Food Chemistry 141 (2013) 4087–4093

Contents lists available at SciVerse ScienceDirect

Food Chemistry
journal homepage: www.elsevier.com/locate/foodchem

Analytical Methods

Simultaneous determination of caffeine, theophylline and theobromine

in food samples by a kinetic spectrophotometric method
Zhenzhen Xia a,b, Yongnian Ni a,b,⇑, Serge Kokot c
a
State Key Laboratory of Food Science and Technology, Nanchang University, Nanchang 330047, China
b
Department of Chemistry, Nanchang University, Nanchang 330031, China
c
School of Chemistry, Physics and Mechanical Engineering, Faculty of Science and Engineering, Queensland University of Technology, Brisbane 4001, Australia

a r t i c l e i n f o a b s t r a c t

Article history: A novel kinetic spectrophotometric method was developed for the simultaneous determination of caf-
Received 5 July 2012 feine, theobromine and theophylline in food samples. This method was based on the different kinetic
Received in revised form 4 June 2013 characteristics between the reactions of analytes with cerium sulphate in sulphuric acid and the associ-
Accepted 26 June 2013
ated change in absorbance at 320 nm. Experimental conditions, the effects of sulphuric acid, cerium sul-
Available online 4 July 2013
phate and temperature, were optimised. Linear ranges (0.4–8.4 lg mL1) for all three analytes were
established, and the limits of detection were: 0.30 lg mL1 (caffeine), 0.33 lg mL1 (theobromine) and
Keywords:
0.16 lg mL1 (theophylline). The recorded data were processed by partial least squares and artificial neu-
Spectrophotometry
Methylxanthines in food
ral network, and the developed mathematical models were then used for prediction. The proposed, novel
Kinetic reaction method was applied to determine the analytes in commercial food samples, and there were no significant
Cerium sulphate differences between the results from the proposed method and those obtained by high-performance
liquid chromatography.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction in human diet, such as cocoa, teas, and chocolate products

Some stimulants are derived from methylxanthines, and they
include caffeine, theobromine and theophylline (Spiller, 1998). Caf-
feine as an alkaloid is found in varying quantities in the seeds,
leaves, and fruit of some plants, where it acts as a natural pesticide
that paralyses and kills certain insects feeding on the plants. It is
most commonly consumed as infusions extracted from the beans
of the coffee plant and the leaves of a tea bush, as well as from var-
ious foods and drinks containing products derived from the kola
nut. Theobromine is a bitter alkaloid of the cacao plant, and it is
found in chocolate as well as in a number of other foods. Theoph-
ylline, also known as dimethylxanthine, is naturally found in tea
and cocoa beans. As a member of the methylxanthine family, it
bears structural and pharmacological similarity to caffeine and
theobromine. The differences in the structures between these com-
pounds are that the NH group in theobromine and theophylline
molecules is replaced in caffeine by an N–CH3 group, and also that
theobromine is the isomer of theophylline. These substances show
various physiological effects on different parts of the body, includ-
ing the central nervous, cardiovascular, gastrointestinal, respira-
tory, and renal systems, and these compounds may also cause
depression and hyperactivity. They are very common chemicals
⇑ Corresponding author at: Department of Chemistry, Nanchang University,
Nanchang 330031, China. Tel./fax: +86 791 83969500.
E-mail addresses: ynni@ncu.edu.cn (Y. Ni), s.kokot@qut.edu.au (S. Kokot).
(Buyuktuncel, 2010). It is therefore, important to assay caffeine,
theobromine and theophylline, because their levels depend on
the genotype and affect the flavour of food and drinks (Brunetto
et al., 2007).
Different analytical techniques and sample preparation proto-
cols have been proposed for the simultaneous determination of
the methylxanthines in foods and biological fluids. Several differ-
ent pretreatments have been employed to minimise the deleteri-
ous effects of complex biofluid matrices with liquid–liquid
extraction (Begas, Kouvaras, Tsakalof, Papakosta, & Asprodini,
2007), solid-phase extraction (Fenske, 2007) and microwaves-as-
sisted extraction (Gonzalez-Nunez & Canizares-Macias, 2011).
Once isolated from the matrix, numerous analytical techniques
have been used for the analysis of these components, including
high-performance liquid chromatography (HPLC) (Ptolemy, Tzio-
umis, Thomke, Rifai, & Kellogg, 2010; Aresta, Palmisano, & Zambo-
nim, 2005), gas chromatography–mass spectrometry (GC–MS)
(Shrivas & Wu, 2007), spectrophotometry (Khanchi et al., 2007;
Khoshayand, Abdollahi, Shariatpahahi, Saadatfard, & Mohammadi,
2008), capillary electrophoresis (CE) (Meinhart et al., 2010), spec-
trofluorimetry (Alves & Poppi, 2009; Moreira et al., 2005), and vol-
tammetry (Sanghavi & Sricastava, 2010; Yang et al., 2010).
Nevertheless, some of these methods, such as chromatography,
are time-consuming, expensive, and need complicated pre-concen-
tration or multi-solvent extraction. Electrochemical methods,

