Nondestructive measurement of squalene in olive oil by near infrared spectroscopy

1 Nondestructive measurement of squalene in olive oil by near infrared 1 spectroscopy 2 José A. Cayuela1, Juan F. García2 3 1Instituto de la Grasa, CSIC 4 Campus of the University ‘Pablo de Olavide’, Ed. 46. 41013 Seville, Spain 5 2Department of Chemical Engineering, University of Seville, C/ Profesor García González, 1, 41012 6 Seville, Spain 7 1Corresponding author: [email protected] 8 ABSTRACT 9 This study sets the basis for developing a rapid technique for measuring olive oil 10 squalene, which is a healthy compound. This technique, based on near infrared 11 spectroscopy, is environmentally friendly. The most suitable wavelength ranges were 12 defined, studying the possible contribution from the visible spectra. For this purpose, 13 Partial Least Squares analysis was independently set up using two optical arrangements, 14 with wavelengths 350-2500 nm and 1100-2300 nm. Models from only near infrared 15 wavelengths gave the best outcomes. The external validation exercise for estimating 16 olive oil squalene was satisfactory, with r2 0.83 and residual predictive deviation 2.31. 17 The results suggest the proposed technique is useful for estimating olive oil squalene 18 content. A sorting test of olive oil in two classes according to its squalene content was 19 carried out, with threshold in 5.0 g.kg-1, using the model built. The success of this 20 classification was 90%. 21 Keywords: classification; environment friendly; NIR; PLS model; ; threshold. 22 23 1. Introduction 24 Squalene (Figure 1) is a triterpene aliphatic hydrocarbon, and it was named because of 25 its profusion in shark liver oil, its richest source, where it reaches 900 g·kg-1. Shark liver 26 oil has long been used as a traditional health food in Japan, with a particular benefit for 27 vascular health (Hamadate et al., 2015). Squalene is widely distributed in nature, 28 especially in vegetable oils such as olive oil, palm oil, wheat-germ oil, amaranth oil, or 29 rice bran oil (Huang, Lin, & Fang, 2009). Therefore, by using different extraction 30 Postprint of LWT - Food Science and Technology Volume 88, February 2018, Pages 103-108 DOI: 10.1016/j.lwt.2017.09.047 2 methods, vegetables or marine animals can be suitable squalene sources (Vázquez, 31 Torres, Fornari, Senorans, & Reglero, 2007). 32 The main part of virgin olive oil is the saponifiable fraction, a lipid matrix of 33 triglycerides, diglycerides and monoglycerides accounting for 985-995 g·kg-1 34 (Civantos, 1999). Squalene is in relatively high quantities within the olive oil minor 35 fraction. Eisner, Iverson, Mozingo, & Firestone (1965) stated that squalene makes up 36 around 85-90% of the hydrocarbon fraction of olive oils. Besides, it makes up 60–75% 37 of the olive oil unsaponifiable fraction, in concentrations between 0.2 and 7.5 g·kg-1 38 (Tiscornia, & Evangelisti, 1982). 39 Figure 1 40 One of the most important differences between olive oil and plant seed oils is squalene. 41 Compared to seed oils, olive oil is an important source of squalene. In other edible 42 vegetable oils, squalene makes up only 0.02–0.3 g·kg-1 (Rao, Newmark, & Reddy, 43 1998). Thus, olive oil contains 7 to 300 fold more squalene than other vegetable oils and 44 up to 5000 fold more than some vegetable foods (Liu, Ahrens, Schreibman, & Crouse, 45 1976). Therefore, virgin olive oil may be a part of the human diet especially rich in 46 squalene. 47 Besides, the squalene content varies widely depending on the olive oil product with a 48 range of 2 to 7 g·kg-1 (Rao, Newmark, & Reddy, 1998). A significant difference 49 between the extra virgin class (EVOO) and the virgin class (VOO) has been reported, 50 with the latter having more squalene than the refined olive oils (Owen, Mier, Giacosa, 51 Hull, Spiegelhalder, & Bartsh, 2000). Nergiz & Çelikkale (2011) showed that refining 52 reduces the squalene content. Furthermore, they pointed out that the major decrease in 53 squalene in vegetable oils within the refining steps occurs during oils’ deodorization. 54 Olive growing techniques (Psomiadou & Tsimidou, 1999), olive fruit variety (Nergiz, & 55 Ünal, 1990) and extraction (Nergiz, & Ünal, 1990; Samaniego-Sánchez, Quesada-56 Granados, López-García de la Serrana, & López-Martínez, 2010) influence the level of 57 squalene. Squalene acts as a weak antioxidant in olive oil (Owen, Mier, Giacosa, Hull, 58 Spiegelhalder, & Bartsh, 2000). Thus, Psomiadou and Tsimidou (1999) proposed that 59 squalene contributes to olive oil stability in a small quantity, even at low temperatures. 