Study of the hydrolysis of a sulfonylurea herbicide using liquid chromatography with diode array detection and mass spectrometry by three-way multivariate curve resolution - Alternating least squares

Article abstract of DOI:10.1021/ac0101656

This research is focused on the development of a novel, automated chemometric method for obtaining relevant chemical information from time-course measurements of an evolving chemical system. This paper describes an investigation of the hydrolysis of Ally, which is a sulfonylurea herbicide. The hydrolysis of this compound is observed at different pHs and temperatures by reversed-phase liquid chromatography using a diode array detector. The data are analyzed using a three-way, multivariate curve resolution technique. Of special interest was the application of a closure constraint in the kinetic dimension followed by the determination of the rate constants for each step of the pathway by using a differential equation solver and nonlinear fitting of the data.

Full text of DOI:10.1021/ac0101656

Anal. Chem. 2001, 73, 4403-4409
S t u d y o f t h e Hyd ro lys is o f a S u lfo n ylu re a He rb ic id e
Us in g Liq u id Ch ro m a t o g ra p h y w it h Dio d e Arra y
De t e c t io n a n d Ma s s S p e c t ro m e t ry b y Th re e -Wa y
Mu lt iva ria t e Cu rve Re s o lu t io n -Alt e rn a t in g Le a s t
S q u a re s
Erns t Be ze m e r a nd Sa ra h Ruta n*
Department of Chemistry, Virginia Commonwealth University, P.O. Box 842006, Richmond, Virginia 23284-2006
5
This research is focused on the development of a novel,
automated chemometric method for obtaining relevant
chemical information from time-course measurements of
an evolving chemical system. This paper describes an
investigation of the hydrolysis of Ally, which is a sulfonyl-
urea herbicide. The hydrolysis of this compound is
observed at different pHs and temperatures by reversed-
phase liquid chromatography using a diode array detector.
The data are analyzed using a three-way, multivariate
curve resolution technique. Of special interest was the
application of a closure constraint in the kinetic dimen-
sion followed by the determination of the rate constants
for each step of the pathway by using a differential
equation solver and nonlinear fitting of the data.
group did not identify the individual species. Some researchers
have characterized some of the products by using microcolumn
liquid chromatography (LC) and fast-atom-bombardment mass
spectrometry (MS) but did not undertake time-dependent experi-
6
ments. Others have used LC with UV-vis detection, but again,
7
,8
these researchers did not obtain kinetic information. A study
in this research group of one specific SU (Glean) using LC with
diode array detection (DAD) showed the applicability of the three-
way chemometric methods for characterizing this system.9
The problem addressed in this research is the extraction of
quantitative kinetic information from poorly resolved chromato-
grams. The quantitative information is necessary to determine the
reaction pathways and rate constants. Chemometric techniques
are used to extract this information from experimental data with
signal overlap. The data from the hydrolysis of Ally, which is a
sulfonylurea herbicide, is used to show the applicability of these
methods.
Sulfonylurea-based (SU) herbicides are widely used for weed
1
control for grain crops such as corn, wheat, and rice. Their high
herbicidal activity and low application rate (10-15 g/ ha of the
active ingredient) have made these herbicides very popular agents
(
THEORY
43 million tons in 1995).1,2 Not all crops are resistant to these
The literature on multiway analysis is not consistent in the
notation of variables of different orders. The notation used in this
paper is as follows: Bold outlined letters (D) are used for three-
dimensional tensors. Bold capital letters (C) are used for matrices.
Small bold letters (p) are used for vectors. Capital italic letters
agents, which results in damage when they are sprayed onto
neighboring crops or when crops are affected by residual quanti-
ties on the field from previous applications. Thus, a better
understanding of the degradation process would help agricultural
scientists learn how to minimize crop damage.
(
K) are used for the size of a dimension and the total number of
measurements. Small italic letters (k) are used for scalars. Small
italic letters with subscripts (drsk) are used to indicate matrix or
tensor elements.
