PDF Archive

Easily share your PDF documents with your contacts, on the Web and Social Networks.

Share a file Manage my documents Convert Recover PDF Search Help Contact



Anal Bioanal Chem 389, 2007, 555 561 .pdf


Original filename: Anal Bioanal Chem 389, 2007, 555-561.pdf

This PDF 1.3 document has been generated by 3B2 Total Publishing System 8.07e/W Unicode / Acrobat Distiller 7.0 (Windows), and has been sent on pdf-archive.com on 03/11/2015 at 02:17, from IP address 71.17.x.x. The current document download page has been viewed 758 times.
File size: 225 KB (7 pages).
Privacy: public file




Download original PDF file









Document preview


Anal Bioanal Chem (2007) 389:555–561
DOI 10.1007/s00216-007-1442-5

ORIGINAL PAPER

A new conceptual approach to assigning biomass
combustion-derived methoxyphenol structures by using a gas
chromatographic retention index model
S. Rayne & N. J. Eggers

Received: 14 January 2007 / Revised: 3 June 2007 / Accepted: 18 June 2007 / Published online: 25 July 2007
# Springer-Verlag 2007

Abstract A new conceptual approach towards iteratively
constructing chromatographic retention time/index models
is presented. The approach is applicable where there is
potential structural uncertainty in a number of members of
the dataset used in constructing the model, and where
limited spectroscopic information is available to guide the
process. The model is demonstrated on a suite of biomass
combustion-derived methoxyphenols for which gas chromatographic polydimethylsiloxane retention index data was
available in the literature, but where there was ambiguity
regarding the identity of several members of the dataset.
The retention property model is populated by sequentially
screening a series of candidate structures that meet basic
mass spectrometric requirements by using a multiple linear
regression model containing molecular and physicochemical
properties that have been previously shown to yield reliable
predictions of chromatographic behaviour within a compound class. The criteria for deciding on the likely structure
(s) out of a suite of candidate structures is based upon the
improved quality of fit the most probable structure gives the
regression model relative to other candidate structures.
Keywords Chromatographic retention . Prediction .
Guaiacyls . Methoxyphenols . Structural determination
Electronic supplementary material The online version of this article
(doi:10.1007/s00216-007-1442-5) contains supplementary material,
which is available to authorized users.
S. Rayne : N. J. Eggers (*)
Chemistry, Earth and Environmental Sciences,
Irving K. Barber School of Arts and Sciences,
The University of British Columbia at Okanagan,
3333 University Way,
Kelowna, British Columbia V1V 1V7, Canada
e-mail: nigel.eggers@ubc.ca

Introduction
Combustion of biomass produces a wide range of compounds, many of which have important flavor/odor and
toxicological properties [1–8], whose identity is of significant interest. However, many of the components of
biomass combustion cannot be isolated and purified in
sufficient quantity to enable structure elucidation by use of
the full suite of spectroscopic tools (e.g. NMR, MS, IR,
UV–visible). Tentative identification is usually based on the
mass spectrum of a peak eluting from a gas or liquidchromatography column, and from a sample that was not
previously fractionated using preparative methods (e.g.
silica/alumina/gel-permeation chromatography). In an even
smaller number of cases an analytical standard of the
candidate compound is available, enabling use of both
chromatographic retention time and the mass spectrum as
evidence for a proposed structure [1, 9–15].
Unfortunately, there are several limitations in our
current biomass combustion products dataset that reduce
confidence in the structures assigned to date. A primary
concern is the coelution of several components within a
single chromatographic peak. Given the number of
potential closely related structures, all having the same
parent mass and similar fragmentation patterns, coelution
remains an issue that is difficult to resolve. Even when an
analytical standard of a candidate compound is available
(and this is rare), if potential coeluting compounds yield
similar molecular ions—and fragmentation is either weakly present and/or ambiguous relative to potential coeluting
structures—then there is a reduced chance of being
confident what proportion of the total chromatographic
peak area is comprised of the candidate structure. This has
clear effects on correct identification and quantification of