60 3 There is multiple scientific evidence on the beneficial effects that the intake of squalene 61 from food has on health (Newmark, 1997; Lasekan, Clayton, Gendron, & Ney, 1990; 62 Smith, 2000; Ostlund, Racette, & Stenson, 2002; Strandberg, Tilvis, & Miettinen, 1990; 63 Smith, 2000; He, & Corke, 2003). However, this feature of olive oil has received little 64 attention in the market so far, since most consumers are unaware on this fact. 65 The olive oil industry has great interest on determining the quality of olive oil, using 66 fast and reliable techniques. Besides, developing non-destructive techniques to reduce 67 the expense of solvents and reagents is increasingly important in an international 68 context of convergence towards environmental sustainability. Among the various non-69 destructive solutions to these needs, near-infrared spectroscopy (NIRS) has made major 70 achievements. NIRS is based on multivariate models in which the spectral data correlate 71 with the analyzed feature. It provides several important advantages, as NIRS needs no 72 solvents or reagents, thus avoiding a major expense, while being environmentally 73 friendly. Additionally, NIRS is a rapid, non-destructive, and potentially multi-parameter 74 method. 75 Several articles on the use of NIRS and chemometrics for the analysis of different olive 76 oil features have been published in the recent years (Nenadis & Tsimidou, 2017). Stand 77 out studies directed to characterizing intact olives and olive paste for optimizing the 78 milling process (Giovenzana et al., 2017), to control the quality of olive pomace oil 79 blended with palm oil used for deep-frying (Ben Hammouda, Zribi, Ben Mansour, 80 Matthaus, & Bouaziz, 2017), as well as for the authentication and detection of fraud 81 (Karunathilaka, Kia, Srigley, Chung, & Mossoba, 2016). Sorting olive oil based on 82 alpha-tocopherol and total tocopherol content using NIRS has been recently reported 83 (Cayuela & García, 2017). The NIRS ability to analyze the major olive oil quality 84 features has been the subject of several studies (Armenta, Garrigues, & De la Guardia, 85 2007; Bendini, Cerretani, Di Virgilio, Belloni, Lercker, & Gallina-Toschi, 2007; Conte, 86 Brussolo, Pizzale, Carazzolo, Meurens, & Pavan, 2003; Costa, Coelho, Gambarra, 87 Bezerra, Harrop, & Ugulino, 2008; Cayuela, Moreda & García, 2013). In fact, NIRS 88 techniques are methods for these routine analyses in a growing number of laboratories. 89 However, the possibility of measuring squalene in olive oil by NIRS has never been 90 reported up to date. Besides, squalene NIR absorption bands have not yet been 91 described, to the best of our knowledge. 92 4 Since the concentration of squalene varies widely among different olive oils, there is an 93 interest in the development of rapid techniques to distinguish olive oils according to its 94 content. In fact, the industry might have an interest in separating olive oils according to 95 different squalene contents. The traditional method for the analysis of squalene in olive 96 oil is GC. However, it is not usually performed in the olive oil industry, since squalene 97 is not considered in the regulation to characterize the quality or purity of olive oil 98 (European Commission, 1991). Therefore, there is a challenge on characterizing olive 99 oil regarding squalene. This work sets up the basis for developing new rapid NIRS 100 techniques for measuring olive oil squalene content. It was convenient to clarify if there 101 are any regions from the olive oil’s visible spectrum contributing to model performance, 102 since pure squalene is a pale yellow liquid. The wavelengths that contribute to 103 predictive models have been defined. 104 2. Material and Methods 105 2.1. Olive Oils 106 A total set of 180 olive oil samples was made up from different origins. High quality 107 Extra Virgin Olive Oils (EVOO) were bought at olive oil specialized shops; this group 108 contributed with 32 samples, of which 27 were varietal and the remaining 5 were 109 mixtures from different varieties. These EVOO were used to elaborate 17 additional 110 coupage samples. Olive oils normally found in the market were also used; this group 111 was composed of 10 EVOOs, 40 Current Olive Oils and 25 Pomace Olive Oils. Olive 112 oil samples were provided also from a collaborator industry, contributing with 14 113 EVOOs, 25 Virgin Olive Oils and 14 Lampante Olive Oils. The characteristics of the 114 olive oil samples are shown in Table S1. 