The two-way data obtained from a hyphenated instrument such
as LC-DAD or LC/ MS are well suited for chemometric analysis.10,11
The UV-vis or mass spectrum in the second dimension contains
Various research groups have investigated the biodegradation
and hydrolysis of these herbicides, but the complete mechanism
has not been determined. The early stages in the metabolic
pathway of these compounds have been elucidated by the use of
3
radiolabeled sulfonylureas; however, the in situ degradation was
not investigated by these workers. The half-lives of the parent
compounds have been reported at various temperatures and pHs,
but most of the degradation products were not identified or
(
(
5) Dinelli, G.; Vicari, A.; Brandolini, V. J. Chromatogr. 1 9 9 5 , 700, 201-207.
6) Reiser, R. W.; Barefoot, A. C.; Dietrich, R. F.; Fogiel, A. J.; Johnson, W. R.;
Scott, M. T. J. Chromatogr. 1 9 9 1 , 91-101.
4
quantified. It has been shown that capillary electrophoresis can
(
7) Braschi, I.; Calamai, L.; Cremonini, M. A.; Fusi, P.; Gessa, C.; Pantani, O.;
be used to separate the degradation products, but that research
Pusino, A. J. Agric. Food Chem. 1 9 9 7 , 45, 4495-4499.
*
Corresponding author. E-mail: srutan@saturn.vcu.edu.
(8) Bossi, R.; Vejrup, K.; Jacobson, C. S. J. Chromatogr. A 1 9 9 9 , 855, 575-
(
1) Beyer, E. M.; Duffy, M. J.; Hay, J. V.; Schlueter, D. D. Herbicides: Chemistry
Degradation and Mode of Action; Marcel Dekker: New York, 1988.
2) Short, P.; Colborn, T. Tox. Indust. Health 1 9 9 9 , 15, 240-275.
582.
(9) Johnson, K.; de Juan, A.; Rutan, S. C. J. Chemom. 1 9 9 9 , 13, 1-11.
(10) de Braekeleer, K.; de Juan, A.; Massart, D. L. J. Chromatogr. 1 9 9 9 , 832,
(
(
3) Sweetser, P. B.; Schow, G. S.; Hutchison, J. M. Pestic. Biochem. Physiol.
67-86.
1
9 8 2 , 17, 18-47.
(11) Salau, J. S.; Honing, M.; Tauler, R.; Barcelo, D. J. Chromatogr. 1 9 9 8 , 795,
(
4) Berger, B. M.; Wolfe, N. L. Environ. Toxicol. Chem. 1 9 9 6 , 15, 1500-1507.
3-12.
1
0.1021/ac0101656 CCC: $20.00 © 2001 American Chemical Society
Analytical Chemistry, Vol. 73, No. 18, September 15, 2001 4403
Published on Web 08/22/2001

unique species-dependent information and can be used to resolve
the overlapping chromatogram. In a simple example with two
such as unimodality, selectivity, nonnegativity, and closure, which
is the application of a mass balance requirement.17
species, the overall wavelength-dependent absorbance (d
by eq 1
j
) is given
If an extra order is present, which in this case is the kinetic
information, additional constraints can be applied on the three-
way data set. A three-way kinetic trilinear data-set, D (T × S ×
K), can be described as the inner product of the retention profiles
R (T × N), the pure component spectra S (S × N), and the kinetic
d ) c × s + c × s
(1)
j
a
aj
b
bj
19
profiles K (K × N), according to eq 6,
D ) R X S X K + E
a b
where c and c are the concentration of species a and b,
respectively, and saj, and sbj, are the molar absorptivities of the
species a and b at a specific wavelength, j.
(6)
In a more general case with N number of species, the overall
wavelength-dependent absorbance is given by eq 2.12
where E is the error tensor, K is defined as the number of kinetic
time points, and N is the number of components.
