predicting collision-induced dissociation spectra: semi

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

Predicting collision-induced dissociation spectra: Semi-empiricalcalculations as a rapid and effective tool in software-aided massspectral interpretation

Patricia Wright1*, Alexander Alex2 and Frank Pullen1

1School of Science, University of Greenwich, Medway Campus, Chatham ME4 4TB, UK2Evenor Consulting Ltd, The New Barn, Mill Lane, Eastry CT13 0JW, UK

RATIONALE: Fifteen molecules were modelled using quantum chemistry, prior to interpreting their collision-induceddissociation (CID) product ion spectra, in a ’blind trial’ to establish if calculated protonation-induced bond elongationcould be used to predict which bonds cleaved during CID. Bond elongation has the potential to be used as a descriptorpredicting bond cleavage.METHODS: The 15 molecules were modelled with respect to protonation-induced bond length changes using DensityFunctional Theory (DFT). Significant bond elongations were highlighted to flag potential bond cleavages. CID production spectra, obtained using positive ion electrospray ionisation (Waters Synapt G1), were interpreted to establish ifobserved 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 polarisedbonds 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 thebonds 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 atomof the bond, inducing conformational changes leading to bond weakening. AM1-calculated bond length changes act as adescriptor for predicting bond cleavage. However, the impetus for cleavage of the unpolarised bonds may be product ionstability rather than bond weakening. Copyright © 2014 John Wiley & Sons, Ltd.

(wileyonlinelibrary.com) DOI: 10.1002/rcm.6870

Mass spectrometry (MS) is not an established rule-baseddiscipline in that the MS performance of compounds, bothin terms of quantitative sensitivity and the qualitativefragmentation behaviour, is difficult to predict even bypractitioners with many years’ experience. This makesinterpretation of collision-induced dissociation (CID) production spectra time-consuming, potentially rate-limiting, andsometimes subjective. In addition, novice users can find massspectral interpretation challenging.There are commercial software packages available to aid

spectral interpretation but, in general, these have the limitationthat they over-predict the number of product ions formed.For example, four major (>5%) product ions were observedin the product ion spectrum of protonated dofetilide, but theWaters’ Mass Fragment software (Waters Corporation,Manchester, UK) predicted 20 possible product ions on thebasis of the accurate mass data and over 100 possible productions for the nominal mass values.[1] The reason for this over-prediction is that many of the software packages lack anyinsight into the specific chemical structure of a given

* Correspondence to: P. A.Wright, School of Science, Universityof Greenwich, Medway Campus, Chatham ME4 4TB, UK.E-mail: [email protected]

Rapid Commun. Mass Spectrom. 2014, 28, 1127–1143

112

compound, so that they may predict product ions structureswhich are not chemically feasible. This is because thepredictions are made on the basis of applying rules (oftenbased on electron ionisation spectra rather than CID)extrapolating from databases and/or applying a ’systemicbond dissociation method’ in which all possible bonds in themolecule are cleaved and the mass of the remaining structurecalculated. Although the predictions made by thesecommercial software packages are generally useful, themethod described in this manuscript of predictingfragmentation by calculating bond elongation is able to narrowdown the number of possible choices significantly andtherefore enables much faster and more efficient interpretationof spectra.

Examples of such commercially available packages includeMass Frontier (Thermo Scientific, San Jose, CA, USA)[2] whichcombines comparison with their database, containing over30 000 fragmentation schemes from the literature and in-house data, with the application of general fragmentation/rearrangement rules. MS Fragmenter (ACDLabs, Toronto,Canada)[3] predicts product ions from the imported parentstructure by applying rules of fragmentation. FragmentiDentificator (FiD)[4] takes the alternative approach ofgenerating all the possible fragments that correspond to theaccurate mass of the observed ions and then ranking in order

Copyright © 2014 John Wiley & Sons, Ltd.

7

P. Wright, A. Alex and F. Pullen

1128

of how likely these fragments are to be formed. EPIC(Elucidation of Product Ion Connectivity)[5] and MetFrag[6]

are both ’systemic bond dissociation’ methods.These programmes assist with mass spectral interpretation

via different approaches, but common to them all is thelimitation that predictions are made on the basis ofassumptions or extrapolations which may not be valid. Thisresults in the prediction of a large number of product ionswhich are not in practice observed in the mass spectra. Forsoftware to be truly effective it needs to make predictionsbased mainly on the properties of the molecule itself withoutrecourse to assumptions. Quantum chemistry offers thepotential to improve the accuracy of in silico product ionpredictions as it describes the behaviour of matter atmolecular, atomic and sub-atomic levels. Quantumcomputational chemistry has been applied in massspectrometry[7] for many years, often used to calculate theenergies of the precursor ions, the product ions and anyintermediates as a way of determining both the most likelyroutes of product ion formation and which product ions arethe most energetically favourable. The approach describedin this manuscript differs from the majority of thesepreviously reported studies regarding the application ofcomputational chemistry to mass spectrometry in that itfocuses on bond length changes as a result of ionisation toidentify the bonds which are likely to cleave.One of the most widely applied quantum chemistry

approaches is Density Functional Theory (DFT)[8] whichcalculates the electronic structure of a given molecule. DFTmodels molecules in the gas phase and so is very well suitedfor determining the behaviour of ions within a massspectrometer. Molecular geometries predicted by DFT areknown to be accurate as they agree closely with experimentalX-ray diffraction data.[9] DFT has been used to great effect torationalise fragmentation based on the thermodynamiceffects that protonation has on the molecule,[10,11] bycalculating the thermodynamically most stable protonatedspecies based on the global minimum energy of the three-dimensional structure, and this information has been usefulin predicting the potential cleavage sites of those differentions. DFT is not routinely used to explain CID product ionmass spectra, however, because the amount of computationalresource required, both in terms of time and computerspecification and in the computational chemistry expertiserequired, limits its accessibility to the mass spectrometrist.The time taken to calculate the geometry of a single, drug-likemolecule can be anything fromminutes to days depending onthe size and flexibility of the molecule.Calculating low-energy geometries and electronic struc-

tures with semi-empirical methods is considerably faster thanwith DFT, and can be undertaken on any desktop PC ofreasonable specification using widely available commercialand academic software. Semi-empirical methods werechampioned by Dewar[12,13] in the 1950s to 1970s, whencomputers were still severely limited by processor speedand memory creating a real need for an approach whichwould allow computational chemistry calculations to beundertaken on realistic time scales.Semi-empirical methods are used to calculate heats of

formation, geometry, dipole moment and ionisation energyas well as chemical reactivity.[14] They give similar results toDFT for calculating bond dissociation energies but they tend

wileyonlinelibrary.com/journal/rcm Copyright © 2014 John Wil

to over-estimate ionisation potentials.[15,16] Semi-empiricalmethods are particularly useful for large molecules whereDFTcalculations take too long. However, the increased speedof calculation with semi-empirical methods is offset by alower accuracy than DFT.

