ionisation occurs which is necessary for this approach.

• There was no single protonated molecular species; the [7] M. Alcamí, O. Mó, M. Yáñez. Computational chemistry: a product ions appear to be formed from a mixture of singly useful (sometimes mandatory) tool in mass spectrometry charged protonated precursors. studies. Mass Spectrom. Rev. 2001, 20, 195. [8] K. Burke. Perspective on density functional theory. J. Chem. 2012 Unpolarised, carbon–carbon bond cleavage apparently Phys. , 136, 150901. [9] A. Saeed, I. Arshad, U. Flörke. Synthesis, crystal structure, and cannot be predicted on the basis of bond length changes. ’ fl – DFT study of N -(2,4-dinitrophenyl)-2- uorobenzohydrazide. We therefore propose that C C bond cleavage is at least J. Chem. 2013. DOI: 10.1155/2013/648121. partly driven by the thermodynamic stability of the resulting [10] P. Wright, A. Alex, T. Nyaruwata, T. Parsons, F. Pullen. products rather than bond length changes during Using density functional theory to rationalise the mass protonation. In addition, product ion intensities did not spectral fragmentation of maraviroc and its metabolites. correlate with the extent of bond elongation and thus could Rapid Commun. Mass Spectrom. 2010, 24, 1025. not be predicted by bond elongation alone. For both [11]A.Alex,S.Harvey,T.Parsons,F.S.Pullen,P.Wright, unpolarised bond cleavage and product ion intensity, J.-A. Riley. Can density functional theory (DFT) be used calculation of product ion stability may be required for them as an aid to a deeper understanding of tandem mass to be rationalised. The relative stability of ions and spectrometric fragmentation pathways? Rapid Commun. fragmentation products can be predicted in principle using Mass Spectrom. 2009, 23, 2619. [12] M. J. S. Dewar, W. Thiel. A semi-empirical model for the both, DFT and AM1. Fortunately, cleavage of non-polar two-center repulsion integrals in the NDDO approximation. bonds is less common than cleavage of polarised bonds and Theor. Chim. Acta 1977, 46, 89. product ion intensity, although very useful in comparing [13] M. J. S. Dewar. A molecular orbital theory of organic spectra with library data, is not critical in interpreting mass chemistry. I. General principles. J. Am. Chem. Soc. 1952, 74, spectra. Therefore, both these limitations of the bond 3341. lengthening approach do not significantly restrict the [14] C. Cojocaru, A. Rotaru, V. Harabagiu, L. Sacaresc. Molecular application of AM1 as a tool for the interpretation of CID structure and electronic properties of pyridylindolizine mass spectra. derivative containing phenyl and phenacyl groups: Comparison between semi-empirical calculations and experimental studies. J. Mol. Struct. 2013, 1034,162. [15] T. R. Sharp. Calculated carbon-hydrogen bond dissociation Acknowledgements enthalpies for predicting oxidative susceptibility of drug substance molecules. Int. J. Pharm. 2011, 418, 304. The British Mass Spectrometry Society (BMSS) provided a [16] E. Klein, M. Matis, V. Lukeš, Z. Cibulková. The applicability small equipment grant for the purchase of the Spartan software of AM1 and PM3 semi-empirical methods for the study of used in this study. Funding for this study was supplied by the N–H bond dissociation enthalpies and ionisation potentials ’ fi Vice-Chancellor sOfce, University of Greenwich, UK. The of amine type antioxidants. Polym. Degrad. Stab. 2006, 91,270. compound CEN025-14 was supplied by Cyclofluidic [17] M. J. S. Dewar, E. G. Zoebisch, E. F. Healy, J. J. P. Stewart. Integrated Discovery (Welwyn Garden City, UK). Development and use of quantum mechanical molecular models. 