- Email: [email protected]
On-line gradient liquid chromatography–Fourier transform infrared spectrometry determination of sugars in beverages using chemometric background correction
Talanta 77 (2008) 779–785
Contents lists available at ScienceDirect
Talanta journal homepage: www.elsevier.com/locate/talanta
On-line gradient liquid chromatography–Fourier transform infrared spectrometry determination of sugars in beverages using chemometric background correction J. Kuligowski, G. Quintás, S. Garrigues, M. de la Guardia ∗ Analytical Chemistry Department, Universitat de València, Edifici Jeroni Mu˜ noz, 50th Dr. Moliner, 46100 Burjassot, Spain
a r t i c l e
i n f o
Article history: Received 30 April 2008 Received in revised form 7 July 2008 Accepted 11 July 2008 Available online 31 July 2008 Keywords: On-line liquid chromatography–Fourier transform infrared spectrometry Background correction based on the use of a reference spectra matrix Carbohydrate analysis
a b s t r a c t An on-line gradient reversed phase liquid chromatography–Fourier transform infrared spectrometry (LC–FTIR) method was developed for the determination of fructose, glucose, sucrose and maltose in beverages. Improved chromatographic resolution was obtained using a linear gradient from 75 to 55% (v/v) acetonitrile in water in 15 min. Changes in the background spectra were corrected employing “univariate background correction based on the use of a reference spectra matrix” (UBC-RSM) and using the ratio of absorbance (AR) at 2256 and 2253 cm−1 for the identification of the eluent spectra within the RSM. The method provided limits of detection in the order of 0.75 mg ml−1 . The precision (as relative standard deviation) ranged between 3.3 and 4.1% for glucose and fructose, respectively at a concentration level of 3.0 mg ml−1 . Quantitative recovery values on spiked samples confirmed the accuracy of the method. A set of samples from the Spanish market were analysed to test the suitability of the procedure. © 2008 Elsevier B.V. All rights reserved.
1. Introduction The term sugar is frequently used to describe monosaccharides as glucose and fructose and disaccharides as sucrose and maltose that are absorbed, digested and fully metabolized [1]. The current interest in the physiological role of carbohydrates, the technological developments in food processing and manufacturing, and the different existing mandates of nutrition labeling (e.g. Food and Drug Administration [2]) have created a need for carbohydrate analysis at different production stages in food industries like in alcoholic and non-alcoholic drinks, fruit juices, sweets or dairy products. Liquid chromatography (LC) has often been employed for sugar analysis in different food matrices where the most commonly used detector is the refractive index detector (RID). However, it shows poor sensitivity, high instability with regard to fluctuations in mobile phase composition and eluent temperature, low selectivity and incompatibility with mobile phase gradients [3,4]. Evaporative light scattering detection (ELSD) is compatible with gradient elution and provides a significant increase in sensitivity as compared with RID, but it is also a low-selective detector. UV detection is not directly applicable for sugar analysis, without a pre- or post-column derivatization of the analytes, due to
∗ Corresponding author. Tel.: +34 96 3544838; fax: +34 96 3544838. E-mail address: [email protected] (M. de la Guardia). 0039-9140/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.talanta.2008.07.036
the low UV-absorbance of these compounds. The short wavelength required for their detection in UV reduces the selectivity of the obtained chromatographic signal increasing the number of possible interferences, thus requiring extensive sample clean-up prior to the detection. Mass spectrometry (MS) is an expensive detection technique which provides high selectivity and sensitivity levels, but its field of application focuses on compounds at trace levels and not in the percentage range as it is the case of the main sugars present in foods. Alternatively, infrared (IR) spectrometry is a versatile analytical tool which can be used for both, qualitative and quantitative determinations regarded as a general detection tool for organic analytes. On-line hyphenated with separation techniques as LC or capillary electrophoresis (CE) increases both, the applicability and the accuracy of IR-based methods by a significant reduction of potential spectral interferences. On the other hand, with limits of detection in the high mg l−1 range, it is clear that this technique cannot meet the current demands of trace analysis. However, IR spectrometry has proven to be a simple and rapid technique for quantitative and qualitative determination of analytes in the percentage level. On-line LC–IR methods carried out under isocratic conditions use a constant eluent reference spectrum to correct the eluent spectral contribution. The advantages of this approach are its simplicity and the good quality of the recovered spectra obtained. Using a constant mobile phase composition, also known as isocratic conditions, the capability of on-line LC–Fourier transform
780
J. Kuligowski et al. / Talanta 77 (2008) 779–785
infrared (FTIR) spectrometry for the determination of sucrose, glucose and fructose in aqueous samples was demonstrated by Vonach et al. [5,6]. Edelmann et al. [7] reported the use of a quantum cascade laser as mid-IR source for the direct determination by on-line LC–IR of glucose and fructose in wine samples. Later, Edelmann et al. [8] used on-line LC–attenuated total reflectance (ATR) measurements for the analysis of organic acids, sugars and alcohols in red wine using multivariate curve resolution–alternating least squares (MCR–ALS) for the quantitative analysis of overlapping compounds. Moreover, a FTIR spectrometer has been on-line coupled with CE for the separation and quantification of sucrose, glucose and fructose in fruit juices [9]. The use of micro-machined flow cells in on-line CE–FTIR enables non-destructive, real time detection of analytes. However from the instrumental point of view the complexity of the required set-up for the CE–FTIR coupling is much higher than that for LC–FTIR. On the other hand, the use of on-line gradient LC–FTIR is still challenging, because the use of a constant reference spectrum is not suitable if the relative concentration of the different IR-absorbing mobile phase components (e.g. water, acetonitrile or methanol) is modified during the chromatographic run [10]. Therefore, in recent years much attention has been paid to background correction in order to increase both, the applicability and the peak separation capabilities of on-line LC–FTIR [11–13]. Recently, Quintás et al. proposed a strategy named “univariate background correction method based on the use of a reference spectra matrix (UBC-RSM)” to perform an automated background correction in continuous liquid flow systems [14]. This approach is based on the assumption that using characteristic absorption bands of the mobile phase, it is possible to correct the background eluent spectral contribution to the overall absorption, by using a previously measured reference spectra matrix which covers a wide range of mobile phase compositions. This background correction method has already been successfully applied using acetonitrile:water (1% acetic acid) [14] and methanol:water [15] mobile phase gradients in reversed phase LC for the determination of atrazine and diuron pesticides in aqueous solutions and for the determination of the critical conditions of polyethyleneglycol, respectively. In this study, FTIR is used as an on-line detector for the LC separation under gradient conditions of four model carbohydrates (fructose, glucose, sucrose and maltose) in beverages in order to: (i) test the suitability of the UBC-RSM approach in the presence of increasing concentrations of sugars; (ii) enable a higher sample throughput than that obtained under isocratic conditions in an equivalent LC–FTIR system and (iii) evaluate the method for the analysis of commercially available beverage samples.
trum of the empty sample compartment as background, with a resolution of 8 cm−1 and a zero filling factor of 2. Zero filling consists in adding zeros on both ends of the interferogram before the Fourier transformation so that the spectral lines have a smoother shape. During on-line LC–FTIR gradient experiments, 25 scans per spectrum were averaged, providing a spectra acquisition frequency of 15 spectra min−1 . For instrumental and measurement control and data acquisition, the OPUS program (Version 4.1) from Bruker was employed. Background correction and data treatment were run under Matlab 7.0 from Mathworks (Natick, USA, 2004) using in-house written Matlab files available from the authors of this paper. d(−)-Fructose, d(+)-glucose, sucrose and maltose-1-hydrate of analytical grade were purchased from Scharlab (Barcelona, Spain). Beverage samples were directly obtained from the Spanish market. 2.2. Sample preparation A volume of homogenized sample between 50 and 500 l was introduced in a 5 ml volumetric flask. Then 3.5 ml of acetonitrile were added and the flask was filled up to volume with water. Before the injection in the chromatographic system the solution obtained was sonicated in an ultrasound water bath for 5 min and filtered through a 0.22 m PFTE membrane. Carbonated liquid samples were previously degassed in an ultrasound water bath for 15 min prior to their dilution. In order to increase the applicability of the method reducing possible interferences, acetonitrile was selected for sample dilution and chromatographic separation because according to a previously published work [16] it precipitates proteins and starch present in sample matrices. 2.3. Univariate background correction based on the use of a reference spectra matrix (UBC-RSM) A detailed description of the UBC-RSM method can be found in a previous work [14]. In brief, the eluent correction method can be divided in five steps: in Step 1 a reference spectra matrix RSM (r, c) is measured. The acquisition of the RSM is carried out in practice by measuring a LC gradient in a defined composition range. This step also include the measurement of the sample matrix SM (z, c). The eluent composition range of the SM should be within the RSM composition interval. Step 2 involves the calculation of the absorbance ratio (AR) at two selected wavenumbers (r1 and r2 ) for each spectrum included in the SM and the RSM. The obtained AR values are characteristic for the mobile phase composition as defined in Eq. (1).