One of the most popular semi-empirical methods is AustinModel 1 (AM1).[17] AM1 performs well in calculating bondlengths, being in good agreement with experimental data(approximately 5% error);[18] however, relative energies ofmolecules are calculated more accurately by DFT.[19] AM1tends to overestimate basicity, having been shown to besomewhat less reliable for calculating proton affinities.[20,21]

DFT generates more accurate heats of formation thanAM1.[22]

The authors have used DFT in previous studies with thepharmaceutical compounds fluconazole, maraviroc anddofetilide to rationalise CID product ion spectra in terms ofbond weakening resulting from conformational changes.[1,10,11]

In general, with a few exceptions,[23] lengthening a bond willcause it to weaken and render it more susceptible tocleavage.[24,25] These three publications reported thatprotonation-induced elongation of bonds did correspond tothe bonds that were actually observed to cleave in the tandemmass (MS/MS) spectra.

In order to further test the hypothesis that bond cleavageduring CID may be predicted by quantum computationalchemistry on the basis that bonds which are calculated toelongate significantly as a result of conformational changesinduced by protonation cleave preferentially during CID, 15pharmaceutical molecules in the mass range 101 to 608 amuwere modelled. Major bond elongations were highlighted toflag potential bond cleavages. The CID mass spectra werethen subsequently interpreted to establish if the predictedbond cleavages had actually occurred. This represented a’blind trial’ of using bond elongation as a descriptorpredicting bond cleavage.

Bond length calculations were undertaken using both DFT(basis set 6.31G**) and AM1. The parameterised approach ofAM1 is generally accepted to give good approximations formolecular geometry,[19] so has the potential to givesufficiently accurate estimates of bond elongation for thisapplication, especially as the absolute values are not required.AM1 calculations run in seconds rather than the hoursrequired for DFT; for example, geometry optimisation ofindole takes 5 s by AM1 but more than 1.5 h by DFT(B3LYP 6.31G).[26] If AM1 were found to give adequateestimates of bond lengths, this would extend the potentialfor this application of quantum computational chemistry tomass spectral interpretation as both the speed and the lackof requirement for specialist computational resource offerthe possibility of routine desktop use by non-expert users.

EXPERIMENTAL

Chemicals

HPLC grade methanol and acetonitrile were supplied byRathburn (Walkerburn, UK). HPLC grade water was obtainedfrom VWR International Ltd (West Chester, PA, USA). Formicacid (99% +) was supplied by Biosolve (Valkenswaard,The Netherlands).

ey & Sons, Ltd. Rapid Commun. Mass Spectrom. 2014, 28, 1127–1143

Predicting CID spectra with semi-empirical calculations

CEN025-014 was donated by Cyclofluidic (Welwyn GardenCity, UK). Sildenafil, doxazosin, ziprasidone and dofetilidewere donated by Pfizer Ltd (Sandwich, UK). All othercompounds were obtained from Sigma-Aldrich (Poole, UK).The structures of all 15 authentic standard compounds areshown in Fig. 1.All compounds were prepared at 1 mg/mL in acetonitrile/

water (between 10 and 100% acetonitrile depending on thecompound solubility), then diluted with 50:50 (v/v)methanol/water 0.1% formic acid to give a final concentrationof 20 μg/mL.

LC/MS

LC/MS data were acquired on a Synapt G1 Q-TOF(quadrupole-time-of-flight) mass spectrometer (WatersCorporation) in ESI positive ion and V mode (resolution15 000 FWHM), calibrated with sodium formate. Leucineenkephalin (MH+556.277) was infused at 5 μL/min as thereference lock mass. Samples (10 μL; 20 μg/mL)) wereintroduced via flow injection (0.5 mL/min 50:50 methanol/water 1% formic acid; no HPLC column). Methanol waschosen as the modifier as it has a lower proton affinity thanthe other common modifier, acetonitrile, potentially enhancingprotonation of the analyte.The following instrumental conditions were applied:

capillary voltage 5 kV; extraction cone voltage 5 V; samplingcone voltage 35 V; transfer collision energy 5 eV; cone gas (N2)flow rate 150 L/h; desolvation gas (N2) flow rate 1800 L/h;source temperature 100°C; desolvation temperature 500°C;trap collision energy 25 to 35 eV (set on a compound bycompound basis to obtain a spread of product ions). Argonwas used as the collision gas.The data acquisition settings were as follows: scan range

m/z 50 to 700; scan time 1 s, data centroid.

Computational modelling

All quantum calculations were undertaken using Spartan’10(Wavefunction, Inc., Irvine, CA, USA). Structures were drawnin ACD Labs Chemsketch (freeware for academic or personaluse), saved as .skc files and then opened in Spartan. Thestarting geometry was obtained using molecular mechanicsMMFF minimum energy geometry optimisation.All compounds were geometry optimised after protonation

at all heteroatoms using both DFT 6.31G** and AM1 with thefollowing preferences: maximum ligand distance 2.00Ǻa;polar area range 1000 kJ/mol; accessible area radius 1.000;’converge’ was selected.All calculations were undertakenwith the explicit hydrogens

(i.e. all hydrogens shown) on the molecule. All calculationswere undertaken locally on a desktop computer of specificationIntel® i7-3370 k CPU @ 3.50Hz, 16GB RAM, 64bit.

aCertain structures were shown to form internal hydrogenbonds when modelled using an initial maximum liganddistance of 3.60Ǻ, the default setting. Internal hydrogenbonding appeared to distort the bond lengths locally andlead to erroneous bond calculations. Therefore, hydrogenbonding was eliminated from the model by reducing themaximum ligand distance to 2.00 Ǻ for all calculations.

Copyright © 2014 JRapid Commun. Mass Spectrom. 2014, 28, 1127–1143

Assignment of product ions

Product ions of greater than approximately 8% abundancewere structurally assigned. All percentages in the assignmenttables are relative to the most abundant product ion (this maynot be the base peak where there is a considerable amount ofunfragmented precursor ion).

Criteria for prediction of bond cleavage

Previous investigations into the fragmentation of protonateddofetilide and four of its analogues[1] indicated that onlybonds which elongated significantly as a result of protonationon a heteroatom at the site of cleavage were observed tocleave. Therefore, only bond elongations of >0.039 Ǻ as aresult of protonation on one of the atoms to which the bondwas connected would be considered as predictive of bondcleavage in this study.