0308-8146/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.foodchem.2013.06.121
4088 Z. Xia et al. / Food Chemistry 141 (2013) 4087–4093
which use modified electrodes, are generally highly sensitivity pro-
cedures but often have poor stability and repeatability. Santos and
Rangel (2012) developed an HPLC method to determine two meth-
ylxanthines. The coupling of a commercial monolithic column to a
traditional low pressure flow injection system, facilitated the anal-
ysis with the use of ultraviolet (UV) detection; a substantial
improvement in analytical results was achieved. The limits of
detection (LODs) were 106 mol L1 for theobromine and theoph-
ylline, and 105 mol L1 for caffeine. In another application (Llo-
rent-Martinez, Garcia-Reyes, Ortega-Barrales, & Molina-Diaz,
2005), a solid-phase ultraviolet sensing system was used to deter-
mine methylxanthines in different kinds of food, beverages and
pharmaceutical samples; the concentration ranges were 1–
16 mg L1 for caffeine, and 1–12 mg L1 for theophylline and
theobromine.
In order to enhance the sensitivity of measurements and sim-
plify the analytical procedures, an alternative method, kinetic
spectrophotometry, was researched and developed for quantita-
tive, simultaneous analysis of complex systems, such as the ternary
mixtures of carbidopa, levodopa and methyldopa (Chamsaz, Safavi,
& Fadaee, 2007), and the determination of ascorbic acid, uric acid
and dopamine (Moghadam, Dadlfarnia, Shabani, & Dhahbazikhah,
2011), with the aid of chemometrics methods. The kinetics meth-
od, based on the difference in reaction rates, especially when the
analytes react with a common reagent, is an efficient way to ana-
lyse simultaneously several analytes without any separation or
masking processes; this approach has been significantly improved
with the use of chemometrics modelling (Chamsaz et al., 2007).
In recent years, chemometrics has been applied for the analysis
of multicomponent systems because, for example, of the develop-
ment of fast data collection with the use of rapid scanning spectro-
photometers (Moghadam et al., 2011). Chemometrics methods
based on principal component regression (PCR), partial least
squares (PLS) and artificial neural networks (ANN), have found
increasing applications for multicomponent kinetic determinations
(Chu & Fan, 2009; Ni, Xia, & Kokot, 2009, 2011).
In this work, a novel kinetic spectrophotometric method, based
on kinetic reactions, was researched and developed for the simul-
taneous determination of caffeine, theobromine and theophylline
in food samples. To interpret the overlapping kinetics data, two
chemometrics methods of data analysis – partial least squares
(PLS) and radial basis function artificial neural networks (RBF-
ANN), were applied for data resolution. Reaction kinetics and the
experimental conditions were investigated in detail, and the devel-
oped calibrations were applied for the analysis of caffeine, theoph-
ylline and theobromine in commercial samples, such as tea, cocoa,
coffee and cola drinks; finally, to verify the proposed method, the
results obtained were compared with those obtained with the
use of a reference method – high performance liquid chromatogra-
phy method (HPLC).
2. Experimental
2.1. Reagents and solutions
All reagents used were of analytical reagent grade, and double
distilled water was used throughout the experiments. A solution
of Ce4+, 2.5  103 mol L1, was prepared by dissolving 0.101 g of
cerium sulphate tetrahydrate (Xilong Chemical Ind., Co., Ltd,
Guangzhou, China) in 5.0 mL 1.0 mol L1 warm sulphuric acid,
and then diluted to 100 mL with water. Stock solutions of analytes
(caffeine, theobromine and theophylline, 100 lg mL1) were pre-
pared by dissolving 10.0 mg of each analyte (Sigma–Aldrich, St.
Louis, MO, USA) in a 3.0 mL sodium hydrate (4.0 mol L1), respec-
tively and then diluted to 100 mL with water. Solution of sulphuric
acid (0.5 mol L1) was prepared by diluting a suitable amount of
concentrated sulphuric acid.
2.2. Apparatus and software
An Agilent 8453 UV–visible photodiode-array spectrophotome-
ter (Aglient Instruments, USA) equipped with a 1.0 cm quartz cuv-
ette was used for measuring the absorbance. All measurements
were performed in a thermostat cell compartment at 70 ± 0.5 °C
with the aid of a model ZC-10 temperature control accessory
(Ningbo Tianheng Instruments Factory, China). An Orion SA 720
digital pH-meter equipped with a combined glass/Ag–AgCl elec-
trode was used for pH adjustment. Micropipettes (Finnpipette,
Labsystems, Finland) were used to deliver solutions less than
1.0 mL, and a stop watch was used to record the cell heating time.
HPLC measurements were performed on an Agilent 1100 series
HPLC–DAD system (Aglient Instruments, USA) equipped with a
vacuum degasser, a quaternary pump, an autosampler, an injector
with a 30 lL loop, an Agilent Zorbax eclipse XDB-C18 column
(250 mm  4.6 mm, 5 lm) with an Agilent Zorbax high pressure
reliance cartridge guard-column (C18, 12.5 mm  4.6 mm, 5 lm)
and a DAD detector. The HPLC system was operated at 25 ± 0.5 °C.
All the programs used in the computing process were written in
MATLAB 6.5 for microsoft windows.
2.3. UV and HPLC procedure
Appropriate amounts of the solutions of caffeine, theobromine
and theopylline or their mixture were transferred into a 1.0 mL
cuvette. Then, taking into account that the total volume required,
was 2.30 mL, 0.3 mL of sulphuric acid (0.5 mol L1) and appropri-
ate amounts of double distilled water were transferred into the
cuvette. This was placed into the cuvette compartment, stirred
and thermostated at 70 °C for 1.5 min. Then, 0.40 mL cerium sul-
phate solution (2.5  103 mol L1) was added to give the final vol-
ume of 2.5 mL, and absorbance measurement was started
immediately. All samples were measured with respect to a distilled
water blank every 4 s between 0 and 600 s at 320 nm (total 150
points). Each reagent was added so as to keep the experimental re-
sults consistent and reproducible within the precision of the mic-
ropipettes, i.e. ±0.1–1.5 lL; thus, the relative error introduced
was <1%.
An HPLC procedure was used as a reference method (Buyuktun-
cel, 2010). All analyte samples and the standard solutions (caffeine,
theobromine and theophylline, Section 2.1) were injected into the
monolithic column via the autosampler of the HPLC instrument
(volume – 30 lL for all samples). A mixture of methanol and phos-
phoric acid (0.1%) was used as the mobile phase, while the flow
rate was set at 1.0 mL min1 at ambient temperature. The gradient
elution was applied as follows: 0–3 min – methanol was at 25%; 3–
5 min, – methanol was from 25% to 40%; 5–10 min – methanol was
at 40%. All analytes were detected at 271 nm, and the retention
times for caffeine, thophylline and thobromine were 7.8 min,
3.9 min, and 5.8 min, respectively. Peak area was used for signals
evaluation, while each sample was filtered through 0.45 lm mem-
brane nylon filters before injection.
2.4. General procedure for the preparation of food samples
Four kinds of commercial sample – cola drinks, caffeine powder,
cocoa powder and tea samples, were purchased from the local
markets in Nanchang city, and were quantitatively analysed for
caffeine, theobromine and theophylline.
Five grams of a tea sample was weighed into a 250 mL beaker,
and 150 mL double distilled water was added (Santos & Rangel,
2012). This beaker was heated on a water bath at 80 °C for 4 h.
 Z. Xia et al. / Food Chemistry 141 (2013) 4087–4093 4089

