In the La bor a tory

Analysis of Soft Drinks: UV Spectrophotometry,

1
Liquid Chrom atogr aphy, and Capillary Electrophoresis
Valerie L. McDevitt, Alejandr a Rodrguez, and Kathryn R. Williams*
Department of Chemistry, University of Florida, P.O. Box 117200 , Gainesville, FL 32611 -7200

An experiment for the undergraduate instrumental

analy- sis laboratory should accomplish several instructional
goals. First, it must demonstrate the capabilities and
limitations of the method, as well as the proper procedures
for data acqui- sition and computation of results. Students
must also be re- minded of the importance of the chemistry
of the sample and how this relates to the analysis.
Application of the method to a commercial product always
helps stimulate student in- terest and teaches the extra
considerations necessary in a real- world analysis. It is also
instructive to analyze the same product by more than one
instrumental method.
Analyses of regular and diet soft drinks fulfill all these
objectives. The samples are common everyday products,
and they may be analyzed by a variety of means. Soft drink
com- ponents have been determined by HPLC with UV
detection (15) for a number of years, and methods
utilizing capillary electrophoresis to determine caffeine (6, 7
) and other com- ponents (8) have been developed.
Although certainly not as useful as LC or CE,
multicomponent UV analysis can also be used if the
product does not contain too many absorbing components.
This paper describes a series of undergraduate experi-
ments using these three instrumental methods for the analysis
of components of public interest in commercial soft drinks:
caffeine, a central nervous system stimulant; sodium
benzoate (determined as benzoic acid), which serves as a
preservative; and the artificial sweetener aspartame. In
addition to teach- ing the physical bases and practical
applications of the three instruments, the experiments stress
the chemical nature of the sample, especially the acid/base
character of the three compounds and the importance of pH
in the design of LC and CE separations. As part of the data
acquisition and analysis, students also determine the
method detection limits (MDL) for the three compounds by
LC and CE. The concepts of MDL, false positives, and false
negatives are especially relevant, considering the current
interest in natural foods.

Multicomponent UV Analysis
Multicomponent spectral analysis is described in stan-
dard analytical texts (9, 10) and is a common experiment in
the undergraduate curriculum (11). In addition to accessing
the software in the Hewlett-Packard 8450A Spectrophotom-
eter, University of Florida students are required to analyze
the data manually using simultaneous equations, as
described in the references. Because the number of equations
must equal or exceed the number of components, manual
data reduction can be unwieldy for a three-component
system. To simplify the math, the analysis is limited to two
of the compounds, caffeine and benzoic acid.

* Corresponding author.
JChemEd.chem.wisc.edu Vol. 75 No. 5 May 1998 Journal of Chemical Education 1
 In the La bor a tory
Figure 1. Structures and UV spectra of 7.83 mg /L caffeine (),
6.55 mg /L benzoic acid ( ), and 18.7 mg /L aspartame (- - -) in 0.01 M
HCl. The structures represent the predominant proto- nated forms at pH
2.0. Spectra show little variation with pH.
Figure 2. Spectrum of a 1:25 dilution of Mello Yello in 0.01 M HCl.
2 Journal of Chemical Education Vol. 75 No. 5 May 1998 JChemEd.chem.wisc.edu
Figure 1 shows the structures and spectra of all three
compounds in 0.01 M HCl. The absorption profiles of
caffeine and benzoic acid are quite different, an indication
that this pair is well suited to the method. In some
instructional settings it may be feasible to use a diet drink
and analyze the aspartame separately by a suitable solid-
phase extraction technique. However, to save time and limit
the focus of the data analysis, a nondiet beverage is used. Of
the several caffeinated drinks that were tested, the best
choice was Mello Yello. Colas do not give acceptable results
because an appreciable colorant band extends into the UV.
The spectrum of Mello Yello is shown in Figure 2. Although
there is a small absorption at 300 nm, due probably to
colorant, the effect on the results is not detrimental to the
overall goals of the experiment.
An essential part of the laboratory experience is the
group prelab quiz. Questions for this experiment focus on the
funda-
mental principles of multicomponent analysis, the quiz, students predict the elution order to be partially ionic aspartame
requirements of the absorbing system, the choice of first, followed by very polar caffeine and less polar benzoic acid. In
wavelengths, and the special features of the HP 8450A actuality,
diode-array instrument.
The procedure calls for the preparation of a series of
caf- feine and benzoic acid standards in 0.01 M HCl, with
con- centrations in the ranges 420 and 210 mg/L,
respectively. Students measure the UV spectrum of each
standard, a syn- thetic unknown prepared by the instructor,
and a 1.000- to 25.00-mL dilution of filtered Mello Yello.
In consultation with the teaching assistant, they choose two
wavelengths and obtain the absorbance data for the
simultaneous equations method. To verify that the two
spectral profiles are additive, a mixed standard containing
known concentrations of both components is also tested,
and the spectrum is plotted on the same sheet as the
spectral sum of the two contributing single-component
standards. In preparation for the subse- quent LC and CE
experiments, students obtain spectra of all components in
the buffer systems to be used.
While in the laboratory, the students obtain the
concen- trations of the two analytes in the synthetic
unknown and Mello Yello from the HP software. For their
laboratory reports, they prepare Beers law plots for each
component at the two wavelengths and use simultaneous
equations to determine the concentrations manually.

