Journal of Oleo Science Copyright ©2015 by Japan Oil Chemists’ Society doi : 10.5650/jos.ess14227 J. Oleo Sci. 64, (3) 255-261 (2015)
Determination of Polar Components in Frying Oils by Fourier-Transform Near-Infrared Spectroscopy Xiumei Chen, Xiuzhu Yu＊ , Yage Wang, Yandie Yang and Jingya Zhang College of Food Science and Engineering, Northwest A&F University, 28 Xinong Road Yangling, 712100, Shaanxi, P. R. China
Abstract: A rapid and convenient method was developed to determine the polar components (PC) of frying oil by Fourier-transform near-infrared (FTNIR) spectroscopy. One hundred twenty six oil samples were used to PC determination by column chromatography and FTNIR spectroscopy combined with partial least-square (PLS) calibration. The optimal PLS calibration was obtained after the Savitzky–Golay smoothing and first derivative treatment performed in the wavelength ranges of 4963 cm−1 to 4616 cm−1, 5222 cm−1 to 5037 cm−1, and 5688 cm−1 to 5499 cm−1. The obtained correlation coefficient (R) was 0.998 and the root mean square error of calibration was 1.0%. The PLS calibration was validated, and the results showed that the highest correlation (R) was 0.997 between reference value and the FTNIR predicted value and the root mean square error of prediction was 1.3%. Therefore, the FTNIR technique can be effectively applied to quantify PC with the advantages of simple operation and no pollution. Key words: frying oil, Fourier-transform near-infrared spectroscopy, polar components, quantification analysis 1 Introduction Frying is one of the most common practices to prepare and manufacture foods1）. Frying oil is exposed at high temperature for long time and generates a series of degradation reactions, including oxidation, polymerization, hydrolysis-pyrolysis, isomerization, and cyclization2, 3）. During frying, the polar compounds（PC）include polar substances such as monoglycerides, diglycerides, free fatty, as well as polar transformation products formed due to high temperature and the presence of air and moisture4, 5）. With prolonged heating time, PC content constantly increases and leads to organoleptic failure, nutritive value decreased6） and can cause mutations and gastrointestinal irritations7）. Most of regulations have set a maximum level of about 25％1PC1）. Therefore, PC is used as an indicator for the quality control of frying oil. Column chromatography（CC）is one of the most widely used techniques and standard method approved by the American Oil Chemists Society（AOCS）to determine PC in frying oil8−10）. In CC analysis, frying oils are separated into nonpolar components and PC; nonpolar components are then eluted. PC is determined by calculating the difference between the weights of the oil samples prepared and added to the column and the weight of the eluted nonpolar fraction6）. However, several factors should be considered11）, for instance, the silica gel used in the column should be stan-
dardized to the correct activity to completely separate PC from triglycerides. In addition, CC is time consuming and destructive to samples, as well as requires hazardous reagents12）. Compared with CC procedure, Fourier-transform near-infrared（FTNIR）technique can be more affordable and more easily operated for food industries. In recent years, a large amount of works regarding rapid determination of oil oxidation have been reported. FTNIR technique in edible oil analyses has been studied because of its advantages, including simple operation, no complex sample pretreatment, rapid determination, and no pollution6, 13）. Buning-Pfaue and Kehrans applied NIR to determine four analytical criteria（acid value, PC, dimeric and polymeric triglycerides）for used deep-frying oils, but the results indicated that was poorer statistical criteria of calibration for PC, dimeeric and polymeric triglycerides14）. Ng et al. have firstly developed a rapid NIR spectroscopic method for PC and free fatty acids （FFA） by using soy-based frying oil, and then used the method for frying various foods application successfully15−17）. In NIR analysis, the representative calibration samples are the most important factor for calibration development13）. In this study, nine oil types of edible oils and their mixtures were selected and samples got through different treatments with the oils. The aim was to develop a robust calibration based on FTNIR to determine the PC content in different frying oils.
＊
Correspondence to: Xiuzhu Yu, College of Food Science and Engineering, Northwest A&F University, 28 Xinong Road Yangling, 712100, Shaanxi, P. R. China E-mail: [email protected]. Accepted November 3, 2014 (received for review October 9, 2014)
Journal of Oleo Science ISSN 1345-8957 print / ISSN 1347-3352 online
http://www.jstage.jst.go.jp/browse/jos/ http://mc.manusriptcentral.com/jjocs 255
X. Chen, X. Yu, Y. Wang
2 Materials and methods 2.1 Materials and reagents Potatoes and edible oils（soy bean oil, peanut oil, corn oil, rapeseed oil, linseed oil, perilla seed oil, sesame oil, olive oil, and sunflower oil）were obtained from a local market. Potassium hydroxide（ Analytical grade）, ethyl alcohol, petroleum ether, and diethyl ether were used for analyses.
following equation: m−（m2−m1） PC＝ ×100％ m Where m is the mass of the oil sample, m1 is the mass of the flask, and m2 is the sum of the masses of the flask and nonpolar fraction. All of the samples were analyzed in duplicate, and the mean value was used for the reported results.
2.2 Sample preparation The potatoes were washed, peeled, and cut into potato chips（about 7 cm×0.5 cm×0.3 cm）, submerged in water to prevent the occurrence of browning. Taking into account that frying oil was not a single type of oil but a mixture of several oils, it was prepared by mixing two or three oils from nine types of oils in a mass ratio of 1:1 （two oils） and 1:1:1（three oils）, respectively. One batch of 76 samples was generated by heating the nine type oils to accelerate oxidation at 105℃ in an oven for a certain period of time18, 19）. Oil samples were collected at 4 h intervals during heating. And another batch of 50 samples was gotten by using mixed oil to fry potato chips at 190℃. Continuous frying lasted for 8 h per day for 3 consecutive days. A 100 ml oil sample was taken every 4 h. No oil was replenished during the frying process. Arbitrary selection was used for dividing 126 samples into calibration set and validation set with the purpose of even distribution of different oxidation degrees from two batches. All of the oil samples were transferred into vials, flushed with nitrogen, and stored at −20℃ to prevent further oxidation for subsequent analysis.