556

Anal Bioanal Chem (2007) 389:555–561

individual compounds in complex, unfractionated environmental extracts such as wood smoke.
Even if coelution is not an issue, and a chromatographically pure peak can be obtained (although this is often very
difficult to both achieve and recognize), the many similar
structures possible for a particular molecular ion can lead to
a low level of confidence in the assigned structure, with or
without the assistance of fragmentation-analysis software
and extensive mass spectral libraries. This is particularly
true for low-resolution mass spectrometry (LR-MS), where
unit mass resolution allows for many possible structural
combinations. As noted above, certain molecular substructures also allow for many similar primary fragmentation
pathways (e.g. β-cleavage in the case of substituted
benzenes to give the stabilized benzylic fragment), so that
major fragmentation analysis is not often unambiguously
conclusive regarding structure. Secondary, tertiary, quaternary, etc., fragmentation is often weak and/or yields
substantial artefacts, and is often not of great assistance
for complex environmental fractions.
The routine use of high-resolution mass spectrometry
(HR-MS) is limited by cost and maintenance issues, and
even though these may be overcome with adequate
resources, many candidate structures are still difficult to
distinguish between at high-resolution. For example, 2methoxy-4-propylphenol and (4-hydroxy-3-methoxyphenyl)acetaldehyde (Fig. 1) are potentially significant
products of biomass combustion, as are many other
methoxyphenols (commonly termed guaiacyls [9–15]).
The molecular m/z ion of these two compounds are
166.10 and 166.06, respectively, meaning a HR-MS would
need to be operating at a resolution ≥5000 to distinguish
between them.
Furthermore, the primary fragmentation path for both
compounds is β-cleavage of the substituent in the para
position. β-Cleavage for both compounds yields the same
fragment (a 2-methoxy-4-methylphenol cation) with an
exact mass m/z of 137.06. Hence, further fragmentation
(e.g. by MSn) of this major β-cleavage product would not
help in distinguishing between the identity of the compounds. The ethyl cation as the minor product from the βcleavage of 2-methoxy-4-propylphenol would have an
exact mass m/z of 43.05, while the acetyl cation from the
O
H

OCH3
OH
2-methoxy-4-propylphenol

OCH3
OH
(4-hydroxy-3-methoxyphenyl)acetaldehyde

Fig. 1 Structures of two potential biomass combustion compounds, 2methoxy-4-propylphenol and (4-hydroxy-3-methoxyphenyl)acetaldehyde

corresponding cleavage of (4-hydroxy-3-methoxyphenyl)
acetaldehyde would have an exact mass m/z of 43.02,
distinguishable at a MS resolution of ≥1500. However,
fragments with m/z values <50 are generally not analyzed
for in GC–MS, because substantial interference from
coeluting contaminants and background noise can be an
issue at low concentrations in the lower mass ranges.
Thus we would argue that, even with HR-MS, distinguishing between two such compounds is difficult. One
option is to obtain authentic standards of each candidate
and analyze them to compare chromatographic retention
times. But if the retention times of the candidates were
equivalent it is likely reliable identification and quantification of the candidates could not be achieved. Different gas
or liquid-chromatography columns could be used, with
longer and more complex elution programs, in an attempt to
separate the compounds, but this may not be successful.
This approach does not address the initial issue of knowing
when two or more compounds of similar structure are
coeluting, and thus making informed decisions on when
and how to perform extensive investigations to separate and
quantify the analyte. Moreover, analytical standards are not
available for the vast majority of environmentally relevant
compounds.
One potential tool for helping to discriminate between
different candidate structures for a chromatographic peak—
and with the guidance of mass spectral data—is the use of
retention time models, which one of the current authors
[16–18] and others [19–25] have shown to be useful for
identifying new compounds in a number of environmental
matrices. Often these retention time models are applied to
gas chromatography, where three major forms of interaction
force determine how long a molecule remains dissolved in
the stationary phase during a gas chromatographic (GC)
analysis. These include dispersion (London), induction
(Debye), and orientation (Keesom) forces [19]. From this
understanding it is known that GC retention can be
adequately modeled as a linear function of molecular
polarizability (α) and/or molecular weight (MW), ionization potential (IP), and the dipole moment (μ) [16–20].
Typically, the approach with retention time models has
been to use an extensive dataset of authentic standards to
populate and validate the model, which can then be used for
identification of new compounds. With many compound
classes, however (e.g. methoxyphenols from biomass
combustion [9–15]), very few authentic standards are
available, because of difficulties in synthesis and purification (particularly for regioisomerism of substituents on a
benzene ring). To help in identifying compounds where no
authentic standards are available and/or to help identify
where significant coelution from other candidate structures
may be an issue, we present here a new conceptual approach
towards iteratively constructing gas-chromatographic reten-

