New Quantitative Structure-Fragmentation ... - ACS Publications

Apr 28, 2014 - Eugen Sisu,. ‡ and Adrian Covaci*. ,§. †“Aurel Vlaicu” University, Arad, 310330, Romania. ‡. University of Medicine and Phar...
1 downloads 0 Views 429KB Size
Subscriber access provided by University of Ulster Library

Article

New Quantitative Structure-Fragmentation Relationship strategy for chemical structure identification using as descriptor the calculated enthalpy of formation for the fragments produced in electron ionization mass spectrometry. Case study: tetrachlorinated biphenyls Nicolae Dinca, Simona Dragan, Mihael Dinca, Eugen Sisu, and Adrian Covaci Anal. Chem., Just Accepted Manuscript • DOI: 10.1021/ac5003728 • Publication Date (Web): 28 Apr 2014 Downloaded from http://pubs.acs.org on May 2, 2014

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Analytical Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 20

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

1

New Quantitative Structure-Fragmentation Relationship strategy for

2

chemical structure identification using as descriptor the calculated

3

enthalpy of formation for the fragments produced in electron ionization

4

mass spectrometry. Case study: tetrachlorinated biphenyls

5 6 7 1

8

2

2

2

3*

Nicolae Dinca , Simona Dragan , Mihael Dinca , Eugen Sisu , Adrian Covaci

9 10 11

1

“Aurel Vlaicu” University, 310330 Arad, Romania

12

2

University of Medicine and Pharmacy “Victor Babes”, 300041 Timisoara, Romania

13

3

Toxicological Center, University of Antwerp, 2610 Wilrijk, Belgium

14 15 16

*-corresponding author:

17

Adrian Covaci, fax: +32-3-265-2722; e-mail: [email protected]

18 19

ACS Paragon Plus Environment

1

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 2 of 20

20

ABSTRACT

21

Differential mass spectrometry correlated with quantum chemical calculations (QCC-∆MS)

22

has been shown to be an efficient tool for the chemical structure identification (CSI) of

23

isomers with similar mass spectra. For this type of analysis, we report here a new strategy

24

based on ordering (ORD), linear correlation (LCOR) algorithms and their coupling, to filter

25

the most probable structures corresponding to similar mass spectra belonging to a group with

26

dozens of isomers (e.g., tetrachlorinated biphenyls, TeCBs). This strategy quantifies and

27

compares the values of enthalpies of formation (∆fH) obtained by QCC for some isobaric

28

ions from the (EI)-MS mass spectra, to the corresponding relative intensities. The result of

29

CSI is provided in the form of lists of decreasing probabilities calculated for all the position-

30

isomeric structures using the specialized software package CSI-Diff-MS Analysis 3.1.1. The

31

simulation of CSI with ORD, LCOR and their coupling of six TeCBs (IUPAC no. 44, 46, 52,

32

66, 74, and 77) has allowed finding the best semi-empirical molecular-orbital methods for

33

several of their common isobaric fragments. The study of algorithms and strategy for the

34

entire group of TeCBs (42 isomers) was made with one of the variants found optimal for the

35

computation of ∆fH using semi-empirical molecular orbital methods of HyperChem: AM1

36

for M+, [M-4Cl]+· ions and RM1 for [M-Cl]+, [M-2Cl]+. The analytical performance of ORD,

37

LCOR and their coupling resulted from the CSI simulation of an analyte of known structure,

38

using a decreasing number of isomeric standards, s = 5, 4, 3, and 2. Compared with the

39

results obtained by classical library search for TeCB isomers, the novel strategies of

40

assigning structures of isomers with very similar mass spectra based on ORD, LCOR and

41

their coupling were much more efficient, because they provide the correct structure at the top

42

of the probability list. Databases used in these CSI do not contain mass spectra as in the case

43

of library search, but series of ∆fH values obtained by QCC. These techniques are capable of

44

relating relative intensities to the chemical structures of analytes via ∆fH of ions which turns

45

out to be a good Quantitative Structure-Fragmentation Relationship (QSFR) descriptor.

46 47 48

Key words: structural identification, differential mass spectrometry, tetrachlorinated

49

biphenyls, quantum chemical calculation, formation enthalpy, structure refining algorithm,

50

Quantitative Structure-Fragmentation Relationship, QSFR descriptor

51 52

ACS Paragon Plus Environment

2

Page 3 of 20

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

53

INTRODUCTION

54

One of the applications of differential mass spectrometry (∆MS) is the chemical structure

55

identification (CSI) of isomers with similar mass spectra using the quantum chemical

56

calculated (QCC) enthalpies of formation (∆fHs) of the ions formed in the mass

57

spectrometer. In this way, structures could be identified by ∆MS for groups of isomers, such

58

as position isomers of nitrobenzophenones or their dimethyl acetals,1-3 of polychlorinated

59

biphenyls (PCBs),4 or endo- and exo-diastereomers of some alcohols.5 In the same context,

60

stereochemical studies of cis- and trans-1,3-dioxane derivatives,6 or α- and β-mannofuranose

61

acetals may be mentioned.7

62

Differential techniques presented in these works use an ordering algorithm (ORD) based

63

on the principle that the more stable an ion is, the more abundant it is in the mass spectrum.