2.4 Spectral acquisition An FTNIR spectrometer （Bruker Optik, GmbH, Ettlingen, Germany）equipped with InGaAs detectors, a 20 W high-intensity tungsten–halogen NIR light source, and OPUS 5.5 software was used for spectral acquisition. The oil samples were mounted on 1.0 mL glass vials（diameter, 5 mm）to obtain the spectra. Separate disposable glass vials were used for each sample. Empty vial was used as the reference before spectral measurements were performed. The temperature of transmission chamber was maintained at 40℃. The spectra were obtained from 12000 cm−1 to 4000 cm−1 by co-adding 32 scans at a resolution of 4 cm−1. All of the samples were analyzed in duplicate, and the average spectrum was considered as the spectrum of each oil sample.
2.3 PC determination The oil samples were measured by column chromatography based on the modification of AOCS official method （AOCS Cd 20-91）. The oil samples was weighed accurately to 0.001 g（m）and then placed in a 50 mL volumetric flask. The oil sample was dissolved in 20 mL of the elution solvent, which was a mixture of light petroleum ether and diethyl ether（87/13） . Approximately 20 mL of this solution was transferred using a volumetric pipet into the column packed with silica gel（particle size, 100 meshes）. PC was adsorbed onto the silica gel, and nonpolar components were eluted by the mobile phase. Approximately 150 mL of the elution solvent was passed through a column at an adjusted flow rate of 2.1 mL/min to 2.5 mL/min. A clean and dry 250 mL flask was weighed accurately to within 0.001 g （m1）and placed under the outlet of the column to collect （s） with the eluate. The eluate was removed from the flask a rotary evaporator by using a water bath at a temperature not greater than 60℃. The flask was dried at 103±2℃ in an oven and then allowed to cool. The flask was weighed . accurately to 0.001 g（m2） The PC content in percent（w/w）is expressed using the
2.5 Calibration and validation Savitzky–Golay smoothing, Norris derivative smoothing, first derivative, and second derivative were applied in the spectra preprocessing by using TQ Analyst 7.2 software. A total of 126 oil samples were randomly divided into calibration and validation sets. A total of 74 oil samples were selected for the calibration set and the 52 remaining oil samples were placed in the validation set. The calibration oil samples included extreme PC values. PLS regression was used to obtain the relationship between the reference and the FTNIR spectra. The values of R, root mean square error of calibration（RMSEC）, and prediction （RMSEP）were used to determine calibration quality. The prediction residual sums of squares（PRESS）was used to select the optimal number of principal components. The calibration with the highest R and the lowest RMSEC and RMSEP is the most appropriate for prediction.
3 Results and discussion 3.1 PC conventional method analysis Before these oils were heated and frying, their PC values as determined by the reference methods ranged from 0.4％ to 2.2％, respectively. All samples exhibited various levels of oil oxidation that ranges from fresh to dispensable and evenly distributed to establish a robust calibration. The PC contents of the calibration oil samples ranged between 0.4％ and 67.6％ with a mean value of 19.4％. Similarly, the
256
J. Oleo Sci. 64, (3) 255-261 (2015)
Determination of Polar Components in Frying Oils by Fourier-Transform Near-Infrared Spectroscopy
PC content of the validation oil samples ranged between 3.3％ and 54.0％ with an average of 21.6％. Evidently, the highest PC content value was higher than the allowable level. The two datasets covered the PC of the qualified and unqualified frying oil samples to have a strong representation and thus obtain accurate results. 3.2 Spectral analysis The FTNIR spectra of the oil samples were obtained from 12000 cm−1 to 4000 cm−1 by using a reflectance NIR spectrometer. Smoothing and derivative computation were performed to provide further details and remove spectrum defects. The FTNIR spectra of the frying oil samples from different spectral processing were presented in Fig. 1. The raw spectra exhibited strong absorption peaks at 8400 cm−1, 7500 cm−1, 5601 cm−1, 5195 cm−1, and 4616 cm−1 wavelengths（Fig. 1a）. The distinct bumps at 8400 and 7500 cm−1 revealed the second overtone of the C–H stretching vibration from –CH＝CH– and the combination
of C–H from –CH2, respectively. The major peaks at 5601 cm−1 and 5195 cm−1 were correlated with the first overtone of the C–H stretching vibration from –CH＝CH– and – CH3. The peak at 4616 cm−1 represented the combination of the C–H stretching vibration from –CH＝CH–13）. These differences are mainly due to the overheated frying, as well as to the formed carbonyl, aldehyde, ketone, hydroperoxides, and free fatty acids in the oil samples16, 20）. After Savitzky-Golay smoothing and first derivative treatment were performed, the acquired spectra became evident（Fig. 1b）compared with the typical raw spectra of the frying oils with different polarities （Fig. 1a）. Moreover, glitches increased and became evident whereas the noise signal was amplified, although spectral resolution was improved using the second derivative（Fig. 1c）. However, no improvement was achieved when the combination of Savitzky–Golay smoothing and second derivative treatments were used （Fig. 1d） .