115 2.2. Reference Analysis 116 Squalene analysis were carried out by Gas Chromatography (GC) according to Lanzón, 117 Albi, Cert, & Gracián (1994), modified according to Moreda, Pérez-Camino, & Cert 118 (2004), at the Instituto de la Grasa (CSIC). Briefly, 0.1 mg of olive oil sample was 119 disposed in a 4 mL screw vial, adding 1 mL of squalane 5 mg·mL-1 as the internal 120 pattern. This was dissolved in heptane to complete a volume of 3 mL and shaked gently 121 by hand. Then 200 L of methanolic 2 mol·L-1 KOH was added, separating the aqueous 122 and lipid phases. The upper phase was collected into a 2 mL chromatography vial and 123 5 then injected into the GC instrument. A GC HP-5890 (Hewlet Packard Enterprise, Palo 124 Alto, USA) equipped with a split/splitless injection system was used with a SP-5 125 capillary column 5% phenylmethylsilicone fused silica, 30 m long, 0.25 mm internal 126 diameter and 0.25 m phase thickness, (Merck, Darmstadt, Germany). Flame ionization 127 detector (FID) and software Chem Station for the recording and processing of data were 128 used. The analyses were conducted with two replicates. The results were given with one 129 significant digit. 130 2.3. Near infra-red spectroscopy 131 Optical arrangements NIRS and VIS/NIRS were used for defining the wavelengths 132 contributing to the predictive models, especially for clarifying the contribution from 133 visible spectra. Besides, using two different instruments allowed checking their results, 134 beyond their comparison. 135 The samples’ spectra were recorded directly from olive oils without any other treatment. 136 The temperature of a body has an important influence on NIR radiation. Therefore, the 137 samples were taken from 4 °C storage and placed at room temperature in the laboratory 138 18h before processing. A thermostatic bath fixed at 33 °C for 30 min held the 20 mL 139 sample containers to ensure temperature stability. The averaged spectrum from two 140 measurements of 50 spectra each was registered, with each sample. The same procedure 141 was used with both optical configurations. 142 For NIRS, the measuring mode was post dispersive transflectance. A Luminar 143 (Brimrose Inc., Maryland, USA) spectrometer was used. This instrument consists of an 144 acousto-optic tunable filter (AOTF) with InGaAs detector (1100-2300 nm). The 145 reference is automatically taken, the scanning speed is 60 ms. The spectrometer is 146 composed of a hand-held unit, equipped with a base for laboratory use. A transflectance 147 probe accessory was used. The probe is in stainless steel, with threaded interchangeable 148 optical path. The spectra were registered as a whole, the spectral variables matching at 2 149 nm intervals. The repeatability of the instrument, expressed as the standard deviation of 150 the average absorbance of five measurements of a white tile, is 6.76 10-4. The signals 151 were captured using Acquire software (Brimrose Corp., Maryland). 152 The VIS/NIRS was carried out using a Labspec (Analytical Spectral Devices Inc., 153 Boulder, USA) spectrometer, with transmittance optical mode consisting of a liquid 154 6 accessory (Ocean Optics, Largo, USA). A quartz spectrophotometric cuvette (Hellma 155 Analytics, Müllheim, Germany) with 10 mm path length held the samples. The whole 156 VIS/NIRS spectrum (350−2500 nm) was registered, with each spectral variable 157 corresponding to a 2 nm interval. The configuration for 50 spectra in continuous 158 acquisition was used. Indico Pro software (Analytical Spectral Devices Inc., Boulder, 159 USA) was used for this purpose. The spectrometer was equipped with three detectors. 160 The detector for the visible range (350-1000 nm) was a fixed reflective holographic 161 diode array with a sensitivity of 512 pixels. A holographic fast scanner InGaAs detector 162 cooled at -25ºC covered the wavelength range of 1000-1800 nm. The same device 163 coupled with a high order blocking filter was used for the interval 1800-2500 nm. The 164 scanning speed was 100 ms, and the acquisition processing time is less than a minute 165 for each sample, all steps included. The repeatability, expressed as the standard 166 deviation of the average absorbance of five measurements of a white tile between 350 167 and 2500 nm, is 6.00 10-4. 168 2.4. Squalene spectrum 169 Squalene of 98% purity (Merck, Darmstadt, Germany) was used as pattern to 170 characterize the squalene VIS/NIR absorption bands. The pattern spectrum was 171 registered only once using the Labspec spectrometer, by averaging four replicates. The 172 rest of the procedure was the same as previously described for olive oil samples. 173 2.5. Chemometry 174 Possible olive oil groups were analyzed by Principal Component Analysis (PCA), 175 which was also used for detecting possible spectral outliers. It was carried out from the 176 olive oils spectra of both optical configurations using The Unscrumbler 9.7 (CAMO 177 Software AS, Oslo, Norway). 