64

Applied to common isobaric ions in similar mass spectra of isomers, ORD can generate a

65

fine quantification of the correlation between the ions’ intensities with their enthalpies of

66

formation. Thus, an ascending series of intensities corresponds to a descending series of

67

enthalpies.4

68

Chemical identification by the interpretation of fragmentation schemes or library search

69

needs diagnostic ions that provide spectrum uniqueness and reduce the risk for false

70

positives.8 In contrast, ∆MS uses only the common ions in the mass spectra. In similar mass

71

spectra, diagnostic ions are most often lacking and the only analytically-exploitable

72

difference between spectra remains the difference between the intensities of common ions. In

73

this case, as shown in the studies mentioned above, CSI using QCC and ∆MS leads to good

74

results. This approach extends the investigative power of mass spectrometry beyond the use

75

of spectral libraries or interpretation of fragmentation schemes for the identification of

76

compounds with similar spectra.4,9

77

Establishing the best QCC methods and calculating series of ∆fH values for ions and

78

radicals is the first step in the strategy of QCC-∆MS analysis because these techniques

79

require thermochemical data as close as possible to the real values that govern the

80

fragmentations in the mass spectrometer. Recently, we reported a method of determining the

81

best series of ∆fH values obtained with semi-empirical QCC methods, which used the

82

simulation of CSI by the ORD algorithm of six chemical standards (TeCBs 44, 46, 52, 66,

83

74, and 77).10 The lists of probabilities of possible structural assignments were obtained by

84

the specialized software package, Chemical Structure Identification by Differential Mass

85

Spectra (CSI-Diff-MS Analysis 3.1.1; BET2 Software, Königsbrunn, Germany).9 The best

86

results were obtained using ∆fH calculated by semi-empirical methods, AM1, MINDO3 and

87

MNDO for M+ ions, RM1 for [M-Cl]+, [M-2Cl]+ ions, PM3 for [M-3Cl]+ and AM1 for [MACS Paragon Plus Environment

3

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 4 of 20

88

4Cl]+.10 Although not considered the most efficient, these calculation methods are faster than

89

other methods. But even so, we performed over 33,000 clicks for each of them, because QCC

90

software is not yet adapted for the rapid creation of thermochemical databases for a large

91

number of fragments. Yet, this is worth the effort, because these databases do not have to be

92

recalculated for each analysis. The series of calculated ∆fH values can be reused in any

93

analysis of the same analytes by MS, the only restriction referring to the mandatory

94

acquisition of spectra under the same conditions and the same instrumental setup.

95

To date, the ORD algorithm was successfully used in analyses where n possible

96

structures were attributed to n isomers with n similar mass spectra (n = 2 to 6). These cases

97

often do not correspond with actual experimental situations in which only a few standards

98

from a large group of isomers are available, and the analyte can have any of the structures

99

proposed for the unknown isomers. We recently built such an analytical situation using as

100

example the TeCBs group for which we already knew several good QCC methods10 and for

101

which there was practical importance of identifying the various isomers. PCBs are a group of

102

209 isomers with varying degree of chlorination which are important environmental11 and

103

food12 contaminants. In addition, epidemiological studies have shown that PCBs can have

104

effects on reproduction and neuro-development, thyroid system, nervous system, immune

105

system, cardiovascular system, leading to disturbances in growth, lipid metabolism, and

106

finally to diabetes and obesity.13

107

Regarding our study, the question that arises is whether ORD can correctly select the ∆fH

108

values (and implicitly the chemical structure) corresponding to the analyte, from the series of

109

values calculated for all 42 possible structures of TeCBs. Can the efficiency of the QCC-

110

∆MS methods for the 6 standards be extended to the entire group of TeCB isomers? What is

111

the minimum number of standards which are isomers with the analyte that can ensure a good

112

analysis? How can the performance of such computational method be improved using

113

complementary algorithms?

114

To answer these questions, we aimed here at investigating: i) the ORD-QCC-∆MS

115

strategy for qualitative analysis; ii) a new algorithm, LCOR-QCC-MS, based on Linear

116

CORrelation of IC with the ∆fH values; and iii) the coupling of ORD-LCOR algorithms for

117

the improvement of analytical performance and their applications for TeCBs.

118 119

EXPERIMENTAL

120

Materials

121

Reference standards of TeCBs (IUPAC no. 44, 46, 52, 66, 74, and 77) were obtained

122

from Dr. Ehrenstorfer Laboratories (Augsburg, Germany) at a concentration of 10 ng/µL in ACS Paragon Plus Environment

4

Page 5 of 20

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

123

isooctane. After appropriate dilution, a mixture containing these six isomers was prepared in

124

isooctane at a concentration of 1 ng/µL.

