RAPID COMMUNICATIONS IN MASS SPECTROMETRY

Rapid Commun. Mass Spectrom. 13, 279–284 (1999)

Deconvolution Gas Chromatography/Mass

Spectrometry of Urinary Organic Acids –
Potential for Pattern Recognition and Automated
Identification of Metabolic Disorders†
John M. Halket1*,2,3, Anna Przyborowska1, Stephen E. Stein4, W. Gary Mallard4, Stephen Down5
and Ronald A. Chalmers6
1
Specialist Bioanalytical Services, Centre for Chemical Sciences, Royal Holloway, University of London, Egham TW20 0EX,
UK
2
Department of Computer Science, Royal Holloway, University of London, Egham TW20 0EX, UK
3
Department of Chemical Pathology, Imperial College of Science, Technology and Medicine, Charing Cross Hospital, London
W6 2HS, UK
4
Physical and Chemical Properties Division, National Institute of Standards and Technology, Gaithersburg, MD 20899, USA
5
HD Science Ltd., 16 Petworth Avenue, Toton, Nottingham NG9 6JF, UK
6
Paediatric Metabolism Unit, Department of Child Health, St. George’s Hospital Medical School, London SW17 0RE, UK

The National Institute of Standards and Technology (NIST) Automated Mass Spectral Deconvolution and
Identification System (AMDIS) is applied to a selection of data files obtained from the gas chromatography/
mass spectrometry (GC/MS) analysis of urinary organic acids. Mass spectra obtained after deconvolution
are compared with a special user library containing both the mass spectra and retention indices of
ethoxime-trimethylsilyl (EO-TMS) derivatives of a set of organic acids. Efficient identification of
components is achieved and the potential of the procedure for automated diagnosis of inborn errors of
metabolism and for related research is demonstrated. Copyright # 1999 John Wiley & Sons, Ltd.
Received 20 September 1998; Revised 27 November 1998; Accepted 10 December 1998

Urinary organic acid profiling by GC/MS is commonly matching algorithms, adjacent peak deconvolution and
carried out to diagnose, confirm and monitor metabolic background subtraction as well as retention index (RI)
disorders in newborns and children. A number of serious comparison. The method provides unambiguous identifica-
conditions are indicated by raised levels of normal tion together with a quantitative indication of the metabolite
components or by the presence of abnormal metabolites.1 levels.3
In many cases, the components are identified by skilled
personnel using a laborious semi-automatic procedure,
sometimes assisted by adapted target compound software, EXPERIMENTAL
where available on the particular system employed. In many
laboratories, over-reliance is placed upon the use of library Sample preparation
search routines and identifications and thus misinterpreta-
Urine samples (equivalent to 5 mmol of creatinine) from
tions and false or missed diagnoses occur. This has become
control infants and children without metabolic disease and
particularly apparent through the participation in external
from patients with a selection of inborn errors of
quality assurance (EQA) schemes of increasing numbers of
metabolism (organic acidurias) – methylmalonic aciduria,
clinical biochemistry laboratories undertaking GC/MS
propionic acidaemia, glutaric aciduria type I, 3-hydroxy-3-
diagnostic work with limited experience and the relatively
methylglutaric aciduria, isovaleric acidaemia and medium-
poor performance of some of these laboratories.2 There is,
chain acyl CoA dehydrogenase (MCAD) deficiency – were
therefore, an increasing need for high quality and unambig-
placed in tubes containing a few crystals of ethoxyamine
uous software-based interpretative assistance.
HCl (to form ethoximes and minimize keto-enol tautomer-
An automated procedure is presented for this purpose: the
ism), made up to 1.0 mL with deionized water (if necessary)
National Institute of Standards and Technology (NIST)
and the pH adjusted to 1.0–1.5 using 50% HCl. After
Automated Mass Spectral Deconvolution and Identification
saturation with NaCl, samples were extracted with 2 mL of
System (AMDIS) designed for Chemical Weapons Con-
ethyl acetate (vortexing 1 min) and centrifuged at 2000 rpm
vention treaty compliance. This software provides decon-
for 5 min at ambient temperature. The upper organic layer
volution, quality matching using advanced spectral
was transferred to a second tube and the procedure repeated
twice. The organic extracts were evaporated to dryness
under a gentle stream of nitrogen, 40 mL of pyridine were
*Correspondence to: J. M. Halket, Specialist Bioanalytical Services, added and the mixture was vortexed and allowed to stand for
Centre for Chemical Sciences, Royal Holloway, University of London,
Egham, Surrey TW20 9LZ, UK. 10 min. 10 mL of a 20 mg/mL tetracosane/hexacosane
† Any mention of commercial products in this paper is for information internal standard mixture were added followed by 200 mL
only; it does not imply recommendation or endorsement by NIST. of N,O-bis(trimethylsilyl)trifluoroacetamide (BSTFA). The