Fig. 1. Effects of (a) concentration of sulphuric acid (cerium sulphate = 3.0  104 mol L1, temperature = 70 °C and time = 600 s), (b) concentration of cerium (sulphuric
acid = 6.0  102 mol L1, temperature = 70 °C and time = 600 s), and (c) temperature (sulphuric acid = 6.0  102 mol L1, cerium sulphate = 4.0  104 mol L1 and
time = 600 s) on absorbance of the reaction between cerium (IV) and caffeine, theobromine and theophylline, (d) absorbance plot of the kinetics reactions with increasing
temperature (sulphuric acid = 6.0  102 mol L1 and cerium sulphate = 4.0  104 mol L1). All analytes = 4.0 lg mL1.

One millilitre of subacetate solution, prepared by dissolving 50.0 g 2.5.2. Radial basis function-artificial neural network
lead subacetate in 100 mL water and filtering after 24 h standing, RBF-ANN modelling has been often applied to analytical
was added into the tea sample and thoroughly mixed. The solution problems and offers many advantages when dealing with the
was then centrifuged at 8000 r min1 for 2 min. The supernatant
layer was transferred into a separating funnel and extracted with
50 mL chloroform, and then the organic phase was evaporated
with the use of a rotary evaporator. The residue was dissolved in
100 mL double distilled water.
Fifty millilitres of cola drink sample was poured into a beaker,
and the dissolved CO2 was expelled by heating (85 °C) for 15 min
(Jafari, Rezaei, & Javaheri, 2011). The solution was then transferred
into a separating funnel and extracted with 50 mL chloroform; the
organic phase was then treated as above.
For the preparation of coffee and cocoa samples (Wang, Jia, &
Xia, 2011), 5.0 g of their powders and 2.0 g MgO were each trans-
ferred into separate 250 mL flasks and 150 mL double distilled
water was added. Each sample was shaken and kept refluxing at
100 °C for 45 min. After cooling to room temperature, each sample
was filtered and the filtrate was transferred into a separating fun-
nel and extracted according to the same method as the tea samples
above.