Fig. 1 FTNIR spectra of the frying oils from different spectral processing. 257
J. Oleo Sci. 64, (3) 255-261 (2015)
X. Chen, X. Yu, Y. Wang
3.3 Abnormal oil sample analysis Abnormal oil samples refer to the spectral data or the concentration of the sample standard with an apparent deviation. The abnormal oil samples caused by improper methods, adverse effects of instruments, or human error can affect calibration accuracy and precision21）. The spectrum outlier option to identify the outliers directly from the spectral distribution difference by calculating Mahalanobis distance was showed in Fig. 2. Abnormal oil samples were classified at the end and marked with different colors. The calibration did not significantly differ in the sample spectra（Fig. 2） ; thus, no outliers were removed. 3.4 Principal components analysis （PCA） PCA reduces the calibration spectral intensity data at several frequencies to a relatively small number of intensities in a transformed full-spectrum coordinate system. Therefore, optimal principal components are determined to fully utilize spectral information and filter out noise22）. PRESS and RMSEC were used as indicators to determine the optimal calibration with different principal components. The low values of PRESS and RMSEC result in more accurate calibration prediction. Table 1 illustrated the contribution rate of the whole
spectra（12000 cm−1 to 4000 cm−1）was 24.1％ under the principal component score of 1. The contribution rate of the analysis region was observed within the ranges of 4963 cm−1 to 4 616 cm−1, 5222 cm−1 to 5037 cm−1, and 5688 cm−1 to 5499 cm−1. The contribution rate was 88.7％ and the RMSEC value was 8.9％. Moreover, as the principal component score was 5, the whole spectra contribution rate was 51.0％ and the contribution rate of the spectral analysis region was 99.8％ with an RMSEC value of 6.4％. As principal component scores lower than 5, the RMSEC value showed a decreasing trend. As principal component score ＞5, the RMSEC value exhibited an increasing trend, indicating that the calibration data exhibited excessive noise that resulted in an overfit23）. The minimum PRESS and RMSEC values corresponded to the principal component score of 5 and resulted in a well-fitted calibration. Therefore, the maximum of principal component score was 5 with the wavelengths of 4963 cm−1 to 4616 cm−1, 5222 cm−1 to 5037 cm−1, and 5688 cm−1 to 5499 cm−1 were used in the PLS to prevent calibration overfitting. 3.5 Different spectral preprocessing procedures The original spectra contained the elements from the equipment, light scattering around the environment, and
Fig. 2 Spectra outliers. Table 1 Scores of PRESS and RMSEC with various principal components. Principal component
Whole spectral contribution rate /%
Analysis of regional spectral contribution rate /%
PRESS
RMSEC /%
0
−
−
15891.258
22.3
1
24.1
88.7
2511.073
8.9
2
32.7
96.7
2353.612
8.6
3
39.5
99.1
2208.474
8.3
4
45.6
99.6
1422.506
6.7
5
51.0
99.8
1304.845
6.4
6
56.0
99.9
1576.813
7.0
7
61.1
99.9
1480.765
6.8
8
65.6
99.9
1425.601
6.7
9
69.1
99.9
1517.864
6.9
10
−
−
2100.525
8.1
258
J. Oleo Sci. 64, (3) 255-261 (2015)
Determination of Polar Components in Frying Oils by Fourier-Transform Near-Infrared Spectroscopy
other adverse factors. These noise signals interfere with the spectral information of the oil samples and affect calibration establishment and precision24）. The Norris smoothing and Savitzky–Golay smoothing eliminate environment noise; derivative computation increases data information and removes spectral defects, such as baseline shifting and peak overlapping25）. Therefore, raw spectral data should be mathematically preprocessed corresponding to Savitzky– Golay smoothing/Norris smoothing and/or first derivative/ second derivative treatments. The different pretreatments used to PC calibrate in the selected spectral ranges of 4963 cm−1 to 4616 cm−1, 5222 cm −1 to 5037 cm −1, and 5688 cm −1 to 5499 cm −1 were shown in Table 2. The value of R was 0.940 when the Norris smoothing （5, 5） was used for PC calibration, indicating that the improved effect was not evident. After Savitzky–Golay smoothing（9, 5）was performed, R was 0.979, which was ＞0.940; thus, this result represented an effectively improved calibration. Using Savitzky–Golay（7, 5） smoothing and first derivative as spectral pretreatments, we obtained an R value of 0.998 and an RMSEC value of 1.0％; an optimal calibration was observed during these three treatments. Thus, the spectral ranges within 4963 cm−1 to 4616 cm−1, 5222 cm−1 to 5037 cm−1, and 5688 cm−1 to （7, 5） smoothing 5499 cm−1 were used with Savitzky–Golay and first derivative treatment as the spectral pretreatment to quantitatively analyze the PC calibration in frying oils. 3.6 PLS calibration development and validation PLS provides a dataset with relevant information and reduced dimensionality; this method eliminates data noise to obtain a more accurate and reproducible calibration13, 26）. PLS calibrations were established using the spectra with the selected wavelength range and mean reference values. The values of R and RMSEC of the calibration were considered as an indicator in quality evaluation. In general, the larger the values of R and RMSEC are, the more predictive ability of the calibration is obtained26）. The regression curve between the reference and the predicted values of PLS calibration was obtained, as shown
in Fig. 3. The number of the oil samples used to PC calibrate was 74. The selected interval spectral wavelength included the spectral ranges from 4963 cm−1 to 4616 cm−1, 5222 cm −1 to 5037 cm −1, and 5688 cm −1 to 5499 cm −1 combined with Savitzky–Golay（7, 5）smoothing and first derivative treatment. PLS calibration yielded an R value of 0.998 with an RMSEC value of 1.0％. The results indicated that PLS calibration was accurate and satisfied the PC determination standard in frying oil samples. The regression curve between the reference and the predicted values was shown in Fig. 4. The number of the validation oil samples was 52. The reference values and the predicted values exhibited significant correlation（Fig. 4）. The R value of this calibration was 0.997 with an RMSEP value of 1.3％. These results indicated the quantitative limit of the proposed method is about 1.3％ which can meet the routine analysis according to AOCS official method cd 20-91. Therefore, FTNIR technique exhibits a satisfactory performance and can be useful for quality control applications.
Fig. 3 R elationship between the reference and the predicted determined PC values.
Table 2 Calibration results with different pretreatments for PC (principle components = 5). Spectral range (cm−1) 4918–4809, 5102–4952, 5673–5638 4898–4809, 5318–5287, 5839–5800 5372–4682, 7274–6032
Pretreatments
R
RMSEC/%
a
0.940
7.1
b
0.979
4.3
NF (5, 5) SF (9, 5) SF (9, 5)
0.969
5.2
4963–4616, 5222–5037, 5688–5499
first derivative+SF (7, 5)
0.998
1.0
4963–4616, 5222–5037, 5688–5499
first derivative+SF (9, 5)
0.987
3.3
a
NF (5, 5) represents the Norris derivative smoothing under segment length is 5 and the gap between segments is 5. b SF (9, 5) represents the smoothing under 9 data points and the 5th polynomial. 259
J. Oleo Sci. 64, (3) 255-261 (2015)
X. Chen, X. Yu, Y. Wang
Fig. 4 R elationship between the reference and the predicted determined PC values. 4 Conclusions Traditional column chromatography is time consuming and involves several chemicals, such as petroleum ether and diethyl ether. Thus, studies have been performed to obtain simpler and more rapid techniques. FTNIR technique is a satisfactory alternative because of its advantages, including simple operation and rapid determination, as well as no hazardous chemicals 16, 17）. An NIR-based method to determine PC in frying oils was successfully developed. A significant correlation was obtained between FTNIR predicted and reference values. PLS calibration showed that the R and RMSEC between reference and FTNIR predicted values of PC in frying oils were 0.998 and 1.0％, respectively. Moreover, validation analysis indicated that the R and RMSEP between the reference and the FTNIR predicted values of PC in frying oils were 0.997 and 1.3％, respectively. These results demonstrated that FTNIR analysis can be effectively applied to evaluate the extent of frying oil degradation rapidly and to monitor the changes occurring in edible oils easily during frying.