CCC 0951–4198/99/040279–06 $17.50 Copyright # 1999 John Wiley & Sons, Ltd.
280 DECONVOLUTION GC/MS OF URINARY ACIDS

reagent volumes were adjusted to allow for variations in together with the added hydrocarbons. All compounds
sample quantities: need not be found. This step created a text file
containing the retention time (RT) and RI values for
Urine creatinine Pyridine Internal standard BSTFA the identified compounds which was then used to
equivalent added added added
calculate the RI values for any component found in
>5 mmoL 40 mL 10 mL 200 mL
step 2, identified or not.
<5 mmoL 20 mL 5 mL 100 mL
<1 mmoL 10 mL 5 mL 50 mL The location of all six standards in a GC/MS
chromatogram by the AMDIS program is illustrated
The tubes were stoppered, vortexed and left overnight or in Fig. 1.
for a minimum of 2 h to derivatize. Contents were
transferred to vials fitted with 200 mL inserts. (2) the mass spectra contained in the data file were then
compared with a special user library extended from 33
mass spectra used in earlier work3 to 111 mass spectra
GC/MS
of the EO-TMS derivatives of metabolites associated
Instrument: Hewlett-Packard 5970B MSD mass spec- with a number of organic acidurias. RI values were
trometer coupled to a HP 5890A gas chromatograph. 4 mL calculated for all components and the difference
split injection (1:70), injection port temperature 230 °C, between the RI value found and that in the library
interface 280 °C. CP-Sil 5 capillary column (25 m  was reported and used to apply a match factor penalty
0.15 mm i.d., 0.12 mm film thickness), program: initial oven to improve the identification.
temperature 90 °C for 1 min, then programmed to increase
to 270 °C at a rate of 10 °C/min and maintained at this Deconvolution was carried out by a new method (S. E.
temperature for 7 min. The electron multiplier voltage was Stein, to be published) incorporating several improvements
2000 V and the instrument was scanned from m/z 50–700 at over the model peak approach4 which extracts ion profiles
1 scan/s. having similar peak shapes. In this original model, the shape
of the most prominent of the maximizing ion chromato-
grams is taken to represent the shape of the actual
Reagents component. Ion chromatograms having this shape are
Ethyl acetate (BDH, HiPerSolv grade), pyridine (BDH, extracted by a simple least-squares procedure. The principal
Aristar grade) and BSTFA (Pierce) were stored desiccated defect of this method is its inability to extract weak signals
at 0–4 °C. All standards were stored at ÿ20 °C. 20 mg/mL reliably because it has no means of effectively establishing
internal standard: 0.2 g of tetracosane (Supelco) and 0.2 g of thresholds for distinguishing signal from noise. The
hexacosane (Fluka) made up to 10 mL in heptane (BDH, improved method incorporates noise analysis in order to
Aristar grade). Stored at 0–4 °C. detect such small signals. A noise factor is determined for
each data file by analysis of the signals in regions of
Data file post-processing by AMDIS constant signal intensity. Components are then perceived by
the scan-to-scan maxima that their ions produce and they are
GC/MS data files were subjected to analysis by AMDIS identified by an optimized spectrum comparison routine
(available with the NIST98 Mass Spectral Database, see which can incorporate GC retention index data. The overall
acknowledgements) in external calibration standard mode. process involves the sequential steps of noise analysis,
In this case, the data files were processed in 2 steps: component perception, mass spectral deconvolution and
compound identification.
(1) the data file was calibrated for retention indices (RI) by Results were output as a customizable list showing
comparison of all spectra with a special library compound name, retention index difference found, a variety
containing the mass spectra and RI values of selected of mass spectral match parameters and an indication of the
compounds normally found in each profile: lactic acid, relative amount of each component (% total signal). The list
phosphoric acid, pyroglutamic acid and citric acid was transferred to a spreadsheet and the amounts of all

Figure 1. Part of AMDIS screen showing automatic location of retention index calibration standards in a
urinary organic acid GC/MS profile: (ethoxime)-trimethylsilyl derivatives of lactic acid (A), phosphoric acid
(B), pyroglutamic acid (C), citric acid (D), added n-tetracosane (E) and added n-hexacosane (F).