112

RESULTS AND DISCUSSION

Comparison of AM1 and DFT for calculating bond lengthchanges

Changes in the bond lengths of the 15 compounds (Fig. 1)resulting from protonation on all heteroatoms were calculatedusing both AM1 and DFT (6-31G** basis set). This representsthe calculation of 4147 bond length changes by eachcomputational method. Using all these data points (bondelongation, contraction and unchanged) a correlation of 0.87(R2= 0.76) was found between these two methods (using the’Correl’ function in Microsoft Excel 2013). Therefore, there is astatistically significant correlation between the two calculationmethods. This correlation is even greater if only the significantbond length increases (>0.039 Ǻ; n = 123) were compared; thecorrelation was increased to 0.96 (R2= 0.88) as shown in Fig. 2.Most importantly, the predictions made as to which bonds arelikely to cleave were the same based on data generated byeither method for all 15 compounds. Considering one of thecompounds as an example, for 1-methyl-2-pyrrolidinol(Table 1), the same bonds were calculated to elongate by>0.039 Ǻ by both computational methods and these were thebonds which were observed to cleave during CID (spectrumshown in Fig. 3 and product ion assignments in Table 2). It isnotable that considering the data in Table 2 the product ionswere formed from two different precursor ions, cations 1 and2 (Table 1), and were not derived from a single chargedmolecular species.

AM1 is far less demanding than DFT both in terms of speedof calculation and in the computational resource required.AM1 calculations typically took less than 30 s, whereas theDFTcalculations took between 15 min and 9 h. This means thatthese bond length calculationsmaybe undertaken routinely by themass spectrometrist as an aid to mass spectral interpretationwithout recourse to specialist computational resource. The speedof calculations also offers the potential for bond length calculationsto be incorporated into commercial mass spectral interpretationsoftware packages to improve the accuracy of the predictions.

Therefore, on the basis of the calculated bond lengths forthese 15 compounds, AM1 has been shown to be fit-for-purpose and only AM1-calculated bond length changes willbe reported and discussed in the rest of this manuscript.

wileyonlinelibrary.com/journal/rcmohn Wiley & Sons, Ltd.

9

mailto:[email protected]

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


1130

Comparison of AM1 and DFT for calculating relativegas-phase basicities

The relative gas-phase basicities were determined bycalculating the global energy minimum for each protonatedform of all the molecules, with the ion of the lowest energy


being the most basic. The energies did not always give thesame order for gas-phase basicities when calculated byAM1 and DFT. An example is shown for ziprasidone inTable 3. This difference in basicity order may result fromthe higher degree of error associated with the energies


Figure 2. Comparison of bond elongations (>0.039 Å; n= 123)calculated by both AM1 and DFTusing 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 only3–4 kcal/mol.[22] Therefore, for ions with energy minimawhich differ by smaller values, the order calculated byAM1 is less reliable than that calculated by DFT. Therefore,in any further discussions in this manuscript only the DFT-calculated relative energies (not bond lengths) will bereferred to.In this study the relative gas-phase basicities were

calculated to elucidate the mechanism of bond cleavage interms of the possibility of proton migration occurringsubsequent to ionisation. Calculation of gas-phase basicitieswas not required for the prediction of bond cleavages asproton migration to the ’dissociative site’[27] appears toinitiate cleavage[1] and so knowledge of the initial site ofionisation was not required. Bond length elongation was

Table 1. Comparison of bond length changes resulting from prC1 and C2, calculated using both DFTand AM1 (Spartan’10). Boof protonation on one of the bonding atoms are highlighted in


the only parameter required to predict bond cleavages forthese 15 compounds and therefore, in practice, only AM1calculations are required for predicting or rationalisingbond cleavage.

Prediction of bond cleavage on basis of calculated bondelongations

For the 15 compounds considered in this study 98 distinctbonds were observed to cleave in the CID product ionspectra. Of these, cleavage of 82 bonds was correctly

otonation of 1-methyl-2-pyrrolidinol at the sites specified asnd elongation calculated to be greater than 0.039Ǻ as a resultthe table


1131

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

Relativeintensity(%)

Experimentalm/z

Proposed ionformula andcalculated

accurate massError(ppm) Proposed structure(s) of ion Bond breaking

100 84.0817 C5H10N 84.0813 4.8 C4,O6

20 56.0506 C3H6N 56.0500 10.0

N5, C7 and/or N5,C3and/or N5,C4 and/orC2,C4 and/or C1,C2and/or C1,C3

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

E (kcal mol�1)

Energy differencebetween most stable

cation and others (kcal mol�1) E (kcal/mol�1)

Energy differencebetween most stable

cation and others (kcal mol�1)

AM1 DFT 6.31G**60.7 Neutral n/a �1233397 Neutral n/a204.3 Cation 3 0 �1233643 Cation 4 0204.3 Cation 4 0 �1233639 Cation 2 4205.4 Cation 2 1 �1233634 Cation 3 9223.5 Cation 1 19 �1233605 Cation 6 26228.2 Cation 6 24 �1233608 Cation 8 36228.9 Cation 8 25 �1233602 Cation 1 39229.8 Cation 5 25 �1233602 Cation 5 41267.2 Cation 7 63 �1233565 Cation 7 78


1132

predicted on the basis of bond elongation alone. Thisrepresents an overall success rate of 84%. Only carbon–carbon bond cleavage was not predicted (n = 16). Asmentioned above, this may be due to the thermodynamicfactors and to product ion stability, which is currently notconsidered in this approach.If polarised bonds only are considered, the success rate for

predicting bond cleavage at a heteroatom is 100% (n= 82).Table 4 summarises the results of this ’blind trial’.The results obtained for ziprasidone (spectrum shown in

Fig. 4; bond length changes in Table 5 and product ionassignments in Table 6) are typical for all 15 compounds.Of the 10 bonds which were observed to cleave, 9were correctly predicted. The bond which was notpredicted to cleave was a carbon–carbon bond (C2–C28).Initial modelling was performed on cations C1 to C7. Thecarbocation C8 (structure shown in Fig. 5) was modelledin retrospect in order to try to rationalise cleavage of C2–C28. Addition of a proton across the double bond in


the carbonyl group to centre the charge on C2 did notresult in a significant elongation of the C2–C28 bond. Thiswas also the case for the other compounds whichunderwent carbon–carbon bond cleavage (cortisone, 5-(p-methyl)phenylhydantoin, reserpine, trichlormethazide andziprasidone); modelling the appropriate carbocationsrather than locating the proton on a heteroatom did notpredict the bond cleavage.

It may be that bond weakening (via lengthening)around heteroatoms results from an increase in polarityof the bond by the addition of proton to the mostelectronegative atom. This is consistent with the BondActivation Rule (BAR) proposed by Alcami;[7,28,29]

the presence of the proton on the electronegative centrepulls the bonding electrons toward the charged centre,reducing the electron density in the bonding region, withcleavage occurring if there is sufficient difference inelectronegativity between the basic centre and the atombonded to it.