Acknowledgements Authors would like to thank the Fundamental Research Funds for the Central Universities （QN 2013057） for the financial support.
References 1）Firestone, D.; Gupta, M.; Warner, K.; White, P. Frying technology and practices. AOCS Press:Champaign, IL.PP. 200-216 （2004） .
2）Saguy, I. S.; Dana, D. Integrated approach to deep fat frying: engineering, nutrition, health and consumer aspects. J. Food Eng. 56, 143-152（2003）. 3）Gertz, C. Chemical and physical parameters as quality indicators of used frying fats. Eur. J. Lipid Sci. Tech. 102, 566-572（2000）. 4）Karakaya, S.; Simsek, S. Changes in total polar compounds, peroxide value, total phenols and antioxidant activity of various oils used in deep fat frying. J. Am. Oil Chem. Soc. 88, 1361-1366（2011）. 5）Dobarganes, M.; Velasco, J.; Dieffenbacher, A. Determination of polar compounds, polymerized and oxidized triacylglycerols, and diacylglycerols in oils and fats: results of collaborative studies and the standardized method（Technical report）. Pure Appl. Chem. 72, 1563-1575（2000）. 6）Hein, M.; Henning, H.; Isengard, H. D. Determination of total polar parts with new methods for the quality survey of frying fats and oils. Talanta 47, 447-454 （1998）. 7）Clark, W.; Serbia, G. Safety aspects of frying fats and oils. Food Tech. 45, 84-89 （1991）. 8）Fritsch, C. Measurements of frying fat deterioration: a brief review. J. Am. Oil Chem. Soc. 58, 272-274 （1981）. 9）AlKahtani, H. A. Survey of quality of used frying oils from restaurants. J. Am. Oil Chem. Soc. 68, 857-862 （1991）. 10）Gertz, C. Routine analysis of deep-frying fats and oils. Lipid Tech. 13, 44-47 （2001）. 11）Erickson, M. D. Deep frying: chemistry, nutrition, and practical applications. vol Ed. 2. AOCS Press: Champaign, IL. PP. 315-320（2006）. 12）Melton, S.; Jafar, S.; Sykes, D.; Trigiano, M. Review of stability measurements for frying oils and fried food flavor. J. Am. Oil Chem. Soc. 71, 1301-1308（1994）. 13）Kuligowski, J.; Carrion, D.; Quintas, G.; Garrigues, S.; Guardia, M. Direct determination of polymerised triacylglycerides in deep-frying vegetable oil by near infrared spectroscopy using partial least squares regression. Food Chem. 131, 353-359 （2012）. 14）Buning-Pfaue, H.; Kehraus, S. Application of near infrared spectroscopy（NIRS）in the analysis of frying fats. Eur. J. Lipid Sci. Tech. 103, 793-797 （2001）. 15）Kazemi, S.; Wang, N.; Ngadi, M.; Prasher, S. O. Evaluation of frying oil quality using VIS/NIR hyperspectral analysis. Int. Comm. Agr. Eng. 7, 1-12 （2005）. 16）Ng, C. L.; Wehling, R. L.; Cuppett, S. L. Method for determining frying oil degradation by near-infrared spectroscopy. J. Agr. Food Chem. 55, 593-597（2007）. 17）Ng, C. L.; Wehling, R. L.; Cuppet, S. L. Near-infrared spectroscopy determination of degradation in vegetable oils used to fry various foods. J. Agr. Food Chem. 59, 12286-12290（2011）.
260
J. Oleo Sci. 64, (3) 255-261 (2015)
Determination of Polar Components in Frying Oils by Fourier-Transform Near-Infrared Spectroscopy
18）Gerde, J.; Hardy, C.; Fehr, W.; White, P. J. Frying performance of no-trans, low-linolenic acid soybean oils. J. Am. Oil Chem. Soc. 84, 557-563 （2007） . 19）Du, R.; Lai, K.; Xiao, Z.; Shen, Y.; Wang, X.; Huang, Y. Evaluation of the quality of deep frying oils with Fourier transform near-infrared and mid-infrared spectroscopy. J. Food Sci. 77, 261-266 （2012） . 20）Gerde, J. A.; Hardy, C. L.; Hurburgh, C. R.; White, P. J. Rapid determination of degradation in frying oils with Near-Infrared spectroscopy. J. Am. Oil Chem. Soc. 84, 519-522 （2007） . 21）Siesler, H. W.; Ozaki, Y.; Kawata, S.; Heise, H. M. Near infrared spectroscopy: principles, instruments, applications. John Wiley and Sons Ltd., New York, pp. 25-30 （2008） . 22）Chu, X. L.; Yuan, H. F.; Lu, W. Z. Progress and applica-
tion of spectral data pretreatment and wavelength selection methods in NIR analytical technique. Progr. Chem. 16, 528-542 （2004）. 23）Chu, X. L.; Xu, Y. P.; Lu, W. Z. Research and application progress of chemometrics methods in near Infrared spectroscopy analysis. Chin. J. Anal. Chem. 36, 350-381（2008）. 24）Zhang, J. Y.; Zhang, J. X.; Yu, X.; Du, S. Qualitative analysis of edible oil oxidation by near Infrared transmission spectroscopy. Food Sci. 4, 43-47（2012）. 25）Barton, I. FE Theory and principles of near infrared spectroscopy. Spectrosc. Eur. 14, 12-18（2002）. 26）Kovalenko, I. V.; Rippke, G. R.; Hurburgh, C. R. Measurement of soybean fatty acids by near-infrared spectroscopy: linear and nonlinear calibration methods. J. Am. Oil Chem. Soc. 83, 421-427 （2006）.