2. Experimental ARs = 2.1. Apparatus and reagents A Dionex (Sunnyvale, CA, USA) P680 high performance liquid chromatograph system, equipped with a Kromasil 100 NH2 column (250 mm × 2 mm, 5 m) fro, Spain) and a sample injection loop of 20 l, was employed for chromatographic separations. Linear acetonitrile:water gradients were run from 75 to 55% acetonitrile (Merck, Darmstadt, Germany) in 15 min. A flow cell with CaF2 and ZnSe windows and a pathlength of 10 m installed on a Bruker (Bremen, Germany) IFS 66/v FTIR spectrometer equipped with a liquid nitrogen refrigerated mercury–cadmium–telluride (MCT) detector, a vacuum system and a dry air purged sample compartment was employed for the FTIR spectra acquisition. The scanner for the interferometer was operated at a HeNe laser modulation frequency of 100 kHz. Spectra were recorded in the range between 4000 and 950 cm−1 using the spec-
yrs1 yrs2
(1)
where yrs1 and yrs2 are the absorbance values at the wavenumbers
r1 and r2 (cm−1 ) measured in the spectra s = (1, . . ., z) for spectra included in the SM and s = (1, . . ., r) for the RSM.In Step 3, for each of the z spectra included in the SM, the most appropriated background spectrum (S,s , s = 1, . . ., z) included in the RSM is located. Step 4 consists of the calculation of a correction factor (KF) which is determined for each sample spectrum. The objective of the KF is to correct slight changes in the spectral intensity of the eluent during the run. The KF is defined as the ratio of absorbance of the sample Ss ) and the previously selected background Ss at wavenumber ϕ (yϕ S,s ) using the following spectrum S,s at a defined wavenumber ϕ (yϕ expression: KFs =
Ss yϕ
S,s yϕ
(2)
J. Kuligowski et al. / Talanta 77 (2008) 779–785
781
Step 5 includes the subtraction to the eluent background spectrum from the sample spectrum using the following expression: CorrectedSs = Ss − KFs S,s
(3)
where CorrectedSs is the background corrected sample spectrum s; Ss is the original sample spectrum; S,s is the background spectrum included in the RSM and KFs is the calculated correction factor for the sample spectrum Ss . The use of the KF is optional and a prior evaluation of its usefulness is recommended. 3. Results and discussion 3.1. FTIR spectra of sugars Fig. 1 shows the spectra of solutions of fructose, glucose, sucrose and maltose dissolved in 70:30 acetonitrile:water. All four studied compounds show the characteristic absorption bands of carbohydrates in the spectral region between 1500 and 950 cm−1 . The bands at 1035, 1015 and 1260 cm−1 were identified to, respectively, the C–O stretch, C–C stretch and C–OH deformation modes [9,17,18]. So, it can be concluded that any of the aforementioned bands are suitable for the quantitation of sugars in beverages. However, in order to do a correct quantification of sugars in LC–FTIR, the changes in the spectrum of the mobile phase during the elution of analytes must be carefully checked to achieve an appropriate chemometric background correction.
Fig. 1. FTIR spectra of sugars in acetonitrile:water (70:30, v/v). Fructose (6.1 mg ml−1 ); glucose (6.0 mg ml−1 ); sucrose (6.2 mg ml−1 ) and maltose (6.0 mg ml−1 ). Spectra are shifted along the y-axis for a better clarity.
Fig. 2. FTIR spectra of an acetonitrile:water (70:30, v/v) mixture (A) and variation of the absorbance at 1639 cm−1 (B). In this case the arrows indicate the sugar elution time windows.
782
J. Kuligowski et al. / Talanta 77 (2008) 779–785
3.2. Changes in the FTIR spectra of the mobile phase components during the elution of the analytes For a proper use of the RSM-based background correction method, the mobile phase spectra should present at least an absorption band which could be used to identify the exact composition of the eluent during the chromatographic run. Special attention must be paid on the selection of the wavenumbers r1 and r2 employed for the calculation of AR because spectral overlapping between the mobile phase and the eluting analytes at the wavenumbers leads to an inaccurate identification of the eluent spectra in the RSM, thus causing severe errors in both, the spectra and elution profiles [14]. Additional difficulties can be created by modification in position and bandwidth due to intermolecular interactions between analytes and the mobile phase components. Fig. 2A shows the spectrum between 4000 and 950 cm−1 of an acetonitrile:water (70:30, v/v) mixture. The spectrum showed four main water absorption bands at ∼3400 cm−1 (stretching band), 2115 cm−1 (combination band), 1639 cm−1 (deformation band) and a latter one starting near 1000 cm−1 (water libration bands), in good agreement with reported data [9,17,18]. When using FTIR detection in aqueous systems, the intense absorptivity of water at ∼3400 cm−1 leaves no detectable light, reducing the signal to noise ratio and obscuring the absorption of the analytes in this spectral region. Because of that, the region above 2400 cm−1 was not used throughout this work. Two main acetonitrile bands were clearly distinguishable at 2252 and 2291 cm−1 , the first corresponding to the CN stretching mode (2 ) while the second one was assigned to a combination of both CH3 bending (3 ) and C–C stretching (4 ) modes [19]. It is well known that IR water bands can be modified by the presence of sugars, increasing considerably the difficulty of performing an adequate IR background correction under gradient conditions. Recent studies of the interaction of water and carbohydrates by Max and Chapados [17,18], have evidenced that the water deformation band (ıHOH ) at 1640 cm−1 increases its intensity and slightly shifts
towards the blue region due to the formation of sugar hydrates, being displaced even several wavenumbers. Furthermore, the spectra and abundances of the different hydrates of fructose, glucose or sucrose in aqueous solutions depend on their concentration [17,18]. To evaluate the effects of gradually increasing carbohydrate concentrations on the spectra of an acetonitrile:water mixture, 20 l of 5 mg ml−1 fructose, glucose, sucrose and maltose solutions were injected in an acetonitrile:water (70:30) flow system. It was found that, at the studied concentrations, isocratic conditions and using a spectral resolution of 8 cm−1 , the position of the maximum of the water deformation band at 1639 cm−1 is constant during the elution of the studied carbohydrates. On the contrary, a slight increase in its intensity can be observed (see Fig. 2B). Therefore, in order to assure the accuracy of the background correction process during the elution of the four analytes it is not recommended to use the band at 1639 cm−1 for the characterization of the mobile phase composition. In contrary no effect on the peak shape of the CN bands in the 2300–2200 cm−1 spectral region was observed, therefore indicating that at this concentration level the spectra of the eluent in this region is not affected by the presence of the analytes. 3.3. Selection of the UBC-RSM background parameters One of the main features of the UBC-RSM method is its simplicity, limiting the user interaction during the background correction to the selection of the wavenumbers r1 and r2 used for the calculation of the AR. For an easy selection of these wavenumbers, all the
Table 1 Absorbance ratios (AR) that provided the lowest noise values in the extracted chromatograms Noise as RMS
1st wavenumber (cm−1 )
2nd wavenumber (cm−1 )
Chromatogram extracted using absorbance values at 2063 cm−1 9.082 × 10−5 2256.5 2252.6 0.00010885 2260.3 2252.6 0.00011515 2264.2 2248.7 0.00012287 2260.3 2248.7 0.00012594 2291.2 2268 Chromatogram extracted using absorbance values at 1640 cm−1 0.0029791 2256.5 2252.6 0.0037747 2291.2 2268 0.0038919 2260.3 2252.6 0.003985 2264.2 2248.7 0.0040928 2268 2244.9 Chromatogram extracted using absorbance values at 1442 cm−1 0.0004805 2256.5 2252.6 0.00053281 2260.3 2256.5 0.0005719 2248.7 2268 0.00058188 2264.2 2252.6 0.00061794 2298.9 2291.2 Chromatogram extracted using absorbance values at 1065 cm−1 0.00041775 2295 2244.9 0.00042011 2291.2 2268 0.00042661 2256.6 2252.6 0.00044693 2298.9 2268 0.00044947 2260.3 2248.7 Noise measured using the root mean square of the extracted chromatogram at 2063, 1640, 1442 and 1065 cm−1 in the background corrected blank gradient injection.
