Beyond current quality indices: Quantitative volatilomics unrevealed cultivar traits, harvesting practices impact, and aroma blueprint of extra-virgin olive oils

Beyond current quality indices: Quantitative volatilomics unrevealed cultivar traits, harvesting practices impact, and aroma blueprint of extra-virgin olive oils Andrea Caratti a , Angelica Fina a , Fulvia Trapani a , Erica Liberto a , Brígida Jim´ enez-Herrera b , Lourdes Arce c , Raquel M. Callej´ on d,* , Chiara Cordero a,* a Dipartimento di Scienza e Tecnologia del Farmaco, Universit` a di Torino, Via Pietro Giuria 9, Turin 10125, Italy b Junta de Andalucía, Instituto Andaluz de Investigaci´ on y Formaci´ on Agraria, Pesquera, Alimentaria y de la Producci´ on Ecol´ ogica (IFAPA), Centro de Cabra, , Antigua Ctra. Cabra-Do˜ na Mencía, km. 2.5 Cabra, C´ ordoba, Spain c Department of Analytical Chemistry, University of Cordoba, Institute of Fine Chemistry and Nanochemistry. International Agrifood Campus of Excellence (ceiA3), Marie Curie Annex Building. Campus de Rabanales, Cordoba 14071, Spain d Departamento de Nutrici´ on y Bromatología, Toxicología y Medicina Legal. Facultad de Farmacia. University of Sevilla, C/ Profesor García Gonz´ alez, 2, Sevilla 41012, Spain ARTICLE INFO Keywords: Organic Extra-virgin olive oil Comprehensive two-dimensional gas chromatography Parallel detection MS/FID Predicted relative response factors Aroma blueprint Accurate quantification ABSTRACT Virgin olive oil, derived from Olea europaea L., is a staple in the Mediterranean diet and is classified into quality categories by EU regulations and the International Olive Council. Extra Virgin Olive Oil (EVOO), the highest quality, is valued for its nutritional and sensory attributes, driving consumer willingness to pay a premium. However, this makes EVOO susceptible to fraud, necessitating robust quality control methods. This study combines advanced untargeted fingerprinting and quantitative volatilomics using headspace solid-phase microextraction (HS-SPME) coupled to comprehensive two-dimensional gas chromatography with parallel mass spectrometry and flame ionization detection (GC×GC-MS/FID) on EVOOs from Hojiblanca and Picual cultivars to define robust markers of quality, named cultivar markers and cultivation practice indicators. The comprehensive analysis identified over 190 peak features and 84 compounds, revealing distinct volatile profiles influenced by cultivar, cultivation methods, and ripening stages. Of the 125 diagnostic volatile features (Fisher value >4) for cultivar discrimination, alkenes, carbonyls, and alcohols predominate. Key compounds, such as (E)-2-octenal, (E)-2-hexenal, and (Z)-3-hexen-1-ol, reflect lipoxygenase pathway activity and maturation stages across cultivars. Quantification confirmed distinct aroma blueprints between cultivars, driven by differences in OAVs of key-odorants. However, minimal differences between odor active markers for organic and conventional cultivation practices, suggest consumers are unlikely to perceive variations in aroma between the two. 1. Introduction Virgin olive oil, derived from Olea europaea L., is central to the Mediterranean diet and primarily produced in the Mediterranean region (Lioupi et al., 2022; MAPAMA, n.d.). Classified by organizations like Codex Alimentarius, International Olive Council, an subjected to EU Regulations (“European Commission Regulation (EC) Reg. 2568/91,” 1991), its quality categories depend on sensory and physicochemical parameters. Extra Virgin Olive Oil (EVOO), the highest