Premium
Multidimensional statistical analysis applied to electron ionization mass spectra to determine steroid stereochemistry
Author(s) -
Le Bizec Bruno,
Antignac JeanPhilippe,
Bertrand Dominique,
Qannari El Mostafa,
Andre François
Publication year - 2005
Publication title -
rapid communications in mass spectrometry
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.528
H-Index - 136
eISSN - 1097-0231
pISSN - 0951-4198
DOI - 10.1002/rcm.1817
Subject(s) - chemistry , mass spectrum , linear discriminant analysis , ionization , mass spectrometry , electron ionization , analytical chemistry (journal) , spectral line , principal component analysis , chromatography , ion , statistics , mathematics , organic chemistry , physics , astronomy
The differentiation of stereoisomers on the basis of their mass spectra only is usually a difficult challenge even when an informative ionization technique such as electron ionization is used; this is particularly the case for steroids. In this study, multivariate statistical techniques have been applied to the mass spectra of derivatized 5 ξ ‐androstane‐3 ξ ,17 ξ ‐diols ( ξ = α , β ) in order to investigate the possibility of discrimination among the different isomers. After collection of the data from the mass spectra (20 replicates for each of the 8 isomers), each ion was considered as a statistical variable and each mass spectrum as an observation. The more discriminative variables (42 out of the 160 initial ones) were selected using the analysis of variance technique (ANOVA). Thereafter, a linear discriminant analysis (LDA) allowed us to set up a predictive model for stereochemistry determination. The two‐dimensional graphical display of the 160 observations on the basis of the canonical variables derived from LDA made it possible to separate the eight isomers. The discrimination of 5 α and 5 β isomers as well as 3 α and 3 β was unambiguous, whereas, the discrimination of 17 α and 17 β epimers was less obvious. The robustness of the model was checked with 40 mass spectra recorded over a 6‐month period on different quadrupole mass spectrometers and under different signal acquisition conditions. The percentage of correct assignment of these ‘unknown’ stereoisomers was 92%; only three 17 α and 17 β epimers were not correctly plotted in the expected zone. Nevertheless, the performance score was better than those observed with traditional mass spectral libraries. Furthermore, this statistical approach allowed us to identify the main fragment ions involved in the discrimination between isomers: m/z 256 and 421 for isomers 5 a ‐5 b ; m/z 241 and 331 for isomers 5 α 3 α ‐5 α 3 β ; m/z 143 and 162 for isomers 5 β 3 α ‐5 β 3 β ; and m/z 255 for epimers 17 α ‐17 β . Copyright © 2005 John Wiley & Sons, Ltd.