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

Compound

Number of bond cleavages

Percentage accuracyof prediction

Type of bondcleavage not

correctly predictedCorrectlypredicted

Observed to cleavebut not predicted

by bond elongation

1-Methyl-2-pyrrolidinol 4 0 100% n/aSulphride 3 0 100% n/aZiprasidone 9 1 90% C-C bondEphredine 3 0 100% n/aDoxazosin 9 0 100% n/aCEN024-014 3 0 100% n/aTrichlormethazide 10 1 90% C-C bondReserpine 9 2 82% C-C bond5-(p-Methylphenyl)-5-phenylhydantoin 2 2 50% C-C bond1,1-Dimethyl biguanide 5 0 100% n/aAmlodipine 8 0 100% n/aCortisone 2 6 25% C-C bondDesipramine 3 0 100% n/aSildenafil 4 0 100% n/aTrimethaprim 6 0 100% n/aTotal 80 12 87% n/a

Figure 4. CID product ion spectrumof protonated ziprasidone([MH]+413).


113

The product ions of ziprasidone were formed from sevendifferent [M+H]+ precursor ions, cations 1 to 7, and were thusnot derived from a single species. This was observed to betrue of all 15 compounds in that their product ions werederived from several protonated precursors, and furtherexemplified by the data for 1-methyl-2-pyrrolidinol shownin Table 2. This is consistent with previous studies whichhighlighted that the precursor ions appear to be a mixtureof ions which are protonated on a number of different basicsites across the molecule.[1,10,11] Other groups have alsoreported that precursor ions appear to be a mixture of ionsprotonated at different positions. Two isobaric ions observedin an MS/MS spectrum could only be assigned if they arederived from precursor ions protonated at different sites,giving rise to different product ions.[30] Kaufmann reportedthat a mixture of singly charged difloxacin species were


formed in the source.[31] Komaromi et al. observed thatN-acetyl-O-methoxyproline exhibits twodistinct fragmentationpathways indicative of the coexistence of several protonatedforms.[32] Komaromi used appropriate DFT calculations tosupport these observations.

As the carbon–carbon bonds are not (or are less) polarised,addition of the proton may have a limited effect. Inparticular, there is less ’incentive’ for the proton to remainassociated with a particular carbon within an unpolarisedbond and it may move along the molecule, possiblyvia hydride shifts. Therefore, formation of a carbocation hasless effect on the polarity of an individual carbon–carbonbond. The carbon–carbon bond cleavage may occur viaan alternative mechanism to protonation-induced bondweakening.

A study of the fragmentation of sulphur–sulphur bondcontaining heterocycles suggests that cleavage ofsulphur–sulphur bonds was driven by the stability of theproduct ion formed.[33] Carbon–carbon bond cleavagemay be analogous to this, especially as all the productions formed via carbon–carbon bond cleavage inthis investigation (i.e. for cortisone, 5-(p-methyl)phenylhydantoin and ziprasidone) showed increasedconjugation and/or increase in planar geometry relative tothe precursor [M+H]+ ions. These proposed product ions areshown in Table 6 for ziprasidone and Table 7 for the othercompounds.

This bond elongation approach over-predicted bondcleavage by 33% (i.e. 32 bonds were predicted to cleave butwere not observed to do so). This is a significantly improvedover-prediction rate compared with many commercialpackages. In one example, Waters Mass Fragment wasobserved to over predict by 400% based on accurate massdata and by over 2000% for nominal mass data.[1] Basingfragmentation predictions on bond lengthening has theadvantage that the predictions are entirely in silico, based on


3

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 ofthe bonding atoms are highlighted in the table


1134

the inherent properties of the molecule itself and will give thesame predictions for nominal mass data as for accurate massdata.From the summary shown in Table 8 it appears that therewere

certain classes of bond which were prone to over-prediction:

• Over half (56%; n=18) of the bonds incorrectly predicted tocleave were to an atom to which one of the other bonds wasobserved to cleave. The bonds that did cleave were elongatedto a significantly greater extent (30–600%) than the bonds tothe same atom which did not cleave in eight of the 18 cases(i.e. 25% of the total over-predicted cleavages). This suggests


that cleavage of multiple bonds to the same atom was notfavoured. In addition, there was a tendency for the mostextended bond to cleave preferentially, but there weresufficient exceptions for this not to be considered as a ’rule’.

• Protonation of nitrogens (n = 6; or 19% of the incorrectlypredicted bond cleavages) within a conjugated systemwas predicted to initiate cleavage, but did not do so. Thismay be due to stabilisation by delocalisation of the chargeacross the conjugated system, resulting in the charge notbeing associated with a single centre. Because the chargeis delocalised, the proton will have less effect on thepolarity, 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

Relativeintensity

Experimentalm/z

Proposed ionformula andcalculated

accurate massError(ppm)

Proposed structure(s)of ion Bond breaking

5% 220.0924 C11H14N3S 7 N13,C12220.0908

100% 194.0331 C10H9ClNO 22 N13,C12194.0373

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

177.0486

20% 166.0427 C9H9ClN 2 N13,C12 C2,C28 C2,N3166.0424

25% 159.0678 C10H9NO 4 N13,C12 C8,Cl19159.0684

8% 131.0738 C9H9N 2S19,C20 N18,C17 N16,C25 N16,C13 N13,C14 N13,C25

131.0735

Figure 5. Structure of ziprasidone carbocation modelled byAM1 for which delta bond length data are reported in Table 5.



113

• Although some sulphonamide cleavage was observedthere was a tendency to over-predict the cleavage of allthe bonds within sulphonamide groups (n = 4 or 13%). Thismay be because sulphonamide bonds are flexible[34] andable to absorb conformational change, and are also ableto delocalise the charge across the sulphonamide groupsuch that it is not strongly associated with any single atom.


5

Table

7.The

prop

osed

assign

men

tsof

prod

uctions

resu

ltingfrom

carbon

-carbo

nbo

ndcleava

gefortrichlormetha

zide,

reserpine,

5-(p-m

ethy

l)-phe

nylhyn

dan

toin

and

cortison

e

Relative

intensity

Exp

erim

ental

m/z

Prop

osed

ion

form

ulaan

dcalculated

accu

rate

mass

Error

(ppm

)Prop

osed

structure(s)

ofion

Bon

dcleave

dCom

poun

d

40%

448.1190

C23H

30NO

84

C10,C12

C24,C23

448.1971

10%

336.1573

C18H

24O

68

C47,C5

C5,C7

C26,N

25C9,C10

336.1600

20%

236.1268

C13H

18NO

38

C22,C21

N13,C14

O32,C33

236.1287

50%

174.0933

C11H

11NO

8C10,C12

C24,C23

174.0919

(Contin

ues)


wileyonlinelibrary.com/journal/rcm Copyright © 2014 John Wiley & Sons, Ltd. Rapid Commun. Mass Spectrom. 2014, 28, 1127–1143

1136

Table

7.(C

ontinu

ed)