125 126

Mass spectrometry

127

An Agilent 6890 series gas chromatograph (Waldbronn, Germany) equipped with a DB-5

128

capillary column (30 m x 0.25 mm x 0.25 mm) was coupled to an Agilent 5973 mass-

129

selective detector (MSD) operating in electron ionization (EI) mode at 70 eV. The heated

130

zone temperature for the injector was set at 260°C, the mass spectrometer interface at 280°C,

131

the quadrupole mass analyzer at 150°C and the ion source at 230°C. Helium was used as a

132

carrier gas at a constant flow rate of 1.3 mL/min and the mixture containing the TeCBs was

133

injected in splitless mode. The oven temperature was programmed at an initial temperature

134

100°C, increased to 250°C at a rate of 22.5°C/min and then further increased to 310°C at

135

5.5°C/min, where it was held for 5 min. The MSD was operated in full-scan acquisition

136

mode between m/z 50 and 400 to obtain the Total Ion Chromatogram (TIC).

137

Automatic background subtraction was applied to obtain clean and interference-free mass

138

spectra. For each isomer, the averaged mass spectrum is obtained on identical intervals (500

139

000 – 2 000 000 relative units) of ion abundance in the front side of the chromatographic

140

peak. The average mass spectra are normalized on TIC (100%), to offer comparable

141

intensities for the common ions of the isomers.9 It is very important that all resulting mass

142

spectra are obtained under identical analytical conditions, so that the differences in intensity

143

between common isobaric ions are due exclusively to structural differences.4 In the CSI-Diff-

144

MS Analysis 3.1.1 software, the tabular mass spectra were imported in format of comma

145

separated values files, *.csv (Figure S-1).

146 147

The common ions for CSI

148

The minimum necessary number of common ions for CSI with ORD is determined based

149

on the number of isomers used in probability matrices. Thus, the maximum number of

150

isomers that can be identified with n ions of each spectrum is 2n.4 For the six TeCB isomers

151

used in this study, three common isobaric ions are enough. Using a larger number of ions

152

than the minimum required can improve the method’s selectivity if their series of ∆fH values

153

have been correctly computed. With a personal computer and the CSI-Diff-MS Analysis

154

3.1.1 software, matrices of 6 spectra x 6 structures for 5 ions can be calculated in less than 10

155

s. Larger matrices may require more powerful computers and/or longer computing periods.

156

The most suitable ions for CSI by QCC-∆MS are those with predictable structures and

157

for which correct thermodynamic data can easily be computed. For TeCBs, the ions M+, [M-

ACS Paragon Plus Environment

5

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 6 of 20

158

Cl]+, [M-2Cl]+, [M-3Cl]+ and [M-4Cl]+ and the corresponding isotopic peaks, e.g., those at

159

m/z 290, 256, 220, 185 and 150, respectively, can be used.10

160 161

The ∆fH database The formation enthalpies (∆fHs) were calculated for the molecule M, the molecular ion

162 163

M and for the common ions produced by successive loss of chlorine atoms: [M-Cl]+, [M-

164

2Cl]+, [M-3Cl]+ and [M-4Cl]+. The geometries of the molecules, ions and radicals were

165

optimized with the force field MM+ and re-optimized with the semi-empirical methods,

166

AM1, MINDO3, MNDO, PM3 and RM1,14,15 using the Restricted Hartree-Fock (RHF)

167

operators for molecules or ions and Unrestricted Hartree-Fock (UHF) for radicals or radical-

168

ions. The convergence limit of Self-Consistent-Field (SCF) was set at 10–5 and the

169

accelerated convergence procedure was used. For the optimization of the geometries, the

170

conjugated gradient method – Polak-Ribiere with a total root-mean-square (RMS) gradient

171

set at 10–2 kcal/(mol·Å) – was used, the molecule being considered in vacuum, since these

172

ions are isolated in a mass spectrometer.16 The semi-empirical molecular-orbital methods

173

(AM1, MINDO3, MNDO, PM3, and RM1) were employed as available in the HyperChem

174

8.0.10 Hypercube, Inc. software.

175 176 177

+

For the considered ions, the fragmentation enthalpies can be obtained using equation (1): ∆fH fragmentation = ∆fH (ion) + n·∆fH (Cl·) + E (electron) - ∆fH (molecule)

(1)

where n is the number of cleaved chlorine atoms (n = 0 to 4).