N
In the ideal case, the data tensor can be unambiguously
decomposed into the corresponding matrices using a direct
trilinear decomposition (DTD) or the generalized rank annihilation
method (GRAM).20,21 However for real data, in which there can
be retention time shifts, peak shape changes, intermolecular
interactions that cause spectral changes, and other effects that
cause the data set to be nontrilinear, these direct approaches
cannot be used, and an iterative method like MCR/ ALS has to
be employed.
The MCR/ ALS algorithm by Tauler and de Juan models only
the predetermined number of components. It uses a principal
component analysis (PCA) compression to represent the data (D)
by using the most significant orthogonal components describing
the data set.22 This ignores information that might be contained
in the smaller principal components and can be a potential
limitation when analyzing nontrilinear data. In that case, the rank
of the data tensor can be higher than the number of real
components. The generalized MCR/ ALS program developed in
our research group does not use this PCA compression.17 It also
allows for a more flexible application of the closure constraint in
all dimensions. A thorough comparison between these two
programs is given in a previous paper.17
One of the multiway constraints, trilinearity, should only be
applied to data having ideal behavior where the three-way data
tensor can be written as the product of three matrices.19 The LC-
DAD kinetic data described in this paper were found to be very
close to trilinear when great care was taken to control the
experimental conditions, such as column and solvent temperature;
however, the generalized ALS algorithm is able to manage
nontrilinearities by relaxing the trilinear constraint for individual
species, thus allowing for some components to be trilinear, while
others are bilinear.
dj ) ci × sij
(2)
∑
i)1
The same equation can be written in matrix notation as eq 3
T
D ) C‚S
(3)
where C is a 2-dimensional array of the concentration profiles of
each component, ST is the transpose of a matrix containing the
spectra of the pure components, and D is the original data matrix.13
Multivariate curve resolution-alternating least squares (MCR/
ALS) algorithms sequentially solve for C and S, as shown in eqs
and 5.14
4
T
T
-1
T
S ) (C ‚C) ‚C ‚D
(4)
(5)
T
-1
C ) D‚(S ‚S) ‚S
The concentration profiles and pure component spectra are
determined by alternating between eqs 4 and 5 until a minimal
T
15
difference between D and (C‚S ) is obtained. Algorithms for
MCR/ ALS have been described by Tauler et al. and Bezemer et
al.16,17 The exact algorithm used in this paper has been described
in the latter.17
To use MCR/ ALS algorithms, an initial estimate for either the
C or the S matrix must be obtained. Evolving factor analysis (EFA)
is the approach used here for the generation of initial estimates
for the concentration profiles. This method is implemented by
using singular value decomposition (SVD) on an increasing
number of columns of the data set.17,18 The resulting singular
values are combined using the chemical assumption that the first
component to start eluting is also the first component to finish
eluting.
EXPERIMENTAL SECTION
Chemicals and Materials. In this research, one specific
member of the SU family is investigated; methyl-2-[[(4-methoxy-
The ALS calculation needs to be constrained to restrict the
range of possible solutions and to result in a chemically reasonable
solution. Several types of constraints can be applied to this model,
6
-methyl-1,3,5-triazin-2-yl)aminocarbonyl]aminosulfonyl]-
benzoate.23 This herbicide is known under the trade name
Metsulfuron methyl or Ally and will be called Ally for the
(
(
12) Harris, D. C. Quantitative Chemical Analysis, 3rd ed.; W. H. Freeman and
Company: 1991.
13) Blyth, T. S.; Robertson, E. F. Matrices and Vector Spaces; Chapman & Hall:
(19) Sanchez, E.; Kowalski, B. R. J. Chemom. 1 9 9 0 , 4, 29-45.
(20) Booksh, K. S.; Lin, Z.; Wang, Z.; Kowalski, B. R. Anal. Chem. 1 9 9 4 , 66,
2561-2569.
(21) Wilson, B. E.; Sanchez, E.; Kowalski, B. R. J. Chemom. 1 9 8 9 , 3, 493-498.