Fig. 3. On-line gradient LC–FTIR spectra obtained from the injection of a sugar standard solution containing 2.5 mg ml−1 of fructose, glucose, sucrose and maltose. (A) Raw spectra between 2200 and 1000 cm−1 . (B) Spectra in the wavenumber range between 1500 and 1000 cm−1 corrected using the UBC-RSM method.
J. Kuligowski et al. / Talanta 77 (2008) 779–785
783
Fig. 4. LC–FTIR chromatograms extracted from the injection of a standard solution containing 5.5 mg ml−1 of fructose, glucose, sucrose and maltose (red) and from the injection of sample 1 (blue). Chromatograms were obtained from the measurement of the area between 1108 and 1069 cm−1 , corrected using a baseline established at 1203 cm−1 . (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
possible ARs within a previously selected range were assayed for an appropriate background selection from the spectra included in the RSM. The employed output function was the noise (measured as RMS) in the extracted chromatograms at four characteristic eluent wavenumbers. So, low chromatographic noise values indicated adequate background correction. In an earlier work [14], for acetonitrile:water (with 1% acetic acid) gradients in the range between (40:60, v/v) and (99:1, v/v), the selection of the AR was made in the spectral region between 2310 and 2180 cm−1 . As a result, the quotient between absorbance values at 2256.3 and 2248.6 cm−1 was selected. In spite of the high similarity between the aforementioned mobile phase and that employed in this work, slight changes in the mobile phase composition might shift the position of eluent bands. To ensure that the AR used throughout this work was appropriate,
the AR selection process described in the aforementioned work [14] was carried out using the absorbance values at 2063, 1640, 1442 and 1065 cm−1 for the evaluation of the chromatographic noise. Table 1 summarizes the five pairs of wavenumbers which provided the lowest noise in the extracted chromatograms using a RSM composed by 986 spectra covering acetonitrile concentrations between 75 and 55% (v/v) acetonitrile and a blank gradient injection formed by 317 spectra of mobile phase solutions in the same concentration range. The noise level in the extracted chromatograms at 2063, 1640, 1442 cm−1 reached the minimum value using the absorbance at 2256.5 and 2252.6 cm−1 for the calculation of the AR, obtaining RMS values between 9 × 10−5 and 2.9 × 10−3 for 2063 and 1640 cm−1 , respectively. On the other hand, the minimum noise in the extracted chromatogram at 1065 cm−1 was achieved using the AR calculated
Fig. 5. Recovered spectra of fructose, glucose, sucrose and maltose at their respective peak apex extracted from the sugar standard solution LC run shown in Fig. 4, compared to reference spectra.
J. Kuligowski et al. / Talanta 77 (2008) 779–785
f
e
c
d
Calibration curve from 10 standard solutions. a and b are the intercept and the slope of the calibration lines. Chromatographic noise measured as the root mean square (RMS) of the chromatographic signal extracted from a background correct blank gradient injection. Limit of detection established as the concentration at which the signal to noise ratio is higher than 3. Limit of detection in beverage samples for a sample volume of 500 l. Relative standard deviation for six-independent measurements carried out at a concentration level of 3.0 mg ml−1 . Confidence interval calculated as IC = sa· tp,n−2 (p = 95%). a
0.4 0.6 0.5 0.4 0.0037 0.0175 0.0175 0.0175 0.4 0.6 0.5 0.4 0.9995 0.9965 0.9962 0.9980 (0.0264 ± 0.0002) (0.0933 ± 0.0018) (0.0921 ± 0.0011) (0.0895 ± 0.0009)
(b ± sb ) (a ± sa ) (CI)
(0.0008 ± 0.0006) (0.0014) (0.000 ± 0.006) (0.014) (0.006 ± 0.006) (0.014) (0.013 ± 0.007) (0.016) Fructose Glucose Sucrose Maltose
b
0.4 0.6 0.5 0.4
LODc (mg ml−1 ) LODd (g/100 ml) LODc (mg ml−1 ) R2 f
Calibration of analytes was made in the previously described experimental conditions. Fructose chromatograms were obtained from the measurement of the spectra area between 1108 and 1069 cm−1 , corrected using a baseline at 1203 cm−1 . Glucose, sucrose and maltose chromatograms were calculated from area measurements between 1177 and 1025 cm−1 , corrected using a baseline at 1177 cm−1 . Table 2 shows the obtained analytical parameters using peak height values from the extracted chromatograms. The linear regression coefficients obtained indicate a good adjustment of data to each calibration curve. Limits of detection in the 0.4–0.6 mg ml−1 range were estimated for each analyte as the concentration at which the chromatographic signal to noise ratio was higher than 3. Precision of six-independent injections of
Table 2 Analytical figures of merit of the on-line LC–FTIR determination of sugars
3.6. Analytical characteristics of the method
Noiseb
3.5. Identification of the analytes To quantify the accuracy of the background correction, a numerical correlation was made through the determination of a “Correlation coefficient (QC)” defined for two spectra (S1 and S2 ) as the ratio from the covariance (Cov(S1 ,S2 )) and the product of the two standard deviations S1 and S2 . Hence, a QC = 100% indicates identical spectra. Using the spectral region between 1250 and 1060 cm−1 , the obtained QC for the spectra shown in Fig. 5 ranged between 96.25 and 98.73%. On the other hand, QCs were also calculated for the corrected spectra obtained from sample injections obtaining also an excellent agreement between extracted and reference spectra (data not shown), which confirmed the identity of carbohydrates, thus increasing the reliability of the results.
4.1 3.3 3.7 3.8
Repeatabilitye (%)
Fig. 3 shows the original (A) and background corrected (B) spectra acquired during the elution of a standard carbohydrate solution of fructose, glucose, sucrose and maltose at a concentration of 3 mg ml−1 for each one. Visual inspection of the 3D graphs reveals that changes in the background signal due to the mobile phase gradient have been corrected in a great extent by the UBC-RSM subtraction procedure. As an example, LC–FTIR chromatograms extracted from the injection of a standard solution containing 5.5 mg ml−1 of fructose, glucose, sucrose and maltose and from the injection of sample 1 are shown in Fig. 4. The depicted chromatograms were obtained from the measurement of the area values of the spectra between 1108 and 1069 cm−1 , corrected using a baseline established at 1203 cm−1 . It can be seen, that the resolution is very good and the specific spectra of the analytes can clearly be distinguished (see Fig. 5). The retention times, established from four injections of standard mixtures of the considered compounds were 11.8 ± 0.2, 12.7 ± 0.2, 15.0 ± 0.2 and 16.5 ± 0.2 min for fructose, glucose, sucrose and maltose, respectively. As it can be seen in the chromatograms of Fig. 4, the lack of sloping baselines and the random distribution of the chromatographic noise support the suitability of the selected background correction conditions for obtaining the chromatograms of sugars during gradient elution.