Rapid Commun. Mass Spectrom. 13, 279–284 (1999) Copyright # 1999 John Wiley & Sons, Ltd.
 DECONVOLUTION GC/MS OF URINARY ACIDS 281

Table 1. Names and retention index values (CP-Sil 5) for “important” compounds selected for final analysis in data matrix. The compounds
marked (*) can be weighted to increase their numerical significance

LACTIC ACID diTMS (1060) *ISOVALERYLGLYCINE diTMS (1510)

2-HYDROXYISOBUTYRIC ACID diTMS (1070) ADIPIC ACID diTMS (1525)
GLYCOLIC ACID diTMS (1072) *TIGLYLGLYCINE TMS (1530)
PYRUVIC ACID EO-TMS (1100) PYROGLUTAMIC diTMS (1540)
OXALIC ACID diTMS (1111) *TIGLYLGLYCINE diTMS (1550)
2-HYDROXYBUTYRIC ACID diTMS (1130) 2-HYDROXYPHENYLACETIC ACID diTMS (1560)
3-HYDROXYPROPIONIC ACID diTMS (1145) *3-METHYLCROTONYLGLYCINE diTMS (1560)
3-HYDROXYISOBUTYRIC ACID diTMS (1160) 3-HYDROXYPHENYLACETIC ACID diTMS (1600)
3-HYDROXYBUTYRIC ACID diTMS (1160) *HEXANOYLGLYCINE TMS (1610)
2-HYDROXYISOVALERIC ACID diTMS (1178) 3-HYDROXY-3-METHYLGLUTARIC ACID triTMS (1610)
MALONIC ACID diTMS (1190) 2-HYDROXYGLUTARIC ACID triTMS (1615)
3-HYDROXYISOVALERIC ACID diTMS (1210) *ACETYLASPARTIC ACID diTMS (1640)
2-METHYL-3-HYDROXYBUTYRIC ACID diTMS (1210) *HEXANOYLGLYCINE diTMS (1650)
METHYLMALONIC ACID diTMS (1230) *SUCCINYLACETONE EO-TMS (1650)
4-HYDROXYBUTYRIC ACID diTMS (1230) 4-HYDROXYPHENYLACETIC ACID triTMS (1665)
3-HYDROXY-n-VALERIC ACID diTMS (1240) *ACETYLASPARTIC ACID triTMS (1680)
N-ACETYLGLYCINE TMS (1250) SUBERIC ACID diTMS (1730)
ETHYLMALONIC ACID diTMS (1280) *OROTIC ACID triTMS (1740)
GLYCEROL triTMS (1300) HOMOVANILLIC ACID diTMS (1760)
SUCCINIC ACID diTMS (1314) *HOMOGENTISIC ACID triTMS (1850)
METHYLSUCCINIC ACID diTMS (1336) *METHYLCITRIC ACID (I) diTMS (1888)
GLYCERIC ACID triTMS (1350) *METHYLCITRIC ACID (II) diTMS (1893)
PROPIONYLGLYCINE TMS (1360) VANILMANDELIC ACID triTMS (1908)
MEVALONOLACTONE TMS (1360) SEBACIC ACID diTMS (1920)
FUMARIC ACID diTMS (1365) 4-HYDROXYPHENYLLACTIC ACID triTMS (1928)
*5-HYDROXYCAPROIC ACID diTMS (1370) 4-HYDROXYPHENYLPYRUVIC ACID EO-TMS (1945)
GLUTARIC ACID diTMS (1400) *3-PHENYLPROPIONYLGLYCINE diTMS (1950)
*BUTYRYLGLYCINE TMS (1420) GALACTITOL hexaTMS (1980)
*PROPIONYLGLYCINE diTMS (1420) 4-HYDROXYPHENYLPYRUVIC ACID diTMS (2060)
3-METHYLGLUTARIC ACID diTMS (1450) 3-HYDROXYSEBACIC ACID triTMS (2087)
3-METHYLGLUTACONIC ACID diTMS (I) (1450) *N-ACETYLTYROSINE diTMS (2110)
*ISOVALERYLGLYCINE TMS (1460) *SUBERYLGLYCINE triTMS (2175)
*BUTYRYLGLYCINE diTMS (1470) *SUBERYLGLYCINE diTMS (2210)
3-METHYLGLUTACONIC ACID diTMS (I) (1495) TETRACOSANE (2400)