2.5. Chemometrics methods

2.5.1. Partial least squares

PLS is a well known multivariate regression method (Martens &
Naes, 1989), and PLS calibration modelling is becoming a very
widely used regression technique for spectroscopic data analysis.
It relates the calibration spectral data matrix to the analyte con-
centration data so as to predict simultaneously several response
variables. Compared with the classical multivariate linear regres-
sion (MLR), PLS is more powerful for dealing with multi-collinear
data, such as the successive data created by spectroscopy. In this
study, the PLS model was constructed by relating the kinetic spec-
trophotometric data to the concentrations of the methylxanthines, Fig. 2. Plot of kinetic behaviour versus time (a) and the absorbance spectra (b) of
the aim being to predict simultaneously caffeine, theobromine and caffeine, theobromine and theophylline against water. sulphuric acid = 6.0  102 -
theophylline in food samples. mol L1, cerium sulphate = 4.0  104 mol L1, and all analytes were 4.0 lg mL1.
4090 Z. Xia et al. / Food Chemistry 141 (2013) 4087–4093

obtain maximum differences between the measured absorbance

curves of the analytes. In this work, the influence of three impor-
tant factors on the reaction rates of caffeine, theobromine and the-
ophylline with Ce4+ were investigated; they were the
concentrations of sulphuric acid and cerium sulphate as well as
temperature.
The effects of sulphuric acid and cerium sulphate were exam-
ined over the range of 2.0  102–1.2  101 mol L1 and
2.6  104–4.4  104 mol L1, respectively (Fig. 1a and b). Almost
no obvious changes of absorbance for caffeine, theobromine and
theophylline were observed with the increasing concentrations of
both cerium sulphate and sulphuric acid. The optimum concentra-
tions selected were 4.0  104 mol L1 cerium sulphate and
6.0  102 mol L1 sulphuric acid; reaction kinetics for the three
analytes were quite stable at these concentrations.
The effect of temperature between 30 and 80 °C was investi-
gated, and in general, largest absorbance differences for the reac-
tions of caffeine, theobromine and theophylline were observed
with increasing temperature (Fig. 1c and d). It was found that
the absorbance of theophylline was almost constant between 40
and 80 °C, while that of caffeine increased almost linearly between
30 and 80 °C. The plot of absorbance changes for theobromine
effectively overlays the caffeine one between 30 and 60 °C and
then levels off to 80 °C. The absorbance versus time plots at differ-
ent temperatures suggested that there was little difference in reac-
tion kinetics behaviour of caffeine and theobromine between 30
and 60 °C (Fig. 1d). Consequently, 70 °C was chosen as the opti-
mum temperature at which clearly discernable differences in the
kinetic rate curves were noted for the reactions involving caffeine,
theobromine and theophylline.

3.2. Reaction kinetics mechanism of the methylxanthines reaction with

Ce4+

The three methylxanthines can reduce Ce4+ in an acidic med-

ium, and this reaction has a maximum absorbance at 320 nm; this
absorbance decreased as the Ce4+ was reduced to Ce3+. However,
the reduction was fastest with the theophylline (Fig. 2a), and the
reaction curves of caffeine and theobromine were also different.
It is these differences that facilitate the simultaneous determina-
tion of the three analytes in their reaction mixtures. However,
the absorbance spectra of the three methylxanthines have very
similar shapes with a maximum absorption at 271 nm (Fig. 2b),
and generally, it is difficult to discriminate the three analytes
quantitatively.
Since Ce4+ was present in excess, the kinetic reactions of these
three compounds ran according to the pseudo-first order reaction
Fig. 3. Kinetic curves of caffeine, theobromine and theophylline with different model. The rate constants for the three analytes were estimated
concentrations at optimum conditions (experimental conditions – Fig. 2).
by fitting the kinetics data obtained from single component sam-
ples of known concentrations, to the equation, A = A0 + a1exp(kt),
non-linearity inherent in such applications. In fact, ANN methods by a suitable regression method (Draper & Smith, 1981).
were designed to work on non-linear, multi-modal, and poorly
understood, improperly or ill-posed problems (Kolmogorov conti- Table 1
nuity theorem) (Castellanos, Martinez Blanco, & Palencia, 2007). Predicted results for synthetic samples by PLS and RBF-ANN methods.
Also, these methods have simple structures, well-established theo-
Methods %RPESa %RPETa
retical basis and fast learning speed. These factors make RBF-ANN
attractive for calibration modelling involving kinetic mixtures. Caffeine Theobromine Theophylline
PLS (3)b 6.0 (106)c 10.3 (110) 2.0 (102) 7.0
RBF-ANN (6,300)d 5.9 (104) 10.1 (109) 2.1 (99) 6.8
3. Results and discussion hP
a m 2
Relative prediction error: RPES ¼ 100  i¼1 ðC ijðfoundÞ  C ijðaddedÞ Þ
.P Pn Pm
n 2 0:5
3.1. Effects of temperature and concentrations of sulphuric acid and i¼1 ðC ijðaddedÞ Þ  (for single component), and RPE T ¼ 100  i¼1 i¼1
.P P
cerium sulphate on reaction kinetics ðC ijðfoundÞ  C ijðaddedÞ Þ2 n
i¼1
n 2 0:5
i¼1 ðC ijðaddedÞ Þ  (for total component).
b
Number of factors used.
P
In general, to perform simultaneous analysis of kinetic c
Mean recovery (%). %Recovery ¼ 100  ni¼1 ½C ijðfoundÞ =C ijðaddedÞ =n.
d
reactions, the optimal reaction conditions should be such so as to Number of iterations and the spread coefficient, respectively.
 Z. Xia et al. / Food Chemistry 141 (2013) 4087–4093 4091