178

If it is accepted that the isomeric ions of TeCBs are formed by the same mechanism, the

179

terms n·∆fH (Cl·) and E (electron) most possibly have the same values for these ions. Upon

180

ordination (ORD) and linear correlation (LCOR), these terms can be neglected without

181

affecting the results in the CSI probability lists. Thus, equation (1) is transformed into (2),

182

and instead of ∆fH fragmentation, the ∆fH (relative) can be used with the same results

183

obtained from:

184

∆fH (relative) = ∆fH (ion) - ∆fH (molecule)

(2)

185

where ∆fH (relative) is the formation enthalpy of the respective ion measured above the level

186

of the molecular enthalpy. When several isomeric ionic structures result for a fragmentation,

187

∆fH minimum was used, because the corresponding ion is the most stable and has the

188

essential contribution to the ionic current (IC).4,10 From here onwards, we have used only the

189

relative minimum of the enthalpy of formation, simply given as ∆fH. The values calculated in

190

this way were imported into the ∆fH library of the CSI-Diff-MS Analysis 3.1.1 software

191

(Figure S-2, Tables S-1 and S-2).

192

ACS Paragon Plus Environment

6

Page 7 of 20

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

193

LCOR algorithm

194

The LCOR (Linear CORrelation) algorithm was designed to complement the ORD

195

algorithm, since the latter cannot estimate how close the calculated ∆fH values are to the

196

experimental ones. LCOR quantifies using a correlation coefficient (R), the inverse linear

197

relationship of the ∆fH values calculated for ions of the same type, with the corresponding

198

relative ionic currents (IC) in the similar mass spectra of isomers (Equation S-1). In this

199

paper, we used IC in place of relative intensity (RI), because its abbreviation can be mistaken

200

for the retention index (RI). The best correlation between ∆fH and IC corresponds to R = -1,

201

which means that in a correct analysis, real structures must be those that provide the best

202

inverse correlation. To run this algorithm, a minimum of three pairs of values (∆fH; IC) is

203

necessary, and consequently, the mass spectra of at least three isomers. The use of a linear

204

correlation function is justified since the ∆fH and IC values lie within narrow intervals and,

205

over a narrow range, any curve can be approximated by a straight line. The fact that a series

206

of increasing enthalpies corresponds to a series of decreasing ionic intensities4 explains the

207

inverse linearity (the negative slope of the regression line).

208 209 210

To achieve the ORD-LCOR coupling, R was converted into probability using the equation (3): PLCOR(%) = 100(1-R)/2

(3)

211

The probability corresponding to the coupling, PORD-LCOR, expresses the degree of

212

simultaneous fulfillment of the conditions of correct succession and proximity (correlation)

213

of the experimental values (IC) to the corresponding theoretical ones (∆fH). PORD-LCOR was

214

calculated with equation (4):

215 216 217 218

PORD-LCOR = PORD · PLCOR

(4)

The total probability (P) results from the simultaneous fulfillment of the conditions ORD and LCOR for all the common ions (m) considered, according to equation (5): P = P1·P2·……·Pm

(5)

219 220

The similarity of mass spectra

221

The similarities of the mass spectra of the standards and analyte, computed from their

222

degrees of overlapping,9 are involved in the selection of the algorithm and of the standards

223

that should be used for the optimum filtration of the structures. After importing the mass

224

spectra, the CSI-Diff-MS program can display the table of similarity that includes the values

225

for all possible pairs of spectra (Figure S-3). For the six TeCB congeners used in this study,

226

the similarity ranges between 76% and 96%. We can distinguish two groups of high

ACS Paragon Plus Environment

7

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 8 of 20

227

similarity: group 1 (PCB 44, 46, and 52) and group 2 (PCB 66, 74, and 77) with values

228

ranging between 92% and 96%.10

229 230

CSI probabilities list

231

We have calculated the probabilities of the CSI list using the CSI-Diff-MS 3.1.1

232

software, the database of enthalpies of formation (∆fH) calculated by HyperChem 8.0.10 and

233

the mass spectra generated by EI-MS (see above). The LCOR algorithm was ran with

234

experimental subroutines of CSI-Diff-MS. With the ORD and LCOR algorithms, the

235

experimental and calculated data are compared in all possible variants of structural

236

assignments and the results are presented as a list of decreasing probabilities of these variants

237

(Figure S-4).9

238

Although the relative error, accuracy and precision are used mainly in quantitative

239

analysis, these analytical parameters can be calculated for each of these probabilities lists

240

using the following equations:

241

Relative error (%) = 100·∆rank S / N

(6)

242

Accuracy (%) = 100(N - ∆rank S) / N

(7)

243

Selectivity (%) = 100(N - rank S + rank P) / N

(8)

244

where rank S is the rank of the real structure in the list of probabilities or the number of

245

structures with probability greater than or equal to that of the real structure, ∆rank S = (rank

246

S – true rank), true rank = 1, rank P represents the number of distinct probabilities among

247

first rank S positions of the list, N is the number of possible structures for the analyte. For the

248

ORD lists, rank S > rank P because several structures offer most often the same probability.