Anal Bioanal Chem (2007) 389:555–561

tion time/index models where there is potential structural
uncertainty in a number of members of the dataset used in
constructing the model.
To help illustrate the approach , we demonstrate the
modelling technique using a suite of biomass combustionderived methoxyphenols for which GC retention index data
using a polydimethylsiloxane column were available in the
literature [11], but where there was ambiguity regarding the
identity of several members of the dataset. The approach
we present here also introduces the concept of mostprobable structure in terms multiple regression modelling
of GC retention properties, and its possible utility in
focussing research and synthetic efforts on the most
relevant target members of this important compound class.

Experimental
DB-1 polydimethylsiloxane retention indices and molecular/β-cleavage m/z ion values for a suite of 18 proposed
substituted methoxyphenol compounds were obtained from
Kjallstrand et al. [11]. Physicochemical properties for these
proposed structures, and candidate structures meeting the
analysis criteria defined below in the Results and discussion
section, were calculated using CambridgeSoft Chem3D
Ultra 6.0 (Cambridge, MA, USA). Molecular structures
were initially drawn in ChemDraw Std 6.0 (CambridgeSoft), copied and pasted into Chem3D Ultra 6.0, and the
molecular configurations were initially optimized by use of
MM2 energy-minimization software with a minimum RMS
gradient of 0.100. The structures were then subjected to
further energy minimization using the AM1 basis set [26] in
the MOPAC2000 algorithm with a minimum RMS gradient
of 0.100 and a closed shell (restricted) wave function.
Following this sequential classical physics semi-empirical
structure-optimization process, physicochemical properties
were calculated by use of MOPAC2000 AM1 software.
Multiple linear regression models were developed using
KyPlot v.2.0 b.13 (Tokyo, JPN), and with molecular weight
(MW), ionization potential (IP), and the dipole moment (μ)
as descriptors. These descriptors were chosen on the basis
of previous literature from one of the current authors [16–
18] and others [20] indicating that they are fundamental to
understanding and predicting gas chromatographic retention properties. We note that the modelling approach shown
here can be applied with any desired (and limited) suite of
potential descriptors, but the approach is not conducive
(with currently available common statistical software) to
examining the relative merits of an unlimited suite of
descriptors to determine the optimized subset of descriptors.
This is in contrast with the common use of stepwise
construction of multiple linear regression models using
forward and backward approaches with or without stepwise