Relative

intensity

Exp

erim

ental

m/z

Prop

osed

ion

form

ulaan

dcalculated

accu

rate

mass

Error

(ppm

)Prop

osed

structure(s)

ofion

Bon

dcleave

dCom

poun

d

8%239.1200

C15H

15N

2O7

N13,C8

C8,N9

239.1184

100%

196.1108

C14H

14N

10C12,C10

N13,C8

196.1126

8%104.0505

C7H

6N5

C12,C14

C12,C10

N13,C8

104.0500

10%

183.9639

C4H

6Cl 2N

2S5

S13,O15

S13,O14

C16,C17

C9,C8

183.9629

(Contin

ues)


wileyonlinelibrary.com/journal/rcmCopyright © 2014 John Wiley & Sons, Ltd.Rapid Commun. Mass Spectrom. 2014, 28, 1127–1143

1137

Table

7.(C

ontinu

ed)

Relative

intensity

Exp

erim

ental

m/z

Prop

osed

ion

form

ulaan

dcalculated

accu

rate

mass

Error

(ppm

)Prop

osed

structure(s)

ofion

Bon

dcleave

dCom

poun

d

15%

258.1617

C17H

22O

21

C7,C5

C28,C5

258.1620

10%

241.1597

C17H

21O

2C7,C8

C28,C5

C2,O1

241.1592

100%

163.1119

C11H

15O

2C23,C25

C9,C8

C5,C28

163.1123

30%

145.1022

C11H

133

C23,C25

C9,C11

C17,O

18145.1017

25%

121.0660

C8H

9O5

C23,C21

C13,C14

121.0653

30%

105.0708

C8H

94

C23,C21

C14,C13

C17,O

18105.0708

15%

93.0700

C7H

95

C23,C21

C14,C13

C17,O

1893.0704


wileyonlinelibrary.com/journal/rcm Copyright © 2014 John Wiley & Sons, Ltd. Rapid Commun. Mass Spectrom. 2014, 28, 1127–1143

1138

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

Compound

Number of bonds predictedto cleave but did not break

(i.e. over-predicted) Type of bond over predicted

1-Methyl-2-pyrrolidinol 0 n/aSulphride 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/aDoxazosin 3 C-O (aliphatic; n = 3): other bond to same oxygen

breaks in preferenceCEN024-014 4 C26-N25 and C31-N25: other bond to N25 broke in

preference (C24-N25) (n = 2) C18-N19: other bond toN19 broke in preference (C20-N19 elongates twice asmuch)(n = 1) C5-N4: other bond to N4 broke in preference(C3-N4 elongates half as much) (n = 1)

Trichlormethazide 0 n/aReserpine 6 N25-C10: other bond to N25 broke in preference

(C24-N25)(n = 2); C-O (aliphatic; n = 6), other bondto 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/aCortisone 0 n/aDesipramine 2 N15 to aromatic carbons (n = 2); two bonds need

be broken to generate leaving groupSildenafil 8 N4-C5 other bond to 029 broke in preference

(N4-S2 elongates by 600% more) (n = 1); C28-O29other 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


113

Product ion intensity

Asummary of factors thatmay influence product ion abundanceis shown in Table 9. There appears to be no correlation betweenthe basicity of the molecule in water (pKa) and formation of themajor product ion; for only five of the 15 compounds was themajor product ion formed by protonation at the most basiccentre in solution. For five compounds, none of the product ionsresulted from protonation at the most basic centre in solution.For all these five compounds, however, the most basic atom inthe gas phase was part of a conjugated system which coulddelocalise the charge. This both stabilises the precursor ion,hence reducing the propensity for fragmentation, and meansthat the charge is not associated with any particular bond. Theresults of this study to date indicate that the charge has to becentred on one of the atoms in the bond to initiate cleavage.Potentially, the number of bonds cleaved to form an ion

may reflect in its relative abundance in that more energy isrequired to cleave multiple bonds. However, the data inTable 9 show that the intensity of the product ions does notappear to depend on the number of bonds broken during itsformation. For only six of the 15 compounds was the majorproduct ion formed by single bond cleavage; the other nineresulted from multiple bond cleavages.


Overall, no correlation was observed between the extentof bond lengthening and the intensity of the product ion.In a previous study using bond length changes to predictthe CID fragmentation of protonated dofetilide, there wasa quantitative relationship between the extent of bondelongation and product ion intensity.[1] However, thislarger study shows that although product ion intensitymay be predictable on the basis of bond lengthening forcertain compounds, it is not valid to apply this approachindiscriminately. Other research groups have successfullyapplied DFT to predict ion intensities for peptides[35] andquinazolines[36] by calculating product ion energies. Thus,it appears that bond weakening may dictate whichpolarised bonds cleave, but it may the relative production stability which determines the relative intensity ofthe product ions formed as a result of these bondcleavages.

There was no obvious correlation between the pKa of themolecule and the appearance of spectra in terms of thenumber and abundance of the ions. For example,trichlormethazide, which has no basic centre, produces sixproduct ions, four of which are major (greater than 30%).Amlodipine contains a primary amino group of pKa 9.5 andgives a product ion spectrum containing eight ions, four of


9

Table

9.Observa

tion

sarou

ndrelative

abun

dan

ceof

prod

uctions

intheCID

spectraof

the[M

+H]+

ions

ofallfi

fteencompo

unds.The

pKawas

calculated

usingMarvin

(Che

mAxo

n,Bud

apest,Hun

gary)[4

6]an

dthega

sph

aseba

sicities

referto

therelative

stab

ilities

(globa

lene

rgyminim

a)of

theproton

ated

species

Com

poun

d

Major

iondue

tosing

lebo

ndcleava

ge?

Major

ion

proton

ated

atmostba

siccentre

(aqu

eous

)?

pKaof

most

basicsite

(aqu

eous

)?

Mostba

sicsites

samein

gas

phasean

dsolution

Atleaston

eprod

uctiondue

toproton

ationat

mostba

sicsite?

App

earanc

eof

spectra

1-Methy

l-2-py

rrolidinol

Yes

No

8.6

Yes

Yes

One

major

ion;

twoions

intotal

Sulphride

Yes

Yes

8.4

Yes

Yes

Twomajor

ions;fou

rions

intotal

Ziprasidon

eYes

Yes

7.1

Yes

Yes

One

major

ion;

four

ions

intotal

Eph

redine

Yes

No

9.5

Yes

Yes

One

major

ion;

threeions

intotal

Dox

azosin

No

No

7.1

Yes

No

Three

major

ions;fi

veions

intotal

CEN024-014

Yes

Yes

8.4

No

Yes

One

major

ion;

four

ions

intotal

Trichlormetha

zide

No

Yes

�4.1

No

Yes

Four

major

ion;

sixions

intotal

Reserpine

Yes

No

7.3

Yes

Yes

Four

major

ion;

eigh

tions

intotal

5-(p-M

ethy

lpheny

l)-5-ph

enylhy

dantoin

No

No

�9.0

Yes

No

Onlyon

emajor

ion

1,1-Dim

ethy

lbigua

nide

No

No

12.6

Yes

No

Multip

lemajor

ions:m

anyba

siccentres?