(22) Gargallo, R.; Tauler, R.; Cuesta Sanchez, F.; Massart, D. L. Trends Anal.
Chem. 1 9 9 6 , 15, 279-286.
1
986.
(
(
(
(
(
14) Mallat, E.; Barcelo, D.; Tauler, E. Chromatographia 1 9 9 7 , 47, 342-350.
15) Tauler, R.; Smilde, A.; Kowalski, B. J. Chemom. 1 9 9 4 , 9, 31-58.
16) Tauler, R. B. D TRAC 1 9 9 3 , 319-327.
17) Bezemer, E. C.; Rutan, S. C. Chemom. Intell. Lab. Sys. 2 0 0 1 , in press.
18) Meader, M.; Zilian, A. Chemom. Intell. Lab. Syst. 1 9 8 8 , 3, 205-213.
(23) Rodriguez, M.; Orescan, D. B. Anal. Chem. 1 9 9 8 , 2710-2717.
4404 Analytical Chemistry, Vol. 73, No. 18, September 15, 2001

remainder of the paper. This compound was generously donated
to us by DuPont de Nemours. All other chemicals were purchased
from Aldrich and used without further purification unless other-
wise stated. The water was purified using a Nanopure ultrapure
water system from Barnstead.
Sample P reparation. The pH 2.00 and pH 4.00 buffers were
made using aspartic acid (pK 1
.00 and pH 5.00 buffers were made using citric acid (pKa1 ) 3.1
and pKa2 ) 4.8).24 These buffer components were chosen for their
low UV-vis absorption and pK values close to the desired pH’s.
a
) 2.0 and pK 2
a
) 3.9).24 The pH
3
a
All buffers had a concentration of 0.0050 M, which is 2 orders of
magnitude larger than the concentration of the analyte, thereby
providing adequate buffer capacity. The pHs of these buffers were
adjusted to the indicated pH using concentrated hydrochloric acid
or solid sodium hydroxide. Particulates in all liquid phases were
removed by 0.45-µm membrane vacuum filtration, and solvents
were degassed by a helium purge.
Fig u re 1 . Time dependence of MS positive ions for Ally (s), OH-
Ally (-‚-‚-‚-), ring-opened Ally (----), benzenesulfonamide (-‚-‚-‚-),
triazine (-‚‚-‚‚-‚‚), OH-triazine (- - - ), and ring-opened triazine
(‚‚‚‚‚‚‚).
For all of the kinetic experiments, 3 mg of Ally was dissolved
in 0.5 mL of acetonitrile. This solution was diluted to 100 mL with
the corresponding buffer solution (pH 2, 3, 4, or 5). Aliquots of
the Ally solution were stored in closed vials and kept in a
waterbath held at 25, 35, 45, or 53 °C, depending on the
experiment. The temperature of the waterbath up to its highest
setting was verified by a calibrated thermometer. At each time
point, a sample was taken, after which the vial was discarded.
P rocedure. The LC-DAD instrument was a Hewlett-Packard
The LC-DAD data were converted to ASCII by a macro
supplied by Hewlett-Packard and transferred to an off-line PC. The
HPLC-DAD data were analyzed using the generalized MCR/ ALS
routine, as described in a previous paper, using MATLAB 5.3 from
MathWorks on a Dell OptiPlex GX1 PC.17
The kinetic profiles resolved by the ALS algorithm were further
analyzed by a nonlinear fitting program called Gepasi.25-28 This
program requires the input of a chemical model, which it uses to
build and solve the differential equations corresponding to the
model. Subsequently, the program fits the kinetic profiles to obtain
rate constants for each of the steps in the degradation pathway.
The built-in genetic algorithm (500 generations; population, 20)
was used to find the optimal rate constants for describing the
kinetic profiles. The precision of the rate constants was determined
by changing the individual rate constants until the sum of squares
error of the fit to the data was 10% above the optimum fit.