557

construction methods based on statistical significance “cutoffs” (e.g. F values).
To test the potential validity of the model, authentic
standards of 1-(4-hydroxy-3-methoxyphenyl)ethanone and
1-(3-hydroxy-4-methoxyphenyl)ethanone were obtained
(Sigma–Aldrich, Oakville, ON, Canada). The Kovat’s
retention indices of these two compounds were determined
on a Hewlett–Packard 5890A gas chromatograph equipped
with a split–splitless injector, a Hewlett–Packard 5970
mass-selective detector (MSD) and an HP 7673A autosampler. Analysis was performed using a DB-1 polydimethylsiloxane capillary column (30 m ×0.25 mm i.d., 0.25 μm
film thickness; J&W Scientific, Folsom, CA, USA). The
split flow was 15 mL min−1. The transfer line from GC to
MSD was set at 300 °C. Following injection the oven
temperature program was increased linearly at 5 ° min−1
from 50 °C to 200 °C. The injector and detector temperatures were isothermal at 220 °C and 250 °C, respectively,
and helium (carrier gas) was used at an inlet pressure of
175 kPa, giving a column flow of 1 mL min−1. MSD
settings were: scan range, 50–600 amu; threshold, 400;
sample rate, 1.1 scans s−1; ionizing potential, 70 eV; and
EM potential, 2000 V. The GC and MSD were controlled,
and MS data collected, by an HP Chemstation running
G1034C software.

Results and discussion
Using the proposed structures in Kjallstrand et al. [11]
(Table 1) and their associated polydimethylsiloxane retention indices (RICH3-Si), a multiple linear regression model
for prediction of the RICH3-Si value based on the molecular
weight (in g mol−1), ionization potential (IP, in eV), and
dipole moment (μ, in Debye) of each compound yielded the
equation (Fig. 2a):
In RICH3 Si ¼ 1:32ð 0:67Þ þ 0:0271ð 0:0611Þ
IP þ 0:0126ð 0:0207Þ μ þ 1:11ð 0:11Þ
In MW

a multiple correlation coefficient (r) of 0.948, a coefficient
of multiple determination (r2) of 0.899, an adjusted r2 value
of 0.877, an F value of 41.6 (Fobs >Fcrit =3.3; p<4×10−7),
an uneven distribution of residuals over the range of
predicted RIs (Fig. 2b), and a negative linear trend in
residuals, p=0.09). Error ranges in the regression equation
are standard errors about the mean for each regression
coefficient. The standard error (SE; square root of the
variance of the residuals) in the model was 0.043,
corresponding to a percent coefficient of variation (%CV)
of 0.60%. None of the variables was significantly correlated
among each other (IP–μ, r2 =0.54; IP–ln MW, r2 =0.009; μ–
ln MW, r2 <10−4), indicating that multicollinearity was not

558

Anal Bioanal Chem (2007) 389:555–561

Table 1 Polydimethylsiloxane retention indices (from Kjallstrand et al. [11]), literature-proposed structures, and revised potential structures for a
suite of eighteen biomass combustion-derived methoxyphenols

Anal Bioanal Chem (2007) 389:555–561

559

Table 1 (continued)

a fault in the model construction. Correlations between the
variables used to construct the model and the ln RICH3-Si
value were: IP, r=0.20; μ, r=0.11; ln MW, r=0.94).
As noted previously, our experience [16–18] and that of
others [20–24] suggests that a significantly higher quality
of fit should be observed for a GC retention index model
using three such fundamental descriptors and a dataset
comprising compounds all in the same class (i.e. substituted
2-methoxyphenols, or “guaiacyls”). Thus, we proposed
construction of a GC-RI model for this suite of compounds
with the objective of using a maximized quality of fit in the
multiple regression model (using r2 values for the model as
the criterion of improvement) to infer likely molecular
structures. In other words, the model is sequentially
populated (“built”) using the compounds yielding the
highest quality of fits.
For each compound used to populate the model, a suite
of potential alternative candidate structures was examined
for the relative quality of fit based on the criteria:
1. the candidate structure must have a molecular ion m/z
value equivalent (within unit resolution) to that
reported in Kjallstrand et al. [11]; and
2. the candidate structure must have a functional group
that, upon β-cleavage (i.e. in the benzylic position),
yields the β-cleavage ion (often the base peak for
substituted methoxyphenols) reported in Kjallstrand
et al. [11].
Potential coeluting compounds that do not meet these
ionization criteria are excluded from model population. The
physicochemical properties of these candidates, and their