quality category, is prized for its nutritional traits and sensory attributes (Caporaso et al., 2015; Cecchi et al., 2021), driving consumer demand and willingness to pay premiums for specific qualities, such as health benefits, organic certification, or geographic origin (Arroyo-Manzanares et al., 2019). However, this high demand also increases the risk of fraud, necessitating stringent quality control measures to maintain consumer confidence (Conte et al., 2020; M. P. Segura-Borrego et al., 2022a; M. Pilar Segura-Borrego et al., 2022b). Characterizing EVOO involves analyzing a wide range of sensory and physicochemical parameters, with a growing need for advanced analytical methods due to the limited effectiveness of current techniques * Corresponding authors. E-mail addresses: [email protected] (A. Caratti), [email protected] (A. Fina), [email protected] (F. Trapani), [email protected] (E. Liberto), [email protected] (B. Jim´ enez-Herrera), [email protected] (L. Arce), [email protected] (R.M. Callej´ on), [email protected] (C. Cordero). Contents lists available at ScienceDirect Journal of Food Composition and Analysis journal homepage: www.elsevier.com/locate/jfca https://doi.org/10.1016/j.jfca.2024.106975 Received 18 July 2024; Received in revised form 19 September 2024; Accepted 11 November 2024 Journal of Food Composition and Analysis 137 (2025) 106975 Available online 14 November 2024 0889-1575/© 2024 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY license ( http://creativecommons.org/licenses/by/4.0/ ). and evolving adulteration methods (Conte et al., 2020). Technological advancements have been embraced by the olive sector to meet quality demands, with producers and researchers focusing on optimizing production under various agronomic and environmental conditions. Efforts aim to ensure the highest quality of olive oils despite these variables (Cecchi et al., 2022; Hong et al., 2017; Quintanilla-Casas et al., 2020b). Aroma is a crucial quality criterion for EVOO, significantly influenced by factors such as olive variety/cultivar, geographic origin, fruit maturation, irrigation, crop season, agronomic practices, and oil extraction conditions (Beltr´ an et al., 2005; Benito et al., 2012; Jim´ enez et al., 2017; Mansouri et al., 2015; Morell´ o et al., 2004; Romero et al., 2003; Vidal et al., 2019). The volatile profile and overall aroma are especially affected by olive variety/cultivar and ripening stage, driving interest in monovarietal oils. Despite promising analytical methods, no official approach exists to authenticate monovarietal EVOOs, highlighting the need for robust authentication methods (Herrera et al., 2012; Jimenez et al., 2015, 2014; Jim´ enez et al., 2017). The type of farming, particularly organic versus conventional methods, is another significant factor due to consumer demand for organic products and strict agricultural regulations (Jim´ enez-Herrera et al., 2019). While organic olive farming has grown, studies comparing organic and conventional EVOOs are limited and often inconclusive due to uncontrolled variables. Future research needs controlled samples to advance knowledge on organic EVOOs, prevent fraud, and provide valuable information to the industry, producers, and consumers. Gas chromatography-mass spectrometry (GC-MS) is widely used to profile olive oil volatile compounds. Combined with headspace (HS) approaches like solid-phase microextraction (SPME) or dynamic headspace (D-HS), it enables high-throughput profiling and monitoring of many aroma-active compounds (Cecchi et al., 2022; Cordero et al., 2015; Quintanilla-Casas et al., 2020b). The introduction of comprehensive two-dimensional GC (GC×GC) has significantly improved the understanding of EVOO’s chemical complexity, providing higher peak capacity, resolution, lower detection thresholds, and more confident analyte identification (Luki´ c et al., 2019; Marriott and Nolvachai, 2021; Purcaro et al., 2014; Ryan and Marriott, 2003). Moreover, adopting non-targeted data processing [e.g., combined untargeted and targeted fingerprinting UT fingerprinting (Magagna et al., 2016)], the investigation of complex fraction of volatile organic compounds (VOCs) by GC×GC has opened to volatilomics, the discipline that connects the chemical information encrypted in the volatilome (or volatome) with biological properties (Broza et al., 2015; Phillips et al., 2013; Stilo et al., 2021a). The volatilome “contains all of the volatile metabolites as well as other volatile organic and inorganic compounds that originate from an organism”(Amann et al., 2014) super-organism, or ecosystem; as part of the metabolome it brings information worthy to be investigated for product valorization and treaciability. Analytical platforms that implement parallel detection, combining mass spectrometry (MS) and flame ionization detector (FID), offer great opportunities allowing for reliable identification and accurate quantification of analytes without external calibration. This enables true quantitative results for robust marker selection and correlation with quality traits (Stilo et al., 2023; Stilo et al., 2021). The use of GC×GC-MS/FID in olive oil research in particular, allows for a comprehensive mapping of detectable VOCs, provides accurate amounts on selected analytes, and enables causal correlation with biological variables such as olive cultivar, ripening stages, geographical origin, agricultural practices, and technological processing (Magagna et al., 2016; Reichenbach et al., 2019; Stilo et al., 2023; Stilo et al., 2021). The quantitative mapping of key-aroma compounds delineates the aroma blueprint of EVOOs, i.e., the unique pattern of odor active compounds responsible of the aroma identity of the product (Granvogl and Schieberle, 2022) This study, for the first time, combines advanced untargeted fingerprinting and quantitative volatilomics using HS-SPME-GC×GCMS/FID on a selection of monocultivar EVOOs to define both robust quality tracers and distinctive chemical blueprints predicting aroma features. Samples were obtained from Hojiblanca and Picual olive cultivars harvested in the Cordoba region (Spain) under different farming procedures (organic and conventional) and ripening stages from November 2021 to January 2022. 2. Materials and methods 2.1. Extra virgin olive oil samples Extra virgin olive oils (EVOOs) were obtained from Picual and Hojiblanca olive cultivars harvested in 2021. Olive trees were subjected to two different cultivation methods, organic and conventional. Trees were located at the Agricultural Research Training Centre in “Cabra”in the province of Cordoba (Spain) [plot coordinates 37◦29’36.4"N 4◦25’45.8"W], which has a continental Mediterranean climate with dry summers and mild winters. Rainfall occurs from autumn to spring, with a mean annual rainfall of 400 mm. The average annual temperature is 17 ◦C, reaching −1.3 ◦C in winter and 43 ◦C in summer. Soils have low depths, being mainly over limestone and siliceous stones. They have a loamy texture and an alkaline pH (8–8.5 range). The electrical conductivity