261
J. Oleo Sci. 64, (3) 255-261 (2015)
Determination of Polar Components in Frying Oils by Fourier-Transform Near-Infrared Spectroscopy Xiumei Chen, Xiuzhu Yu＊ , Yage Wang, Yandie Yang and Jingya Zhang College of Food Science and Engineering, Northwest A&F University, 28 Xinong Road Yangling, 712100, Shaanxi, P. R. China
Abstract: A rapid and convenient method was developed to determine the polar components (PC) of frying oil by Fourier-transform near-infrared (FTNIR) spectroscopy. One hundred twenty six oil samples were used to PC determination by column chromatography and FTNIR spectroscopy combined with partial least-square (PLS) calibration. The optimal PLS calibration was obtained after the Savitzky–Golay smoothing and first derivative treatment performed in the wavelength ranges of 4963 cm−1 to 4616 cm−1, 5222 cm−1 to 5037 cm−1, and 5688 cm−1 to 5499 cm−1. The obtained correlation coefficient (R) was 0.998 and the root mean square error of calibration was 1.0%. The PLS calibration was validated, and the results showed that the highest correlation (R) was 0.997 between reference value and the FTNIR predicted value and the root mean square error of prediction was 1.3%. Therefore, the FTNIR technique can be effectively applied to quantify PC with the advantages of simple operation and no pollution. Key words: frying oil, Fourier-transform near-infrared spectroscopy, polar components, quantification analysis 1 Introduction Frying is one of the most common practices to prepare and manufacture foods1）. Frying oil is exposed at high temperature for long time and generates a series of degradation reactions, including oxidation, polymerization, hydrolysis-pyrolysis, isomerization, and cyclization2, 3）. During frying, the polar compounds（PC）include polar substances such as monoglycerides, diglycerides, free fatty, as well as polar transformation products formed due to high temperature and the presence of air and moisture4, 5）. With prolonged heating time, PC content constantly increases and leads to organoleptic failure, nutritive value decreased6） and can cause mutations and gastrointestinal irritations7）. Most of regulations have set a maximum level of about 25％1PC1）. Therefore, PC is used as an indicator for the quality control of frying oil. Column chromatography（CC）is one of the most widely used techniques and standard method approved by the American Oil Chemists Society（AOCS）to determine PC in frying oil8−10）. In CC analysis, frying oils are separated into nonpolar components and PC; nonpolar components are then eluted. PC is determined by calculating the difference between the weights of the oil samples prepared and added to the column and the weight of the eluted nonpolar fraction6）. However, several factors should be considered11）, for instance, the silica gel used in the column should be stan-
dardized to the correct activity to completely separate PC from triglycerides. In addition, CC is time consuming and destructive to samples, as well as requires hazardous reagents12）. Compared with CC procedure, Fourier-transform near-infrared（FTNIR）technique can be more affordable and more easily operated for food industries. In recent years, a large amount of works regarding rapid determination of oil oxidation have been reported. FTNIR technique in edible oil analyses has been studied because of its advantages, including simple operation, no complex sample pretreatment, rapid determination, and no pollution6, 13）. Buning-Pfaue and Kehrans applied NIR to determine four analytical criteria（acid value, PC, dimeric and polymeric triglycerides）for used deep-frying oils, but the results indicated that was poorer statistical criteria of calibration for PC, dimeeric and polymeric triglycerides14）. Ng et al. have firstly developed a rapid NIR spectroscopic method for PC and free fatty acids （FFA） by using soy-based frying oil, and then used the method for frying various foods application successfully15−17）. In NIR analysis, the representative calibration samples are the most important factor for calibration development13）. In this study, nine oil types of edible oils and their mixtures were selected and samples got through different treatments with the oils. The aim was to develop a robust calibration based on FTNIR to determine the PC content in different frying oils.
＊
Correspondence to: Xiuzhu Yu, College of Food Science and Engineering, Northwest A&F University, 28 Xinong Road Yangling, 712100, Shaanxi, P. R. China E-mail: [email protected]. Accepted November 3, 2014 (received for review October 9, 2014)
Journal of Oleo Science ISSN 1345-8957 print / ISSN 1347-3352 online
http://www.jstage.jst.go.jp/browse/jos/ http://mc.manusriptcentral.com/jjocs 255
X. Chen, X. Yu, Y. Wang
2 Materials and methods 2.1 Materials and reagents Potatoes and edible oils（soy bean oil, peanut oil, corn oil, rapeseed oil, linseed oil, perilla seed oil, sesame oil, olive oil, and sunflower oil）were obtained from a local market. Potassium hydroxide（ Analytical grade）, ethyl alcohol, petroleum ether, and diethyl ether were used for analyses.
following equation: m−（m2−m1） PC＝ ×100％ m Where m is the mass of the oil sample, m1 is the mass of the flask, and m2 is the sum of the masses of the flask and nonpolar fraction. All of the samples were analyzed in duplicate, and the mean value was used for the reported results.
2.2 Sample preparation The potatoes were washed, peeled, and cut into potato chips（about 7 cm×0.5 cm×0.3 cm）, submerged in water to prevent the occurrence of browning. Taking into account that frying oil was not a single type of oil but a mixture of several oils, it was prepared by mixing two or three oils from nine types of oils in a mass ratio of 1:1 （two oils） and 1:1:1（three oils）, respectively. One batch of 76 samples was generated by heating the nine type oils to accelerate oxidation at 105℃ in an oven for a certain period of time18, 19）. Oil samples were collected at 4 h intervals during heating. And another batch of 50 samples was gotten by using mixed oil to fry potato chips at 190℃. Continuous frying lasted for 8 h per day for 3 consecutive days. A 100 ml oil sample was taken every 4 h. No oil was replenished during the frying process. Arbitrary selection was used for dividing 126 samples into calibration set and validation set with the purpose of even distribution of different oxidation degrees from two batches. All of the oil samples were transferred into vials, flushed with nitrogen, and stored at −20℃ to prevent further oxidation for subsequent analysis.