Liquid Chromatography and Capillary Electrophoresis

Initially, the LC and CE analyses were combined into a
single experiment. Although this was feasible in terms of
labo- ratory time, the average student failed to grasp all the
instru- mental and chemical concepts, and the experiment
was split. The three-week soft drink module is scheduled in
the order UV, LC, CE.
In addition to the fundamental concepts of LC and
CE, the oral quizzes and written reports stress the
chemistry of the system, especially the importance of pH in
the two sepa- ration processes. The laboratory manual gives
students the literature values for the pKas of benzoic acid
(4.202 at zero ionic strength [12]) and aspartame (2.96 and
7.37 at 0.15 M ionic strength [13]). The actual dissociation
constant for caf- feine is not available in standard
references. According to The Merck Index, the pH of a 1%
solution is 6.9 (14 ). A 0.01 M solution in highly purified
water was prepared in this labo- ratory. The readings for the
solution and the water were both 7 within experimental
error. Thus, the students are told that caffeines pKb is ca.
14.
An eluent mixture of 45% methanol/55% 0.025 M
aqueous phosphate, pH 3.0, gives the best separation of the
three components without exposing the bonded octadecyl
(C18) stationary phase to excessive acidity. To understand
the relationship of the mobile phase composition to the
separation, students must first recognize that components
should be neutral to interact with the octadecyl stationary
phase and that compounds are expected to elute in order of
decreasing polarity. As shown above, caffeine (Caf) is such a
weak base that it is neutral at pHs higher than about 1. In
the pH 3.0 mobile phase, the benzoic acid is also in its
neutral protonated form (HBenz). On the basis of the pKa
data, aspartame exists as an equimolar mixture of the fully
protonated HAsp+ and zwitterionic HAsp forms. In the oral
the aspartame elutes shortly after the caffeine,
as shown in Figure 3. The most logical
explanation is the presence of 45% methanol,
which may increase the fraction of aspartame
in the HAsp form (i.e., pKa may decrease).
1
Also, benzoic acids long retention time is
indicative of considerable interaction of the
phenyl group with the stationary phase, and
this ef- fect may also occur with aspartame.
The CE experiment is performed on a
Hewlett-Packard 3DCapillary Electrophoresis
system, which has an automatic sample
changer and a diode-array detector, using an
applied potential of 20 kV with the cathode at
the outlet. With this configuration, the
migration order is cations first in order of
decreasing electrophoretic mobility (ep),
neutrals next as a group, and anions last in
order of increasing ep. To separate the three
components, a 0.025 M borate buffer, pH 9.4,
is used. At this high pH only Caf is neutral;
aspartame and ben- zoate are both anionic
(Asp and Benz). Because of its larger size, Asp should
have a lower ep than Benz. Therefore, stu-
Figure 3. Liquid chromatogram of a caffeine / benzoic acid / aspar-
tame mixture obtained on a 15 cm 4.5 mm Higgins Analytical
column packed with 5 m Hiasil C18, using a flow rate of 1.0
mL/ min and a mobile phase composition of 45% methanol / 55%
0.025 M aqueous phosphate, pH 3.0. Other instrument compo-
nents included a TSP P200 pump, a Valco injector with a 20-L
loop, a TSP UV 100 detector set to 218 nm, and a SpectraPhysics
SP4270 integrator.
Figure 4. Electropherogram of a caffeine / benzoic acid /
aspartame mixture obtained on a Hewlett-Packard 3DCE system
using a 0.025 M borate buffer, pH 9.4. The 33 cm 50 m
capillary was oper- ated at 20 kV. The detection wavelengths were
272 nm for caf- feine, 229 nm for benzoate, and 210 nm for
aspartame. Before the laboratory period, the capillary was flushed
with 0.10 M NaOH, water, and the borate buffer. Buffer flushes were
included after every third sample.
 dents expect Caf to migrate fastest, followed by Asp, with Table 1 . Spectral Data for Caffeine, Benzoic Acid,
Benz last. As shown in Figure 4, this order is indeed and Aspartame
observed. a
Compound (n (L/ g (103L/ mol
In addition to explaining the pH effects and predicting m cm)
cm)
the elution/migration orders, students are asked to use the Caffeine 2 51.1 9.93
UV spectra to choose wavelengths for the LC and CE 1
2 25.9 5.02
analyses during the prelab quiz. For the LC analysis, the 2 48.9 9.50
7
chosen wave- length is 218 nm. As Figure 1 shows, Benzoic Acid 2 52.8 6.45
although the compo- nents all absorb appreciably at 218 1
2 90.1 11.00
nm, this wavelength does not correspond to max for any of 2 7.66 0.935
the three compounds. The diode-array capability of the CE 7
Aspartame 2 29.3 8.61
system allows an optimum wavelength to be used for each 1
2 11.09 3.26
component (210 for Asp, 272 for Caf, and 229 for Benz). 1
In the laboratory, students analyze the same solutions
on both instruments. First, individual solutions of each method (sample pretreatment, injection, peak quantitation).
com- ponent in water are injected to determine the elution This is considered to be a truer representation of the detec-
and migration times. On the CE, students also verify the tion limit than the value obtained according to IUPAC rec-
peak assignments from the spectra obtained by the diode- ommendations (17, 18), which utilizes only the
array detector. They next inject a series of mixed standards, fluctuations in the instrument signal. The full EPA protocol
each containing known concentrations of all three is slightly modified (see Appendix) to meet the time
components, for preparation of the three calibration plots. constraints of the student laboratory.