76. AM1: a new general purpose quantum mechanical molecular model. J. Am. Chem. Soc. 1985, 107, 390.2. [18] C. Cojocaru, A. Rotaru, V. Harabagiu, L. Sacarescu. Molecular REFERENCES structure and electronic properties of pyridylindolizine derivative containing phenyl and phenacyl groups: [1] P. Wright, A. Alex, S. Harvey, T. Parsons, F. Pullen. Comparison between semi-empirical calculations and Understanding collision-induced dissociation of dofetilide: experimental studies. J. Mol. Struct. 2013, 1034. a case study in the application of density functional theory [19] G. Zheng, S. Irle, K. Morokuma. Performance of the DFTB as an aid to mass spectral interpretation. Analyst 2013, method in comparison to DFT and semi-empirical methods 138, 6861. for geometries and energies of C20–C86 fullerene isomers. ™ [2] Mass Frontier . Available: http://www.highchem.com/ Chem. Phys. Lett. 2005, 412, 210. massfrontier/mass-frontier.html. [20] S. Rajak, N. Islam, D. Ghosh. Modeling of the chemico- [3] ACD/MS Fragmenter. ACDLabs, Toronto, Canada. ACD/ physical process of protonation of molecules entailing some Labs.com. quantum chemical descriptors. J. Quantum Inf. Sci. 2011, 1,87. [4] M. Heinonen, A. Rantanen, T. Mielikäinen, M. Heinonen, [21] P. J. Amorim Madeir, P. D. Vaz, R. J. N. Bettencourt da Silva, A. Rantanen, T. Mielikäinen. FiD: a software for ab initio M. H. Florêncio. Can semi-empirical calculations help solve structural identification of product ions from tandem mass mass spectrometry problems? Protonation sites and proton spectrometric data. Rapid Commun. Mass Spectrom. 2008, affinities of amino acids. Chempluschem 2013, 78, 1149. 22, 3043. [22] P. E. M. Siegbahn. The performance of hybrid DFT for [5] S. Wolf, S. Schmidt, M. Müller-Hannemann, S. Neumann. In mechanisms involving transition metal complexes in silico fragmentation for computer assisted identification of enzymes. J. Biol. Inorg. Chem. 2006, 11, 695. metabolite mass spectra. BMC Bioinformatics 2010, 11, 148. [23] E. Kraka, D. Creme. Characterization of CF bonds with [6] H. Horai, M. Arita, S. Kanay, Y. Nihe, T. Ikeda, K. Suwa, multiple-bond character: bond lengths, stretching force Y. Ojima, K. Tanaka, S. Tanaka, K. Aoshima, Y. Oda, constants, and bond dissociation energies. Chemphyschem Y. Kakazu, M. Kusano, T. Tohge, F. Matsuda, Y. Sawada, 2009. 10, 686. M. Yokota Hirai, H. Nakanishi, K. Ikeda, N. Akimoto, [24] L. Pauling. The dependence of bond energy on bond length. T. Maoka, H. Takahashi,T. Ara, N. Sakurai,H. Suzuki, J. Phys. Chem. 1954, 58, 662. D. Shibata,S. Neumann, T. Iida, K. Tanaka, K. Funatsu, [25] L. Pauling. The Nature of the Chemical Bond and the Structure F. Matsuura, T. Soga, R. Taguchi, K. Saito, T. Nishioka. of Molecules and Crystals: An Introduction to Modern Structural

1142 MassBank: a public repository for sharing mass spectral Chemistry, (3rd edn.). Cornell University Press, Ithaca NY, data for life sciences. J. Mass Spectrom. 2010, 45, 703. 1960.