Amlodipine

No

No

9.5

Yes

Yes

Four

major

ions;e

ight

ions

intotal

Cortisone

No

No

�3.2

Yes

Yes

Twomajor

ions;sev

enions

intotal

Desipramine

No

Yes

10.0

No

Yes

Twomajor

ions;fou

rions

intotal

Silden

afil

No

No

6.0

Yes

No

Three

major

ions;fi

veions

intotal

Trim

etha

prim

No

No

7.2

Yes

No

Seve

nmajor

ions;n

ineions

intotal



1140

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

Proton migration

Proton migration to a thermodynamically less favourable sitehas been proposed to be required to rationalise the formationof certain product ions. For example, the loss of ammoniafrom peptidic amides required protonation on the nitrogenalthough the oxygen is both the most energetically favouredprotonation site and the observed initial protonationsite.[37–39] Penicillin shows cleavage of the β-lactam bondafter transfer of the proton from the carbonyl to the lactamnitrogen.[40,41] In addition, dibenzyl ethers,[27] the pharmaceuticalcompound maraviroc,[10] dialkylphosphoric acid esters[42] andthiourea/urea compounds[43] have all been reported toexhibit product ions in their mass spectra generated followingproton migration from the most thermodynamically likelysite. It has been proposed that the energy for proton migrationis obtained by the transfer of kinetic to internal energyduring ion molecule collisions,[21] probably in the collision cellduring CID.

The results of this larger scale study support the hypothesisthat proton migration from the initial site of ionisation to the’dissociative’ sitemaybe required to initiate fragmentation:[1,27]

• As expected, the greatest effect in terms of conformationalchange, and hence bond length changes, occurred in theimmediate area around the protonation site. In a few cases,however, protonation did result in significant bondelongation remote from the protonation site (shown inTable 10). None of these bond length changes gave rise tothe product ions observed in the CID spectra, reinforcingthe proposal that the proton needed to be adjacent to thesite of cleavage for fragmentation to occur.

• The spectra of four compounds (doxazosin, reserpine,1,1-dimethyl biguanide and sildenafil) did not containany product ions derived from precursors protonated atthe most basic site (Table 9). As all spectra were obtainedvia ESI, the original ionisation site is likely to be the centrewith the highest pKa. Reviewing the literature for evidenceof gas-phase protonation during ESI, in many cases directgas-phase ionisation is proposed because it is difficult torationalise the product ions in the spectra of certaincompounds by protonation on the most basic site insolution. For example, it has been proposed thatgas-phase ionisation via ion-molecule reactions plays amajor role in ESI, such as by proton transfer from gaseousammonium ions to analytes of higher proton affinity.[44]

However, for all these four compounds the most basic siteis the same in both the solution and gas phase. Therefore,direct gas-phase ionisation at the less basic, dissociativesite is thermodynamically unlikely. Thus, the protonneeds to move from the site of greatest basicity (i.e. theinitial ionisation site) to the dissociative site to initiatefragmentation.

This raises the possibility that some cases of charge-remotefragmentation[45] reported in the literature may, in fact,represent proton migration followed by charge-directedfragmentation.


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

Compound BondCalculated increasein bond length (Ǻ)

Site of protonationcausing bond length increase

Observed to cleaveduring CID?

Sulphride S2-N4 0.335 N7 NoS2-C5 0.040 O14 NoC13-O14 0.103 O9 No

Doxazosin C12-N16 0.052 O23 NoTrichlormethazide S2-C5 0.070 N10 No

S2-C5 0.049 N12 NoS2-C5 0.490 O14 NoS2-C5 0.815 O15 No

Reserpine O1-C2 0.880 O49 NoSildenafil S2-N4 0.043 N7 No

C10-N4 0.057 O29 NoS2-C11 0.092 O29 NoS2-N4 0.078 O29 NoN4-C5 0.067 O29 No


114

CONCLUSIONS

This study (15 compounds, 98 observed bond cleavages andover 8000 bond length calculations) has confirmed thatsignificant bond elongation (>0.039 Å) may be used as adescriptor for cleavage of polarised bonds during CID byflagging which bonds have been weakened as a result ofstructural changes due to protonation. This approach achieved100% success rate in the prediction of polarised bond cleavage.Moreover, it has been shown that the semi-empiricalcomputational approach AM1 can be used to calculate thesebond length changes as it gives very similar results to thoseobtained by DFT. Most studies to date applying computationalchemistry to mass spectral data have used DFT, which iscomputationally demanding both in terms of calculation timeand in the computing power required. This has limited thespread of the application by mass spectrometrists ofcomputational chemistry to the prediction or rationalisationof mass spectral fragmentation. The evidence that AM1 canused to predict bond cleavage opens up this approach to manymore scientists. AM1 calculations may take only seconds andbe undertaken on a standard computer, rather than on anextremely high specification server which is often used forDFTcalculations. The speed of the AM1 calculations also offersthe potential for their incorporation into commercial softwareto improve the ’chemical sense’ of these packages and reducethe over-prediction of product ions. Over-prediction of bondcleavage was only 34% in this study, a significant improvementover the over-prediction of product ion formation by manyspectral interpretation software packages.Even without software packages tailored specifically for

mass spectral interpretation, the procedure for using thecurrently available computational chemistry packages as anaid to predicting and/or explaining fragmentation of these15 protonated molecules was found to be straightforward:

• Structures for each neutral molecule and the correspondingmolecules protonated at all possible sites were generated.The basicity of the protonation sites did not need to beconsidered and so protonation at all heteroatoms neededto be modelled.


• These structures were imported into a semi-empirical AM1computational chemistry software package which wasused to calculate all the bond lengths in the neutral andprotonated molecules. Internal hydrogen bonds may leadto misleading results so these needed to be eliminatedduring modelling.

• The bond lengths calculated for the protonated moleculeswere subtracted from the corresponding bonds in theneutral molecule to obtain the bond length changes.

• Significant increases in bond length (in this case >0.039 Å),as a result of protonation on one of the atoms forming thebond,maybe consideredpredictive of polarisedbond cleavage.

The behaviour of these 15 compounds is consistent with themodel of CID fragmentation that has been proposed inprevious publications:

• Protonation caused conformational changes whichresulted in bond length changes, which was accuratelycalculated using quantum chemistry based computationalsoftware.

• Weakening of bonds is indicated by lengthening of bondsand significant bond length increases (>0.039 Å) weakenthe bond to such an extent that it is preferentially brokenduring CID.