2.4 Spectral acquisition An FTNIR spectrometer （Bruker Optik, GmbH, Ettlingen, Germany）equipped with InGaAs detectors, a 20 W high-intensity tungsten–halogen NIR light source, and OPUS 5.5 software was used for spectral acquisition. The oil samples were mounted on 1.0 mL glass vials（diameter, 5 mm）to obtain the spectra. Separate disposable glass vials were used for each sample. Empty vial was used as the reference before spectral measurements were performed. The temperature of transmission chamber was maintained at 40℃. The spectra were obtained from 12000 cm−1 to 4000 cm−1 by co-adding 32 scans at a resolution of 4 cm−1. All of the samples were analyzed in duplicate, and the average spectrum was considered as the spectrum of each oil sample.
2.3 PC determination The oil samples were measured by column chromatography based on the modification of AOCS official method （AOCS Cd 20-91）. The oil samples was weighed accurately to 0.001 g（m）and then placed in a 50 mL volumetric flask. The oil sample was dissolved in 20 mL of the elution solvent, which was a mixture of light petroleum ether and diethyl ether（87/13） . Approximately 20 mL of this solution was transferred using a volumetric pipet into the column packed with silica gel（particle size, 100 meshes）. PC was adsorbed onto the silica gel, and nonpolar components were eluted by the mobile phase. Approximately 150 mL of the elution solvent was passed through a column at an adjusted flow rate of 2.1 mL/min to 2.5 mL/min. A clean and dry 250 mL flask was weighed accurately to within 0.001 g （m1）and placed under the outlet of the column to collect （s） with the eluate. The eluate was removed from the flask a rotary evaporator by using a water bath at a temperature not greater than 60℃. The flask was dried at 103±2℃ in an oven and then allowed to cool. The flask was weighed . accurately to 0.001 g（m2） The PC content in percent（w/w）is expressed using the
2.5 Calibration and validation Savitzky–Golay smoothing, Norris derivative smoothing, first derivative, and second derivative were applied in the spectra preprocessing by using TQ Analyst 7.2 software. A total of 126 oil samples were randomly divided into calibration and validation sets. A total of 74 oil samples were selected for the calibration set and the 52 remaining oil samples were placed in the validation set. The calibration oil samples included extreme PC values. PLS regression was used to obtain the relationship between the reference and the FTNIR spectra. The values of R, root mean square error of calibration（RMSEC）, and prediction （RMSEP）were used to determine calibration quality. The prediction residual sums of squares（PRESS）was used to select the optimal number of principal components. The calibration with the highest R and the lowest RMSEC and RMSEP is the most appropriate for prediction.
3 Results and discussion 3.1 PC conventional method analysis Before these oils were heated and frying, their PC values as determined by the reference methods ranged from 0.4％ to 2.2％, respectively. All samples exhibited various levels of oil oxidation that ranges from fresh to dispensable and evenly distributed to establish a robust calibration. The PC contents of the calibration oil samples ranged between 0.4％ and 67.6％ with a mean value of 19.4％. Similarly, the
256
J. Oleo Sci. 64, (3) 255-261 (2015)
Determination of Polar Components in Frying Oils by Fourier-Transform Near-Infrared Spectroscopy
PC content of the validation oil samples ranged between 3.3％ and 54.0％ with an average of 21.6％. Evidently, the highest PC content value was higher than the allowable level. The two datasets covered the PC of the qualified and unqualified frying oil samples to have a strong representation and thus obtain accurate results. 3.2 Spectral analysis The FTNIR spectra of the oil samples were obtained from 12000 cm−1 to 4000 cm−1 by using a reflectance NIR spectrometer. Smoothing and derivative computation were performed to provide further details and remove spectrum defects. The FTNIR spectra of the frying oil samples from different spectral processing were presented in Fig. 1. The raw spectra exhibited strong absorption peaks at 8400 cm−1, 7500 cm−1, 5601 cm−1, 5195 cm−1, and 4616 cm−1 wavelengths（Fig. 1a）. The distinct bumps at 8400 and 7500 cm−1 revealed the second overtone of the C–H stretching vibration from –CH＝CH– and the combination
of C–H from –CH2, respectively. The major peaks at 5601 cm−1 and 5195 cm−1 were correlated with the first overtone of the C–H stretching vibration from –CH＝CH– and – CH3. The peak at 4616 cm−1 represented the combination of the C–H stretching vibration from –CH＝CH–13）. These differences are mainly due to the overheated frying, as well as to the formed carbonyl, aldehyde, ketone, hydroperoxides, and free fatty acids in the oil samples16, 20）. After Savitzky-Golay smoothing and first derivative treatment were performed, the acquired spectra became evident（Fig. 1b）compared with the typical raw spectra of the frying oils with different polarities （Fig. 1a）. Moreover, glitches increased and became evident whereas the noise signal was amplified, although spectral resolution was improved using the second derivative（Fig. 1c）. However, no improvement was achieved when the combination of Savitzky–Golay smoothing and second derivative treatments were used （Fig. 1d） .