identified components (columns) normalized to the reported which do not contribute to the diagnostic process. The
abundance of the n-tetracosane internal standard. The data derivatives of significant or diagnostic metabolites corre-
matrices thus obtained were subjected to statistical analysis sponding to the remaining rows are listed in Table 1.
either in raw form or after editing by removal of rows Alternatively, weighting (generally 10x) was applied to
corresponding to components such as citric acid or urea rows corresponding to components such as the acylglycines,
of great diagnostic significance but which can be present at
relatively low levels (marked by * in Table 1) particularly in
the asymptomatic patient or one in remission.
The importance of RI values is illustrated in Fig. 2 which
compares the TMS derivatives of methylmalonic (A,
important metabolite) and succinic (B) acid. The spectra
are difficult to distinguish by simple library matching, RI
comparison being essential.
Computing was carried out on 200 MHz Pentium and 400
M Hz Pentium II computers. Principal components analysis
(PCA) was performed using Cornerstone for Windows
Release 2.1 (BBN Corporation, USA) on these computers or
on a DEC Alpha 2100 using a standard library PCA routine.

RESULTS
Examples of mass spectral deconvolution are shown in Figs
3 and 4. Figure 3 shows a GC/MS peak containing
ethylmalonic acid covered by an excess of phosphoric acid
(A). The mass spectrum of ethylmalonic acid (B) maximiz-
ing at a retention time of 7.373 min is extracted from the
Figure 2. Library mass spectra of the trimethylsilyl derivatives of peak and stripped of most of the phosphoric acid covering it.
methylmalonic (A) and (B) succinic acid. Only a small residue at m/z 299 remains and the library

Copyright # 1999 John Wiley & Sons, Ltd. Rapid Commun. Mass Spectrom. 13, 279–284 (1999)
282 DECONVOLUTION GC/MS OF URINARY ACIDS

Figure 3. AMDIS deconvolution of a GC/MS peak containing ethylmalonic acid covered by an excess of
phosphoric acid: (A) mass spectrum at 7.38 min, (B) mass spectrum extracted at 7.373 min, (C) library mass
spectrum of ethylamonic acid.

search match (C) factor is very good. The broken lines in the As a result of internal RI calibration on each data file, the
extracted spectrum (B) indicate a degree of uncertainty in differences in retention index between analyte and library
their deconvolution. For m/z 73, this is due to its presence at spectrum are generally very small thereby providing
significant intensity in both components. Similarly, Fig. 4 considerable confidence in identification, essential for
shows a typical ion chromatogram plot for the separation of automation.
the di- and tri-TMS derivatives of suberyglycine observed in In addition, AMDIS is adept at detecting very low levels
the urine extract from a patient with MCAD deficiency. of metabolites, a particularly important feature as certain
Improved mass spectral matching was achieved for the metabolites, for example, the important acylglycines, can be
separated components. present at almost baseline levels in the asymptomatic

Rapid Commun. Mass Spectrom. 13, 279–284 (1999) Copyright # 1999 John Wiley & Sons, Ltd.
 DECONVOLUTION GC/MS OF URINARY ACIDS 283