249

For the LCOR lists, rank S = rank P because two structures with the same probability cannot

250

exist. Precision is estimated by the probabilities in the lists.

251 252

Determining the best series of enthalpies

253

The best series of ∆fH values, calculated with various QCC methods, can be determined

254

by simulating the CSI by the studied algorithm, for standards from the respective group of

255

isomers. The ∆fH values which lead to the correct assignment of the structures with the

256

highest rank, selectivity and probability were found to be the most appropriate. For a

257

accurate, selective and precise CSI, even mixtures of ∆fH values calculated by different

258

methods can be used, provided that the same method is used for a certain ion type.10

259 260

ACS Paragon Plus Environment

8

Page 9 of 20

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

261

RESULTS AND DISCUSSION

262

Optimal semi-empirical method for LCOR

263

ORD and LCOR are operationally different and it would be useful to know if there are

264

series of ∆fH values that could provide the best CSI simultaneously for the two algorithms

265

and the ions used. In this case, the same set of values could be used to run both algorithms,

266

and the analysis time would be substantially reduced. The scores obtained for LCOR, shown

267

comparatively with those of ORD in Table 1, are encouraging because it can be noted that

268

there are ∆fH series calculated by two methods (bold/grey formatted), AM1 for ions M+, [M-

269

4Cl]+ and RM1 for [M-Cl]+, [M-2Cl]+, which provides good results for both algorithms.

270

Further in the study, we used the optimal series of ∆fH values corresponding to these two

271

methods for all TeCB isomers. The ∆fH series for M+ resulting from MINDO3, MNDO and

272

PM3 are only slightly better than those of AM1. However, their use would not justify the

273

computational load for three methods. Since LCOR did not provide good results for the ion

274

[M-3Cl]+ by any of the semi-empirical methods, it has not been used.

275 276

CSI with ORD-QCC-∆MS

277

The study of the analytical performance of ORD was carried out by simulating the CSI of

278

an analyte of known structure, using a decreasing number of isomeric standards s = 5, 4, 3,

279

and 2. The role of the standards and analyte was played in turn by all the 6 TeCBs. Only for

280

CSI with 5 standards, all six possible cases were run, while for the other situations when less

281

than 5 standards were used, representative variants were selected. The analyte was sought

282

among the 42-s possible isomeric structures. The selection of its structure was achieved by

283

establishing by ORD the most appropriate ∆fH values for the ions intensities in its mass

284

spectrum.

285

The results in the CSI probability lists obtained by ORD are shown in Tables 2, 3, 4 and

286

5, column 3. The first number is the rank of the probability corresponding to the correct

287

structure (rank P). The second is the number of isomers that have a probability greater than

288

or equal to that of the correct structure (rank S) or confounding structures number. The third

289

number is the probability (PORD) corresponding to the degree of ordering of the values ∆fH-

290

IC. The best result < rank P / rank S / P > is < 1 / 1 / 100% >.

291

For each list, the relative error, accuracy and selectivity were calculated using the

292

equations (6), (7) and (8). Their average values, grouped in three situations determined by the

293

similarity of the spectra of the standards and analyte (cases a, b and c), are given in Table 6.

294

The three analytical situations which can occur are:

ACS Paragon Plus Environment

9

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

295 296

Page 10 of 20

(a) when both the standards and the analyte have mass spectra with high similarities (e.g. above 92%), that is, they belong to the same similarity group;

297

(b) the standards belong to one similarity group, and the analyte to the other;

298

(c) the standards come from both similarity groups, having diverse similarities (e.g. 75-

299

96%).

300

Certainly, the calculated analytical parameters can give an overview of the quality or

301

shortcomings of the analysis, but what matters is whether the filtering algorithm succeeds in

302

bringing the real structure on the first place or at least among the first places of the list. In

303

other words, rank S is the most important parameter for CSI.

304

For ORD, selectivity is the one that limits the performance. It increases in the order (b)
6 for the six standards),4 filtering of structures with ORD, LCOR and their coupling was

355

much more efficient. Thus, the ORD-LCOR structural filter could successfully complement

356

the library search of isomers with similar mass spectra. Since differential algorithms do not

357

use databases of standard spectra, but databases of ∆fH calculable for all isomeric structures,

358

important steps toward a de novo structural analysis are made, which would require minimal

359

pre-knowledge.17 The CSI-Diff-MS software platform offers the possibility to perform this

360

type of analysis with ORD and LCOR.

361 362

CONCLUSIONS

363

Although this study was conducted on TeCBs using EI-MS spectra and semi-empirical

364

QCC methods, there is no reason for us to believe that the kinetic and thermodynamic laws ACS Paragon Plus Environment

11

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 12 of 20

365

of fragmentation confirmed as implicitly functional on this exercise, do not occur and cannot

366

be pursued for analytical purposes also for other groups of isomers, using other instruments,

367

conditions or QCC methods. Thus, we can consider that the CSI by QCC-MS-∆MS strategy

368

with ORD and LCOR algorithms described here has a high degree of generality. In this

369

study, ∆fH was shown to be a good descriptor of fragmentations, and the differential

370

techniques capable of relating the relative ionic current to the chemical structures of analytes,

371

thus representing true Quantitative Structure-Fragmentation Relationship methods.