178 Multivariate Partial Least Squares (PLS) analysis was performed from the spectral 179 variables of near infrared (NIRS) and visible-near-infrared (VIS/NIRS), using the 180 squalene reference analysis as a dependent variable. Transmittance spectral data were 181 averaged to 8 nm intervals and transformed into absorbance, then, mean normalization 182 (MN), standard normal variate normalization (SNV), and first (D1SG) and second 183 (D2SG) Savitzsky−Golay derivatives treatments were carried out. The PLS models for 184 squalene were built from the averaged and treated spectrum using The Unscrumbler 9.7. 185 7 The full cross internal validation (FCV) procedure was used. The outliers were 186 identified as samples showing significant high residuals, according to The Unscrumbler 187 9.7. Scores plot were displayed from the regression overview plot, then selecting the 188 warning list option. The outlier list was displayed by clicking the outliers button. The 189 residuals are the differences between the predicted and the analyzed values. 190 Two independent multivariate calibration models for squalene prediction (M1 and M2) 191 were established from the spectra obtained with NIRS and VIS/NIRS. The models’ 192 principal components (PCs) were fixed after the tests using 10 PCs at first. 193 The calibration set for PLS models excluded the external validation set. It was defined 194 as one third from the 180 olive oil samples available, counting from the first. Sixty 195 samples were taken by including one of each three, from the data base of The 196 Unscrambler 9.7., in the same order as the samples were included. This validation set 197 was randomly formed, since each sample was randomly registered from a completely 198 independent olive oil batch, even when the mechanical selection by the software it was 199 not random. The squalene concentrations range in the validation set were similar to that 200 range of the calibration, as it is shown forward. Those wavelengths whose correlation 201 with squalene content was closer to zero were removed in successive PLS cycles, using 202 The Unscrambler 9.7. The variable selection ended in the last cycle that improved the 203 squared coefficient of cross validation of the calibration (R2CV). This procedure 204 provided the spectral variables selected for the PLS models. To assess model’s fitness, 205 the standard error of calibration (SEC) and the closeness between their squared 206 coefficient of calibration (R2) and R2CV were considered. 207 2.6. Model Performance Assessment 208 Calibration models were assessed by external validation exercises. For this purpose, 209 squalene content was predicted in a previously reserved set formed by 60 olive oil 210 samples which did not participate in the multivariate models. The model performance 211 was assessed according to the r2 from the external validation exercises, which 212 corresponds to the simple linear regression between the analyzed and predicted values. 213 At the same time, the residual predictive deviation (RPD) from the external validation 214 exercise was considered. The RPD was defined (Fearn, 2002) as the ratio between the  215 from the reference data of the validation set and the standard error of performance 216 8 (SEP). Also, the separate analysis of the calibrations of NIRS and VIS/NIRS allowed 217 confirming their predictive ability. 218 2.7. Classification Tests 219 Classification tests of olive oil according to their squalene content were conducted by 220 using the PLS model as a qualitative discrimination technique. The spectral data from 221 only the configuration providing the best yields were used for this purpose. Two classes 222 of olive oils, High Squalene (HS), with squalene concentration above 5.0 g·kg-1, and the 223 other one Low Squalene (LS), with squalene concentration below or equal to 5.0 g·kg-1, 224 were fixed for classification tests into two squalene levels. These two classes were 225 defined according to the squalene mean of the total sample set analyzed, whose value 226 was this threshold. The technique performance assessment was by its success grade. 227 This was expressed as the percentage of samples in which the predicted and actual 228 classifications coincided. 