[26] C. K. Kim, S. G. Cho, C. K. Kim, M.-R. Kim, H. W. Lee. using density functional calculations. Rapid Commun. Mass Prediction of physicochemical properties of organic Spectrom. 2013, 27,964. molecules using semi-empirical methods. Bull. Korean Chem. [37] K. A. Cox, S. J. Gaskell, M. Morris, A. Whiting. Role of the Soc. 2013, 34, 1043. site of protonation in the low-energy decompositions of [27] Y.-P. Tu. Dissociative protonation and fragmentation: retro- gas-phase peptide ions. J.Am.Soc.MassSpectrom.1996, 7, Friedel–Crafts reactions of heterocyclic drug and metabolite 522. molecules in mass spectrometry. Int. J. Mass Spectrom. 2012, [38] L. R. Snyder, J. W. Dolan. High-Performance Gradient Elution: 316/318, 40. The Practical Application of the Linear-Solvent-Strength Model. [28] M. Alcamí, O. Mó, M. Yáñez, J.-L. M. Abboud, J. Elguero. John Wiley, New Jersey, 2007. Bond activation by protonation in the gas phase. Chem. [39] H.-Y. Lin, D. P. Ridge, E. Uggerud, T. Vulpius. Unimolecular Phys. Lett. 1990, 172, 471. chemistry of protonated formamide. mass spectrometry and [29] M. Alcamí, O. Mó, M. Yáñez. Modeling intrinsic basicities ab Initio quantum chemical calculations. J. Am. Chem. Soc. and acidities. J. Phys. Org. Chem. 2002, 15, 174. 1994, 116, 2996. [30] S. W. Holman, P. Wright, G. J. Langley. High-throughput [40] C. K. Fagerquist, A. R. Lightfield. Confirmatory analysis of approaches towards the definitive identification of beta-lactam antibiotics in kidney tissue by liquid pharmaceutical drug metabolites. 2. An example of how chromatography/electrospray ionization selective reaction unexpected dissociation behaviour could preclude correct monitoring ion trap tandem mass spectrometry. Rapid assignment of sites of metabolism. Rapid Commun. Mass Commun. Mass Spectrom. 2003, 17, 660. Spectrom. 2009, 23, 2017. [41] C. K. Fagerquist, A. R. Lightfield, S. J. Lehotay. [31] A. Kaufmann, P. Butcher, K. Maden, M. K. Widmer, Confirmatory and quantitative analysis of beta-lactam K. Giles, D. Uría. Are liquid chromatography/electrospray antibiotics in bovine kidney tissue by dispersive solid- tandem quadrupole fragmentation ratios unequivocal phase extraction and liquid chromatography-tandem mass confirmation criteria? Rapid Commun. Mass Spectrom. 2009, spectrometry. Anal. Chem. 2005, 77, 1473. 23, 985. [42] Y. Zeng, X. Chen , D. Zhao, H. Li, Y. Zhang, X. Xiao. [32] I. Komáromi, A. Somogyi, V. H. Wysocki. Proton migration Estimation of pKa values for carboxylic acids, alcohols, and its effect on the MS fragmentation of N-acetyl OMe phenols and amines using changes in the relative Gibbs free proline: MS/MS experiments and ab initio and density energy. Fluid Phase Equilib. 2012, 313, 148. functional calculations. Int. J. Mass Spectrom. 2005, 241, 315. [43] F. Falvo, L. Fiebig, F. Dreiocker, R. Wang, P. B. Armentrout, [33] J.-L. M. Abboud, M. Esseffar, M. Herreros, O. Mó, M. T. Molina, M. Schäfer. Fragmentation reactions of thiourea- and urea- R.Notario,M.Yáñez.ProtonationofS4,S6,andS8sulfur compounds examined by tandem MS-, energy-resolved cycles. A quantitative study. J. Phys. Chem. A 1998, 102, CID experiments, and theory. Int. J. Mass Spectrom. 2012, 7996. 330/332, 124. [34] C. Baldauf, R. Günther, H.-J. Hofmann. Conformational [44] R. Guevremont, K. W. M. Siu, J. C. Y. Blanc, S. S. Berman. properties of sulfonamido peptides. J. Mol. Struct. Are the electrospray mass spectra of proteins related to their THEOCHEM 2004, 675, 19. aqueous solution chemistry? J. Am. Soc. Mass Spectrom. 1992, [35] T. Pechan, S. R. Gwaltney. Calculations of relative intensities 3, 216. of fragment ions in the MSMS spectra of a doubly charged [45] C. Cheng, M. L. Gross. Applications and mechanisms of penta-peptide. BMC Bioinformatics 2012, 13, S13. charge-remote fragmentation. Mass Spectrom. Rev. 2000, 19, [36] A. Galezowska, M. W. Harrison, J. M. Herniman, 398. C.-K. Skylaris, G. J. Langley. A predictive science approach [46] G. Pirok, N. Máté, J. Varga, J. Szegezdi, M. Vargyas, to aid understanding of electrospray ionisation tandem mass S. Dóránt, F. Csizmadia. Making ’real’ molecules in virtual spectrometric fragmentation pathways of small molecules space. J. Chem. Inf. Model. 2006, 46, 563. 1143