structures, are given in the Electronic Supplementary
Information, Tables S1– S18, inclusive.
To build the model, a matrix of corresponding IP, μ, ln
MW, and ln RICH3-Si values was assembled with the
original structures as proposed in Kjallstrand et al. [11].
Multiple linear regression of this matrix yields the model
discussed above and shown in Fig. 2a. Candidate compounds were tested in order of their reported ln RICH3-Si
value, and the model was populated by moving from earlier
eluting unknown structures to later eluting compounds.
Each candidate structure for a particular ln RICH3-Si value
was placed in the data matrix with its corresponding IP and
μ values (note, ln MW does not change within a suite of
candidate structures), and multiple linear regression was
performed.
The model’s r2 value with the candidate structure was
then compared with the r2 value from the original structure
proposed in Kjallstrand et al. [11], and Δr2 was calculated.
For each candidate structure examined the Δr2 value
relative to the original structure proposed in Kjallstrand et
al. [11] is reported in the Electronic Supplementary
Information, Tables S1– S18. The criteria for determining
the likely structure(s) at a particular ln RICH3-Si were:
1. the candidate structure(s) must yield a Δr2 >0; and
2. no two ln RICH3-Si values can predict the same
candidate structure.
Where Δr2 =0, it is likely the candidate structure(s) may
co-elute with the original proposed structure from Kjallstrand
et al. [11] but, because of the small error in the predictive
ability of any GC retention index model, this co-elution

560

Anal Bioanal Chem (2007) 389:555–561

Fig. 2 Plot of observed versus
predicted polydimethylsiloxane
retention indices and the residuals over the range of predicted
polydimethylsiloxane retention
indices using a multiple linear
regression model based on the
eighteen methoxyphenol structures reported in Kjallstrand et
al. [11] and shown in Table 1
(for plots (a) and (b), respectively) and the revised structures
based on the current study as
shown in Table 1 (for plots (c)
and (d), respectively). Error
bars are 95% confidence limits
on the predicted retention index

cannot be confirmed without synthesizing all the potential
co-eluting candidate structures and examining their actual
polydimethylsiloxane retention properties.
To confirm the modelling approach was not sensitive to
the direction in which it was populated (i.e. from low to
high RICH3-Si values), the model was initially constructed
from low to high RICH3-Si values, and then each suite of
candidate structures was randomly re-examined (in terms
of the order selected for each RICH3-Si value) starting with
the set of values obtained by use of the initially
constructed model. Although this error-checking approach
resulted in slight changes to the magnitudes of the Δr2
values, it did not result in a change in the relative Δr2
values within a suite of candidate structures, thereby
validating the approach.
Based on this method of model population, we have
provided in Table 1 a tentative suite of revised structures that
correspond with the RICH3-Si values given in Kjallstrand
et al. [11]. Using these revised structures (eight of the
eighteen original structures proposed in Kjallstrand et al. [11]
were retained) and their associated RICH3-Si values, a
multiple linear regression model for the prediction of the
RICH3-Si value based on the ln MW, IP, and μ of each
compound yielded the equation (Fig. 2c):
ln RICH3 Si ¼ 1:07ð 0:25Þ þ 0:105ð 0:019Þ
IP 0:0390ð 0:0046Þ
μ þ 1:05ð 0:04Þ ln MW
a r value of 0.9912, a r2 value of 0.983, an adjusted r2 value
of 0.979, an F value of 262 (Fobs >Fcrit =3.3; p<10−11), and