Fig. 1 FTNIR spectra of the frying oils from different spectral processing. 257
J. Oleo Sci. 64, (3) 255-261 (2015)
X. Chen, X. Yu, Y. Wang
3.3 Abnormal oil sample analysis Abnormal oil samples refer to the spectral data or the concentration of the sample standard with an apparent deviation. The abnormal oil samples caused by improper methods, adverse effects of instruments, or human error can affect calibration accuracy and precision21）. The spectrum outlier option to identify the outliers directly from the spectral distribution difference by calculating Mahalanobis distance was showed in Fig. 2. Abnormal oil samples were classified at the end and marked with different colors. The calibration did not significantly differ in the sample spectra（Fig. 2） ; thus, no outliers were removed. 3.4 Principal components analysis （PCA） PCA reduces the calibration spectral intensity data at several frequencies to a relatively small number of intensities in a transformed full-spectrum coordinate system. Therefore, optimal principal components are determined to fully utilize spectral information and filter out noise22）. PRESS and RMSEC were used as indicators to determine the optimal calibration with different principal components. The low values of PRESS and RMSEC result in more accurate calibration prediction. Table 1 illustrated the contribution rate of the whole
spectra（12000 cm−1 to 4000 cm−1）was 24.1％ under the principal component score of 1. The contribution rate of the analysis region was observed within the ranges of 4963 cm−1 to 4 616 cm−1, 5222 cm−1 to 5037 cm−1, and 5688 cm−1 to 5499 cm−1. The contribution rate was 88.7％ and the RMSEC value was 8.9％. Moreover, as the principal component score was 5, the whole spectra contribution rate was 51.0％ and the contribution rate of the spectral analysis region was 99.8％ with an RMSEC value of 6.4％. As principal component scores lower than 5, the RMSEC value showed a decreasing trend. As principal component score ＞5, the RMSEC value exhibited an increasing trend, indicating that the calibration data exhibited excessive noise that resulted in an overfit23）. The minimum PRESS and RMSEC values corresponded to the principal component score of 5 and resulted in a well-fitted calibration. Therefore, the maximum of principal component score was 5 with the wavelengths of 4963 cm−1 to 4616 cm−1, 5222 cm−1 to 5037 cm−1, and 5688 cm−1 to 5499 cm−1 were used in the PLS to prevent calibration overfitting. 3.5 Different spectral preprocessing procedures The original spectra contained the elements from the equipment, light scattering around the environment, and
Fig. 2 Spectra outliers. Table 1 Scores of PRESS and RMSEC with various principal components. Principal component
Whole spectral contribution rate /%
Analysis of regional spectral contribution rate /%
PRESS
RMSEC /%
0
−
−
15891.258
22.3
1
24.1
88.7
2511.073
8.9
2
32.7
96.7
2353.612
8.6
3
39.5
99.1
2208.474
8.3
4
45.6
99.6
1422.506
6.7
5
51.0
99.8
1304.845
6.4
6
56.0
99.9
1576.813
7.0
7
61.1
99.9
1480.765
6.8
8
65.6
99.9
1425.601
6.7
9
69.1
99.9
1517.864
6.9
10
−
−
2100.525
8.1
258
J. Oleo Sci. 64, (3) 255-261 (2015)
Determination of Polar Components in Frying Oils by Fourier-Transform Near-Infrared Spectroscopy
other adverse factors. These noise signals interfere with the spectral information of the oil samples and affect calibration establishment and precision24）. The Norris smoothing and Savitzky–Golay smoothing eliminate environment noise; derivative computation increases data information and removes spectral defects, such as baseline shifting and peak overlapping25）. Therefore, raw spectral data should be mathematically preprocessed corresponding to Savitzky– Golay smoothing/Norris smoothing and/or first derivative/ second derivative treatments. The different pretreatments used to PC calibrate in the selected spectral ranges of 4963 cm−1 to 4616 cm−1, 5222 cm −1 to 5037 cm −1, and 5688 cm −1 to 5499 cm −1 were shown in Table 2. The value of R was 0.940 when the Norris smoothing （5, 5） was used for PC calibration, indicating that the improved effect was not evident. After Savitzky–Golay smoothing（9, 5）was performed, R was 0.979, which was ＞0.940; thus, this result represented an effectively improved calibration. Using Savitzky–Golay（7, 5） smoothing and first derivative as spectral pretreatments, we obtained an R value of 0.998 and an RMSEC value of 1.0％; an optimal calibration was observed during these three treatments. Thus, the spectral ranges within 4963 cm−1 to 4616 cm−1, 5222 cm−1 to 5037 cm−1, and 5688 cm−1 to （7, 5） smoothing 5499 cm−1 were used with Savitzky–Golay and first derivative treatment as the spectral pretreatment to quantitatively analyze the PC calibration in frying oils. 3.6 PLS calibration development and validation PLS provides a dataset with relevant information and reduced dimensionality; this method eliminates data noise to obtain a more accurate and reproducible calibration13, 26）. PLS calibrations were established using the spectra with the selected wavelength range and mean reference values. The values of R and RMSEC of the calibration were considered as an indicator in quality evaluation. In general, the larger the values of R and RMSEC are, the more predictive ability of the calibration is obtained26）. The regression curve between the reference and the predicted values of PLS calibration was obtained, as shown
in Fig. 3. The number of the oil samples used to PC calibrate was 74. The selected interval spectral wavelength included the spectral ranges from 4963 cm−1 to 4616 cm−1, 5222 cm −1 to 5037 cm −1, and 5688 cm −1 to 5499 cm −1 combined with Savitzky–Golay（7, 5）smoothing and first derivative treatment. PLS calibration yielded an R value of 0.998 with an RMSEC value of 1.0％. The results indicated that PLS calibration was accurate and satisfied the PC determination standard in frying oil samples. The regression curve between the reference and the predicted values was shown in Fig. 4. The number of the validation oil samples was 52. The reference values and the predicted values exhibited significant correlation（Fig. 4）. The R value of this calibration was 0.997 with an RMSEP value of 1.3％. These results indicated the quantitative limit of the proposed method is about 1.3％ which can meet the routine analysis according to AOCS official method cd 20-91. Therefore, FTNIR technique exhibits a satisfactory performance and can be useful for quality control applications.
Fig. 3 R elationship between the reference and the predicted determined PC values.
Table 2 Calibration results with different pretreatments for PC (principle components = 5). Spectral range (cm−1) 4918–4809, 5102–4952, 5673–5638 4898–4809, 5318–5287, 5839–5800 5372–4682, 7274–6032
Pretreatments
R
RMSEC/%
a
0.940
7.1
b
0.979
4.3
NF (5, 5) SF (9, 5) SF (9, 5)
0.969
5.2
4963–4616, 5222–5037, 5688–5499
first derivative+SF (7, 5)
0.998
1.0
4963–4616, 5222–5037, 5688–5499
first derivative+SF (9, 5)
0.987
3.3
a
NF (5, 5) represents the Norris derivative smoothing under segment length is 5 and the gap between segments is 5. b SF (9, 5) represents the smoothing under 9 data points and the 5th polynomial. 259
J. Oleo Sci. 64, (3) 255-261 (2015)
X. Chen, X. Yu, Y. Wang
Fig. 4 R elationship between the reference and the predicted determined PC values. 4 Conclusions Traditional column chromatography is time consuming and involves several chemicals, such as petroleum ether and diethyl ether. Thus, studies have been performed to obtain simpler and more rapid techniques. FTNIR technique is a satisfactory alternative because of its advantages, including simple operation and rapid determination, as well as no hazardous chemicals 16, 17）. An NIR-based method to determine PC in frying oils was successfully developed. A significant correlation was obtained between FTNIR predicted and reference values. PLS calibration showed that the R and RMSEC between reference and FTNIR predicted values of PC in frying oils were 0.998 and 1.0％, respectively. Moreover, validation analysis indicated that the R and RMSEP between the reference and the FTNIR predicted values of PC in frying oils were 0.997 and 1.3％, respectively. These results demonstrated that FTNIR analysis can be effectively applied to evaluate the extent of frying oil degradation rapidly and to monitor the changes occurring in edible oils easily during frying.