• The proton had to be located on one of the atoms (the mostelectronegative) involved in the bond for cleavage to occur;bond elongation remote from the protonation site did notlead to bond cleavage.

• Protonation at the most basic sites (liquid and gas phase)did not necessarily lead to bond cleavage. Therefore, forsome compounds, the proton appears to have migratedfrom the primary site of protonation during ionisation toa thermodynamically less stable site to initiate cleavage.

• As the proton may migrate from the protonation site duringionisation, calculation of basicity (gas- and liquid-phase)wasunnecessary for fragmentation predictions; the only requiredcalculation to predict cleavage of a polarised bond wasthe bond length change. It is the protonation site which hasthegreatest effect onadjacent bond lengths rather than the centreat which ionisation occurs which is necessary for this approach.


1


1142

• There was no single protonated molecular species; theproduct ions appear to be formed from a mixture of singlycharged protonated precursors.

Unpolarised, carbon–carbon bond cleavage apparentlycannot be predicted on the basis of bond length changes.We therefore propose that C–C bond cleavage is at leastpartly driven by the thermodynamic stability of the resultingproducts rather than bond length changes duringprotonation. In addition, product ion intensities did notcorrelate with the extent of bond elongation and thus couldnot be predicted by bond elongation alone. For bothunpolarised bond cleavage and product ion intensity,calculation of product ion stability may be required for themto be rationalised. The relative stability of ions andfragmentation products can be predicted in principle usingboth, DFT and AM1. Fortunately, cleavage of non-polarbonds is less common than cleavage of polarised bonds andproduct ion intensity, although very useful in comparingspectra with library data, is not critical in interpreting massspectra. Therefore, both these limitations of the bondlengthening approach do not significantly restrict theapplication of AM1 as a tool for the interpretation of CIDmass spectra.

AcknowledgementsThe British Mass Spectrometry Society (BMSS) provided asmall equipment grant for the purchase of the Spartan softwareused in this study. Funding for this study was supplied by theVice-Chancellor’s Office, University of Greenwich, UK. Thecompound CEN025-14 was supplied by CyclofluidicIntegrated Discovery (Welwyn Garden City, UK).

REFERENCES

[1] P. Wright, A. Alex, S. Harvey, T. Parsons, F. Pullen.Understanding collision-induced dissociation of dofetilide:a case study in the application of density functional theoryas an aid to mass spectral interpretation. Analyst 2013,138, 6861.

[2] Mass Frontier™. Available: http://www.highchem.com/massfrontier/mass-frontier.html.

[3] ACD/MS Fragmenter. ACDLabs, Toronto, Canada. ACD/Labs.com.

[4] M. Heinonen, A. Rantanen, T. Mielikäinen, M. Heinonen,A. Rantanen, T. Mielikäinen. FiD: a software for ab initiostructural identification of product ions from tandem massspectrometric data. Rapid Commun. Mass Spectrom. 2008,22, 3043.

[5] S. Wolf, S. Schmidt, M. Müller-Hannemann, S. Neumann. Insilico fragmentation for computer assisted identification ofmetabolite mass spectra. BMC Bioinformatics 2010, 11, 148.

[6] H. Horai, M. Arita, S. Kanay, Y. Nihe, T. Ikeda, K. Suwa,Y. Ojima, K. Tanaka, S. Tanaka, K. Aoshima, Y. Oda,Y. Kakazu, M. Kusano, T. Tohge, F. Matsuda, Y. Sawada,M. Yokota Hirai, H. Nakanishi, K. Ikeda, N. Akimoto,T. Maoka, H. Takahashi,T. Ara, N. Sakurai,H. Suzuki,D. Shibata,S. Neumann, T. Iida, K. Tanaka, K. Funatsu,F. Matsuura, T. Soga, R. Taguchi, K. Saito, T. Nishioka.MassBank: a public repository for sharing mass spectraldata for life sciences. J. Mass Spectrom. 2010, 45, 703.


[7] M. Alcamí, O. Mó, M. Yáñez. Computational chemistry: auseful (sometimes mandatory) tool in mass spectrometrystudies. Mass Spectrom. Rev. 2001, 20, 195.

[8] K. Burke. Perspective on density functional theory. J. Chem.Phys. 2012, 136, 150901.

[9] A. Saeed, I. Arshad, U. Flörke. Synthesis, crystal structure, andDFT study of N’-(2,4-dinitrophenyl)-2-fluorobenzohydrazide.J. Chem. 2013. DOI: 10.1155/2013/648121.

[10] P. Wright, A. Alex, T. Nyaruwata, T. Parsons, F. Pullen.Using density functional theory to rationalise the massspectral fragmentation of maraviroc and its metabolites.Rapid Commun. Mass Spectrom. 2010, 24, 1025.

[11] A. Alex, S. Harvey, T. Parsons, F. S. Pullen, P. Wright,J.-A. Riley. Can density functional theory (DFT) be usedas an aid to a deeper understanding of tandem massspectrometric fragmentation pathways? Rapid Commun.Mass Spectrom. 2009, 23, 2619.

[12] M. J. S. Dewar, W. Thiel. A semi-empirical model for thetwo-center repulsion integrals in the NDDO approximation.Theor. Chim. Acta 1977, 46, 89.

[13] M. J. S. Dewar. A molecular orbital theory of organicchemistry. I. General principles. J. Am. Chem. Soc. 1952, 74,3341.

[14] C. Cojocaru, A. Rotaru, V. Harabagiu, L. Sacaresc. Molecularstructure and electronic properties of pyridylindolizinederivative containing phenyl and phenacyl groups:Comparison between semi-empirical calculations andexperimental studies. J. Mol. Struct. 2013, 1034, 162.

[15] T. R. Sharp. Calculated carbon-hydrogen bond dissociationenthalpies for predicting oxidative susceptibility of drugsubstance molecules. Int. J. Pharm. 2011, 418, 304.

[16] E. Klein, M. Matis, V. Lukeš, Z. Cibulková. The applicabilityof AM1 and PM3 semi-empirical methods for the study ofN–H bond dissociation enthalpies and ionisation potentialsof amine type antioxidants. Polym. Degrad. Stab. 2006, 91, 270.

[17] M. J. S. Dewar, E. G. Zoebisch, E. F. Healy, J. J. P. Stewart.Development and use of quantum mechanical molecularmodels. 76. AM1: a newgeneral purpose quantummechanicalmolecular model. J. Am. Chem. Soc. 1985, 107, 390.2.

[18] C. Cojocaru, A. Rotaru, V. Harabagiu, L. Sacarescu.Molecularstructure and electronic properties of pyridylindolizinederivative containing phenyl and phenacyl groups:Comparison between semi-empirical calculations andexperimental studies. J. Mol. Struct. 2013, 1034.