1
2
090 system. The eluate spectra (1 scan/ sec) were observed from
12 to 330 nm at 2 nm intervals. The mobile phase for the LC-
DAD experiments consisted of 50% acetonitrile and 50% of the
pH 2 aqueous buffer mixed by the instrument. The flow rate was
0
.4 mL/ min, and the column temperature was controlled at 38
°
C.
The ESI-MS experiments without prior chromatographic
separation were carried out on a Nermag 3010 MS in positive ion
mode using a 2800 V electrospray potential and a 0.5 µL/ min flow-
rate. The internal standard was 2-chlorobenzenesulfonamide. The
hydrolysis was observed in a 0.01 M HCl solution, as opposed to
the pH 2 aspartate buffer.
The LC/ MS experiments were performed using a Hewlett-
Packard 1050 LC pump coupled to a Micromass Quattro LC mass
detector via an electrospray interface (ESI) in positive ion mode.
The mobile phase for the LC/ MS experiments was 50% acetonitrile
and 50% purified water. Both phases contained 0.1% formic acid.
The flow rate for these experiments was 0.2 mL/ min. Both the
LC-DAD and LC/ MS separations were performed on a Phenom-
enex LUNA C18 column (150 × 4.6 mm) with 5-µm particles.
Samples were injected using a 20-µL sample loop. Both of the
MS experiments used alternate aqueous acids to avoid interfer-
ence from the aspartate buffer ions.
RESULTS AND DISCUSSION
The determination of the kinetic parameters was accomplished
by the chemometric analysis of the LC-DAD data. However, before
the degradation pathways and reaction rates were determined
using the LC-DAD data, the chemical species involved were
identified. To do this, two MS experiments were performed. In
the first experiment, aliquots of the reaction mixture were directly
injected into the electrospray source, whereas in the second,
several representative aliquots of the reaction mixture were
separated chromatographically before being analyzed by MS using
the LC/ MS setup.
ESI-MS Analysis. Figure 1 shows the relative abundance
ratios of the degradation products relative to the internal standard
for this direct MS experiment for the first 50 h of the experiment.
The reaction was observed for a total of 4 days, but no additional
information was revealed at longer reaction times. Although the
signal-to-noise ratio is low, the profiles clearly showed three
intermediates under these conditions. The first intermediate had
a maximum abundance at ∼4 h, the second had its highest
Data Analysis Methods. The ESI-MS data were binned from
-0.50 to +0.49 for each nominal mass-to-charge ratio. Further-
more, intensities were summed for components that were also
present as solvent complexes, sodiated ions, or dimers (i.e., M+,
+
+
+). Last, the intensity ratio between this total
M(s) , M(Na) , (M)
2
component intensity and the internal standard was calculated. The
same procedure was used to treat the LC/ MS data.
(
25) Mendes, P. Gepasi, version 3.22, http:/ / gepasi.dbs.aber.ac.uk/ softw/ ge-
pasi.html, 2000.
(
26) Mendes, P. Comput. Appl. Biosci. 1 9 9 3 , 9, 563-571.
(
24) Harris, D. C. Quantitative Chemical Analysis, 2nd ed.; W. H. Freeman and
(27) Mendes, P. Trends Biochem. Sci. 1 9 9 7 , 22, 361-363.
(28) Mendes, P.; Kell, D. B. Bioinformatics 1 9 9 8 , 14, 869-883.
Company: New York, 1982.
Analytical Chemistry, Vol. 73, No. 18, September 15, 2001 4405

Ta b le 1 . S t ru c t u re s a n d Ca p a c it y Fa c t o rs
a
b
5
0% ACN and 50% pH 2.00 0.0050M aspartate mobile phase 50% ACN and 50% aqueous mobile phase with 0.1% formic acid.
abundance at ∼8 h, and the third, at ∼24 h. These experiments
were also performed in parallel using the LC-DAD in order to
determine the influence of the change in buffer for the reaction.
Changing the reaction buffer from pH 2.00 aspartate to pH 2.00
HCl had no effect on the reaction kinetics.