0.5–6.0 0.5–5.2 0.5–4.9 0.5–5.2
Linear range (mg ml−1 )
3.4. On-line LC–FTIR separation of carbohydrates
0.5–6.0 0.5–5.2 0.5–4.9 0.5–5.2
Linear range (mg/100 mla )
from absorbance values at 2295 and 2249.5 cm−1 . The difference between the obtained noise using this pair of wavenumbers and that obtained using the absorbance ratio between 2256.5 and 2252.6 cm−1 is only 2.1% and cannot be considered as statistically significant. Based on the foregoing data, the quotient between the absorbance values at 2256.5 and 2252.6 cm−1 was selected for the background correction of further measurements.
Calibration curvea [y = a + bC (mg ml−1 )]
784
J. Kuligowski et al. / Talanta 77 (2008) 779–785
785
Table 3 On-line LC–FTIR determination of sugars in commercial samples Sample
Presentation
Found concentration (g/100 ml)
1 2 3 4 5 6 7
Soft drink Soft drink Soft drink (sugar free) Soft drink Soft drink Energy drink Fruit juice
Label content (g/100 ml)
Fructose
Glucose
Sucrose
Maltose
Total
5.14 6.39
7.84 4.75
13.0 11.1 0.00 9.0 7.3 9.8 10.1
12.9 11.6 0.00 8.9 7.9 10.7 12.4
Table 4 Recovery study of fructose, glucose, sucrose and maltose in spiked commercial samples Sample
Presentation
1 2 4 5 6 7
Soft drink Soft drink Soft drink Soft drink Energy drink Fruit juice a
g/100 ml founda (% recovery)
g/100 ml added Fructose
Glucose
Sucrose
Maltose
Fructose
2.54 2.06 1.00 0.97 1.43 2.20
1.90 2.36 0.93 0.90 0.97 2.18
2.98 3.33 0.97 1.30 1.07 1.74
1.84 3.34 1.19 1.10 1.17 1.57
7.69 8.36 4.48 1.13 1.78 5.92
± ± ± ± ± ±
Glucose
0.08 (100.4) 0.04 (95.6) 0.06 (98.0) 0.08 (100.0) 0.09 (101.4) 0.02 (97.7)
9.84 7.19 5.11 1.76 4.06 5.90
± ± ± ± ± ±
0.09 (105.1) 0.09 (103.5) 0.04 (99.8) 0.09 (98.9) 0.04 (101.9) 0.08 (101.4)
Sucrose 2.92 3.34 2.39 7.54 7.45 4.32
± ± ± ± ± ±
Maltose 0.014 (98.1) 0.02 (100.2) 0.02 (104.0) 0.06 (98.7) 0.03 (95.8) 0.08 (98.1)
1.84 3.39 1.28 1.02 1.15 1.65
± ± ± ± ± ±
0.05 (100.2) 0.09 (101.3) 0.05 (99.1) 0.08 (92.5) 0.07 (98.4) 0.07 (104.9)
Recoveries from two-independent replicates.
a standard calibration solution containing 3.0 mg ml−1 of each analyte provided relative standard deviations ranging between 3.3 and 4.1% for glucose and fructose, respectively. This obtained sensitivity and precision results together with the good correlation factors obtained for the recovered analyte spectra suggested that they were appropriate for the quantitative determination and identification of sugars in beverages. 3.7. Determination of sugars in beverages The applicability of the proposed method was evidenced by the analysis of a set of 7 commercially available samples purchased from the Spanish market. Sample treatment was reduced to an adequate dilution and filtration of samples before their injection in the chromatographic system. Table 3 lists the different obtained sugar contents from duplicate analysis of diluted samples. Additionally, a recovery study was carried out in six of the analysed samples. The employed spiked concentration ranges were: 0.97–2.54 (g/100 ml), 0.90–2.36 (g/100 ml), 0.97–3.133 (g/100 ml) and 1.10–3.34 (g/100 ml) for fructose, glucose, sucrose and maltose, respectively. Mean recovery values ranged between 105.1 and 92.5% (see Table 4). 4. Conclusions A simple on-line gradient LC method in combination with FTIR detection and chemometrics has been developed for the determination of four characteristic carbohydrates in beverages. Fructose, glucose, sucrose and maltose could be separated, identified and quantified in commercial samples with limits of detection between 0.4 and 0.6 mg ml−1 . The use of the univariate method for the selection of the background from a reference spectra matrix permitted an accurate chemometric eluent correction yielding distinguishable spectra of good quality for the studied analytes with correlation factors between 96.25 and 98.73%. The main advantages provided by the proposed on-line LC–FTIR approach include instrumental simplicity and low cost (due to the possibility of using commercial flow cells).
Acknowledgements Authors acknowledge the financial support of Ministerio de Educación y Ciencia (Project CTQ205-05604, FEDER) and Direcció General d’Investigació I Transferència Tecnològica de la Generalitat Valenciana (Project ACOMP/2007/131). J.K. acknowledges the “V Segles” grant provided by the University of Valencia to carry out this study. G.Q. is grateful for a post-doctoral grant (“Ayudas para estancias de doctores en centros de investigación de excelencia de la Comunidad Valenciana”) from the Conselleria de Industria, Generalitat Valenciana (Spain). References [1] M.L. Wheeler, F.X. Pi-Sunyer, J. Am. Diet Assoc. 108 (2008) S34. [2] FDA Nutrition Labeling Manual, U.S. Food and Drug Administration, Center for Food Safety and Applied Nutrition, 1998. [3] A. Cáceres, S. Cárdenas, M. Gallego, A. Rodríguez, M. Valcárcel, Chromatographia 52 (2000) 314. [4] W.L. Qian, Z. Khan, D.G. Watson, J. Fearnley, J. Food Comp. Anal. 21 (1) (2008) 78. [5] R. Vonach, B. Lendl, R. Kellner, J. Chromatogr. A 824 (1998) 159. [6] R. Vonach, B. Lendl, R. Kellner, Anal. Chem. 69 (1997) 4286. [7] A. Edelmann, C. Ruzicka, J. Frank, B. Lendl, W. Schrenk, E. Gornik, G. Strasser, J. Chromatogr. A 934 (2001) 123. [8] A. Edelmann, K. Diewok, J. Rodriguez-Baena, B. Lendl, Anal. Bioanal. Chem. 376 (2003) 92. [9] M. Kölhed, B. Karlberg, Analyst 130 (2005) 772. [10] G.W. Somsen, C. Gooijer, U.A.Th. Brinkman, J. Chromatogr. A 856 (1999) 213. [11] K. István, R. Rajkó, G. Keresztury, J. Chromatogr. A 1104 (1-2) (2006) 154. [12] R.J. Dijkstra, H.F.M. Boelens, J.A. Westerhuis, F. Ariese, U.A.Th. Brinkman, C. Gooijer, Anal. Chim. Acta 519 (2004) 129. [13] H.F.M. Boelens, R.J. Dijkstra, P.H.C. Eilers, F. Fitzpatrick, J.A. Westerhuis, J. Chromatogr. A 1057 (2004) 21. [14] G. Quintás, B. Lendl, S. Garrigues, M. de la Guardia, J. Chromatogr. A 1190 (2008) 102. [15] J. Kuligowski, G. Quintás, S. Garrigues, M. de la Guardia, Anal. Chim. Acta 624 (2) (2008) 278. [16] J.T. Gotsick, R.F. Benson, J. Liq. Chromatogr. 14 (1991) 1887. [17] J.J. Max, C. Chapados, J. Phys. Chem. A 111 (2007) 2679. [18] J.J. Max, C. Chapados, J. Phys. Chem. A 105 (2001) 10681. [19] T. Takamuku, M. Tabata, A. Yamaguchi, J. Nishimoto, M. Kumamoto, H. Wakita, T. Yamaguchi, J. Phys. Chem. B 102 (1998) 8880.
