Sorting olive oil based on alpha-tocopherol and total tocopherol content using near-infra-red spectroscopy (NIRS) analysis

1 Sorting olive oil based on alpha-tocopherol and total tocopherol 1 content using Near-Infra-Red Spectroscopy (NIRS) analysis 2 José A. Cayuela1, Juan F. García2 3 1Instituto de la Grasa, CSIC 4 Campus Universitario Pablo de Olavide, Ed. 46. 41013 Sevilla, Spain 5 2Department of Chemical Engineering, University of Seville, C/ Profesor García González, 1 6 41012 Sevilla Spain 7 1Corresponding author: jacayu[email protected] 8 Abstract 9 Olive oil is an important vitamin E source, which shows a wide variation range. 10 Therefore the interest on distinguish classes. In this study, we assessed models based on 11 partial least squares (PLS) and discriminant analysis (PLS-DA), using near-infrared 12 spectroscopy (NIRS). Estimating the  –tocopherol and total tocopherols contents by 13 using the PLS models were suitable according to the predicting exercises, which gave 14 residual predictive deviations 2.37 and 2.01. Sorting test of olive oil in two classes by 15  -tocopherol with the PLS model provided 99.9% success. The PLS-DA assessment for 16 the same purpose gave coefficients of predictive specificity and sensitivity for the high 17  -tocopherol class 0.96 and 0.84, respectively. The data proves the feasibility of 18 estimating the olive oil-tocopherol or total tocopherols contents by using NIRS. 19 Besides, these techniques can be helpful rapid methods in the industry for sorting olive 20 oils according to their vitamin E content. They are friendly to the environment, which is 21 important. 22 Key words:  -tocopherol, classification, olive oil, NIRS, tocopherols, vitamin E. 23 *Manuscript Click here to view linked References Postprint of Journal of Food Engineering, Volume 202, June 2017, Pages 79–88 DOI: 10.1016/j.jfoodeng.2017.01.015 2 Abbreviations: AOTF, acousto-optic tunable filter; CSIC, Spanish Council for Scientific Research; 1DSG, first 24 Savitzsky-Golay derivative; 2DSG, second Savitzsky-Golay derivative; EVOO, extra virgin olive oil; FCV, full cross 25 validation; H  T, High  -Tocopherol class; InGaAs, detector using In, Ga and As chemical elements; iPLS, interval 26 partial least squares; LT, Low -Tocopherol class; MN, mean normalization; NIR, near infrared; NIRS, near 27 infrared spectroscopy; PCA, principal component analysis; PLS, partial least squares; PLS-DA, partial least squares 28 discriminant analysis; PLSf, PLS factors; r, regression coefficient between the predicted and reference values; R, 29 calibration coefficient; RCV, calibration coefficient from the internal cross validation; RPD, residual predictive 30 deviation; SEC, standard error of calibration; Vis/NIR, visible and near infrared; Vis/NIRS, visible and near infrared 31 spectroscopy; Vit E, vitamin E. 32 33 1. Introduction 34 Olive oil is a prominent source of vitamin E (Vit E), since it provides between 10 and 35 150 mg of  -tocopherol each 100 g. In fact, vegetable oils are the richest source of Vit 36 E according to food composition tables (Souci, Fachman & Kraut, 1994; USDA, 2007). 37 Pomace olive oil shows higher quantities than those above indicated (Perona, Arcemis, 38 Ruíz, & Catalá, 2005). A daily consumption of 25 g Virgin Olive Oil (VOO) would 39 provide almost 30% of the recommended intake of this vitamin (EFSA, 2015, 40 www.nal.usda.gov/fnic/foodcomp/ search). 41 Tocopherols are present in variable quantities within the minor olive oil fraction. As an 42 example, Beltrán et al. (2010) reported total tocopherols content ranging between 8.4 43 and 46.3 mg/100 g within thirty olive cultivars, which were monitored during fruit 44 ripening for three consecutive crop seasons. Other leading sources of Vit E as sunflower 45 peeled seed and almond, contribute around 37 mg/100 g and 26 mg/100 g (Souci, 46 Fachman & Kraut, 1994). 47 Tocopherols are methylated phenols many of which have Vit E activity. In nature, Vit E 48 exists as at least eight naturally occurring compounds, including  -,  -,  - and  -49 tocopherol and  -,  -,  - and  -tocotrienol. The  -tocopherol is the most biologically 50 active and occurs naturally as one isomer (Dutta & Dutta, 2003). It is the main isomer in 51 Europe, where the main dietary sources are olive and sunflower oils (Wagner, Kamal-52 3 Eldin, & Elmadfa, 2004). In contrast, γ-tocopherol is the most common form in the 53 American diet because of a higher intake of soybean and corn oil (Jiang, Christen, 54 Shigenaga, & Ames, 2001; Wagner, Kamal-Eldin, & Elmadfa, 2004). 55 Vit E is antioxidant and it has regulatory cellular and molecular roles. As antioxidant, 56 Vit E inhibits lipid oxidation in food and organisms (Lodge, Traber, Elsner, & 57 Brigelius-Flohé, 2000) by stopping the radical oxygen species (ROS) chain reaction 58 (Wiseman, Tijburg, & Van de Put, 2002). As a result, it prevents the peroxidation of 59 polyunsaturated fatty acids from cellular and subcellular membranes. Also, -60 tocopherol plays an important role as antioxidant in olive oil stability (Deiana et al., 61 2002). The biological activity of tocopherols makes them compounds that play an 62 essential role against aging (Shaidi, 2004; Jessup, Horne, Yarandi, & Quindry, 2004). 