is low, as is the organic matter content and the carbon/nitrogen ratio. Twenty olive trees were randomly selected (always among the most loaded to guarantee sampling) of the Picual and Hojiblanca cultivars from conventional cultivation and twenty trees of the two cultivars from organic cultivation. Organic cultivation was conducted according to EU Regulations and certified by Instituto Andaluz de Investigaci´ on y Formaci´ on Agraria, Pesquera, Alimentaria y de la Producci´ on Ecol´ ogica (IFAPA) (Junta de Andalucía) (“Regulation (EU) 2018/848 of the European Parliament and of the Council of 30 May 2018 on organic production and labelling of organic products and repealing Council Regulation (EC) No 834/2007,”2018). Olive harvesting was done by randomly hand-picking healthy fruits (without any type of infection or physical damage), beginning in November 2021 and ending in January 2022, and comprising different harvest periods. It conforms to a total of 3 ripening stages: 15 of November (stage I), 13 of December (stage II), and 1 of January (stage III). The ripening index (RI) of each harvest was determined according to the methodology proposed by Uceda and Frias (Uceda and Frias, 1975). Table 1 EVOO samples characteristics and coding. Sample coding Cultivar Cultivation method Olives harvesting Maturity stage Ripening Index EVOO_1 Picual Conventional November 15 I 3.5 EVOO_2 Picual Conventional December 13 II 4.8 EVOO_3 Picual Conventional January 1 III 5.7 EVOO_4*Picual Conventional December 13 II 4.8 EVOO_5 Picual Organic November 15 I 4.1 EVOO_6 Picual Organic December 13 II 4.8 EVOO_7 Picual Organic January 1 III 5.9 EVOO_9 Hojiblanca Conventional November 15 I 1.4 EVOO_10 Hojiblanca Conventional December 13 II 3.8 EVOO_11 Hojiblanca Conventional January 1 III 4.0 EVOO_12*Hojiblanca Conventional December 13 II 3.8 EVOO_13 Hojiblanca Organic November 15 I 1.2 EVOO_14 Hojiblanca Organic December 13 II 2.6 EVOO_15 Hojiblanca Organic January 1 III 4.7 * Technological replicates from a different batch A. Caratti et al. Journal of Food Composition and Analysis 137 (2025) 106975 2 The fruit RI values for each modality are shown in Table 1. In each harvesting stage, 200 kg of olives were collected without any discriminant selection before immediate processing in the experimental mill of IFAPA Venta del Llano (Mengíbar, Ja´ en, Spain) to obtain the oils. Thus, the extraction of the oils was carried out in a two-phase semi-industrial continuous system (Il Molinetto, Gruppo Pieralisi, Italy) equipped with a hammer mill, a thermo-blender, and a horizontal centrifuge. The 200 kg of olives were ground and then, the resulting paste was beaten for 45 min at 28 ◦C, after which the oil was separated by centrifuging the beaten paste at ≈3250 G. The oils obtained were decanted and filtered. Samples for volatilome screening were immediately sent to the University of Turin laboratory in dark-glass bottles. Before analysis they were kept at −18◦C away from UV exposure. Sample descriptions and details are reported in Table 1. Olive oils were also subjected to the evaluation of the sensory panel of the Priego de C´ ordoba Protected Denomination of Origin (PDO), following the standards of the International Olive Council (IOC) (Ríos-Reina et al., 2021). According to sensory results, all samples considered in this work resulted “Extra-Virgin”Olive Oils. 2.2. Reference standards and solvents Pure reference compounds for identity confirmation of marker volatiles and potent odorants listed in Table 2,n-alkanes (n-C7 to n-C25) for Linear Retention Indices (I T ) calibration, and α /β-thujone used as internal standards for response stability check were obtained from Merck (Milan, Italy). The external standard for predicted relative response factors (RRF) based on combustion enthalpies calibration was hexanal (99 % purity, Merck). The reference solution was prepared in cyclohexane (99 % purity, Merck) at a final concentration of 10.00 g/L and then diluted in dibutyl phthalate (99 % purity, Merck) to suitable concentrations for linearity evaluation. 