Contents lists available at ScienceDirect
Talanta journal homepage: www.elsevier.com/locate/talanta
On-line gradient liquid chromatography–Fourier transform infrared spectrometry determination of sugars in beverages using chemometric background correction J. Kuligowski, G. Quintás, S. Garrigues, M. de la Guardia ∗ Analytical Chemistry Department, Universitat de València, Edifici Jeroni Mu˜ noz, 50th Dr. Moliner, 46100 Burjassot, Spain
a r t i c l e
i n f o
Article history: Received 30 April 2008 Received in revised form 7 July 2008 Accepted 11 July 2008 Available online 31 July 2008 Keywords: On-line liquid chromatography–Fourier transform infrared spectrometry Background correction based on the use of a reference spectra matrix Carbohydrate analysis
a b s t r a c t An on-line gradient reversed phase liquid chromatography–Fourier transform infrared spectrometry (LC–FTIR) method was developed for the determination of fructose, glucose, sucrose and maltose in beverages. Improved chromatographic resolution was obtained using a linear gradient from 75 to 55% (v/v) acetonitrile in water in 15 min. Changes in the background spectra were corrected employing “univariate background correction based on the use of a reference spectra matrix” (UBC-RSM) and using the ratio of absorbance (AR) at 2256 and 2253 cm−1 for the identification of the eluent spectra within the RSM. The method provided limits of detection in the order of 0.75 mg ml−1 . The precision (as relative standard deviation) ranged between 3.3 and 4.1% for glucose and fructose, respectively at a concentration level of 3.0 mg ml−1 . Quantitative recovery values on spiked samples confirmed the accuracy of the method. A set of samples from the Spanish market were analysed to test the suitability of the procedure. © 2008 Elsevier B.V. All rights reserved.
1. Introduction The term sugar is frequently used to describe monosaccharides as glucose and fructose and disaccharides as sucrose and maltose that are absorbed, digested and fully metabolized [1]. The current interest in the physiological role of carbohydrates, the technological developments in food processing and manufacturing, and the different existing mandates of nutrition labeling (e.g. Food and Drug Administration [2]) have created a need for carbohydrate analysis at different production stages in food industries like in alcoholic and non-alcoholic drinks, fruit juices, sweets or dairy products. Liquid chromatography (LC) has often been employed for sugar analysis in different food matrices where the most commonly used detector is the refractive index detector (RID). However, it shows poor sensitivity, high instability with regard to fluctuations in mobile phase composition and eluent temperature, low selectivity and incompatibility with mobile phase gradients [3,4]. Evaporative light scattering detection (ELSD) is compatible with gradient elution and provides a significant increase in sensitivity as compared with RID, but it is also a low-selective detector. UV detection is not directly applicable for sugar analysis, without a pre- or post-column derivatization of the analytes, due to
∗ Corresponding author. Tel.: +34 96 3544838; fax: +34 96 3544838. E-mail address: [email protected] (M. de la Guardia). 0039-9140/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.talanta.2008.07.036
the low UV-absorbance of these compounds. The short wavelength required for their detection in UV reduces the selectivity of the obtained chromatographic signal increasing the number of possible interferences, thus requiring extensive sample clean-up prior to the detection. Mass spectrometry (MS) is an expensive detection technique which provides high selectivity and sensitivity levels, but its field of application focuses on compounds at trace levels and not in the percentage range as it is the case of the main sugars present in foods. Alternatively, infrared (IR) spectrometry is a versatile analytical tool which can be used for both, qualitative and quantitative determinations regarded as a general detection tool for organic analytes. On-line hyphenated with separation techniques as LC or capillary electrophoresis (CE) increases both, the applicability and the accuracy of IR-based methods by a significant reduction of potential spectral interferences. On the other hand, with limits of detection in the high mg l−1 range, it is clear that this technique cannot meet the current demands of trace analysis. However, IR spectrometry has proven to be a simple and rapid technique for quantitative and qualitative determination of analytes in the percentage level. On-line LC–IR methods carried out under isocratic conditions use a constant eluent reference spectrum to correct the eluent spectral contribution. The advantages of this approach are its simplicity and the good quality of the recovered spectra obtained. Using a constant mobile phase composition, also known as isocratic conditions, the capability of on-line LC–Fourier transform
780
J. Kuligowski et al. / Talanta 77 (2008) 779–785
infrared (FTIR) spectrometry for the determination of sucrose, glucose and fructose in aqueous samples was demonstrated by Vonach et al. [5,6]. Edelmann et al. [7] reported the use of a quantum cascade laser as mid-IR source for the direct determination by on-line LC–IR of glucose and fructose in wine samples. Later, Edelmann et al. [8] used on-line LC–attenuated total reflectance (ATR) measurements for the analysis of organic acids, sugars and alcohols in red wine using multivariate curve resolution–alternating least squares (MCR–ALS) for the quantitative analysis of overlapping compounds. Moreover, a FTIR spectrometer has been on-line coupled with CE for the separation and quantification of sucrose, glucose and fructose in fruit juices [9]. The use of micro-machined flow cells in on-line CE–FTIR enables non-destructive, real time detection of analytes. However from the instrumental point of view the complexity of the required set-up for the CE–FTIR coupling is much higher than that for LC–FTIR. On the other hand, the use of on-line gradient LC–FTIR is still challenging, because the use of a constant reference spectrum is not suitable if the relative concentration of the different IR-absorbing mobile phase components (e.g. water, acetonitrile or methanol) is modified during the chromatographic run [10]. Therefore, in recent years much attention has been paid to background correction in order to increase both, the applicability and the peak separation capabilities of on-line LC–FTIR [11–13]. Recently, Quintás et al. proposed a strategy named “univariate background correction method based on the use of a reference spectra matrix (UBC-RSM)” to perform an automated background correction in continuous liquid flow systems [14]. This approach is based on the assumption that using characteristic absorption bands of the mobile phase, it is possible to correct the background eluent spectral contribution to the overall absorption, by using a previously measured reference spectra matrix which covers a wide range of mobile phase compositions. This background correction method has already been successfully applied using acetonitrile:water (1% acetic acid) [14] and methanol:water [15] mobile phase gradients in reversed phase LC for the determination of atrazine and diuron pesticides in aqueous solutions and for the determination of the critical conditions of polyethyleneglycol, respectively. In this study, FTIR is used as an on-line detector for the LC separation under gradient conditions of four model carbohydrates (fructose, glucose, sucrose and maltose) in beverages in order to: (i) test the suitability of the UBC-RSM approach in the presence of increasing concentrations of sugars; (ii) enable a higher sample throughput than that obtained under isocratic conditions in an equivalent LC–FTIR system and (iii) evaluate the method for the analysis of commercially available beverage samples.
trum of the empty sample compartment as background, with a resolution of 8 cm−1 and a zero filling factor of 2. Zero filling consists in adding zeros on both ends of the interferogram before the Fourier transformation so that the spectral lines have a smoother shape. During on-line LC–FTIR gradient experiments, 25 scans per spectrum were averaged, providing a spectra acquisition frequency of 15 spectra min−1 . For instrumental and measurement control and data acquisition, the OPUS program (Version 4.1) from Bruker was employed. Background correction and data treatment were run under Matlab 7.0 from Mathworks (Natick, USA, 2004) using in-house written Matlab files available from the authors of this paper. d(−)-Fructose, d(+)-glucose, sucrose and maltose-1-hydrate of analytical grade were purchased from Scharlab (Barcelona, Spain). Beverage samples were directly obtained from the Spanish market. 2.2. Sample preparation A volume of homogenized sample between 50 and 500 l was introduced in a 5 ml volumetric flask. Then 3.5 ml of acetonitrile were added and the flask was filled up to volume with water. Before the injection in the chromatographic system the solution obtained was sonicated in an ultrasound water bath for 5 min and filtered through a 0.22 m PFTE membrane. Carbonated liquid samples were previously degassed in an ultrasound water bath for 15 min prior to their dilution. In order to increase the applicability of the method reducing possible interferences, acetonitrile was selected for sample dilution and chromatographic separation because according to a previously published work [16] it precipitates proteins and starch present in sample matrices. 2.3. Univariate background correction based on the use of a reference spectra matrix (UBC-RSM) A detailed description of the UBC-RSM method can be found in a previous work [14]. In brief, the eluent correction method can be divided in five steps: in Step 1 a reference spectra matrix RSM (r, c) is measured. The acquisition of the RSM is carried out in practice by measuring a LC gradient in a defined composition range. This step also include the measurement of the sample matrix SM (z, c). The eluent composition range of the SM should be within the RSM composition interval. Step 2 involves the calculation of the absorbance ratio (AR) at two selected wavenumbers (r1 and r2 ) for each spectrum included in the SM and the RSM. The obtained AR values are characteristic for the mobile phase composition as defined in Eq. (1).