63 Improving the olive oil industry and its economy, continuously stimulates the search for 64 new technologies. The main goals are the product’s quality, for which tocopherols 65 content is important. The olive oil industry has great interest on checking the quality 66 using fast and reliable techniques, both to simplify the product handling, as well as for 67 the classification and labeling. Among the various nondestructive solutions to these 68 needs, near-infrared spectroscopy (NIRS) has made major achievements. The ability of 69 NIRS for analyzing the major quality features of olive oil, such as free acidity or 70 peroxides value, has been reported (Mailer, 2004; Bendini et al., 2007, Conte et al., 71 2008, among others). In fact, NIRS techniques are methods for these routine analyses in 72 a growing number of laboratories. NIRS is based on correlating the spectral data with 73 the analyzed feature, using multivariate models. It has several important advantages, as 74 NIRS does not need solvents or reagent, thus avoiding a major expense. It protects the 75 environment, increasingly important. Besides, NIRS techniques are rapid and can be 76 4 multi-parameter. Must note that using NIRS techniques needs calibrating for a defined 77 purpose, using a specific spectrometer. As well, it needs a procedure for periodic 78 validation. 79 The more usual methods for olive oil tocopherols analysis is HPLC, according to 80 standards ISO 9936 or IUPAC 2432, although there are not official methods. Analyzing 81 tocopherols in animal feeds by using NIRS (González-Martín et al., 2006a; González-82 Martín et al., 2006b) has been reported. As well, Ayerdi-Gotor et al. (2007) suggested 83 that NIRS could be useful to discriminate sunflower seeds according to their tocopherol 84 levels. Besides, Szłyk et al. (2005) reported NIRS determination of -tocopherol in 85 several edible oils, after extraction with ethanol. Measuring olive oil tocopherols 86 directly without sample preparation by using NIRS has not been studied up to date at 87 the best of our knowledge. 88 Besides, from a consumer’s perspective there is increasing interest in information about 89 food bioactive compounds content. Its satisfaction is a major target for the food industry 90 according to official rules (EU, 2011). The industry may wish distinguishing olive oil 91 according to tocopherols content, rather than standing a detailed tocopherols analysis. 92 This technique can provide sorting the produce, avoiding the mix of batches with 93 different tocopherols content. Later, it allows labeling the olive oil according to their 94 Vit E, as well as diversify the product. This labeling holds a potential added value, 95 either in economic terms or in trade competitiveness. 96 This study had as a first purpose to prove a new NIRS technique using PLS models, for 97 estimating  -tocopherol and total tocopherols content in olive oil. A second objective 98 was to test two alternative techniques for sorting olive oils according to their -99 tocopherol content, based on quantitative PLS and qualitative PLS-DA. These 100 5 techniques can be helpful as rapid methods for sorting olive oils in the industry 101 according to Vit E content. Thus they may be useful to the competitiveness of the 102 product. 103 2. Material and Methods 104 2.1. Olive Oils 105 An important reason for the robustness of NIRS calibrations is the statistical range of 106 the analyte. Olive oil samples were taken from different sources to guarantee that their 107 characteristics were sufficiently broad enough. High quality Extra Virgin Olive Oils 108 (EVOO) were acquired on special markets, this group contributing with 16 samples, 11 109 of which were varietal and the remaining 5 were coupages from different varieties. 110 Olive oils normally found in the market were included, this group composed by 80 111 EVOO and 80 not virgin olive oils. Olive oil samples were provided also from a 112 collaborator industry, contributing with 20 EVOO and 25 virgin olive oils. Additionally, 113 28 lampante and 30 pomace olive oils were used to increase the diversity of –114 tocopherols (, , or  –tocopherol). Besides, a research project provided 48 EVOO 115 samples which were used for the optical configuration transflectance measurements, 116 described later. These were extracted in the Instituto de la Grasa (CSIC) from the olives 117 using a laboratory mill (MC2, Seville, Spain) based on the Abencor system (Martínez et 118 al., 1975). The samples in total used were 327. 119 2.2. Spectral Acquisition 120 The spectrum of every sample were acquired using two spectrometers with different 121 features, for standing results with their optical configurations. The spectrometers used 122 6 were Labspec (Analytical Spectral Devices Inc., Boulder) and Luminar 5030 (Brimrose 123 Corp., Maryland). 