2.3. Multiple headspace solid phase microextraction: devices and conditions Volatiles from EVO oils were extracted by HS-SPME with a divinylbenzene/carboxen/polydimethyl siloxane (DVB/CAR/PDMS) fiber (d f 50/30 μ m; 2 cm length) from Supelco (Bellefonte, PA, USA) according to a previously optimized procedure (Stilo et al., 2019). The SPME fiber was conditioned before use as recommended by the manufacturer. Sampling was carried out on 0.100 ±0.003 g of oil, precisely weighed in a 20 mL headspace vial, and kept at 40 ◦C for 60 min under constant agitation. The very low amount of sample was chosen to match for HS linearity conditions for most of the characteristic analytes of the EVO oil volatilome (Stilo et al., 2021). After extraction, the SPME device was automatically transferred to the split/splitless injection port of the GC×GC system, kept at 250 ◦C, and thermal desorption was for 5 min. Samples were analyzed in four replicates randomly distributed over two weeks. Multiple headspace extraction by SPME (MHS-SPME) of samples and calibration solutions were conducted by applying the above-indicated conditions, and the number of consecutive extraction steps was set to four, achieving an almost exhaustive extraction of the analytes under study (Stilo et al., 2021). 2.4. Comprehensive two-dimensional gas chromatography: instrument setup and conditions Automated MHS-SPME was performed by a multipurpose sampler, model MPS-2 (Gerstel, Mülheim a/d Ruhr, Germany), installed on a GC×GC system equipped with a reverse-inject differential-flow modulator based on capillary flow technology™(Agilent Technologies, Little Falls, DE, USA). The Agilent 7890B GC unit was coupled to an Agilent 5977B equipped with a high-efficiency source (HES) and fast quadrupole MS analyzer (Agilent Technologies). The MS was operating in electron ionization mode at 70 eV. Ion source and transfer-line temperatures were set at 280 ◦C, and the quadrupole temperature was set at 240 ◦C. The scan range was set between 40 and 250 m/z, achieving a data acquisition frequency of 30 Hz. Parallel detection was by a fast FID with base temperature held at 280 ◦C; H 2 flow 40 mL/min, air flow 350 mL/min, and sampling frequency 200 Hz. The column set was configured as follows: first dimension ( 1 D) HeavyWax™column (100 % polyethylene glycol - PEG; 20 m ×0.18 mm dc ×0.18 μ m d f ) coupled with second dimension ( 2 D) DB17 column (50 % phenyl-methylpolysiloxane; 1.8 m ×0.18 mm d c ×0.18 μ m d f ), both from Agilent Technologies. The connection between the 2 D column and deactivated silica capillaries (Agilent Technologies) toward MS (0.5 m ×0.1 mm d c ) and FID (1.1 m x 0.18 mm d c ) for parallel detection was by a three-way unpurged capillary microfluidic splitter (G3181B, Agilent Technologies). The resulting split ratio was 70:30 FID/MS. The GC split/splitless injector port was set at 250 ◦C and operated in pulsed-split mode (250 kPa overpressure applied to the injection port until 2 min) with a split ratio 1:20. The carrier gas