Olive oil quality classification and measurement of its organoleptic attributes by 1 untargeted GC-MS and multivariate statistical-based approach

8 The capabilities of dynamic headspace entrainment followed by thermal desorption in 9 combination with gas chromatography (GC) coupled to single quadrupole mass spectrometry 10 (MS) have been tested for the determination of volatile components of olive oil. This 11 technique has shown a great potential for olive oil quality classification by using an 12 untargeted approach. The data processing strategy consisted of three different steps: 13 component detection from GC-MS data using novel data treatment software PARADISe, a 14 multivariate analysis using EZ-Info, and the creation of the statistical models. The great 15 amount of compounds determined enabled not only the development of a quality 16 classification method as a complementary tool to the official established method “PANEL 17 TEST” but also a correlation between these compounds and different types of defect. 18 Classification method was finally validated using blind samples. An accuracy of 85 % in oil 19 classification was obtained, with 100% of extra virgin samples correctly classified. 20


25
Olive oil quality is a matter of concern for consumers and producers. It establishes the 26 differences between the products with poor attributes and the products with outstanding 27 features, as well as it contributes to set oil prizes. For this reason and to avoid fraud, many 28 times linked to this specific product (Jabeur, Zribi, & Bouaziz, 2016;Kalogiouri, Aalizadeh, 29 and their derivatives, with different concentrations and odour thresholds. To this extent, 48 qualitative and quantitative analysis of volatile organic compounds (VOCs) has been an 49 important issue of scientific interest for the organoleptic characterization of olive oil. 50 Although PANEL TESTs are quite well trained in distinguishing these differences with an 51 impressive precision, such methodology is rather expensive and remarkably time-consuming. 52 In this scenario, a more objective methodology, based on instrumental responses, could be 53 presented as cheaper and faster alternative approach to PANEL TESTs and could also be 54 useful as a complementary tool to prevent fraud due to sample adulteration by means of 55 quality mislabeling. 56 Dynamic headspace with sorbent trapping (DHS) together with gas chromatography (GC) 57 coupled to mass spectrometry (MS) in full scan mode is a well-known technique that has  together with data acquisition has to be carefully optimized through the use of specialized 91 software to automatically obtain valuable markers (chromatographic peaks and masses) from 92 raw data. As no compounds are selected in advance, chromatography must be robust and has 93 to pursue the best peak resolution possible. Also, data acquisition has to be performed in full-   Internal standard toluene-d8 (tol-d8) ≥ 99% was purchased from Sigma Aldrich (Germany). External standards of volatile compounds used for signal deviation correction (Z-3-hexenal, 123 hexanal, E,E-2,4-hexadienal, 6-methyl-5-hepten-2-one, 6-methyl-5-hepten-2-ol, E,E-2,4-124 heptadienal, R-limonene, 2-isobutylthiazole, guaiacol, E-2-octenal, linalool, 2-125 phenylethanol, methyl salicylate, α-terpineol, β-cyclocitral, Z-citral, E-citral, E,E-2,4-  Olive oil samples were allowed to defrost at room temperature before analysis. Then, they 150 were aliquoted in 4 different 10 mL vials. One aliquot was immediately used to perform the 151 extraction and the remaining ones were stored at 4 ºC.  to find the optimal number of components, looking for a good model fitting (over 95%), noise 261 removal and low residuals, with a core consistency over 95%. Also model overfitting was 262 avoided while obtaining well resolved peaks. As an example, the capabilities of PARADISe 263 for compound detection and noise reduction are displayed in Figure 1. In Figure 1(A), the 264 total ion chromatogram shows a very complex interval, with three presumable compounds.

265
The residuals in this case were up to 10 6 . After selecting 5 different components ( Figure   266 1(B)), residuals were lowered by two orders of magnitude, and the algorithm detected four  i.e with a priori unknown quality. Figure S5 shows a confusion matrix presenting the results  As a final step, PARADISe automatically compares deconvoluted spectra with NIST library 315 (in this case NIST08 (NIST, Maryland, USA)), giving the best fitted candidate for each peak.

316
In order to add more confidence to identification, retention index for each compound was 317 calculated using a C7-C30 alkane mixture which was injected along with the rest of the PLOT graph for each case and to inspect them looking for endpoints, especially in the part 334 of the defect, to see which compounds were highly related to each negative attribute.

335
Applying this methodology to each defect, a group of compounds were considered as 336 responsible of the bad quality of virgin/lampante olive oils, which are summarized in Table   337 2. The results show the great potential of this technique for the identification of defect-related 338 compounds, as well as for the discrimination of samples according to their defect. These   The high pre concentration factor obtained by DHS-TD has allowed the detection of a huge 374 number of volatile compounds in olive oil at trace levels. PARADISe has demonstrated huge 375 capabilities for robust peak detection. Thanks to its special algorithm (PARAFAC2), 376 extremely clean mass spectra has been provided. This has been very useful for tentative 377 identification of unknown compounds when matching their spectra with NIST libraries and 378 also for resolving coeluting peaks.

379
The developed methodology has permitted to obtain an enhanced quality classification 380 model, with a 100% discrimination of extra samples, and an overall 86% accuracy for the 381 three different classes, which reveals it as a very important complement to the PANEL TEST.

382
As a final remark, the method has allowed also to putatively identify and completely identify