124 The temperature of a body has an important influence on the NIR radiation it reflects 125 and absorbs, thus it is decisive in NIRS (Jiang et al., 2008). Therefore, the samples were 126 taken from 4 °C storage and placed in the laboratory 18 h before processing. Before 127 recording spectra with both spectrometers, a water bath (Nahita, London, United 128 Kingdom) fixed at 33 °C for 30 min. held the 20 mL sample containers, assuring the 129 temperature stability. 130 Wavelengths in the visible spectrum can carry useful information related to the analyzed 131 parameter, since the olive oil tocopherols are pale yellow color (Budavari, 1989). This 132 was the purpose for using Labspec (Analytical Spectral Devices Inc., Boulder, 133 Colorado, USA), a Visible/NIR (Vis/NIR) spectrometer equipped with three detectors. 134 The detector for the visible range (350-1000 nm) is a fixed reflective holographic diode 135 array with a sensitivity of 512 pixels. A holographic fast scanner InGaAs detector 136 cooled at -25 ºC covers the wavelength range of 1000-1800 nm. The same coupled with 137 a high order blocking filter runs for the 1800-2500 nm interval. The instrument equips 138 internal shutters and automatic offset correction, the scanning speed is 100 ms. The 139 repeatability of the instrument, expressed as standard deviation on the average 140 absorbance of five measures of a white tile between 350 and 2500 nm, is 6.00 10-4 cm-1 141 mol-1.With the Labspec, the spectra were registered by transmittance from each sample 142 of EVOO directly, without any other treatment. A Hellma quartz spectrophotometric 143 cuvette with 10 mm path length held the samples while their averaged spectra were 144 acquired. The whole spectrum Vis/NIR (350−2500 nm) was registered, each spectral 145 variable matching to a 1 nm interval. Configuration for 50 spectra in continuous 146 acquisition was used, each spectral variable matching to 1 nm interval. Indico Pro 147 7 software (Analytical Spectral Devices Inc., Boulder, Colorado, USA) was used for this 148 purpose. The registering time is less than a minute for each sample spectrum, all steps 149 included. 150 Luminar 5030 (Brimrose Corp., Baltimore, Mayland, USA)is an AOTF (acousto-optic 151 tunable filter) NIR spectrophotometer, equipped with a transflectance post dispersive 152 optical configuration and InGaAs (1100-2300 nm) detector. The reference spectrum is 153 taken automatically by the instrument, similarly to an UV-Vis spectrophotometer dual 154 beam. The beam divides before leaving the instrument, and a small portion is sent to a 155 second detector that makes the reference. The scanning speed in Luminar 5030 is 60 ms. 156 The spectrometer has a hand-held unit, equipped with a base for optional use in the 157 laboratory. The instrument’s liquid probe accessory was used. The liquid probe is in 158 stainless steel, with a threaded interchangeable optical path.The NIR radiation coming 159 from the optical fiber goes through the liquid sample, reflects in the polished stainless 160 steel surface of the optical path's cylinder. Then, it back through the liquid sample, 161 heading spectrometer detector. Thus, this is transflectance. The spectrometer set is 162 complete with its computer unit. The whole spectrum was registered, each spectral 163 variable matching to a 2 nm interval. The repeatability of the instrument, expressed as 164 standard deviation on the average absorbance of 1100 to 2300 nm of five measures of a 165 white tile, is 6.76 10-4 cm-1 mol-1. The signals are acquired with software Acquire 166 (Brimrose Corp., Baltimore, Maryland, USA). Averaged spectrum were obtained for 167 each sample, resulting from a total 100 spectra matching to two measures of 50 spectra 168 each, it for both spectrometers. 169 2.3. Reference Analysis 170 8 The analysis ofand tocopherols were carried out by HPLC according to 171 standards ISO 9936 and IUPAC 2432, in the Instituto de la Grasa (CSIC). Briefly, 100 172 mg of oil was diluted with hexane to 10 mL final volume. The prepared solution was 173 filtered with Minisart RC15 (0.45 µm pore size, Hannover, Germany), and then a 174 portion (20 µL) of the final solution was subjected to the FLD-normal Phase HPLC. The 175 HPLC system consisted of a HP 1100 series (Hewlett-Packard, strasse 8, 76337 176 Waldbrown, Germany), G 1311A Quat pump, G1316A column oven, G1313A ALS 177 injector. Silice 5 µm, 250x4 mm (Lichrospher 5 µ, sil 60 A, 626077-1) was used as the 178 HPLC column. The mobile phase it was hexane/isopropilic alcohol (99:1). The flow 179 rate was adjusted to 1.0 mL/min and the temperature was maintained at 25ºC. 180 Tocopherols were detected by HP-FLD detector (excitation 290 nm, emission 330 nm, 181 HP1100 series, G1321-95002, Germany). All peak areas were registered using HP 182 Chemistation program 2010. 183 Calibration curves were established from patterns with different concentrations of 184  and total tocopherols between 1 and 11.5 mg/mL in n-hexane. The standards 185 curves were linear, with a concentration range of 1-10 mg/L. These were admitted when 186 r2 ≥ 0.999, respectively. The -tocopherols concentrations in olive oil were calculated by 187 extrapolating the peak areas within the calibration curves. The results are expressed as 188 ppm, with one decimal. 