Table 2
Analysis of the three methylxanthines in commercial samples with the use of the RBF-ANN calibration and the HPLC reference method.

Sample (mg g1) Contents detected Added (mg g1) Found (mg g1) Recovery (%)
a
CF TB TP CF TB TP CF TB TP CF TB TP
Kinetic method by PLS
Geen tea 1.07 ± 0.02b NDc ND 0.17 0.17 0.50 1.35 ± 0.02 0.13 ± 0.00 0.49 ± 0.02 108 76 98
Red tea 1.23 ± 0.02 ND ND 0.17 0.25 0.33 1.68 ± 0.03 0.26 ± 0.01 0.36 ± 0.01 120 104 109
Black tea 1.30 ± 0.01 0.09 ± 0.01 ND 0.25 0.17 0.42 1.50 ± 0.01 0.20 ± 0.01 0.47 ± 0.01 97 98 111
Olong tea 1.30 ± 0.01 0.12 ± 0.02 ND 0.17 0.33 0.50 1.17 ± 0.02 0.42 ± 0.02 0.52 ± 0.01 80 93 104
Cola 1 0.22 ± 0.01 ND ND –d – –
Cola 2 0.28 ± 0.02 ND ND – – –
Cocoa powder ND 1.56 ± 0.03 ND 5.50 0.00 2.75 4.39 ± 0.04 1.61 ± 0.02 2.63 ± 0.03 80 103 96
Coffee powder 2.67 ± 0.03 ND ND 1.50 1.50 1.50 5.22 ± 0.03 1.37 ± 0.02 1.46 ± 0.02 125 91 87
HPLC method
Geen tea 1.00 ± 0.01 ND ND 0.17 0.17 0.50 1.21 ± 0.02 0.13 ± 0.00 0.50 ± 0.01 103 76 100
Red tea 1.21 ± 0.01 ND ND 0.17 0.25 0.33 1.54 ± 0.02 0.25 ± 0.02 0.38 ± 0.00 112 100 115
Black tea 1.38 ± 0.01 0.08 ± 0.00 ND 0.25 0.17 0.42 1.58 ± 0.01 0.23 ± 0.01 0.48 ± 0.02 97 92 114
Olong tea 1.25 ± 0.02 0.10 ± 0.00 ND 0.17 0.33 0.50 1.42 ± 0.01 0.42 ± 0.02 0.50 ± 0.01 100 98 100
Cola 1 0.17 ± 0.00 ND ND – – –
Cola 2 0.21 ± 0.01 ND ND – – –
Cocoa powder ND 1.50 ± 0.02 ND 5.50 0.00 2.75 5.00 ± 0.03 1.50 ± 0.03 2.50 ± 0.03 91 100 91
Coffee powder 2.50 ± 0.03 ND ND 1.50 1.50 1.50 3.75 ± 0.02 1.25 ± 0.02 1.50 ± 0.02 94 83 100
a
CF, TB and TP represent caffeine, theobromine and theophylline, respectively.
b
Average ± standard deviation (SD) of the three replicate measurements for each test sample.
c
Not detected.
d
Not performed.

The synergistic effect of the analytes is the main cause for erro-
neous results in analysis of kinetic reactions, and this phenomenon
has attracted some attention in attempts to minimise this problem.
Some years ago, it was demonstrated that this effect could be re-
duced or eliminated with the use of chemometrics methods (Ni,
Liu, & Kokot, 2000).
3.3. Oxidation reactions of the three methylxanthines
were 0.30, 0.33 and 0.16 lg mL1 (corresponding to 1.5  106,
8.7  107 and 1.8  106 mol L1) for the same three analytes,
respectively. The LODs of the proposed methods were similar or
better than those reported in a recent HPLC study (Santos & Rangel,
2012).
3.5. Simultaneous prediction of the three methylxanthine analytes in
synthetic mixtures – PLS and RBF-ANN models