2. Experimental ARs = 2.1. Apparatus and reagents A Dionex (Sunnyvale, CA, USA) P680 high performance liquid chromatograph system, equipped with a Kromasil 100 NH2 column (250 mm × 2 mm, 5 m) fro, Spain) and a sample injection loop of 20 l, was employed for chromatographic separations. Linear acetonitrile:water gradients were run from 75 to 55% acetonitrile (Merck, Darmstadt, Germany) in 15 min. A flow cell with CaF2 and ZnSe windows and a pathlength of 10 m installed on a Bruker (Bremen, Germany) IFS 66/v FTIR spectrometer equipped with a liquid nitrogen refrigerated mercury–cadmium–telluride (MCT) detector, a vacuum system and a dry air purged sample compartment was employed for the FTIR spectra acquisition. The scanner for the interferometer was operated at a HeNe laser modulation frequency of 100 kHz. Spectra were recorded in the range between 4000 and 950 cm−1 using the spec-
yrs1 yrs2
(1)
where yrs1 and yrs2 are the absorbance values at the wavenumbers
r1 and r2 (cm−1 ) measured in the spectra s = (1, . . ., z) for spectra included in the SM and s = (1, . . ., r) for the RSM.In Step 3, for each of the z spectra included in the SM, the most appropriated background spectrum (S,s , s = 1, . . ., z) included in the RSM is located. Step 4 consists of the calculation of a correction factor (KF) which is determined for each sample spectrum. The objective of the KF is to correct slight changes in the spectral intensity of the eluent during the run. The KF is defined as the ratio of absorbance of the sample Ss ) and the previously selected background Ss at wavenumber ϕ (yϕ S,s ) using the following spectrum S,s at a defined wavenumber ϕ (yϕ expression: KFs =
Ss yϕ
S,s yϕ
(2)
J. Kuligowski et al. / Talanta 77 (2008) 779–785
781
Step 5 includes the subtraction to the eluent background spectrum from the sample spectrum using the following expression: CorrectedSs = Ss − KFs S,s
(3)
where CorrectedSs is the background corrected sample spectrum s; Ss is the original sample spectrum; S,s is the background spectrum included in the RSM and KFs is the calculated correction factor for the sample spectrum Ss . The use of the KF is optional and a prior evaluation of its usefulness is recommended. 3. Results and discussion 3.1. FTIR spectra of sugars Fig. 1 shows the spectra of solutions of fructose, glucose, sucrose and maltose dissolved in 70:30 acetonitrile:water. All four studied compounds show the characteristic absorption bands of carbohydrates in the spectral region between 1500 and 950 cm−1 . The bands at 1035, 1015 and 1260 cm−1 were identified to, respectively, the C–O stretch, C–C stretch and C–OH deformation modes [9,17,18]. So, it can be concluded that any of the aforementioned bands are suitable for the quantitation of sugars in beverages. However, in order to do a correct quantification of sugars in LC–FTIR, the changes in the spectrum of the mobile phase during the elution of analytes must be carefully checked to achieve an appropriate chemometric background correction.
Fig. 1. FTIR spectra of sugars in acetonitrile:water (70:30, v/v). Fructose (6.1 mg ml−1 ); glucose (6.0 mg ml−1 ); sucrose (6.2 mg ml−1 ) and maltose (6.0 mg ml−1 ). Spectra are shifted along the y-axis for a better clarity.
Fig. 2. FTIR spectra of an acetonitrile:water (70:30, v/v) mixture (A) and variation of the absorbance at 1639 cm−1 (B). In this case the arrows indicate the sugar elution time windows.
782
J. Kuligowski et al. / Talanta 77 (2008) 779–785
3.2. Changes in the FTIR spectra of the mobile phase components during the elution of the analytes For a proper use of the RSM-based background correction method, the mobile phase spectra should present at least an absorption band which could be used to identify the exact composition of the eluent during the chromatographic run. Special attention must be paid on the selection of the wavenumbers r1 and r2 employed for the calculation of AR because spectral overlapping between the mobile phase and the eluting analytes at the wavenumbers leads to an inaccurate identification of the eluent spectra in the RSM, thus causing severe errors in both, the spectra and elution profiles [14]. Additional difficulties can be created by modification in position and bandwidth due to intermolecular interactions between analytes and the mobile phase components. Fig. 2A shows the spectrum between 4000 and 950 cm−1 of an acetonitrile:water (70:30, v/v) mixture. The spectrum showed four main water absorption bands at ∼3400 cm−1 (stretching band), 2115 cm−1 (combination band), 1639 cm−1 (deformation band) and a latter one starting near 1000 cm−1 (water libration bands), in good agreement with reported data [9,17,18]. When using FTIR detection in aqueous systems, the intense absorptivity of water at ∼3400 cm−1 leaves no detectable light, reducing the signal to noise ratio and obscuring the absorption of the analytes in this spectral region. Because of that, the region above 2400 cm−1 was not used throughout this work. Two main acetonitrile bands were clearly distinguishable at 2252 and 2291 cm−1 , the first corresponding to the CN stretching mode (2 ) while the second one was assigned to a combination of both CH3 bending (3 ) and C–C stretching (4 ) modes [19]. It is well known that IR water bands can be modified by the presence of sugars, increasing considerably the difficulty of performing an adequate IR background correction under gradient conditions. Recent studies of the interaction of water and carbohydrates by Max and Chapados [17,18], have evidenced that the water deformation band (ıHOH ) at 1640 cm−1 increases its intensity and slightly shifts
towards the blue region due to the formation of sugar hydrates, being displaced even several wavenumbers. Furthermore, the spectra and abundances of the different hydrates of fructose, glucose or sucrose in aqueous solutions depend on their concentration [17,18]. To evaluate the effects of gradually increasing carbohydrate concentrations on the spectra of an acetonitrile:water mixture, 20 l of 5 mg ml−1 fructose, glucose, sucrose and maltose solutions were injected in an acetonitrile:water (70:30) flow system. It was found that, at the studied concentrations, isocratic conditions and using a spectral resolution of 8 cm−1 , the position of the maximum of the water deformation band at 1639 cm−1 is constant during the elution of the studied carbohydrates. On the contrary, a slight increase in its intensity can be observed (see Fig. 2B). Therefore, in order to assure the accuracy of the background correction process during the elution of the four analytes it is not recommended to use the band at 1639 cm−1 for the characterization of the mobile phase composition. In contrary no effect on the peak shape of the CN bands in the 2300–2200 cm−1 spectral region was observed, therefore indicating that at this concentration level the spectra of the eluent in this region is not affected by the presence of the analytes. 3.3. Selection of the UBC-RSM background parameters One of the main features of the UBC-RSM method is its simplicity, limiting the user interaction during the background correction to the selection of the wavenumbers r1 and r2 used for the calculation of the AR. For an easy selection of these wavenumbers, all the
Table 1 Absorbance ratios (AR) that provided the lowest noise values in the extracted chromatograms Noise as RMS
1st wavenumber (cm−1 )
2nd wavenumber (cm−1 )
Chromatogram extracted using absorbance values at 2063 cm−1 9.082 × 10−5 2256.5 2252.6 0.00010885 2260.3 2252.6 0.00011515 2264.2 2248.7 0.00012287 2260.3 2248.7 0.00012594 2291.2 2268 Chromatogram extracted using absorbance values at 1640 cm−1 0.0029791 2256.5 2252.6 0.0037747 2291.2 2268 0.0038919 2260.3 2252.6 0.003985 2264.2 2248.7 0.0040928 2268 2244.9 Chromatogram extracted using absorbance values at 1442 cm−1 0.0004805 2256.5 2252.6 0.00053281 2260.3 2256.5 0.0005719 2248.7 2268 0.00058188 2264.2 2252.6 0.00061794 2298.9 2291.2 Chromatogram extracted using absorbance values at 1065 cm−1 0.00041775 2295 2244.9 0.00042011 2291.2 2268 0.00042661 2256.6 2252.6 0.00044693 2298.9 2268 0.00044947 2260.3 2248.7 Noise measured using the root mean square of the extracted chromatogram at 2063, 1640, 1442 and 1065 cm−1 in the background corrected blank gradient injection.