189 The determination was made twice in each sample, and these repeats averaged. The 190 values in which there was an error of ± 10% were eliminated. The tocopherols contents 191 are reported in mg/kg of olive oil (ppm). Analytical sensitivity of tocopherols was 2 192 ppm. The time needed for the analysis was about 20 min. 193 2.4. Chemometrics and Calibration Procedure 194 9 Principal Component Analysis (PCA) was carried out from the olive oil spectra using 195 The Unscrumbler 9.7 (CAMO Software AS, Oslo, Norway). It was used both for the 196 NIR and Vis/NIR wavelengths for analyzing the possible bundling of olive oils and 197 detecting possible spectral outliers. The outliers were detected in PCA as those samples 198 with high residuals, according to the procedure described by the software previously 199 referred. Quantitative tocopherols models were built from the spectral variables visible 200 (Vis) and near infrared (NIR), by Partial Least Squares (PLS). The data from analysis of 201 virgin olive oils were used as reference. Transmittance spectral data were reduced to 8 202 nm intervals by average, transformed to absorbance and mean normalized (MN). Then, 203 treatments by first (1DSG) and second (2DSG) Savitzsky−Golay derivatives were 204 tested. These derivatives were both carried out with 3 smoothing points, and their 205 polynomial order was 2. The full cross internal validation (FCV) procedure was used. 206 The calibration set for models development excluded an external validation set. We 207 took one out of each three olive oil samples from the third one, to stand one third of the 208 total samples. Therefore, this external validation set does not engage in the 209 multivariate models. The calibration and validation sample sets at beginning were the 210 same for all the tests. Calibration models for predicting  and total tocopherols 211 were established from the spectra gained with Luminar and Labspec. The models’ PLS 212 factors were set after tests, using 15 at first. These treatments were made by using The 213 Unscrumbler 9.7. 214 Selecting the spectral variables involved in the models was made by consecutive cycles 215 removing those which the contribution were closer to zero. Variable selection ended in 216 the last cycle that improved the statistical model R2CV. Model fitness was assessed by 217 the closeness between their R2 and R2CV and by the standard error of calibration (SEC). 218 16 Vis/NIRS. As mentioned previously, the purpose for using these olive oils types was to 357 improve the robustness of the calibrations. Those samples showing relation of predicted 358 to analyzed values higher to 2, as well as the same for the inverse, were separated in 359 these calibrations only from such classes. The same happened for the NIRS and 360 Vis/NIRS calibrations of total tocopherols, since  -tocopherol is the major in them, 361 applying the same treatment. On the contrary, in the  -tocopherol and  -tocopherol 362 calibrations, some EVOO samples showed bad fit while the fact before described with 363 lampante and pomace olives oils did not occurs. This probably is because of the higher 364  -tocopherol and  -tocopherol values in some pomace and lampante samples, which we 365 did check. These facts explained the different sizes of the calibration sets for both 366 optical modes and types of tocopherols shown in Table 2. 367 Table 2 368 The NIRS model for  -tocopherol (M1) provided R = 0.95, RCV = 0.94 and SEC = 369 36.14. The calibration statistics from the model for  -tocopherol held with Vis/NIR 370 provided R = 0.94, RCV = 0.92 and SEC = 33.90. 371 On total tocopherols, the model from NIRS (M2) yields R = 0.92, RCV = 0.88 and SEC = 372 57.15. Besides, the statistics from the Vis/NIR were R = 0.91, RCV = 0.89 and SEC = 373 43.83. 374 The general mathematical equation of M1 and M2 models are [2]: 375 Yp = B0 + ∑ (Bi . Xi) [2] 376 Where 377 17 Yp = values predicted of -tocopherol or total tocopherols respectively in M1 and M2 378 (ppm). 379 B0 = B0 coefficient. 380 Bi = Coefficient of math treated spectrum for each contributing wavelength. 381 Xi = Absorbance of the sample for each contributing wavelength (cm-1 mol-1 nm-2). 382 The B0 and Bi values are given by the software. 383 The Figure 4 represents the multivariate regressions for -tocopherol (a) and total 384 tocopherols (b). 385 According to the data above, both optical configurations showed similar calibration 386 fitness, with small differences in their statistics. Therefore it is reasonable to expect 387 similar working from both. 388 Figure 4 389 3.6. External Validation Exercises 390 The statisticals results obtained from the external validations tests by to predicting 391    and totaltocopherols on the reserved set of 107 samples reserved are shown 392 in Table 3. As these statisticals results have shown, the models with the transflectance 393 optical configuration from with NIR wavelengths NIR only, performed better than the 394 Vis/NIR transmittance model. This fact happened for  ,  and total tocopherols. For  -395 tocopherols, both configurations provided low performance, Vis/NIRS showing r = 0.41 396 versus r = 0.26 with NIR. It is reasonable to think this last is because  tocopherol is a 397 minor compound in olive oil. 398 18 The validation using M1 model (V1) for -tocopherol is plotted in Figure 5a. The 399 model performance is defined by the r value 0.91 and RPD = 2.37 from this external 400 validation exercise. The NIRS model for total tocopherols was tested for predicting the 401 validation set (V2). This external validation, depicted in Figure 5b, gave r = 0.90 and 402 RPD = 2.01. 