372 373

Acknowledgments

374

The authors acknowledge BET2 Software for the LCOR experimental subroutines of the

375

CSI-Diff-MS software. Part of this work was supported by the Romanian National Authority

376

for Scientific Research (CNCS-UEFISCDI) through project PN-II-PCCA-2011-142.

377 378

Supporting Information Available

379

This material is available free of charge via the Internet at http://pubs.acs.org

380 381

References

382

(1) Dinca, N.; ȘiȘu, E.; ȘiȘu, I.; Oprean, I.; Csunderlik, C.; Mracec, M. Rev. Roum. Chim.

383 384 385 386 387

2002, 47(3-4), 379-385. (2) Dinca, N.; ȘiȘu, E.; ȘiȘu, I.; Csunderlik, C.; Oprean. I. Rev. Chim.-Bucharest 2002, 53(5), 332-336. (3) Dinca, N.; ȘiȘu, E.; Mracec, M.; Oprean, I.; Sander, O. Rev. Roum. Chim. 2004, 49(3-4), 331-338.

388

(4) Dinca, N. In Applications of Mass Spectrometry in Life Safety; Popescu, C.; Zamfir, A.D.;

389

Dinca N., Ed.; NATO Public Diplomacy Division & Springer: Dordrecht, 2008; pp. 221-

390

233.

391 392

(5) Dinca, N.; Stanescu, M.D.; ȘiȘu, E.; Mracec, M. Rev. Chim.-Bucharest 2004, 55(5), 347-350.

393

(6) Harja, F.; Bettendorf, C.; Grosu, I.; Dinca, N. In Applications of Mass Spectrometry in

394

Life Safety; Popescu, C.; Zamfir, A.D.; Dinca N., Ed.; NATO Public Diplomacy

395

Division & Springer: Dordrecht, 2008; pp. 185-191.

396 397 398

(7) Rafaila, M.; Pascariu, M.C.; Gruia, A.; Penescu, M.; Purcarea, V.L.; Medeleanu, M.; Rusnac, L.M.; Davidescu, C.M. Farmacia, 2013, 61(1), 116-126. (8) Stein, S. Anal. Chem. 2012, 84(17), 7274−7282.

ACS Paragon Plus Environment

12

Page 13 of 20

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

399 400

(9) Patent Number: DE102005028944-A1, Assignee: C. Bettendorf, Inventors: C. Bettendorf, N. Dinca, 2007; http://www.bet2-soft.de

401

(10) Dinca, N.; Covaci, A. Rapid Commun. Mass Sp. 2012, 26(17), 2033-2040.

402

(11) Covaci, A.; Gheorghe, A.; Voorspoels, S.; Maervoet, J.; Steen Redekker, E.; Blust, R.;

403

Schepens, P. Environ. Int. 2005, 31(3), 367-375.

404

(12) Voorspoels, S.; Covaci, A.; Neels, H. Environ. Toxicol. Pharm. 2008, 25(2), 179-182.

405

(13) Hamers, T.; Kamstra, J. H.; Cenijn, P. H.; Pencikova, K.; Palkova, L.; Simeckova, P.;

406

Vondracek, J.; Andersson, P. L.; Stenberg, M.; Machala, M. Toxicol. Sci. 2011,

407 408 409

121(1), 88–100. (14) Dewar, M.J.S.; Zoebisch, G.E.; Healy, F.E.; Stewart, J.J.P. J. Am. Chem. Soc. 1985, 107(13), 3902-3909.

410

(15) Stewart, J.J.P. J. Comput. Aid. Mol. Des. 1990, 4(1), 1-103.

411

(16) Holmes, J.L.; Aubry, C.; Mayer, P.M. Assigning Structures to Ions in Mass

412 413

Spectrometry; CRC Press: Boca Raton, 2006. (17) Kind, T.; Fiehn, O. Bioanal. Rev. 2010, 2(1-4), 23-60.

414 415

ACS Paragon Plus Environment

13

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 14 of 20

416

Table 1. Ranking of semi-empirical QCC methods for the ions of TeCB standards based on

417

the scores obtained using the ORD and LCOR algorithms of the CSI-Diff-MS software. For

418

ORD, a higher number of stars corresponds to a higher degree of confidence in the calculated

419

∆fH series.10 For LCOR, rank S and probability (rank S / PLCOR%) are shown. The calculated

420

∆fH values are the better the closer rank S (out of 720 possible variants) is to 1, and the

421

probability to 100%. The optimal ∆fH series used in this paper are formatted grey.