The values were determined by calculating the average and
standard deviation of a minimum of 5 runs. The peaks from the
LC/ MS experiments and the LC-DAD experiments using the 50%
ACN and 50% pH 2.00 aspartate mobile phase were matched on
the basis of the capacity factor and according to similarities in
the corresponding kinetic and spectral profiles. All of the low-
molecular-weight components had small capacity factors. These
components were presumed to be charged under these acidic
mobile phase conditions. The large difference in the capacity factor
of Ally versus OH-Ally was also due to the use of an acidic mobile
phase. Ally is uncharged under these mobile phase conditions,
whereas OH-Ally is ionized, and thus, is far less retained by the
LC/ MS Analysis. In the LC/ MS experiments, the aliquots
were taken at time points when the reactant, intermediates, and
products were predicted to be at their peak concentrations. The
separation conditions used for the LC-DAD experiments could
not be used because of interference from the buffer ions.
Furthermore, this MS detector required a lower flow rate than
the 0.4 mL/ min used for the other experiments. This instrument
did not have a thermostated column compartment, nor did it have
an absorption detector; therefore, all samples were analyzed in
parallel using the LC-DAD instrument using the same mobile
phase as used for the LC/ MS instrumental setup. Constraints on
instrumental availability did not allow for complete kinetic studies
to be obtained using this technique.
C
18 stationary phase.
In comparing the capacity factors between the formic acid and
aspartate mobile phases, some changes in the retention order
occurred. As pointed out by Bosch et al., even though the pH
aspartate aqueous phase was lower than that of the formic acid
aqueous phase, in the presence of 50% ACN, the trend in acidities
may be reversed.29,30
Table 1 shows the capacity factors under both sets of LC
conditions for the important species in the degradation pathway.
4406 Analytical Chemistry, Vol. 73, No. 18, September 15, 2001

Fig u re 2 . Typical LC-DAD chromatogram. Region 1 contains the system peak and is not analyzed. Region 2 contains four components.
Regions 3, 4, and 5 contain one component each.
Ta b le 2 . Ra t e Co n s t a n t s a n d Fit t in g Erro rs fo r Ally De g ra d a t io n P a t h w a y
fit error
temp,
°C
k1 ,
k2 ,
k3 ,
k4 ,
k5 ,
k6 ,
k7 ,
ALS,
%
GEPASI,
%
10- min
4
-1
10-4 min
-1
10-4 min
-1
10-4 min
-1
10-4 min
-1
10-4 min
-1
10-4 min
-1
pH
2
3
4
5
2
2
2
2
45
45
45
45
25
35
45
53
61 ( 19
23 ( 6
33 ( 13
19 ( 8
43 ( 11
15 ( 4
5.7
0.2
0.13
0.054
<0.01
0.02
5.7
4.9
0.26
0.34
2.1
0.26
7.6
1.3
0.05
0.33
1.0
<0.01
<0.01
1.3
0.13
3.3
0.36
2.8
<0.01
0.03
0.13
0.01
24
27
22
34
25
32
24
32
4.0
0.12
1.5
0.18
11
2.1
4.0
3.2
15 ( 3
1.9 ( 0.9
1.1 ( 0.2
1.0 ( 0.7
20 ( 9
11 ( 3
2.9 ( 0.4
9 ( 3
44 ( 18
61 ( 19
120 ( 40
1.2 ( 0.2
10 ( 3
14 ( 5
43 ( 11
110 ( 30
33 ( 13
70 ( 30
4.9
19
22
1.9
Chemometric Analysis. Figure 2 shows a typical DAD
chromatogram of an aliquot of the Ally solution at pH 2 after 4 h
of reaction at 45 °C. Region 1 contains the system peak. Region
envelopes for the experiment when both components are present.
Once again, two-way ALS was performed to refine the coarse initial
estimates. Three-way ALS was carried out on this part of the data
set using this matrix with two components as an initial estimate
for each of the experiments. The two components were con-
strained to nonnegativity and vertical unimodality.31 The vertical
unimodality constraint was necessary to resolve these severely
overlapped peaks with very similar spectral profiles. This proce-
dure was repeated for the second peak envelope.