The standards, which are prepared by the instructor before
the laboratory period, have concentrations in the ranges Results
40200 mg/L for caffeine, 25150 mg/L for benzoic acid,
and 75600 mg/L for aspartame in water. The remaining The wavelengths chosen for the UV analysis are the
samples are unknowns: filtered soft drinks, including Mello maxs for the two components: 229 nm (benzoic acid) and
Yello and several diet drinks, a synthetic unknown, and a 272 nm (caffeine). Both compounds obey Beers law over
solution of one packet of Equal (aspartame) in 100 mL of the con- centration ranges used in the analysis. Table 1
water. As described fur- ther below, students determine the summarizes the absorptivity data, including values at 218
method detection limits for the three compounds. To nm and 210 nm (aspartame only), which are used in the
obtain the necessary data, they prepare seven replicate HPLC and CE experiments.
dilutions of one of the standards and obtain the peak areas Typical analytical results are presented in Table 2. Results
using the same separation conditions as the drink samples. for the synthetic unknown show excellent agreement with
Data analysis for both experiments starts with preparation the instructors values, although the caffeine result for Mello
of six calibration plots of peak area versus concentration for Yello is somewhat high. As stated above, this is probably
each component by each method. The least-squares due to a small absorbance for Mello Yello at 300 nm, where
equations are used to evaluate the concentrations of the the individual components do not absorb. This
components in the unknowns and the aspartame content of a undoubtedly ex- tends into the UV and accounts for the
packet of Equal. high caffeine result. In the conclusion section of their report,
To emphasize the relationship of the detection systems students are expected
to the UV spectrophotometer, students are also asked to
cal- culate the ratios (Caf:Asp and HBenz:Asp)
of the slopes of the LC calibration plots.
They observe that these ratios are close to EPA method takes into consideration the statis- tical fluctuations of
the corresponding ratios of the absorptivi- the entire analytical
ties at the LC wavelength (218 nm)
obtained from the UV spectra. The CE
report in- cludes comparisons of the LC
and CE re- sults for the same samples and
results for Mello Yello by all three
methods.
Analysis of the same compounds by
more than one method provides an excel-
lent opportunity to evaluate detection lim-
its and compare the values for different in-
struments. For this exercise, students are
asked to determine the method detection
limit according to the method specified by
the Environmental Protection Agency (15,
16 ) for caffeine, aspartame, and benzoic
acid by both LC and CE. As explained
further in the Appendix to this paper, the
 Table 2 . Results of Caffeine / Benzoic Acid / Asp 156.9 1 153
Aspartame Analyses 7
Diet Coke Caf 127 1 133
Valuea Experimental Benz 3
1 149
Sample Compou Result
nd (mg /L) UVb Asp 521 5 496
0
UV Syn Unk Caf 11.12 11.09 / Diet Pepsi Caf 101 1 89
Benz 4.15 11.38/ 4.24
4.15 Benz 0
1 162
Mello Yello Caf 148 167 / 176 Asp 6
4 475
Benz 208 / 219 d 6
Equal Asp 367 3 317
LC / CE Syn Caf 101.8 a
Instructors value for synthetic unknowns; manufacturers value 5for commercial prod-
Unk Benz 78.6 ucts, if available. bResults by HP software / simultaneous equations. cAverage of 2 deter-
d
minations. Solution prepared by dissolving 1 packet in 100 mL of water.
to comment on how well the Mello Yello system meets the reinforce students knowledge of acid/base behavior and stress the
requirements for multicomponent analysis, and the extra importance of fundamental chemistry in the design of an analytical
ab- sorption is an obvious point to address. Thus, this method.
analytical interference is actually very useful instructionally.
There is also some concern about possible interactions
between the two components, because the solubility of
caffeine in pharmaceutical preparations is known to increase
in the presence of sodium benzoate (19). However, the
additivity test described above produces spectra that match
almost perfectly, and a published UV analysis for
pharmaceutical mixtures of caf- feine, sodium benzoate,
phenacetin, and Pyramidon showed good agreement, with
relative errors of less than 3% (14 ).
The LC and CE calibration plots are also very linear,
and the results for the synthetic unknown are generally
quite good, with relative errors less than 10% (less than
4%, if aspartame by LC and caffeine by CE are excluded).
For the most part, LC and CE analyses of the drinks agree
with each other, the major exception being caffeine in Mello
Yello, which produces a significantly higher value by CE. Of
the Mello Yello analyses, the LC result is closest to the
manufacturers value. The high CE result may be due to
another neutral co- migrating with the caffeine, but the
spectral profile shows no obvious deviation from that of
pure caffeine.
The LC and CE values for caffeine in Diet Coke may
also be compared to those reported previously in this Journal:
92 mg/L (3) and 134 mg/L (5 ) by LC; 124 (7 ) by CE.
Agree- ment with the latter two values is very satisfactory.
The MDLs (mg/L, M) by LC are 1.2, 6.1 for
caffeine; 0.57, 4.7 for benzoic acid; and 1.4, 4.9 for
aspartame. The
MDLs by CE are 1.7, 8.5 for caffeine; 0.29, 2.4 for benzoic
acid; and 2.2, 7.5 for aspartame. The LC and CE values are
quite close to each other, and all six values are within 2 and
8.5 on a micromolar basis. The two caffeine MDLs are also
remarkably close to the previously reported value of 1.9 mg/L
by CE (7 ).