an even distribution of residuals over the range of predicted
RIs (Fig. 2d; no linear trend in residuals, p=0.44). Error
ranges in the regression equation are standard errors about
the mean for each regression coefficient. The SE in the
model was 0.018, corresponding to a %CV of 0.25%. None
of the variables was significantly correlated among each
other (IP–μ, r2 =0.13; IP–ln MW, r2 =0.03; μ–ln MW, r2 =
0.004), indicating that multicollinearity was not a fault in the
model construction. Correlations between the variables used
to construct the model and the ln RICH3-Si value were: IP, r=
0.24; μ, r=−0.31; ln MW, r=0.94).
The regression analysis based on the revised structures given
in Table 1 yields a significantly stronger model (Δr2 =+0.084,
no trending in residuals) relative to the model using the
structures originally proposed in Kjallstrand et al. [11]. In
addition, the regression model shown in Fig. 2a based on the
structures originally proposed in Kjallstrand et al. [11] has a
non-unity slope (0.86±0.08; ±SE) and non-zero y-intercept
(202±112) whereas the model based on the revised structures
shown in Fig. 2c has a slope with a standard error range that
includes unity (0.97±0.03; ±SE) and zero y-intercept within
the standard error of calculation (38±49). However, the
current model more clearly demonstrates the potentially wide
number of coeluting compounds that may be present in
samples collected from biomass combustion, necessitating
further synthetic and analytical studies to better determine
their relative contributions.
We examined the potential validity of the conceptual
model by using authentic commercially available standards
of 1-(4-hydroxy-3-methoxyphenyl)ethanone and 1-(3-

Anal Bioanal Chem (2007) 389:555–561

hydroxy-4-methoxyphenyl)ethanone to test our prediction
of the likely compound(s) eluting with a retention index of
1435. We found that, under our conditions on a DB-1
column as specified above, 1-(3-hydroxy-4-methoxyphenyl)ethanone had a retention index of 1438±5 (error
bars are the range from duplicate trials) and 1-(4-hydroxy-3methoxyphenyl)ethanone had a retention index of 1452±3.
Our retention index of 1452±3 for 1-(4-hydroxy-3-methoxyphenyl)ethanone is similar to the value of 1447 reported by
Nagarajan et al. [27]. However, the substantial peak tailing
for 1-(3-hydroxy-4-methoxyphenyl)ethanone led to incomplete resolution of this analyte from 1-(4-hydroxy-3methoxyphenyl)ethanone, with separation occurring at
approximately 40% of peak height. Thus, where one of
these analytes is present in lower quantities relative to the
other analyte it may be difficult to distinguish (and
quantify) the less abundant compound. The results seem to
validate our approach and the hypothesis that alternate
regioisomers and isobaric members to the guaiacols must
be regarded as potentially significant contributors to overall
lignin-derived combustion signatures. Evidence for these
non-conventional lignin combustion products is appearing in
the literature [28], thereby warranting the application of
computationally derived prediction models to help identify
new compounds in complex environmental matrices.

Conclusions
We present here what is, to the best of our knowledge, a
new conceptual approach towards iteratively constructing
gas-chromatographic retention time/index models where
there is potential structural uncertainty in a number of
members of the dataset used in constructing the model. The
model is demonstrated on a suite of methoxyphenols for
which polydimethylsiloxane retention index data were
available in the literature, but where there was ambiguity
regarding the identity of several members of the dataset.
The retention property model is populated by sequentially
screening a series of candidate structures that meet basic
mass spectrometric requirements (i.e. suitable molecular
and β-cleavage ions) using a multiple linear regression
model containing molecular (e.g. molecular weight) and
physicochemical (e.g. ionization potential, dipole moment)
properties that have been previously shown in the literature
to yield reliable predictions of chromatographic behaviour.
The criteria for deciding on the likely structure, out of a
suite of candidate structures, is based upon the improved
quality of fit the likely structure gives the regression model
relative to other candidate structures (as measured by the
maximum Δr2 value). This approach should, with other
structural elucidation tools, help further refine our under-

561

standing regarding potential molecular structures of unknown compounds for which analytical standards may not
be currently available, or where co-elution could mean that
a significant portion of an observed chromatographic peak
does not arise from a single compound.