Acknowledgements Authors would like to thank the Fundamental Research Funds for the Central Universities （QN 2013057） for the financial support.
References 1）Firestone, D.; Gupta, M.; Warner, K.; White, P. Frying technology and practices. AOCS Press:Champaign, IL.PP. 200-216 （2004） .
2）Saguy, I. S.; Dana, D. Integrated approach to deep fat frying: engineering, nutrition, health and consumer aspects. J. Food Eng. 56, 143-152（2003）. 3）Gertz, C. Chemical and physical parameters as quality indicators of used frying fats. Eur. J. Lipid Sci. Tech. 102, 566-572（2000）. 4）Karakaya, S.; Simsek, S. Changes in total polar compounds, peroxide value, total phenols and antioxidant activity of various oils used in deep fat frying. J. Am. Oil Chem. Soc. 88, 1361-1366（2011）. 5）Dobarganes, M.; Velasco, J.; Dieffenbacher, A. Determination of polar compounds, polymerized and oxidized triacylglycerols, and diacylglycerols in oils and fats: results of collaborative studies and the standardized method（Technical report）. Pure Appl. Chem. 72, 1563-1575（2000）. 6）Hein, M.; Henning, H.; Isengard, H. D. Determination of total polar parts with new methods for the quality survey of frying fats and oils. Talanta 47, 447-454 （1998）. 7）Clark, W.; Serbia, G. Safety aspects of frying fats and oils. Food Tech. 45, 84-89 （1991）. 8）Fritsch, C. Measurements of frying fat deterioration: a brief review. J. Am. Oil Chem. Soc. 58, 272-274 （1981）. 9）AlKahtani, H. A. Survey of quality of used frying oils from restaurants. J. Am. Oil Chem. Soc. 68, 857-862 （1991）. 10）Gertz, C. Routine analysis of deep-frying fats and oils. Lipid Tech. 13, 44-47 （2001）. 11）Erickson, M. D. Deep frying: chemistry, nutrition, and practical applications. vol Ed. 2. AOCS Press: Champaign, IL. PP. 315-320（2006）. 12）Melton, S.; Jafar, S.; Sykes, D.; Trigiano, M. Review of stability measurements for frying oils and fried food flavor. J. Am. Oil Chem. Soc. 71, 1301-1308（1994）. 13）Kuligowski, J.; Carrion, D.; Quintas, G.; Garrigues, S.; Guardia, M. Direct determination of polymerised triacylglycerides in deep-frying vegetable oil by near infrared spectroscopy using partial least squares regression. Food Chem. 131, 353-359 （2012）. 14）Buning-Pfaue, H.; Kehraus, S. Application of near infrared spectroscopy（NIRS）in the analysis of frying fats. Eur. J. Lipid Sci. Tech. 103, 793-797 （2001）. 15）Kazemi, S.; Wang, N.; Ngadi, M.; Prasher, S. O. Evaluation of frying oil quality using VIS/NIR hyperspectral analysis. Int. Comm. Agr. Eng. 7, 1-12 （2005）. 16）Ng, C. L.; Wehling, R. L.; Cuppett, S. L. Method for determining frying oil degradation by near-infrared spectroscopy. J. Agr. Food Chem. 55, 593-597（2007）. 17）Ng, C. L.; Wehling, R. L.; Cuppet, S. L. Near-infrared spectroscopy determination of degradation in vegetable oils used to fry various foods. J. Agr. Food Chem. 59, 12286-12290（2011）.
260
J. Oleo Sci. 64, (3) 255-261 (2015)
Determination of Polar Components in Frying Oils by Fourier-Transform Near-Infrared Spectroscopy
18）Gerde, J.; Hardy, C.; Fehr, W.; White, P. J. Frying performance of no-trans, low-linolenic acid soybean oils. J. Am. Oil Chem. Soc. 84, 557-563 （2007） . 19）Du, R.; Lai, K.; Xiao, Z.; Shen, Y.; Wang, X.; Huang, Y. Evaluation of the quality of deep frying oils with Fourier transform near-infrared and mid-infrared spectroscopy. J. Food Sci. 77, 261-266 （2012） . 20）Gerde, J. A.; Hardy, C. L.; Hurburgh, C. R.; White, P. J. Rapid determination of degradation in frying oils with Near-Infrared spectroscopy. J. Am. Oil Chem. Soc. 84, 519-522 （2007） . 21）Siesler, H. W.; Ozaki, Y.; Kawata, S.; Heise, H. M. Near infrared spectroscopy: principles, instruments, applications. John Wiley and Sons Ltd., New York, pp. 25-30 （2008） . 22）Chu, X. L.; Yuan, H. F.; Lu, W. Z. Progress and applica-
tion of spectral data pretreatment and wavelength selection methods in NIR analytical technique. Progr. Chem. 16, 528-542 （2004）. 23）Chu, X. L.; Xu, Y. P.; Lu, W. Z. Research and application progress of chemometrics methods in near Infrared spectroscopy analysis. Chin. J. Anal. Chem. 36, 350-381（2008）. 24）Zhang, J. Y.; Zhang, J. X.; Yu, X.; Du, S. Qualitative analysis of edible oil oxidation by near Infrared transmission spectroscopy. Food Sci. 4, 43-47（2012）. 25）Barton, I. FE Theory and principles of near infrared spectroscopy. Spectrosc. Eur. 14, 12-18（2002）. 26）Kovalenko, I. V.; Rippke, G. R.; Hurburgh, C. R. Measurement of soybean fatty acids by near-infrared spectroscopy: linear and nonlinear calibration methods. J. Am. Oil Chem. Soc. 83, 421-427 （2006）.
261
J. Oleo Sci. 64, (3) 255-261 (2015)