Fig. 3. On-line gradient LC–FTIR spectra obtained from the injection of a sugar standard solution containing 2.5 mg ml−1 of fructose, glucose, sucrose and maltose. (A) Raw spectra between 2200 and 1000 cm−1 . (B) Spectra in the wavenumber range between 1500 and 1000 cm−1 corrected using the UBC-RSM method.
J. Kuligowski et al. / Talanta 77 (2008) 779–785
783
Fig. 4. LC–FTIR chromatograms extracted from the injection of a standard solution containing 5.5 mg ml−1 of fructose, glucose, sucrose and maltose (red) and from the injection of sample 1 (blue). Chromatograms were obtained from the measurement of the area between 1108 and 1069 cm−1 , corrected using a baseline established at 1203 cm−1 . (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
possible ARs within a previously selected range were assayed for an appropriate background selection from the spectra included in the RSM. The employed output function was the noise (measured as RMS) in the extracted chromatograms at four characteristic eluent wavenumbers. So, low chromatographic noise values indicated adequate background correction. In an earlier work [14], for acetonitrile:water (with 1% acetic acid) gradients in the range between (40:60, v/v) and (99:1, v/v), the selection of the AR was made in the spectral region between 2310 and 2180 cm−1 . As a result, the quotient between absorbance values at 2256.3 and 2248.6 cm−1 was selected. In spite of the high similarity between the aforementioned mobile phase and that employed in this work, slight changes in the mobile phase composition might shift the position of eluent bands. To ensure that the AR used throughout this work was appropriate,
the AR selection process described in the aforementioned work [14] was carried out using the absorbance values at 2063, 1640, 1442 and 1065 cm−1 for the evaluation of the chromatographic noise. Table 1 summarizes the five pairs of wavenumbers which provided the lowest noise in the extracted chromatograms using a RSM composed by 986 spectra covering acetonitrile concentrations between 75 and 55% (v/v) acetonitrile and a blank gradient injection formed by 317 spectra of mobile phase solutions in the same concentration range. The noise level in the extracted chromatograms at 2063, 1640, 1442 cm−1 reached the minimum value using the absorbance at 2256.5 and 2252.6 cm−1 for the calculation of the AR, obtaining RMS values between 9 × 10−5 and 2.9 × 10−3 for 2063 and 1640 cm−1 , respectively. On the other hand, the minimum noise in the extracted chromatogram at 1065 cm−1 was achieved using the AR calculated
Fig. 5. Recovered spectra of fructose, glucose, sucrose and maltose at their respective peak apex extracted from the sugar standard solution LC run shown in Fig. 4, compared to reference spectra.
J. Kuligowski et al. / Talanta 77 (2008) 779–785
f
e
c
d
Calibration curve from 10 standard solutions. a and b are the intercept and the slope of the calibration lines. Chromatographic noise measured as the root mean square (RMS) of the chromatographic signal extracted from a background correct blank gradient injection. Limit of detection established as the concentration at which the signal to noise ratio is higher than 3. Limit of detection in beverage samples for a sample volume of 500 l. Relative standard deviation for six-independent measurements carried out at a concentration level of 3.0 mg ml−1 . Confidence interval calculated as IC = sa· tp,n−2 (p = 95%). a
0.4 0.6 0.5 0.4 0.0037 0.0175 0.0175 0.0175 0.4 0.6 0.5 0.4 0.9995 0.9965 0.9962 0.9980 (0.0264 ± 0.0002) (0.0933 ± 0.0018) (0.0921 ± 0.0011) (0.0895 ± 0.0009)
(b ± sb ) (a ± sa ) (CI)
(0.0008 ± 0.0006) (0.0014) (0.000 ± 0.006) (0.014) (0.006 ± 0.006) (0.014) (0.013 ± 0.007) (0.016) Fructose Glucose Sucrose Maltose
b
0.4 0.6 0.5 0.4
LODc (mg ml−1 ) LODd (g/100 ml) LODc (mg ml−1 ) R2 f
Calibration of analytes was made in the previously described experimental conditions. Fructose chromatograms were obtained from the measurement of the spectra area between 1108 and 1069 cm−1 , corrected using a baseline at 1203 cm−1 . Glucose, sucrose and maltose chromatograms were calculated from area measurements between 1177 and 1025 cm−1 , corrected using a baseline at 1177 cm−1 . Table 2 shows the obtained analytical parameters using peak height values from the extracted chromatograms. The linear regression coefficients obtained indicate a good adjustment of data to each calibration curve. Limits of detection in the 0.4–0.6 mg ml−1 range were estimated for each analyte as the concentration at which the chromatographic signal to noise ratio was higher than 3. Precision of six-independent injections of
Table 2 Analytical figures of merit of the on-line LC–FTIR determination of sugars
3.6. Analytical characteristics of the method
Noiseb
3.5. Identification of the analytes To quantify the accuracy of the background correction, a numerical correlation was made through the determination of a “Correlation coefficient (QC)” defined for two spectra (S1 and S2 ) as the ratio from the covariance (Cov(S1 ,S2 )) and the product of the two standard deviations S1 and S2 . Hence, a QC = 100% indicates identical spectra. Using the spectral region between 1250 and 1060 cm−1 , the obtained QC for the spectra shown in Fig. 5 ranged between 96.25 and 98.73%. On the other hand, QCs were also calculated for the corrected spectra obtained from sample injections obtaining also an excellent agreement between extracted and reference spectra (data not shown), which confirmed the identity of carbohydrates, thus increasing the reliability of the results.
4.1 3.3 3.7 3.8
Repeatabilitye (%)
Fig. 3 shows the original (A) and background corrected (B) spectra acquired during the elution of a standard carbohydrate solution of fructose, glucose, sucrose and maltose at a concentration of 3 mg ml−1 for each one. Visual inspection of the 3D graphs reveals that changes in the background signal due to the mobile phase gradient have been corrected in a great extent by the UBC-RSM subtraction procedure. As an example, LC–FTIR chromatograms extracted from the injection of a standard solution containing 5.5 mg ml−1 of fructose, glucose, sucrose and maltose and from the injection of sample 1 are shown in Fig. 4. The depicted chromatograms were obtained from the measurement of the area values of the spectra between 1108 and 1069 cm−1 , corrected using a baseline established at 1203 cm−1 . It can be seen, that the resolution is very good and the specific spectra of the analytes can clearly be distinguished (see Fig. 5). The retention times, established from four injections of standard mixtures of the considered compounds were 11.8 ± 0.2, 12.7 ± 0.2, 15.0 ± 0.2 and 16.5 ± 0.2 min for fructose, glucose, sucrose and maltose, respectively. As it can be seen in the chromatograms of Fig. 4, the lack of sloping baselines and the random distribution of the chromatographic noise support the suitability of the selected background correction conditions for obtaining the chromatograms of sugars during gradient elution.