229 3. Results and Discussion 230 3.1. Olive Oil Spectrum 231 Near-infrared spectra show various overlapping bands, due to the first and second 232 overtones and a combination of the fundamental vibrations, mainly carbon–hydrogen 233 (Shenk, Workman, & Westerhaus, 2001). Assigning the major visible absorption bands 234 of olive oil was done by Moyano, el ndez, Alba, & Heredia (2008). Olive oil spectra 235 from the samples analyzed in this work, shown in Figure 2, are consistent with the 236 previously indicated reports. A first minor peak occured near 415 nm. This area suits 237 the wavelengths of oil absorption for dark blue colored light. It could be due mainly to 238 carotenoids, as well as to pheophytin A, pheophorbide A and pyropheophytin A. A 239 second peak was near 450 nm, which corresponds to blue light absorption, a 240 characteristic of carotenoids. A third peak appeared around 670 nm, which coincides 241 with chlorophylls absorption (Moyano, el ndez, Alba, & Heredia, 2008). A broad 242 absorbance band showed around 1220 nm, probably due to second overtones of C–H 243 and CH=CH– stretching vibrations of oil. A high intensity absorbance peak occured 244 around 2300 nm, caused by a combination of fundamental vibrations of the C-H groups 245 (Hourant, Baeten, Morales, Meurens, & Aparicio, 2000). The squalene VIS/NIR spectra 246 registered in this study are shown in Figure 3. The major differences in the spectrum of 247 9 squalene with respect to olive oil correspond to the visible zone, since squalene is 248 almost colorless. On the contrary, in the NIR region the peaks with squalene were 249 practically the same as in olive oil, without remarkable differences. As a hydrocarbon 250 molecule, a part of the squalene C–H and CH=CH– stretching vibrations may be 251 overlapping those of triglycerides, whose structure is also carbon–hydrogen. However, 252 the spatial configuration of the squalene molecule is clearly different from that of 253 triglycerides due to its specific bonds. Moreover, the latter have oxygen, while squalene 254 lacks this chemical element. Therefore, NIR absorption intensities may reflect such 255 differences among different olive oils, despite the fact that they maybe not explicit in 256 the spectrum shape. 257 Figure 2 –Figure 3 258 3.2. Population Characterization 259 The values of the squalene reference analysis of the calibration and external validation 260 sets are gathered in Table 1. As can be seen, a wide squalene variation integrates into 261 the calibrations, ranging from 1.01 g·kg-1 to 10.15 g·kg-1. The statistical mean of the 262 calibration sets analyzed was 5.02 g·kg-1 of squalene. The squalene content in the olive 263 oils analyzed in this work showed a range even wider than those described in the 264 literature (Tiscornia & Evangelisti, 1982; Rao, Newmark & Reddy, 1998; Owen, Mier, 265 Giacosa, Hull, Spiegelhalder, & Bartsh, 2000). This range reflects the extent of 266 variation of squalene in this product, which accounts for the interest of distinguishing 267 olive oils that are helpful to health due to their high squalene content. 268 Table 1 269 3.3. Principal Component Analysis 270 The olive oils spectra PCA analysis of both spectrometers showed the absence of 271 sample groups. The PCA of NIRS, shown in Figure 4, stands out for showing three 272 samples widely separated from the major group. Two more samples appeared separated 273 from the major group, both consistent with the HS class. The remaining samples did not 274 show any consistent grouping trend. In the PCA of the VIS/NIRS spectra, shown in 275 Figure 5, only two samples appear clearly separated. One of these samples matches the 276 16 Tiscornia, E. F. M., & Evangelisti, F. (1982). Chemical compsqualenetion of olive oil 440 and its variations induced by refining. Rivista Italiana delle Sostanze Grasse, 59, 519–441 56. 442 Rao, C. V., & Reddy, B. S. (1993). Modulating effect of amount and types of dietary fat 443 on ornithine decarboxylase, tyrsqualenene protein kinase and prostaglandins production 444 during colon carcinogenesis in male F344 rats. Carcinognesis, 14, 1327–1333. 445 Vazquez, L., Torres, C. F., Fornari, T., Senorans, F. J., & Reglero, G. (2007). Recovery 446 of squalene from vegetable oil sources using countercurrent supercritical carbon dioxide 447 extraction. Journal of Supercritical Fluids, 40, 59–66. 448 449 Figure captions 450 Figure 1. Squalene structure. 451 Figure 2. Olive oil visible and near infrared spectra from the samples analyzed. 452 Figure 3. Visible and near infrared spectra of the squalene pettern (purity 98.0%). 453 Figure 4. Principal Component Analysis of the near infrared spectra (1100-2300 nm) 454 from the analyzed olive oils. 455 Figure 5. Principal Component Analysis of the visible and near infrared spectra (350-456 2500 nm) from the analyzed olive oils. 457 Figure 6. Spectral variables contributing to the squalene near infrared model (M1). 458 Figure 7. Partial Least Squares quantitative squalene model (M1). 459 Figure 8. Prediction exercise of the olive oil squalene content (V1). 460 461 Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Table 1 N  σ Range Calibration 118 5.10 2.19 1.01-10.15 Validation 59 4.88 2.33 1.22-10.02 Table 1. Statistics of the olive oil squalene content (g·kg-1) of the calibration and external validation sets. N, sample set size; , mean; σ, standard deviation.