Several equilibrium studies to characterise the different Ce4+
species in aqueous H2SO4 media have been carried out (Das,
2001). The predominant equilibria are:
Ce4þ þ HSO4 ¼ CeðSO4 Þ2þ þ Hþ
CeðSO4 Þ2þ þ HSO4 ¼ CeðSO4 Þ2 þ Hþ
CeðSO4 Þ2 þ HSO4 ¼ HCeðSO4 Þ3
Given the variety of cerium species in aqueous H2SO4, there is
little tendency for Ce4+ to hydrolyse, and thus, Ce(SO4)2+ is the ma-
jor species in this medium. Consequently, the postulated redox
reaction between the methylxanthine analytes and cerium (IV) is:
MethylxanthinesðRedÞ þ CeðSO4 Þ2þ þ Hþ
! Ce3þ þ MethylxanthinesðOxÞ þ H2 O
Literature suggests that the C@N bonds in the methylxanthine
structure were converted to C–N, and the C–H bond was then oxi-
dised to C@O (Sanghavi & Sricastava, 2010).
3.4. Calibrations for the analysis of the three methylxanthine
Calibration graphs for single methylxanthines were constructed
from the measured kinetics data at different concentrations of the
three analytes obtained at 320 nm according to Beer’s law (condi-
tions – Section 3.1). Kinetics data for each analyte are summarised
in Fig. 3; the individual linear calibration models were established
at a selected time (600 s; parameters – insert Fig. 3). The obtained
correlation coefficients (0.9998, 0.9992 and 0.9996) suggested
good linearity over the concentration range of 0.4–8.2 lg mL1
for caffeine, theobromine and theophylline, respectively. The LODs
The orthogonal array design was selected for the building of the
calibration models because this approach minimises the number of
calibration samples required for the model without any loss of statis-
tical rigor. For the simultaneous prediction of three analytes sixteen
ð1Þ
calibration samples were required to build a four-level model with
concentrations of 0.4, 2.8, 5.2 and 6.8 lg mL1 for caffeine, theobro-
ð2Þ mine and theophylline, respectively, and these data were organised
into a concentration matrix, C (16  3). These levels were selected to
ð3Þ allow for a wide distribution of concentrations, which also covered
the range of levels found in real samples. Then, the kinetics data were
obtained according to the experimental procedures (Section 2.3),
and arranged in a kinetics data matrix, A (16  150), where 150 rep-
resents the sampling time points from 0 to 600 s at intervals of 4 s.
An equation between these two data matrices, A and C, was estab-
lished: A = C  K (Ni, Deng, & Kokot, 2010).
Two different types of chemometrics methods, PLS and RBF-
ð4Þ
ANN – linear and nonlinear modelling, respectively, were used to
investigate the analytical data from the three methylxanthine ana-
lytes. Relative standard error of prediction for single or combined
analyte results, RPES and RPET, were calculated, and the mean
recoveries (%) were used to investigate the reliability of the meth-
od (Table 1). Of the two types of model constructed, PLS and RBF-
ANN, the latter performed somewhat better (Table 1), and thus, it
was applied for the analysis of the real samples.
3.6. Analysis of methylxanthines in real samples
The novel kinetic spectrophotometric method described in this
work was applied for the determination of the three analytes from
cola drinks, tea samples, cocoa powder and caffeine powder. For
this analysis, the real samples were treated as previously described
(Section 2.4), and then a suitable volume of each extract was
4092 Z. Xia et al. / Food Chemistry 141 (2013) 4087–4093
analysed spectrophotometrically. The results of such measure-
ments were then processed with the use of the RBF-ANN calibra-
tions (Table 2); recoveries (%) obtained by standard additions to
each sample are listed and compared with those found by an HPLC
reference method procedure so as to ascertain the performance of
the novel method.
Another approach was used to compare the performance the
novel kinetic spectrophotometric method and the reference HPLC
one (Miller & Miller, 2000, chap. 5); regression analysis was per-
formed on the results for the three methylxanthines and plotted
on an X (HPLC) versus Y (kinetic spectrophotometry) graph. Inter-
cept (a), slope (b), and the correlation coefficient (r) of this regres-
sion line were estimated. In the ideal case, where results from the
two methods are the same, then a = 0, b = 1 and the correlation
coefficient, r = 1. Generally, deviations from the ideal situation
can occur in a number of different ways, e.g. different background
signals and systematic errors. In practise, tests are performed by
determining confidence limits for a and b, commonly at 95% signif-
icance. In this work, the calculated correlation coefficients (n = 6)
and the confidence limits with a t-value of 2.78 (4 degree of free-