0.5–6.0 0.5–5.2 0.5–4.9 0.5–5.2
Linear range (mg ml−1 )
3.4. On-line LC–FTIR separation of carbohydrates
0.5–6.0 0.5–5.2 0.5–4.9 0.5–5.2
Linear range (mg/100 mla )
from absorbance values at 2295 and 2249.5 cm−1 . The difference between the obtained noise using this pair of wavenumbers and that obtained using the absorbance ratio between 2256.5 and 2252.6 cm−1 is only 2.1% and cannot be considered as statistically significant. Based on the foregoing data, the quotient between the absorbance values at 2256.5 and 2252.6 cm−1 was selected for the background correction of further measurements.
Calibration curvea [y = a + bC (mg ml−1 )]
784
J. Kuligowski et al. / Talanta 77 (2008) 779–785
785
Table 3 On-line LC–FTIR determination of sugars in commercial samples Sample
Presentation
Found concentration (g/100 ml)
1 2 3 4 5 6 7
Soft drink Soft drink Soft drink (sugar free) Soft drink Soft drink Energy drink Fruit juice
Label content (g/100 ml)
Fructose
Glucose
Sucrose
Maltose
Total
5.14 6.39
7.84 4.75
13.0 11.1 0.00 9.0 7.3 9.8 10.1
12.9 11.6 0.00 8.9 7.9 10.7 12.4
Table 4 Recovery study of fructose, glucose, sucrose and maltose in spiked commercial samples Sample
Presentation
1 2 4 5 6 7
Soft drink Soft drink Soft drink Soft drink Energy drink Fruit juice a
g/100 ml founda (% recovery)
g/100 ml added Fructose
Glucose
Sucrose
Maltose
Fructose
2.54 2.06 1.00 0.97 1.43 2.20
1.90 2.36 0.93 0.90 0.97 2.18
2.98 3.33 0.97 1.30 1.07 1.74
1.84 3.34 1.19 1.10 1.17 1.57
7.69 8.36 4.48 1.13 1.78 5.92
± ± ± ± ± ±
Glucose
0.08 (100.4) 0.04 (95.6) 0.06 (98.0) 0.08 (100.0) 0.09 (101.4) 0.02 (97.7)
9.84 7.19 5.11 1.76 4.06 5.90
± ± ± ± ± ±
0.09 (105.1) 0.09 (103.5) 0.04 (99.8) 0.09 (98.9) 0.04 (101.9) 0.08 (101.4)
Sucrose 2.92 3.34 2.39 7.54 7.45 4.32
± ± ± ± ± ±
Maltose 0.014 (98.1) 0.02 (100.2) 0.02 (104.0) 0.06 (98.7) 0.03 (95.8) 0.08 (98.1)
1.84 3.39 1.28 1.02 1.15 1.65
± ± ± ± ± ±
0.05 (100.2) 0.09 (101.3) 0.05 (99.1) 0.08 (92.5) 0.07 (98.4) 0.07 (104.9)
Recoveries from two-independent replicates.
a standard calibration solution containing 3.0 mg ml−1 of each analyte provided relative standard deviations ranging between 3.3 and 4.1% for glucose and fructose, respectively. This obtained sensitivity and precision results together with the good correlation factors obtained for the recovered analyte spectra suggested that they were appropriate for the quantitative determination and identification of sugars in beverages. 3.7. Determination of sugars in beverages The applicability of the proposed method was evidenced by the analysis of a set of 7 commercially available samples purchased from the Spanish market. Sample treatment was reduced to an adequate dilution and filtration of samples before their injection in the chromatographic system. Table 3 lists the different obtained sugar contents from duplicate analysis of diluted samples. Additionally, a recovery study was carried out in six of the analysed samples. The employed spiked concentration ranges were: 0.97–2.54 (g/100 ml), 0.90–2.36 (g/100 ml), 0.97–3.133 (g/100 ml) and 1.10–3.34 (g/100 ml) for fructose, glucose, sucrose and maltose, respectively. Mean recovery values ranged between 105.1 and 92.5% (see Table 4). 4. Conclusions A simple on-line gradient LC method in combination with FTIR detection and chemometrics has been developed for the determination of four characteristic carbohydrates in beverages. Fructose, glucose, sucrose and maltose could be separated, identified and quantified in commercial samples with limits of detection between 0.4 and 0.6 mg ml−1 . The use of the univariate method for the selection of the background from a reference spectra matrix permitted an accurate chemometric eluent correction yielding distinguishable spectra of good quality for the studied analytes with correlation factors between 96.25 and 98.73%. The main advantages provided by the proposed on-line LC–FTIR approach include instrumental simplicity and low cost (due to the possibility of using commercial flow cells).
Acknowledgements Authors acknowledge the financial support of Ministerio de Educación y Ciencia (Project CTQ205-05604, FEDER) and Direcció General d’Investigació I Transferència Tecnològica de la Generalitat Valenciana (Project ACOMP/2007/131). J.K. acknowledges the “V Segles” grant provided by the University of Valencia to carry out this study. G.Q. is grateful for a post-doctoral grant (“Ayudas para estancias de doctores en centros de investigación de excelencia de la Comunidad Valenciana”) from the Conselleria de Industria, Generalitat Valenciana (Spain). References [1] M.L. Wheeler, F.X. Pi-Sunyer, J. Am. Diet Assoc. 108 (2008) S34. [2] FDA Nutrition Labeling Manual, U.S. Food and Drug Administration, Center for Food Safety and Applied Nutrition, 1998. [3] A. Cáceres, S. Cárdenas, M. Gallego, A. Rodríguez, M. Valcárcel, Chromatographia 52 (2000) 314. [4] W.L. Qian, Z. Khan, D.G. Watson, J. Fearnley, J. Food Comp. Anal. 21 (1) (2008) 78. [5] R. Vonach, B. Lendl, R. Kellner, J. Chromatogr. A 824 (1998) 159. [6] R. Vonach, B. Lendl, R. Kellner, Anal. Chem. 69 (1997) 4286. [7] A. Edelmann, C. Ruzicka, J. Frank, B. Lendl, W. Schrenk, E. Gornik, G. Strasser, J. Chromatogr. A 934 (2001) 123. [8] A. Edelmann, K. Diewok, J. Rodriguez-Baena, B. Lendl, Anal. Bioanal. Chem. 376 (2003) 92. [9] M. Kölhed, B. Karlberg, Analyst 130 (2005) 772. [10] G.W. Somsen, C. Gooijer, U.A.Th. Brinkman, J. Chromatogr. A 856 (1999) 213. [11] K. István, R. Rajkó, G. Keresztury, J. Chromatogr. A 1104 (1-2) (2006) 154. [12] R.J. Dijkstra, H.F.M. Boelens, J.A. Westerhuis, F. Ariese, U.A.Th. Brinkman, C. Gooijer, Anal. Chim. Acta 519 (2004) 129. [13] H.F.M. Boelens, R.J. Dijkstra, P.H.C. Eilers, F. Fitzpatrick, J.A. Westerhuis, J. Chromatogr. A 1057 (2004) 21. [14] G. Quintás, B. Lendl, S. Garrigues, M. de la Guardia, J. Chromatogr. A 1190 (2008) 102. [15] J. Kuligowski, G. Quintás, S. Garrigues, M. de la Guardia, Anal. Chim. Acta 624 (2) (2008) 278. [16] J.T. Gotsick, R.F. Benson, J. Liq. Chromatogr. 14 (1991) 1887. [17] J.J. Max, C. Chapados, J. Phys. Chem. A 111 (2007) 2679. [18] J.J. Max, C. Chapados, J. Phys. Chem. A 105 (2001) 10681. [19] T. Takamuku, M. Tabata, A. Yamaguchi, J. Nishimoto, M. Kumamoto, H. Wakita, T. Yamaguchi, J. Phys. Chem. B 102 (1998) 8880.