Conclusions
In the conclusion to the CE report (i.e., after all three
experiments have been performed), students are asked to
com- pare the advantages and limitations of the three
methods. Factors such as the number of analyzable
components, MDLs, sample size, ease of performance, and
accuracy (compared to manufacturers values) are discussed.
The most notable consideration is the number of
components, which is clearly a limitation in UV
multicomponent analysis. Good students note that sample
size is another important factor. Although the MDLs are
about equal for LC versus CE, the 20-L injection loop on
the LC can be rinsed and filled with about 50 L of filtered
drink, whereas the standard CE autosampler vials hold
about 500 L. Thus, even though most of the sample can be
recovered from the autosampler vial, the CE analy- sis requires
a total sample volume about ten times that for LC.
The analysis of soft drink components stimulates student
interest and is instructionally useful. The experiments dem-
onstrate the analytical use of the three instruments, as well
as their advantages and limitations. The MDL determinations
teach an important procedure frequently used in the work-
place but not included in most texts. The experiments also
Acknowledgment 10. Harris, D. C. Quantitative Chemical Analysis, 4th ed.; Freeman:
New York, 1995; p 526.
Purchase of the capillary electrophoresis 11. Williams, K. R.; Cole, S. R.; Boyette, S. E.; Schulman, S. G. J.
system was facilitated by grant #DUE Chem. Educ. 1990, 67, 535.
9650497 from the National Science 12. Martell, A. E.; Smith, R. M. Critical Stability Constants;
Foundation. Plenum: New York, 1974; Vol. 3, p 16.
13. Martell, A. E.; Smith, R. M. Critical Stability Constants;
Note Plenum: New York, 1974; Vol 5, p 111.
1. Presented at the annual meeting of the 14. The Merck Index, 11th ed; Budavari, S., Ed.; Merck: Rahway, NJ,
Southeast Association of Analytical Chemists, Athens, 1989; p 248.
GA, October 1996, and the Annual Meeting of the 15. U.S. Environmental Protection Agency. In Code of Federal Regu-
Florida Sections of the American Chemical Society, lations; Part 136, Title 40, Appendix B, Revision 1.11, U.S.
Orlando, FL, May 1997. Gov- ernment Printing Office: Washington, DC, 1990; pp 537
539.
Literature Cited 16. Harris, D. C. Quantitative Chemical Analysis, 4th ed.; Freeman:
New York, 1995; p 84.
1. Smyly, D. S.; Woodward, B. B.; Conrad, E. C. J.A.O.A.C. 1976, 17. Winefordner, J. D.; Long, G. L. Anal. Chem. 1983, 55, 712A
5 724A.
9 18. Skoog, D. A.; Leary, J. J. Principles of Instrumental Analysis, 4th
, ed.; Saunders: Fort Worth, 1992; pp 78.
19. Machek, G.; Lorenz, F. Scientia Pharmaceutica 1966, 34, 213231.
1
4 Appendix