422 Semi-empirical method AM1 MINDO3

MNDO PM3 RM1

ORD-∆MS LCOR-MS ORD-∆MS LCOR-MS ORD-∆MS LCOR-MS ORD-∆MS

M+·

[M-Cl]+

[M-2Cl]+

[M-3Cl]+

[M-4Cl]+

** 11 / 98.8%

11 / 94.4% ** 9 / 94.6% -

58 / 76.4% -

** 1 / 99.8%

** 9 / 98.5%

* 55 / 82.0% 38 / 88.0% -

230 / 64.9%

** 10 / 96.8%

LCOR-MS ORD-∆MS

* 7 / 95.0% -

LCOR-MS

32 / 94.2%

12 / 94.3%

67 / 86.2%

* 68 / 75.7%

* 10 / 94.3%

* 11 / 95.9%

** 34 / 77.1%

*** 4 / 94.9%

**** 1 / 99.5%

* 85 / 74.2%

* 19 / 89.1% 206 / 65.0% 151 / 69.0% 134 / 75.0%

423 424

ACS Paragon Plus Environment

14

Page 15 of 20

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

425

Table 2. Results of structural filtering using five TeCB standards and the calculated ∆fH

426

database. For each of the 6 variants of selecting the standard and analyte out of the 6

427

possible, there are N = 42 – 5 = 37 possible structures for the analyte. The first number

428

represents rank P, the second is rank S (confounding structures), and the third is probability,

429

(rank P / rank S / P).

430 5 standards

Analyte

Case

1 PCBs 46, 52, 66, 74, 77 PCBs 44, 52, 66, 74, 77 PCBs 44, 46, 66, 74, 77 PCBs 44, 46, 52, 74, 77 PCBs 44, 46, 52, 66, 77 PCBs 44, 46, 52, 66, 74

2 PCB 44 PCB 46 PCB 52 PCB 66 PCB 74 PCB 77

3 c c c c c c

ORD Algorithm 4 1 /1/ 95% 1 /2/ 95% 1 /1/ 95% 1 /2/ 95% 1 /2/ 95% 1 /1/ 95%

LCOR Algorithm 5 2 /2/ 93.2% 1 /1/ 93.2% 1 /1/ 93.2% 3 /3/ 93.2% 1 /1/ 93.2% 1 /1/ 93.2%

ORD-LCOR Coupling 6 1 /1/ 88.5% 1 /1/ 88.5% 1 /1/ 88.5% 1 /1/ 88.5% 1 /1/ 88.5% 1 /1/ 88.5%

431 432 433

ACS Paragon Plus Environment

15

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 16 of 20

434

Table 3. Results < rank P / rank S / P > of the analysis with four TeCB standards using the

435

calculated ∆fH database. A number of 6 variants of the standards and the analyte were

436

selected from 30 possible. For each variant there are N = 42 – 4 = 38 possible structures for

437

the analyte.

438 4 standards

Analyte

Case

1

2 PCB 74 PCB 77 PCB 44 PCB 74 PCB 44 PCB 52

3 c c c c c c

PCBs 44, 46, 52, 66 PCBs 46, 52, 66, 77 PCBs 46, 66, 74, 77

ORD Algorithm 3 1 /2/ 95% 1 /1/ 97% 1 /1/ 97% 1 /2/ 95% 1 /1/ 95% 1 /1/ 95%

LCOR Algorithm 4 1 /1/ 92.1% 1 /1/ 92.7% 2 /2/ 92.1% 1 /1/ 92.4% 2 /2/ 98.2% 1 /1/ 92.4%

ORD-LCOR Coupling 5 1 /1/ 88.1% 1 /1/ 89.3% 1 /1/ 89.3% 1 /1/ 87.8% 1 /1/ 93.3% 1 /1/ 87.8%

439 440

ACS Paragon Plus Environment

16

Page 17 of 20

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

441

Table 4. Results < rank P / rank S / P > of the analysis with three TeCB standards using the

442

calculated ∆fH database. A number of 12 variants of the standards and the analyte were

443

selected from 60 possible. For each variant there are N = 42 – 3 = 39 possible structures for

444

the analyte.

445 3 standards

Analyte

Case

1

2 PCB 66 PCB 74 PCB 77 PCB 52 PCB 74 PCB 77 PCB 46 PCB 52 PCB 66 PCB 44 PCB 46 PCB 52

b b b c c c c c c b b b

PCBs 44, 46, 52

PCBs 44, 46, 66

PCBs 44, 74, 77

PCBs 66, 74, 77

ORD Algorithm 3 1 /25/ 95% 1 /25/ 95% 1 /25/ 95% 1 /1/ 95% 1 /2/ 95% 1 /1/ 100% 1 /3/ 95% 1 /7/ 91% 1 /2/ 91% 1 /20/ 91% 1 /13/ 91% 1 /20/ 91%