403 According to several authors (Williams, 2014; Esbensen, Geladi, & Larsen, 2014), RPD 404 interpretation must consider the material analyzed. There is general agreement that 405 models with RPD higher to 10 are excellent (Fearn, 2002, Williams, 2014, Fearn, 2015, 406 Esbensen, Geladi, & Larsen, 2014). RPD between 5 and 10 holds high accuracy and 407 aptness for analysis analogous to that of the reference methods. Those with values 408 between 2 and 5 offer the accuracy wanted for providing good estimations (Williams, 409 2014). Other authors consider good RPD greater than two (Barlocco et al., 2006). One 410 Sshould note RPD is essentially the same statistic r (Fearn, 2002; Minasny & 411 McBratney, 2013). The prediction exercise for -tocopherol, V1 (Figure 8), provided r = 412 0.91 and RPD = 2.37, while for total tocopherols V2 (Figure 9) gave r = 0.90 and RPD = 413 2.00. These statistics prove that referred models are useful for estimating  - and total 414 tocopherols in olive oils. 415 416 Table 3 417 Figure 5 418 3.7. Sorting Tests 419 The olive oil  -tocopherol content was classified by using the M1 PLS model on the 420 validation set previously reserved. This classification was according to the classes High 421 19  -Tocopherol (H  T), with -tocopherol higher to 200 ppm, and Low -Tocopherol 422 (LT), lower or equal to 200 ppm. The result it was 91 samples correctly sorted out of 423 the 104 samples and a success of 99.9%. This procedure has been registered in the 424 Spanish Patent and Trademark Office as the patent P201531729., currently in priority 425 year. The sorting test results and for predicted and reference values of -tocopherol are 426 in the Appendix. 427 The iPLS method did select four spectral windows for the PLS-DA model, which gave 428 the best calibration statistics, shown in Table 4. We must note in the case of PLS-DA 429 models there is not a simple regression vector to use, as in PLS. The sorting test two 430 olive oil classes according to -tocopherol, by using this PLS-DA model with threshold 431 at 200 ppm, gave Sn = 0.85 and Sp = 0.96. The results from the validation exercises 432 carried out with PLS-DA models both from the whole spectrum and from the iPLS 433 variable selection, are gathered in Table 4. One can notice in these data, based on the 434 statistical analysis results, that these models perform better Can note in these data the 435 statistics points out a better working when using the whole spectrum. Therefore, the 436 validation exercises did not prove the seeming a considerable improvement by iPLS 437 method. The PLS-DA method for distinguishing the proposed classes of olive oils 438 according to their -tocopherol content was successful. PLS-DA achievement is 439 similar, although, it is in whole overall slightly lower than that from PLS. PLS yields a 440 continuous quantitative variable continuous, which we use to set a discrete variable of 441 two classes, while PLS-DA classifies using directly this discrete class variable. Thus, 442 comparing results from PLS-DA against with PLS quantitative classification is difficult. 443 Should note that more reference information involves in the quantitative PLS model 444 than in PLS-DA, this last built directly from the discrete variable. Therefore, PLS 445 20 supplies more information, implying that PLS can provide greater accuracy and 446 robustness. The results of this study agree with that. 447 Table 4 448 4. Conclusions 449 The models using only NIR wavelengths predicted better the total tocopherols, -450 tocopherol and -tocopherol than that using wavelengths from the visible spectrum. 451 Predicting -tocopherol with the NIRS model provided a valid estimate of the Vit E in 452 the olive oils. The validation exercises of sorting olive oils according to-tocopherol 453 by the quantitative PLS model affords success higher to 99%. The same purpose carried 454 out by PLS-DA yields similar, although slightly lower. The data proves the feasibility of 455 estimating the olive oil-tocopherol or total tocopherols contents by using NIRS. It 456 may provide useful information on product labeling. At the same time, these rapid 457 techniques can be helpful in the industry for sorting olive oils according to their Vit E. 458 The use of these techniques needs a suitable software, and calibrating for a specific 459 spectrometer. As well, it requires a procedure for periodic validation. 460 Acknowledgements 461 We are thankful to the European Regional Development Fund, the Ministry of Economy 462 and Competitiveness of Spain and the Spanish Council for Scientific Research, funding 463 the Recupera2020-1.4.4. project, within which this research has been conducted. 