The two chemometrically resolved profiles for each of the
smaller peak envelopes in region 2 were combined to give four
profiles. The retention profiles of these four peaks as resolved by
the ALS algorithm are shown in Figure 3A, and the corresponding
spectral profiles are shown in Figure 3B. These four profiles were
subsequently merged with the profiles of the three baseline-
separated components that were isolated earlier to form a three-
way initial guess tensor consisting of seven components. No extra
components were used to account for background effects, because
an eighth component did not improve the fit.
2
contains four components. Regions 3, 4, and 5 contain only a
single component. To apply three-way chemometric methods to
these data, a set of initial estimates had to be generated for the
ALS algorithm. There were three prominent peaks that are
baseline-separated in regions 3, 4, and 5, respectively. Initial
estimates for these three components were generated by perform-
ing EFA on each peak individually, followed by two-way alternating
least squares. Nonnegativity on both the spectral and concentra-
tion profiles were the only constraints applied. Subsequent two-
way ALS analysis resulted in a closer approximation to the true
spectral and chromatographic responses than the initial EFA
result.
The earlier eluting components in region 2 were treated in a
different manner. The envelope was divided in two smaller
envelopes, each containing the significant part of two of the four
overlapped components, and the following procedure was applied
to both envelopes. EFA was performed on each of the smaller
Nonnegativity constraints were applied to the retention,
spectral, and kinetic profiles of all the components. Vertical
(
(
29) Bosch, E.; Bou, P.; Allemann, H.; Roses, M. Anal. Chem 1 9 9 6 , 68, 3651-
657.
30) Bosch, E.; Espinosa, S.; Roses, M. J. Chromatogr. 1 9 9 8 , 824, 137-146.
3
(31) de Juan, A. Ph.D. dissertation, Universitat de Barcelona, Barcelona, Spain,
1997.
Analytical Chemistry, Vol. 73, No. 18, September 15, 2001 4407

peaks was due to a single component. The closure constraint was
implemented in the algorithm developed here and resulted in
obtaining absolute concentrations rather than relative intensities.
It forces the sum of the concentrations of the included components
in the mass balance to a set value. The components that included
the triazine ring or the ring-opened triazine were included in one
closure, and the second closure contained the components that
were assumed to include the benzenesulfonamide unit. The
application of the trilinearity constraint and the closure constraint
resulted in the ALS fit error shown in Table 2. Without these
constraints, the fit error was under 10% for most experiments.
This indicates that the data are not perfectly trilinear; therefore,
the quantification of the peak area was inaccurate, resulting in a
less accurate implementation of the closure constraint and, thus,
a poorer fit to the data.
Nonlinear Kinetic Fitting of the Rate Constants. From
these experiments, a tentative degradation mechanism was
determined and is shown in Figure 4. The kinetic profiles found
by three-way analysis of the LC-DAD were fit to a model
describing this scheme using Gepasi differential equation solver.
Figure 5 shows a comparison between the chemical model with
the fitted rate constants and the kinetic profiles resolved by ALS.
It indicates three intermediates with reaction times for their
respective peak concentration that correspond to the maximum
peak concentrations seen in the ESI-MS experiments.
The fitting results for all of the reaction conditions are
summarized in Table 2. At low pHs and high temperatures, the
main pathway for degradation was hydrolysis of the methyl group
on the triazine ring followed by ring opening; however, there was
a trend that indicated that at higher pHs and lower temperatures,
the predominant pathway could become the hydrolysis of the
Fig u re 3 . Retention profiles (A) of OH-Ally (-‚-‚-‚-), triazine
-‚‚-‚‚-‚‚), OH-triazine (- - -), and ring-opened triazine (‚‚‚‚‚‚‚).