[19] G. Zheng, S. Irle, K. Morokuma. Performance of the DFTBmethod in comparison to DFT and semi-empirical methodsfor geometries and energies of C20–C86 fullerene isomers.Chem. Phys. Lett. 2005, 412, 210.

[20] S. Rajak, N. Islam, D. Ghosh. Modeling of the chemico-physical process of protonation of molecules entailing somequantum chemical descriptors. J. Quantum Inf. Sci. 2011, 1, 87.

[21] P. J. AmorimMadeir, P. D. Vaz, R. J. N. Bettencourt da Silva,M. H. Florêncio. Can semi-empirical calculations help solvemass spectrometry problems? Protonation sites and protonaffinities of amino acids. Chempluschem 2013, 78, 1149.

[22] P. E. M. Siegbahn. The performance of hybrid DFT formechanisms involving transition metal complexes inenzymes. J. Biol. Inorg. Chem. 2006, 11, 695.

[23] E. Kraka, D. Creme. Characterization of CF bonds withmultiple-bond character: bond lengths, stretching forceconstants, and bond dissociation energies. Chemphyschem2009. 10, 686.

[24] L. Pauling. The dependence of bond energy on bond length.J. Phys. Chem. 1954, 58, 662.

[25] L. Pauling. The Nature of the Chemical Bond and the Structureof Molecules and Crystals: An Introduction to Modern StructuralChemistry, (3rd edn.). Cornell University Press, Ithaca NY,1960.


http://www.highchem.com/massfrontier/mass-frontier.html

http://www.highchem.com/massfrontier/mass-frontier.html


[26] C. K. Kim, S. G. Cho, C. K. Kim, M.-R. Kim, H. W. Lee.Prediction of physicochemical properties of organicmolecules using semi-empirical methods. Bull. Korean Chem.Soc. 2013, 34, 1043.

[27] Y.-P. Tu. Dissociative protonation and fragmentation: retro-Friedel–Crafts reactions of heterocyclic drug and metabolitemolecules in mass spectrometry. Int. J. Mass Spectrom. 2012,316/318, 40.

[28] M. Alcamí, O. Mó, M. Yáñez, J.-L. M. Abboud, J. Elguero.Bond activation by protonation in the gas phase. Chem.Phys. Lett. 1990, 172, 471.

[29] M. Alcamí, O. Mó, M. Yáñez. Modeling intrinsic basicitiesand acidities. J. Phys. Org. Chem. 2002, 15, 174.

[30] S. W. Holman, P. Wright, G. J. Langley. High-throughputapproaches towards the definitive identification ofpharmaceutical drug metabolites. 2. An example of howunexpected dissociation behaviour could preclude correctassignment of sites of metabolism. Rapid Commun. MassSpectrom. 2009, 23, 2017.

[31] A. Kaufmann, P. Butcher, K. Maden, M. K. Widmer,K. Giles, D. Uría. Are liquid chromatography/electrospraytandem quadrupole fragmentation ratios unequivocalconfirmation criteria? Rapid Commun. Mass Spectrom. 2009,23, 985.

[32] I. Komáromi, A. Somogyi, V. H. Wysocki. Proton migrationand its effect on the MS fragmentation of N-acetyl OMeproline: MS/MS experiments and ab initio and densityfunctional calculations. Int. J. Mass Spectrom. 2005, 241, 315.

[33] J.-L.M.Abboud,M. Esseffar,M.Herreros, O.Mó,M. T.Molina,R. Notario, M. Yáñez. Protonation of S 4, S 6, and S 8 sulfurcycles. A quantitative study. J. Phys. Chem. A 1998, 102,7996.

[34] C. Baldauf, R. Günther, H.-J. Hofmann. Conformationalproperties of sulfonamido peptides. J. Mol. Struct.THEOCHEM 2004, 675, 19.

[35] T. Pechan, S. R. Gwaltney. Calculations of relative intensitiesof fragment ions in the MSMS spectra of a doubly chargedpenta-peptide. BMC Bioinformatics 2012, 13, S13.

[36] A. Galezowska, M. W. Harrison, J. M. Herniman,C.-K. Skylaris, G. J. Langley. A predictive science approachto aid understanding of electrospray ionisation tandem massspectrometric fragmentation pathways of small molecules


using density functional calculations. Rapid Commun. MassSpectrom. 2013, 27, 964.

[37] K. A. Cox, S. J. Gaskell, M. Morris, A. Whiting. Role of thesite of protonation in the low-energy decompositions ofgas-phase peptide ions. J. Am. Soc. Mass Spectrom. 1996, 7,522.

[38] L. R. Snyder, J. W. Dolan. High-Performance Gradient Elution:The Practical Application of the Linear-Solvent-Strength Model.John Wiley, New Jersey, 2007.

[39] H.-Y. Lin, D. P. Ridge, E. Uggerud, T. Vulpius. Unimolecularchemistry of protonated formamide. mass spectrometry andab Initio quantum chemical calculations. J. Am. Chem. Soc.1994, 116, 2996.

[40] C. K. Fagerquist, A. R. Lightfield. Confirmatory analysis ofbeta-lactam antibiotics in kidney tissue by liquidchromatography/electrospray ionization selective reactionmonitoring ion trap tandem mass spectrometry. RapidCommun. Mass Spectrom. 2003, 17, 660.

[41] C. K. Fagerquist, A. R. Lightfield, S. J. Lehotay.Confirmatory and quantitative analysis of beta-lactamantibiotics in bovine kidney tissue by dispersive solid-phase extraction and liquid chromatography-tandem massspectrometry. Anal. Chem. 2005, 77, 1473.

[42] Y. Zeng, X. Chen , D. Zhao, H. Li, Y. Zhang, X. Xiao.Estimation of pKa values for carboxylic acids, alcohols,phenols and amines using changes in the relative Gibbs freeenergy. Fluid Phase Equilib. 2012, 313, 148.

[43] F. Falvo, L. Fiebig, F. Dreiocker, R. Wang, P. B. Armentrout,M. Schäfer. Fragmentation reactions of thiourea- and urea-compounds examined by tandem MS-, energy-resolvedCID experiments, and theory. Int. J. Mass Spectrom. 2012,330/332, 124.

[44] R. Guevremont, K. W. M. Siu, J. C. Y. Blanc, S. S. Berman.Are the electrospray mass spectra of proteins related to theiraqueous solution chemistry? J. Am. Soc. Mass Spectrom. 1992,3, 216.

[45] C. Cheng, M. L. Gross. Applications and mechanisms ofcharge-remote fragmentation. Mass Spectrom. Rev. 2000, 19,398.

[46] G. Pirok, N. Máté, J. Varga, J. Szegezdi, M. Vargyas,S. Dóránt, F. Csizmadia. Making ’real’ molecules in virtualspace. J. Chem. Inf. Model. 2006, 46, 563.


1143

predicting collision-induced dissociation spectra: semi

Documents