1 The EPA defines the MDL as the minimum
9 concentrationthat can be measured and reported with 99% con-
. fidence that the analyte concentration is greater than zero and is
2. Gillyon, E. C. P. Chromatogr. Newslett. 1980, 8, 5051. determined from an analysis of a sample in a given matrix
3. Delaney, M. F.; Pasko, K. M.; Mauro, D. M.; contain- ing the analyte (15 ). The latter part of the definition
Gsell, D. S.; Korologos, P. C.; Morawski, J.;
indicates that the MDL must be determined by an actual analysis
Krolikowski, L. J.; Warren, F. V., Jr. J. Chem.
Educ. 1985, 62, 618620. and is strictly valid only for the particular sample conditions. The
4. DiNunzio, J. E. J. Chem. Educ. 1985, 62, 446447. code also gives the complete protocol, but, for the readers
5. Strohl, A. N. J. Chem. Educ. 1985, 62, 447448. convenience, a brief explanation is included here.
6. Mabrouk, P. A.; Marzilli, L. A. Presented at the First, the analyst must have an estimate of the MDL. The
212th National ACS Meeting, Orlando, August code lists several types of estimates, but the two most applicable to
1996; Paper No. 276; see CHED Newslett. 1996, these analyses are (i) the concentration giving a signal roughly 2.5
Fall.
to 5 times the baseline noise (method usually used by the
7. Conte, E. D.; Barry, E. F.; Rubinstein, H. J. Chem. Educ. 1996,
73, 11691170. students), and
8. Schuster, R.; Gratzfeld-Hsgen, A. Hewlett- (ii) the lowest concentration in the linear response range (i.e., the
Packard Publ. #12- 5963-1122E; Hewlett- concentration at which the calibration plot shows a noticeable
Packard: Palo Alto, 1994. change in slope). Next, a standard containing 1 to 5 times the
9. Skoog, D. A.; Leary, J. J. Principles of approxi-
Instrumental Analysis, 4th ed.; Saunders: Fort
Worth, 1992; p 164.
mate MDL is prepared, and seven or more aliquots are analyzed
by the usual laboratory procedure. Because of the time required
to dissolve the solid compounds, the procedure is modified to
have
students prepare and inject seven replicate dilutions of one of the
mixed calibration standards.
The analyte concentration is evaluated from the least-squares
response equation (and suitable additional calculations, if applicable).
The MDL is given by: MDL = t99 s
where s is the standard deviation of the replicate concentration mea-
surements and t99 is Students t (one-sided) at the 99% confidence
level (98% confidence level if a table of two-sided ts is used) for
N
1 degrees of freedom (N = the number of replicates).