The spectral profiles (B) of OH-Ally (-‚-‚-‚-), triazine
-‚‚-‚‚-‚‚), OH-triazine (- - -) and ring-opened triazine (‚‚‚‚‚‚‚).
sulfonylurea bridge, as is often reported in the literature.1
(
Other research groups have studied Ally at a range of pHs
and temperatures and have reported the half-lives or pseudo-first-
order rate constants.1,4 The half-life of Ally for this study can be
(
calculated by dividing the natural log of two by the sum of k
; however, the only direct comparison that was available was
for the hydrolysis at pH 5 and 45 °C.1,4 Beyer et al. reported a
1
and
unimodality as well as trilinearity was applied to all components.31
The local rank was also restricted, because each of the latter three
k
3
Fig u re 4 . Proposed degradation pathway of Ally.
4408 Analytical Chemistry, Vol. 73, No. 18, September 15, 2001

Fig u re 5 . Time dependence of Ally at pH 2 and 45 °C and its degradation products as the ALS kinetic profiles (symbols) and the chemical
model with the rate constants fit by Gepasi (lines) for Ally (s b), OH-Ally (-‚-‚-‚- 9), ring-opened Ally (------ +), benzenesulfonamide
(-‚-‚-‚- *), triazine (-‚‚-‚‚-‚‚ O), OH-triazine (- - - 4), and ring-opened triazine (‚‚‚‚‚‚‚ ×).
half-life of 2.1 days under these conditions, but our results
indicated a half-life of 1.2 ( 0.3 days. Our other measurements
were not directly comparable to any literature value because of
mismatches in both temperature and pH.
and higher temperature. In addition, at these conditions, the
predominant degradation pathway proceeded via the hydrolysis
of the triazine-methoxy group, followed by ring opening, whereas
direct hydrolysis of the sulfonylurea bridge was the minor
pathway. This was different from what has been demonstrated to
happen under ambient conditions, where the predominant pathway
is hydrolysis of the sulfonylurea-bridge.1
The application of three-way curve-resolution methods to
kinetic data is of great advantage when studying reaction pathways
of environmental pollutants. The chemometric methods can
resolve the kinetic profiles of the reacting species from interfering
components, such as those found in soil and water samples. And
even though this paper demonstrates the resolution of kinetic
profiles from a degradation reaction, it can easily be used to
observe the temporal or seasonal profiles of those components
in the field. In addition, this approach is easily generalized to any
number of kinetic studies in fields beyond environmental science,
including pharmaceutical, biomedical, chemical, and materials
sciences.
The calculated optimal fit was verified by changing the value
of the different rate constants and simulating the kinetic profiles.
In all cases, the sum of squares error increased upon both positive
and negative changes in the rate constants. The precision of the
first three rate constants was determined by changing each rate
constant individually and determining when the sum of squares
error was 10% worse than the best fit. The last four rate constants
showed several local minimums. Those values are much less
precise, and it was not possible to determine the precision of their
values.
CONCLUSION
This paper has shown the applicability of three-way chemo-
metric methods to kinetic studies performed by LC-DAD and uses
a general approach for the generation of the initial estimates. The
overlapped components in the chromatogram were resolved, and
all species were identified with the aid of MS and LC/ MS
experiments. The implementation of the closure constraint in the
third dimension resulted in the determination of the absolute
concentrations rather than relative ones. This was necessary to
be able to implement the kinetic modeling procedure. This
modeling was implemented using the Gepasi program, which
calculated the rate constants that give the best fit to the kinetic
profiles determined by ALS. The rates of degradation determined
by this method agreed with the half-lives reported in the literature
for acidic conditions.1,4 The overall rates increased with lower pH
ACKNOWLEDGMENT
This work was supported by DuPont through their donation
of the herbicide. This research was sponsored by the Jeffress
Trust (J-517) and by grant 9709437 from the National Science
Foundation
Received for review February 6, 2001. Accepted June 29,
2001.
AC0101656
Analytical Chemistry, Vol. 73, No. 18, September 15, 2001 4409