patient, making manual detection difficult. Trace levels of

tiglyglycine and orotic acid were detected in the profile
obtained from a patient with mild propionic acidaemia (data
not shown). This detection was much more difficult with
conventional extracted ion profiling.
The reliable identifications achieved by AMDIS enable a
normally complex data file to be reduced to an abbreviated
chromatographic profile (Fig. 5) containing only data of
diagnostic significance or potential diagnostic significance
and with potential for automated interpretation.
Such abbreviated profiles obtained by AMDIS are
suitable for analysis by multivariate statistical techniques
such as principal components analysis (PCA) or learning
machine methods. Figure 6 shows a PCA plot (Cornerstone)
distinguishing between a series of 20 controls (C1-20), 10
patients with MCAD deficiency (M1-10), 3 patients with
methylmalonic aciduria (MM1-3), one patient with iso-
Figure 4. Ion abundance profiles created during the AMDIS valeric acidaemia (II) and 3 patients with propionic
deconvolution of a GC/MS peak containing the di- (19.02 min) and acidaemia (P1-3). This data matrix had incorporated no
tri-TMS (18.95 min) derivatives of suberylglycine (MCAD sample).
weighting factor on the components indicated in Table 1.
The MMA samples are closely grouped in the diagram, a
consequence of the presence of large amounts of methyl-
malonic acid dominating the chromatograms and account-
ing for much of the variance. The MCAD deficiency
samples are more widely spread from the normal grouping,
a consequence of the diverse nature of the chromatograms,
i.e. variations in the relative amounts of hexanoylglycine,
suberylglycine and 3-phenylpropionylglycine. Samples
from the same MCAD deficiency patient are indicated by
(A) when presenting symptoms (but not acutely unwell) and
(B) when in remission. Thus, a simple graphical representa-
tion such as that obtained by PCA can assist the analyst to
distinguish ‘normal’ and abnormal cases, thereby prompting
further investigation. Weighting of the important com-
pounds (marked * in Table 1) enabled all controls to be
clearly distinguished from MCAD deficiency samples and
Figure 5. An ‘abbreviated chromatogram’ based upon AMDIS
identified components and containing only data of diagnostic
this effect will be included in a further communication.
significance. Although the present data were obtained after adjustment of

Figure 6. PCA plot distinguishing between a series of 20 controls (C1-20), 10 patients with MCAD
deficiency (M1-10), 3 patients with methylmalonic aciduria (MM1-3), one patient with isovaleric
acidaemia (I1) and 3 patients with propionic acidaemia (P1-3).

Copyright # 1999 John Wiley & Sons, Ltd. Rapid Commun. Mass Spectrom. 13, 279–284 (1999)
284 DECONVOLUTION GC/MS OF URINARY ACIDS

the urine samples to constant creatinine, the method gave A powerful research tool is available, particularly as
similar results with extracted neonatal urine samples multivariate statistical methods can allow the ready
without such adjustment.3 However, adjustment of the inclusion of clinical as well as further biochemical
urine concentration in this way facilitates quantitative parameters, including acylcarnitines.
analysis and ‘fine-tuning’ of the methodology. The data
matrices are then available for analysis by more sophisti-
cated techniques including learning machines which will Acknowledgements
benefit from the results obtained from large sample
Thanks are due to Mark G. Jones for skilled technical assistance. Some
numbers. In this way, the machine will learn which profiles PCA work was carried out in the laboratory of Professor A.
correspond to cases of metabolic disease and enable their Gammerman, Department of Computer Science, Royal Holloway,
automatic detection. University of London, Egham, UK. The AMDIS program is available
For more general analysis, a larger user library of organic as part of the NIST98 mass spectral database for Windows; details may
acid and drug metabolite spectra can be employed. AMDIS be obtained from NIST, Gaithersburg, MD 20899, USA (tel.: 301-975-
2208, fax: 301-926-0416, e-mail: srdata@nist.gov) or from HD
facilitates the creation and editing of such user libraries. Science, Nottingham NG9 6JF, UK (tel.: ‡44 115 946 9066, fax: ‡44
The advantages of AMDIS include a user-friendly 115 946 8848, e-mail: 100410.3722@compuserve.com).
interface, an extensively tested deconvolution algorithm,
potential for reduced analysis times, convenient creation/
edit of user libraries, convenient retention index calculation REFERENCES
and calibration, integration with NIST/EPA/NIH library,
mass spectrometer data system independence (most major 1. R. A. Chalmers and A. M. Lawson, Organic Acids in Man,
Chapman & Hall, London (1985).
file formats and new ones being added) and provision of 2. J. R. Bonham, M. Downing, R. J. Pollitt, M. J. Manning, K. H.
reliable data for machine recognition. Carpenter, S. E. Olpin, J. C. Allen and E. Worthy, Ann. Clin.
Biochem. 31, 129 (1994).
3. J. M. Halket, A. Przyborowska, B. Yang and S. E. Stein, Proc. 45th
CONCLUSIONS ASMS Conf. Mass Spectrom. Allied Topics, Palm Springs, CA, p.
1152 (1997).
The identification capabilities of AMDIS provide a basis for 4. R. G. Dromey, M. J. Stefik, T. C. Rindfleisch and A. M. Duffield,
the automation of organic acid screening and interpretation. Anal. Chem. 12, 1368 (1976).

Rapid Commun. Mass Spectrom. 13, 279–284 (1999) Copyright # 1999 John Wiley & Sons, Ltd.