Acknowledgements This study was supported by the British
Columbia Wine Grape Council, the Investment Agriculture Foundation of British Columbia, and the Western Diversification Program. S.
Rayne thanks the Natural Sciences and Engineering Research Council
(NSERC) of Canada for financial support.

References
1. Guillen MD, Ibargoitia ML (1999) J Sci Food Agric 79:1889–
1903
2. Guillen MD, Manzanos MJ, Zabala L (1995) J Agric Food Chem
43:463–468
3. Guillen MD, Ibargoitia ML (1996) J Sci Food Agric 72:104–110
4. Galletti GC, Carnacini A, Bocchini P, Antonelli A (1995) Rapid
Commun Mass Spectrom 9:1331–1334
5. Maga JA (1988) Smoke in food processing. CRC Press, Boca
Raton, FL, USA
6. Sjostrom E (1993) Wood chemistry: fundamentals and applications. Academic Press, San Diego, CA, USA
7. Tesfaigzi Y, Singh SP, Foster JE, Kubatko J, Barr EB, Fine PM,
McDonald JD, Hahn FF, Mauderly JL (2002) Toxicol Sci 65:115–
125
8. Larson TV, Koenig JQ (1994) Ann Rev Public Health 15:133–156
9. Hawthorne SB, Krieger MS, Miller DJ, Mathiason MB (1989)
Environ Sci Technol 23:470–475
10. Rogge WF, Hildemann LM, Mazurek MA, Cass GR, Simoneit
BRT (1998) Environ Sci Technol 32:13–22
11. Kjallstrand J, Ramnas O, Petersson G (1998) J Chromatogr A
824:205–210
12. Ehara K, Takada D, Saka S (2005) J Wood Sci 51:256–261
13. Hawthorne SB, Miller DJ, Barkley RM, Krieger MS (1988)
Environ Sci Technol 22:1191–1196
14. Hawthorne SB, Miller DJ, Langenfeld JJ, Krieger MS (1992)
Environ Sci Technol 26:2251–2262
15. Edye LA, Richards GN (1991) Environ Sci Technol 25:1133–1137
16. Rayne S, Ikonomou MG (2002) Anal Chem 74:5263–5272
17. Rayne S, Ikonomou MG (2003) Anal Chem 75:1049–1057
18. Rayne S, Ikonomou MG (2003) J Chromatogr A 1016:235–248
19. Vernon F (1978) Dev Chromatogr 1:1–39
20. Ong VS, Hites RA (1991) Anal Chem 63:2829–2834
21. Stanton DT, Jurs PC (1989) Anal Chem 61:1328–1332
22. Stanton DT, Jurs PC (1990) Anal Chem 62:2323–2329
23. Georgakopoulos CG, Kiburis JC, Jurs PC (1991) Anal Chem
63:2021–2024
24. Katritzky AR, Ignatchenko ES, Barcock RA, Lobanov VS,
Kareison M (1994) Anal Chem 66:1799–1807
25. Zhao HX, Zhang Q, Xue XY, Liang XM, Kettrup A (2005) Anal
Bioanal Chem 382:1304–1310
26. Dewar MJS, Zoebisch EG, Healy EF, Stewart JJP (1985) J Am
Chem Soc 107:3902–3909
27. Nagarajan S, Rao LJM, Guirudutt KN (2001) Flav Fragr J 16:27–29
28. del Rio JC, Martinez AT, Gutierrez A (2007) J Anal Appl Pyrol
79:33–38


Related documents


anal bioanal chem 389 2007 555 561
anal chem 75 2003 1049 1057
j chromatogr a 1016 2003 235 248
anal chem 74 2002 5263 5272
j environ sci health a 44 2009 866 879
med chem res 19 2010 864 901


Related keywords