464 Bibliography 465 Altman, D.G., & Bland, J.M. (1994). Diagnostic tests. 1: Sensitivity and specificity. 466 British Medical Journal, 308, 1552. 467 21 Ayerdi-Gotor, A., Farkas, E., Berger, M., Labalette, F., Centis, S., Daydé, J., Calmon, 468 A. (2007). Determination of tocopherols and phytosterols in sunflower seeds by NIR 469 spectrometry. European Journal of Lipid Science and Technology, 109, 525-530. 470 Barlocco, N., Vadell, A., Ballesteros, F., Galietta, G., Cozzolino, D., 2006. Predicting 471 intramuscular fat, moisture and Warner-Bratzler shear force in pork muscle using near 472 infrared reflectance spectroscopy. Animal Science, 82, 111-116. 473 Beebe, K.R., & Kowalski, B.R. (1987). An introduction to multivariate calibration and 474 analysis. Analytical Chemistry, 59, 1007A-1017A. 475 Beltrán, G., Jiménez, A., Del Rio, C., Sánchez, S., Martínez, L., Uceda, M., & 476 Aguilera, M.P. (2010). Variability of Vit E in virgin olive oil by agronomical and 477 genetic factors. Journal of Food Composition Analysis, 23, 633-639. 478 Bendini, A., Cerretani, L., Di Virgilio, F., Belloni, P., Lercker, G., & Gallina-Toschi, T. 479 (2007). In-process monitoring in industrial olive mill by means of FT-NIR. European 480 Journal of Lipid Science and Technology, 109, 498−504. 481 Budavari, S. Ed. The Merck Index. An encyclopedia of chemical, drugs and biologicals. 482 11th ed. Merck & CO., Inc. Rahway, 1989. 483 Cimato, A., Moldi, G., Mattei, A., Niccolai, M., & Alessandri, S. (1991). La 484 caraterizzazione dell’olio extravergine. Consorzio Regionale Olivo extra vergine di 485 oliva ‘Tipico Toscano’, pp 1-77. Montelpulciano, Italy. 486 Conte, L.S., Brussolo, G., Pizzale, L., Carazzolo, A., Meurens, M., & Pavan, O. (2003). 487 Application of near infrared reflectance analysis to olive oil production quality control. 488 Rivista Italiana delle Sostanze Grasse, 80, 213−217. 489 22 Costa, A.F., Coelho, M.J., Gambarra, F.F., Bezerra, S.R., Harrop, R.K., & Ugulino, 490 M.C. (2008). NIR spectrometric determination of quality parameters in vegetable oils 491 using PLS and variable selection. Food Research International, 41, 341−348. 492 Deiana, M., Rosa, A., Cao, C.F., Pirisi, F.M., Bandino, G., & Dessi, M.A. (2002). Novel 493 approach to study oxidative stability of extra virgin olive oils: Importance of alpha-494 tocopherol concentration. Journal of Agricultural and Food Chemistry, 50, 4342–4346. 495 Dutta, A., & Dutta, S.K. (2003). Vit E and its role in the prevention of atherosclerosis 496 and carcinogenesis: a review. Journal of the American College of Nutrition, 22, 258-497 268. 498 European Food Safety Authority. (2015). Scientific Opinion on Dietary Reference 499 Values for Vit E as α-tocopherol. European Food Safety Authority Journal, 13, 4149. 500 European Union. Regulation (EU) No 1169/2011 of the European Parliament and of 501 the Council of 25 October 2011 on the provision of food information to 502 consumers. Official Journal of the European Union, November 22, 2011. 503 Esbensen, K.H., Geladi, P., & Larsen, A. (2014). The RPD myth. NIR News, 25(5), 24. 504 Fearn, T. (2002). Assesing calibration: SEP, RPD, RER and R2. NIR News, 13(6), 12-505 14. 506 Fearn, T. (2015). The Library of Babel (validation). NIR News, 26(8), 23. 507 García, D.L., Baeten, V., Fernández, J.A., & Tena, N. (2013). Infrared, Raman, and 508 fluorescence spectroscopies: Methodologies and applications. In R. Aparicio, & J. 509 Harwood (Eds.), Handbook of olive oil: Analysis and properties (pp. 335-393). New 510 York: Springer. 511 23 García, J.F. (2015). Optical path length and wavelength selection using VisNIR 512 spectroscopy for olive oil free acidity determination. International Journal of Food 513 Science and Technology, 50, 1461–1467. 514 González-Martín, I., Hernández-Hierro, J.M., Bustamante-Rangel, M., Barros-Ferreiro, 515 N. (2006a). Near-infrared spectroscopy (NIRS) reflectance technology for the 516 determination of tocopherols in alfalfa. Analytical and Bioanalytical Chemistry, 386, 517 1553–1558. 518 González-Martín, I., González-Cabrera, J.M., Bustamante-Rangel, M., Delgado-519 Zamarreño, M.M. (2006b). Near infrared spectroscopy (NIRS) reflectance technology 520 for determination of tocopherols in animal feeds. Analytica Chimica Acta, 558, 132–521 136. 522 Harwood, J., & Aparicio, R. (2000). Handbook of Olive Oil: Analysis and Properties, 523 Gaithersburg: Aspen Publishers (Chapter 10). 524 Hourant, P., Baeten, V., Morales, M.T., Meurens, M., & Aparicio, R. (2000). Oil and fat 525 classification by selected bands of near-infrared spectroscopy. Applied Spectroscopy, 526 54, 1168−1174. 527 ISO. Standard 9936. Animal and vegetable fats and oils. Determination of tocopherol 528 and tocotrienol contents by high-performance liquid chromatography. Geneva, 2016. 529 Jiang, H.Y., Xie, L.J., Peng, Y.S., Yin, Y.B. (2008). Study on the influence of 530 temperature on near infrared spectra. Guang Pu Xue Yu Guang Pu Fen Xi, 28(7), 1510-531 1513. 532 24 Jiang, Q., Christen, S., Shigenaga, M.K., & Ames, B.N. (2001). Gamma-tocopherol, the 533 major form of Vit E in the US diet, deserves more attention. American Journal of 534 Clinical Nutrition, 74, 714–722. 