LCOR Algorithm 4 6 /6/ 90.3% 4 /4/ 90.7% 2 /2/ 89.0% 1 /1/ 90.3% 1 /1/ 98.5% 1 /1/ 98.6% 1 /1/ 98.9% 1 /1/ 91.0% 3 /3/ 97.5% 5 /5/ 97.5% 3 /3/ 97.8% 3 /3/ 96.5%

ORD-LCOR Coupling 5 6 /6/ 85.8% 4 /4/ 86.2% 2 /2/ 84.5% 1 /1/ 85.8% 1 /1/ 93.5% 1 /1/ 98.6% 1 /1/ 93.9% 1 /1/ 82.8% 1 /1/ 88.7% 5 /5/ 88.7% 3 /3/ 89.0% 3 /3/ 87.8%

446 447 448

ACS Paragon Plus Environment

17

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 18 of 20

449

Table 5. Results < rank P / rank S / P > of the analysis with two TeCB standards using the

450

calculated ∆fH database. A number of 20 variants of the standards and the analyte were

451

selected from 60 possible. For each analytical variant there are N = 42 - 2 = 40 possible

452

structures for the analyte.

453 2 standards

Analyte

Case

1

2 PCB 52 PCB 66 PCB 74 PCB 77 PCB 46 PCB 52 PCB 74 PCB 77 PCB 44 PCB 52 PCB 66 PCB 77 PCB 44 PCB 46 PCB 52 PCB 66 PCB 46 PCB 44 PCB 77 PCB 74

a b b b c c c c c c c c b b b a a a a a

PCBs 44, 46

PCBs 44, 66

PCBs 46, 74

PCBs 74, 77 PCBs 44, 52 PCBs 46, 52 PCBs 66, 74 PCBs 66, 77

ORD Algorithm 3 1 /1/ 91% 1 /28/ 100% 1 /28/ 100% 1 /28/ 100% 1 /3/ 100% 1 /14/ 91% 1 /2/ 91% 1 /1/ 100% 1 /1/ 100% 1 /1/ 100% 1 /14/ 91% 1 /14/ 91% 1 /20/ 91% 1 /20/ 91% 1 /21/ 91% 1 /2/ 83% 1 /2/ 91% 1 /1/ 91% 1 /1/ 83% 1 /2/ 83%

LCOR Algorithm 4 1 /1/ 54.8% 5 /5/ 99.1% 4 /4/ 99.1% 2 /2/ 99.6% 2 /2/ 99.1% 1 /1/ 89.2% 1 /1/ 98.7% 2 /2/ 98.2% 2 /2/ 99.1% 1 /1/ 91.2% 2 /2/ 98.5% 1 /1/ 99.0% 6 /6/ 98.7% 4 /4/ 99.0% 5 /5/ 97.4% 11 /11/ 51.3% 3 /3/ 54.8% 1 /1/ 54.8% 1 /1/ 51.3% 1 /1/ 51.3%

ORD-LCOR Coupling 5 1 /1/ 49.9% 5 /5/ 99.1% 4 /4/ 99.1% 2 /2/ 99.6% 1 /1/ 99.1% 1 /1/ 81.2% 1 /1/ 89.8% 1 /1/ 98.2% 1 /1/ 99.1% 1 /1/ 91.2% 1 /1/ 89.6% 1 /1/ 90.1% 6 /6/ 89.8% 4 /4/ 90.1% 5 /5/ 88.6% 8 /8/ 88.6% 3 /3/ 49.9% 1 /1/ 49.9% 1 /1/ 42.6% 1 /1/ 42.6%

454 455 456

ACS Paragon Plus Environment

18

Page 19 of 20

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

457

Table 6. Averages of the analytical parameters corresponding to the ORD, LCOR algorithms

458

and their coupling in the 3 cases of similarity of the standards and analyte, for the variants

459

shown in Tables 2-5. The values formatted bold/grey correspond to the analytical variants

460

recommended for the analysis of TeCBs.

461 Case

Similarity of standards

2 a b c a b c

3 high high various high high various

Similarity of the analyte with the standards 4 high small various high small various

Average selectivity (%)

a b c

high high various

Average precision (%)

a b c a b c

Analytical parameter 1 Average relative error (%) Average accuracy (%)

Rank S interval

ORD Algorithm

LCOR Algorithm

ORDLCOR Coupling

5 1.25 55 6.3 98.75 45 93.7

6 5 7.8 1.1 95 92.2 98.9

7 3.75 7.8 0 96.25 92.2 100

high small various

98.8 45 94.5

100 100 100

100 100 100

high high various

high small various

87 94.2 95.2

53 96.2 94.9

54.0 96.2 90.3

high high various

high small various

1-2 13-28 1-14

1-11 2-6 1-3

1-8 2-6 1

462

ACS Paragon Plus Environment

19

Analytical Chemistry

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 ACS Paragon Plus Environment

Page 20 of 20