dom) are: r = 0.9164, a = 0.055 ± 2.159 and b = 1.033 ± 0.627 for
caffeine; r = 0.9996, a = 0.030 ± 0.046 and b = 1.102 ± 0.046 for
theobromine; r = 0.9989, a = 0.039 ± 0.145 and b = 1.052 ± 0.069
for theophylline. From these figures it is clear that the calculated
correlation coefficient, slope and intercept do not differ signifi-
cantly from the ideal values of 1, 1 and 0, respectively, and thus,
the results obtained from the proposed kinetics method showed
good agreement for most samples with that from the HPLC mea-
surements. The differences between the mean %Recovery of the
proposed methods and HPLC, for quantification of each analytes
in real samples, was less than 2% for caffeine, 2.7% for theobromine
and 2.5% for theophylline. It is also evident that the %Recoveries for
each analyte in the samples by the proposed method were in the
range from 80% to 125%. These results strongly suggest that the dif-
ferential kinetics method with the aid of chemometrics presented
in this paper, under established experimental conditions, could
be successfully applied to the simultaneous determination of caf-
feine theobromine and theophylline in food samples.
4. Conclusion
A novel spectrophotometric method, based on reaction kinetics,
was researched and developed for the simultaneous analysis of
three common methylxanthines – caffeine, theobromine and the-
ophylline, present in synthetic and real samples. The underlying
chemistry involved the kinetics of the redox reaction between
the methylxanthines and Ce4+ at the 320 nm; decrease in absor-
bance of the Ce4+ band in the presence of the analytes was ob-
served. A satisfactory calibration model was developed for the
simultaneous prediction of the three analytes with the use of the
RBF-ANN chemometrics method (%RPES (range: 4–7%) and %RPET
(6%)). The proposed method was applied for the determination of
the methylxanthines in real samples, and was verified with the
use of the HPLC method. The results were satisfactory on the basis
of %Recovery and the HPLC verification trial. This method there-
fore, is a potential alternative to the commonly used HPLC tech-
nique; it is low-cost, and uses easily available chemicals and
instrumentation.
Acknowledgements
The authors are grateful for financial support from the National
Natural Science Foundation of China (NSFC-201065007), and the
State Key Laboratory of Food Science and Technology of Nanchang
University (SKLF-ZZA-201302 and SKLF-ZZB-201303).
References
Alves, J. C. L., & Poppi, R. J. (2009). Simultaneous determination of acetylsalicylic
acid, paracetamol, and caffeine using solid-phase molecular fluorescence and
parallel factor analysis. Analytica Chimica Acta, 642, 212–216.
Aresta, A., Palmisano, F., & Zambonim, C. G. (2005). Simultaneous determination of
caffeine, theobromine, theophylline, paraxanthine and nicotine in human milk
by liquid chromatography with diode array UV detection. Food Chemistry, 93,
177–181.
Begas, E., Kouvaras, E., Tsakalof, A., Papakosta, S., & Asprodini, E. K. (2007). In vivo
evaluation of CYP1A2, CYP2A6, NAT-2 and xanthine oxidase activities in a Greek
population sample by the RP–HPLC monitoring of caffeine metabolic rations.
Biomededical Chromatography, 21, 190–200.
Brunetto, M. R., Gutierrez, L., Delgado, Y., Gallignani, M., Zambrano, A., Gomez, A.,
et al. (2007). Determination of theobromine, theophylline and caffeine in cocoa
samples by a high-performance liquid chromatographic method with on-line
sample cleanup in a switching-column system. Food Chemistry, 100, 459–
467.
Buyuktuncel, E. (2010). Simultaneous determination of theobromine, phraxanthine,
theophylline and caffeine in urine by reversed-phase high-performance liquid
chromatography with diode array UV detection. Analytical Letters, 43,
2518–2542.
Castellanos, A., Martinez Blanco, A., & Palencia, V. (2007). Applications of radial
basis neural networks for area forest. International Journal ‘‘Information Theories
& Applications’’, 14(3), 218–222.
Chamsaz, M., Safavi, A., & Fadaee, J. (2007). Simultaneous kinetic-
spectrophotometric determination of carbidopa, levodopa and methyldopa in
the presence of citrate with the aid of multivariate calibration and artificial
neural networks. Analytical Chimica Acta, 603, 140–146.
Chu, N., & Fan, S. H. (2009). Sequential injection kinetic spectrophotometric
determination of quaternary mixtures of carbamate pesticides in water and
fruit samples using artificial neural networks for multivariate calibration.
Spectrochimica Acta A, 74, 1173–1181.
Das, A. K. (2001). Kinetic and mechanistic aspects of metal ion catalysis in cerium
(IV) oxidation. Coordination Chemistry Reviews, 213, 307–325.
Draper, N., & Smith, H. (1981). Applied regression analysis (2nd ed.). New York:
Wiley.
Fenske, M. (2007). Caffeine determination in human saliva and urine by TLC and
ultraviolet absorption densitometry. Chromatographia, 65, 233–238.
Gonzalez-Nunez, L. N., & Canizares-Macias, M. P. (2011). Focused microwaves-
assisted extraction of theobromine and caffeine from cacao. Food Chemistry, 129,
1819–1824.
Jafari, M. T., Rezaei, B., & Javaheri, M. (2011). A new method based on electrospray
ionisation ion mobility spectrometry (ESI-IMS) for simultaneous determination
of caffeine and theophylline. Food Chemistry, 126, 1964–1970.
Khanchi, A. R., Mahani, M. K., Hajihosseini, M., Maragheh, M. G., Chaloosi, M., & Bani,
F. (2007). Simultaneous spectrophotometric determination of caffeine and
theobromine in Iranian tea by artificial neural networks and its comparison
with PLS. Food Chemistry, 103, 1062–1068.
Khoshayand, M. R., Abdollahi, H., Shariatpahahi, M., Saadatfard, A., & Mohammadi,
A. (2008). Simultaneous spectrophotometric determination paracetamol,
ibuprofen and caffeine in pharmaceuticals by chemometric method.
Spectrochimica Acta A, 70, 491–499.
Llorent-Martinez, E. J., Garcia-Reyes, J. F., Ortega-Barrales, P., & Molina-Diaz, A.
(2005). Solid-phase ultraviolet sensing system for determination of
methylxanthines. Analytical Bioanalytical Chemistry, 382, 158–163.
Martens, H., & Naes, T. (1989). Multivariate calibration. Chichester: Wiley.
Meinhart, A. D., Bizzotto, C. S., Ballus, C. A., Prado, M. A., Bruns, R. E., Filho, J. T., et al.
(2010). Optimisation of a CE method for caffeine analysis in decaffeinated
coffee. Food Chemistry, 120, 1155–1161.
Miller, J. N., & Miller, J. C. (2000). Statistics and chemometrics for analytical chemistry
(4th ed.). England: Pearson Education Limited.
Moghadam, M. R., Dadlfarnia, S., Shabani, A. M. H., & Dhahbazikhah, P. (2011).
Chemometric-assisted kinetic-spectrophotometric method for simultaneous
determination of ascorbic acid, uric acid, and dopamine. Analytical Biochemistry,
410, 289–295.
Moreira, A. B., Dias, I. L. T., Neto, G. O., Zagatto, E. A. G., Ferreira, M. M. C., & Kubota, L.
T. (2005). Solid-phase spectrofluorimetric determination of acetylsalicylic acid
and caffeine in pharmaceutical preparations using partial least-squares
multivariate calibration. Talanta, 67, 65–69.
Ni, Y. N., Deng, N., & Kokot, S. (2010). A simple kinetic spectrophotometric method
for simultaneous determination of tetracyclines by use of chemometrics.
Analytical Methods, 2, 1302–1309.
Ni, Y. N., Liu, C., & Kokot, S. (2000). Simultaneous kinetic spectrophotometric
determination of acetaminophen and Phenobarbital by artificial neural
networks and partial least squares. Analytica Chimica Acta, 419, 185–
196.
Ni, Y. N., Xia, Z. Z., & Kokot, S. (2011). A kinetic spectrophotometric method for
simultaneous determination of phenol and its three derivatives with the aid of
artificial neural networks. Journal of Hazardous Materials, 192, 722–729.
Ni, Y. N., Xiao, W. Q., & Kokot, S. (2009). A differential kinetic spectrophotometric
method for determination of three sulphanilamide artificial sweeteners with
the aid of chemometrics. Food Chemistry, 113, 1339–1345.
Ptolemy, A. S., Tzioumis, E., Thomke, A., Rifai, S., & Kellogg, M. (2010). Quantification
of theobromine and caffeine in saliva, plasma and urine via liquid
 Z. Xia et al. / Food Chemistry 141 (2013) 4087–4093
chromatography–tandem mass spectrometry: A single analytical protocol
applicable to cocoa intervention studies. Journal of Chromatography B, 878,
409–416.
Sanghavi, B. J., & Sricastava, A. K. (2010). Simultaneous voltammetric determination
of acetaminophen, aspirin and caffeine using an in situ surfactant-modified
multiwalled carbon nanotube paste electrode. Electrochimica Acta, 55,
8638–8648.
Santos, J. R., & Rangel, A. O. S. S. (2012). Development of a chromatographic low
pressure flow injection system: Application to the analysis of methylxanthines
in coffee. Analytica Chimica Acta, 715, 57–63.
Shrivas, K., & Wu, H. F. (2007). Rapid determination of caffeine in one drop of
beverages and foods using drop-to-drop solvent microextraction with gas
chromatography/mass spectrometry. Journal of Chromatography B, 1170, 9–14.
4093
Spiller, M. A. (1998). The chemical components of coffee. In G. A. Spiller (Ed.),
Caffeine (pp. 97–161). Boca Raton: CRC Press.
Wang, J. L., Jia, L., & Xia, M. (2011). Determination of caffeine, theobromine and
theophylline in drinks by HPLC. Modern Food Science and Technology, 27,
114–116.
Yang, S. L., Yang, R., Li, G., Qu, L. B., Li, J. J., & Yu, L. L. (2010). Nafion/multi-wall
carbon nanotubes composite film coated glassy carbon electrode for sensitive
determination of caffeine. Journal of Electroanalytical Chemistry, 639, 77–82.