535 Jessup, J.V., Horne, C., Yarandi, H., & Quindry, J. (2003). The effects of endurance 536 exercise and Vit E on oxidative stress in the elderly. Biological Research for Nursery, 5, 537 47–55. 538 Yong-Ho, K. (2004). Quantification of Tocopherol and Tocotrienol Content in Rice 539 Bran by Ner Infrated Reflectance Spectroscopy. Korean Journal of Crop Science, 49, 540 211-215. 541 Lodge, J.K., Traber, M.G., Elsner, A., & Brigelius-Flohé, R. (2000). A rapid method for 542 the extraction and determination of Vit E metabolites in human urine. Journal of Lipid 543 Research, 41, 148–154. 544 Mailer, R.J. (2004). Rapid evaluation of olive oil quality by NIR reflectance 545 spectroscopy. Journal of the American Oil Chemists Society, 81, 823−827. 546 Martínez, J.M., Muñoz, E., Alba, J., Lanzón, A. (1975). Report on the use of the 547 Abencor olive oil yields analyser. Grasas y Aceites 26, 379-385. 548 inasny, B., cBratney, A. (2013). Why you don’t need to use RPD. Pedometron 33, 549 14-15. 550 Moyano, M.J., el ndez, A.J., Alba, J., & Heredia, F.J. (2008). A comprehensive study 551 on the colour of virgin olive oils and its relationship with their chlorophylls and 552 carotenoids indexes (I): CIEXYZ non-uniform colour space. Food Research 553 International, 41, 505−512. 554 25 Office of Disease Prevention and Health Promotion of the U.S.A. Dietary Guidelines. 555 (2016). http://health.gov/dietaryguidelines/2015/guidelines/ Accessed 02/26/2016. 556 Perona, J.S., Arcemis, C., Ruíz, V., Catalá, A. (2005). Effect of Dietary High-Oleic-557 Acid Oils that are Rich in Antioxidants on Microsomal Lipid Peroxidation in Rats. 558 Journal of Agricultural and Food Chemistry, 53, 730-735. 559 Sayago, A., Marín, M.I., Aparicio, R., & Morales, M.T. (2007). Vitamina E y aceites 560 vegetales. Grasas y Aceites, 58(1), 74-86. 561 Shaidi, F. (2004). Functional foods: Their role in health promotion and disease 562 prevention. Journal of Food Science, 69, 146–149. 563 Shenk, J.S., Workman, J.J., & Westerhaus, M.O. (2001). Application of NIR 564 spectroscopy to agricultural products. In D. A. Burns, & C. W. Ciurcak (Eds.), 565 Handbook of Near Infrared Analysis (pp. 419-474). New York: Marcel Dekker. 566 Souci, S.W., Fachmann, W., & Kraut, H. (1994). Food Composition and Nutrition 567 Tables. (5th edition)., Medpharm Scientific Publishers: Boca Raton: CRC Press. (Olive 568 oil). 569 Szłyk, E., Szydłowska-Czerniak, A., Kowalczyk-Marzec, A. (2005). NIR spectroscopy 570 and partial least-squares regression for determination of natural alpha-tocopherol in 571 vegetable oils. Journal of Agricultural and Food Chemistry, 53(18), 6980-6987. 572 Uceda, M., & Hermoso, M. (2001). La calidad del aceite de olive. In D. Barranco, R. 573 Fernández-Escobar, & L. Rallo (Eds.), El cultivo del olivo (pp. 589-614). 4th edition. 574 Madrid: Mundi-Prensa. 575 Table 1. Statistics of the tocopherols contents (ppm) in the calibration and external validation sets. Calibration N = 218 Validation N = 109 Tocopherols  Range  Range Total 209.0 142.6 64.2 -1078.0 209.5 124.3 63.1 - 762.3  179.7 111.8 54.5 - 755.9 187.9 111.9 55.2 - 466.4  2.4 1.7 0.7 - 14.1 2.4 1.4 0.5 - 10.0  17.8 13.4 2.5 - 103.8 17.7 12.1 1.8 - 73.4  ,  -tocopherol;    tocopherol;    tocopherol (  and  -tocopherols, HPLC analyzed) Total, Total tocopherols (arithmetically calculated by sum of  and  -tocopherols). Table 1 Table 2. Statistics of PLS quantitative models. Total Tocopherols  -tocopherol  -tocopherol  -tocopherol NIR Vis/NIR NIR Vis/NIR NIR Vis/NIR NIR Vis/NIR N 213 197 206 189 211 195 211 197 Tr. D2SG D2SG D2SG D1SG D1SG D2SG D2SG D1SG PLSf 15 5 15 9 14 10 10 10 R 0.92 0.91 0.95 0.94 0.92 0.87 0.64 0.66 RCV 0.88 0.89 0.91 0.92 0.88 0.85 0.52 0.54 SEC 57.15 43.83 36.14 33.90 5.34 4.54 0.58 0.59 PLS, Partial Least Squares; N, samples number; Tr., treatment; PLSf, PLS factors; R, calibration coefficient; RCV, calibration coefficient from the internal cross validation; SEC, standard error of calibration. NIR, spectra acquired with Luminar (1100-2300 nm); Vis/NIR, spectra acquired with Labspec (350-2500 nm); D2SG, Savitzsky-Golay second derivative; D1SG, Savitzsky-Golay first derivative. Table 2 Table 3. Statistics of predicting exercises with PLS quantitative models. PLS, Partial Least Squares; N, samples number; r, regression coefficient between the predicted and reference values; SEP, standard error of prediction; RPD, residual predictive deviation; NIR, spectra acquired with Luminar (1100-2300 nm); Vis/NIR, spectra acquired with Labspec (350-2500 nm). Total Tocopherols  -tocopherol  -tocopherol  -tocopherol NIR Vis/NIR NIR Vis/NIR NIR Vis/NIR NIR Vis/NIR N 107 91 104 93 106 101 108 102 r 0.90 0.80 0.91 0.86 0.88 0.75 0.26 0.41 SEP 61.84 76.26 47.21 58.28 6.33 8.12 1.35 1.32 RPD 2.01 1.63 2.37 1.92 1.91 1.49 1.04 1.06 Table 3 Table 4. PLS-DA for -tocopherol from NIR spectra. PLS-DA, Partial Least Squares Discriminant Analysis; H  T, high  -tocopherol; L  T, low  -tocopherol. iPLS, interval Partial Least Squares; Sn-Cal, calibration sensitivity; Sp-Cal, calibration specifity; Sn-CV, sensitivity in the cross validation; Sp-CV, specifity in the cross validation; Sn-Pred, prediction sensitivity; Sp-Pred, prediction specifity. (1) iPLS variables: 1302-1342 nm; 1390-1422 nm; 1830-1862 nm; 2030-2062 nm. Note: The PLSDA classification can be represented as a PLS predicting between class A and B, by using y = 0 or y = 1. The threshold value where y predict greater than or less than the same, it determines to which class a sample belongs to: P(y,1) Probability that a sample is class 1 = --------------- (P(y,1)+P(y,0)) where y is the y value predicted from the PLSDA model for the sample in question, P(y,1) is the probability of measuring the given y value for a class 1 sample and P(y,0) is the probability of measuring the y value for a class 0 sample. The two probabilities used above (P(y,1), P(y,0)) are estimated from the y-values observed in the calibration data. Calibration N = 218 Validation N = 109 H  T L  T H  T L  T H  T L  T H  T L  T Spectrum whole iPLS variables(1) Spectrum whole iPLS variables(1) Sn-Cal 0.85 0.96 0.86 0.95 Sn-Pred 0.84 0.96 0.84 0.94 Sp-Cal 0.96 0.85 0.95 0.86 Sp-Pred 0.96 0.84 0.94 0.84 Sn-CV 0.85 0.95 0.86 0.95 Sp-CV 0.95 0.85 0.95 0.86 Table 4 1 Figure Captions Figure 1 - Olive oil Vis/NIR spectra from all samples analyzed. Figure 2 - Spectral variables contributing to the -tocopherol NIRS model (M1). Figure 3 – PCA from olive oils spectra a) 1100-2300 nm. ; b) 350-2500 nm. - tocopherol (ppm). Figure 4 - PLS models a) -tocopherol (M1); b) Total tocopherols (M2). Figure 5 - Prediction exercises a) -tocopherol (V1); b) Total tocopherols (V2). Figure Captions Sample Com. Class . Class Predicted Reference Success 1 EVOO-Com. 178.06 182.28 * 2 EVOO-Com. 209.83 203.60 * 3 EVOO-Com. 232.84 177.54 4 EVOO-Com. 176.70 222.43 5 EVOO-Com. 158.90 196.05 * 6 EVOO-Com. 159.98 175.16 * 7 EVOO-Com. 141.16 185.89 * 8 EVOO-Com. 140.81 111.61 * 9 EVOO-Com. 231.50 116.96 10 EVOO-Com. 170.19 135.92 * 11 EVOO-Com. 210.55 166.77 12 Lamp.-Ind. 130.28 148.30 * 13 Lamp.-Ind. 136.50 162.89 * 14 Lamp.-Ind. 97.49 160.52 * 15 Lamp.-Ind. 162.68 123.11 * 16 Lamp.-Ind. 139.15 117.76 * 17 Lamp.-Ind. 128.84 136.89 * 18 Lamp.-Ind. 136.42 123.39 * 19 Lamp.-Ind. 173.68 241.04 20 Lamp.-Ind. 278.16 219.58 * 21 OO-Com. 168.37 204.14 22 OO-Com. 115.44 82.00 * 23 OO-Com. 74.50 144.60 * 24 OO-Com. 92.41 147.34 * 25 OO-Com. 118.78 146.56 * 26 OO-Com. 148.47 82.24 * 27 OO-Com. 105.22 96.62 * 28 OO-Com. 129.14 117.97 * 29 OO-Com. 136.58 125.34 * 30 OO-Com. 176.28 147.01 * 31 OO-Com. 136.71 134.45 * 32 OO-Com. 130.78 130.20 * 33 OO-Com. 140.61 127.82 * 34 OO-Com. 146.71 148.38 * 35 EVOO-Com. 111.01 154.41 * 36 EVOO-Com. 155.82 108.36 * 37 EVOO-Com. 101.54 140.74 * 38 EVOO-Com. 104.06 116.38 * 39 EVOO-Com. 106.08 91.64 * 40 EVOO-Com. 154.96 107.72 * 41 EVOO-Com. 150.62 141.78 * 42 EVOO-Com. 192.42 153.95 * 43 EVOO-Com. 226.64 143.22 44 EVOO-Com. 151.92 200.12 45 EVOO-Com. 153.47 204.50 46 EVOO-Com. 215.50 156.86 47 EVOO-Com. 225.43 143.39 48 EVOO-Com. 81.19 114.21 * 49 EVOO-Com. 137.00 97.84 * 50 EVOO-Com. 99.55 95.83 * 51 EVOO-Com. 69.35 92.83 * 52 EVOO-Com. 198.85 120.29 * Sample Com. Class . Class Predicted Reference Success 53 EVOO-Com. 97.69 106.85 * 54 EVOO-Com. 39.37 107.68 * 55 EVOO-Com. 40.50 94.41 * 56 EVOO-Com. 108.78 91.97 * 57 EVOO-Com. 60.00 55.21 * 58 EVOO-Com. 82.09 68.24 * 59 EVOO-Com. 92.55 75.16 * 60 OO-Com. 211.54 120.00 61 OO-Com. 164.76 119.72 * 62 OO-Com. 136.50 119.06 * 63 OO-Com. 169.23 121.55 * 64 OO-Com. 61.30 149.12 * 65 OO-Com. 110.94 97.18 * 66 OO-Com. 73.66 111.65 * 67 OO-Com. 95.45 84.72 * 68 OO-Com. 205.08 117.42 69 OO-Com. 127.05 96.79 * 70 OO-Com. 124.06 116.71 * 71 OO-Com. 113.88 69.68 * 72 OO-Com. 144.07 108.46 * 73 POO-Com. 346.01 335.07 * 74 POO-Com. 403.76 320.42 * 75 POO-Com. 375.95 251.43 * 76 POO-Com. 400.16 368.52 * 77 POO-Com. 329.63 387.37 * 78 POO-Com. 341.90 392.60 * 79 POO-Com. 382.61 395.72 * 80 POO-Com. 415.17 389.16 * 81 POO-Com. 341.62 309.31 * 82 EVOO-R.P. 356.22 379.60 * 83 EVOO-R.P. 392.87 333.00 * 84 EVOO-R.P. 368.48 466.40 * 85 EVOO-R.P. 396.30 440.70 * 86 EVOO-R.P. 402.21 406.50 * 87 EVOO-R.P. 347.94 408.60 * 88 EVOO-R.P. 434.57 355.50 * 89 EVOO-R.P. 387.07 465.00 * 90 EVOO-R.P. 372.48 465.30 * 91 EVOO-R.P. 389.44 372.50 * 92 EVOO-R.P. 410.46 453.90 * 93 EVOO-R.P. 419.33 386.60 * 94 EVOO-R.P. 380.58 347.60 * 95 EVOO-R.P. 381.06 349.30 * 96 VOO-Ind. 195.72 166.30 * 97 VOO-Ind. 146.74 162.72 * 98 VOO-Ind. 151.62 137.75 * 99 VOO-Ind. 180.04 115.30 * 100 VOO-Ind. 155.90 95.41 * 101 VOO-Ind. 167.82 122.87 * 102 VOO-Ind. 101.46 125.06 * 103 VOO-Ind. 137.82 116.45 * 104 VOO-Ind. 105.70 161.13 * Appendix. Sorting test using PLS model M1. Predicted and reference values of -tocopherol, ppm; EVOO-Com., extra virgin olive oil commercial; Lamp.-Ind., lampante olive oil industrial; OO-Com., current olive oil commercial; POO-Com., pomace olive oil commercial; EVOO-R.P., extra virgin olive oil from research project; VOO-Ind., virgin olive oil industrial; Success (*), coincidence in sorting the olive oil as high or low in -tocopherol. It is considered high -tocopherol for contents greater than 200 ppm and low-tocopherol for less or equal to this value. Appendix