was helium at a nominal flow of 0.4 mL/min along the 1 D column and 10 mL/min along the 2 D column. The oven temperature program was set as: from 40 ◦C (2.29 min) to 240 ◦C (11’) at 3.06 ◦C/min. The modulation period (P M ) was set at 3 s and pulse time at 250 ms. The n-alkanes liquid sample solution for I T determination was analyzed under the following conditions: split/splitless injector in split mode, split ratio 1:50, injector temperature 250 ◦C, and injection volume 1 μ L. 2.5. Analytes identification criteria Analyte identification was by combining retention data (carried out using experimental I T with ±10 units tolerance vs. tabulated ones) and comparing electron ionization (EI)-MS spectral signature with reference spectra in commercial and in-house databases by using the NIST identity search algorithm with direct match factor (DMF) and reverse match factor (RMF) scores threshold ≥900. 2.6. Analytical data acquisition, processing and mining 2.6.1. Combined untargeted and targeted (UT) fingerprinting by smart templates A template is a pattern of 2D peaks and/or graphic objects (features) created over a reference chromatogram or image (single or cumulative). This template is used to identify similar patterns of 2D peaks in a set of analyzed chromatograms or images (Bressanello et al., 2018; Reichenbach et al., 2019). The process involves specific matching functions that establish correspondences between features across multiple chromatograms. Specificity is achieved by defining confidence thresholds for retention times and MS spectral similarity to account for variable peak detection, and by using suitable transform functions to correct for retention time inconsistencies between runs (Squara et al., 2023b; Stilo et al., 2019). Once the correspondences between features are established, the template’s metadata (chemical name, retention times, mass spectra, informative ions, and their relative ratios) are transferred to candidate peaks and/or graphic objects (peak regions) in the analyzed chromatogram. This feature template, which includes untargeted reliable peaks and peak regions, is obtained through a fully automated workflow in GC Image Investigator™(GC Image™, GC Image LLC). The workflow consists of the following steps: (a) Match peak patterns between all chromatogram pairs from a set of representative samples, using optimized parameters: S/N threshold of 50, MS constraint of 700 for both the direct match factor (DMF) and reverse match factor (RMF) based on the NIST similarity algorithm (Squara et al., 2023b; Technology, 2005). (b) Select reliable peaks across the analyzed chromatograms, using a relaxed reliability criterion that includes peaks matching at least 50 % +1 of the chromatograms (Reichenbach et al., 2013). This template of A. Caratti et al. Journal of Food Composition and Analysis 137 (2025) 106975 3 Table 2 Identified compounds in the volatile fraction of EVOOs together with analytical information. See text for identification criteria. Compound Name 1 D t R min SD 2 D t R sec SD Exp I T Lit. I T MW Formulae PRRF β(±SD) Odor quality OT (ng/g)* Analytes subjected to accurate quantification           Propanal 4.83 0.073 0.42 0.019 792 786 58.1 C 3 H 6 O 1.38 0.90 (±0.07) Fresh, fruity, malty 9.4 [1] Ethyl acetate 6.44 0.028 0.37 0.070 897 898 88.1 C 4 H 8 O 2 1.60 0.92 (±0.08) Fruity, sweet, winey 940 [3] 1-Penten−3-one 10.15 0.016 0.68 0.049 1017 1021 84.1 C 5 H 8 O 1.12 0.65 (±0.05) Pungent, spicy 1.6 [1] α -Pinene 10.24 0.087 1.68 0.055 1017 1017 136.2 C 10 H 16 0.79 0.93 (±0.07) Herbal, woody,terpenic 274 [4] Hexanal (Internal Standard for FID) 12.57 0.027 0.91 0.048 1079 1080 100.2 C 6 H 12 O - 0.73 (±0.07) Green, grass 300 [1] β-Pinene 13.34 0.062 1.85 0.067 1097 1100 136.2 C 10 H 16 0.79 0.92 (±0.06) Herbal, pine - (E)−2-Pentenal 14.45 0.011 0.73 0.049 1123 1131 84.1 C 5 H 8 O 1.12 0.75 (±0.02) Pungent, apple-like 300 [3] 2-Pentanol 14.99 0.016 0.44 0.052 1136 1138 88.1 C 5 H 12 O 1.02 0.88 (±0.11) Musty, fermented 380 [2] (Z)−3-Hexenal 15.14 0.020 0.77 0.050 1139 1133 98.1 C 6 H 10 O 1.04 0.56 (±0.03) Green, grassy 1.7 [1] δ−3-Carene 15.16 0.040 1.77 0.085 1139 1144 136.2 C 10 H 16 0.79 0.9 (±0.04) Citrus, pine, herbal 770 [4] 1-Penten−3-ol 15.75 0.006 0.42 0.052 1153 1157 86.1 C 5 H 10 O 1.07 0.57 (±0.02) Pungent, butter 400 [3] Heptanal 16.99 0.024 1.02 0.062 1181 1181 114.2 C 7 H 14 O 0.96 0.93 (±0.05) Citrus-like, fatty 500 [1] Limonene 17.48 0.027 1.67 0.050 1192 1195 136.2 C 10 H 16 0.79 0.92 (±0.06) Citrus, terpenic 250 [2] 3-Methyl−1-butanol 17.80 0.006 0.45 0.050 1199 1211 88.1 C 5 H 12 O 1.02 0.63 (±0.03) Malty, ethereal 100 [3] Eucalyptol 17.88 0.036 1.86 0.089 1201 1205 154.2 C 10 H 18 O 0.90 0.94 (±0.03) Herbal, minty 15 [2] (E)−2-Hexenal 18.40 0.006 0.79 0.051 1212 1216 98.1 C 6 H 10 O 1.04 0.6 (±0.02) Bitter almond, green, fruity 320 [1] Hexyl acetate 21.05 0.014 1.15 0.050 1271 1275 144.2 C 8 H 16 O 2 1.09 0.7 (±0.08) Fruity, pear-like 1040 [3] Octanal 21.70 0.006 1.10 0.053 1285 1289 128.2 C 8 H 16 O 0.92 0.94 (±0.04) Citrus-like, fatty 140 [1] (Z)−2-Penten−1-ol 22.50 0.006 0.42 0.052 1303 1306 86.1 C 5 H 10 O 1.07 0.57 (±0.02) Green, almond 250 [2] (E)−2-Penten−1-ol 22.84 0.022 0.40 0.047 1310 1313 86.1 C 5 H 10 O 1.07 0.68 (±0.03) Mushroom, earthy 250 [2] (Z)−3-Hexenyl acetate 23.01 0.025 1.01 0.047 1315 1317 142.2 C 8 H 14 O 2 1.12 0.55 (±0.04) Green 200 [1] (E)−2-Heptenal 23.17 0.025 0.86 0.031 1318 1320 112.2 C 7 H 12 O 0.99 0.81 (±0.05) Green, fatty 1200 [1] 6-Methyl−5-hepten−2-one 23.80 0.006 0.95 0.054 1332 1339 126.2 C 8 H 14 O 0.95 0.69 (±0.03) Pungent, green, fruity-like 1000 [3] 1-Hexanol 24.30 0.044 0.48 0.054 1343 1348 102.2 C 6 H 14 O 0.97 0.66 (±0.02) Fruity, banana 400 [3] (Z)−3-Hexen−1-ol 25.69 0.022 0.46 0.048 1374 1380 100.2 C 6 H 12 O 1.00 0.6 (±0.01) Banana, fresh, grass 1100 [5] (E,E)−2,4-Hexadienal 26.30 0.006 0.63 0.051 1388 1397 96.1 C 6 H 8 O 1.08 0.68 (±0.12) Green, soapy 270 [2] Nonanal 26.35 0.006 1.13 0.054 1389 1390 142.2 C 9 H 18 O 0.90 0.95 (±0.02) Fatty, waxy, pungent 610 [1] (E)−2-Hexen−1-ol 26.55 0.011 0.62 0.050 1394 1398 100.2 C 6 H 12 O 1.00 0.7 (±0.07) Green, grass 500 [1] (E)−2-Octenal 27.70 0.009 0.40 0.054 1420 1427 126.2 C 8 H 14 O 0.95 0.92 (±0.04) Fatty, nutty 120 [1] Acetic acid 28.51 0.023 0.24 0.051 1440 1440 60.1 C 2 H 4 O 2 3.62 0.73 (±0.09) Sour, vinegary 350 [1] (E,E)−2,4-Heptadienal 29.20 0.006 0.70 0.055 1456 1461 110.2 C 7 H 10 O 1.02 0.86 (±0.05) Fatty, green, oily 710 [1] α -Copaene 30.55 0.016 2.37 0.052 1489 1491 204.4 C 15 H 24 0.78 0.94 (±0.06) Woody - Benzaldeyde 31.59 0.022 0.66 0.051 1513 1524 106.1 C 7 H 6 O 0.92 0.91 (±0.04) Almond, burnt sugar, erthy 60 [4] Popanoic acid 32.15 0.013 0.26 0.055 1526 1534 74.1 C 3 H 6 O 2 2.06 0.81 (±0.05) Acidic, pungent 720 [4] (E)−2-Nonenal 32.26 0.026 0.91 0.058 1530 1534 140.2 C 9 H 16 O 0.92 0.95 (±0.02) Fatty, green, soapy 140 [1] 1-Octanol 33.01 0.022 0.55 0.055 1548 1553 130.2 C 8 H 18 O 0.90 0.86 (±0.03) Nut, mushroom 27 [2] Butanoic acid 35.75 0.006 0.25 0.053 1616 1613 88.1 C 4 H 8 O 2 1.60 0.85 (±0.05) Cheesy, sour 34 [1] (E,E)- α -Farnesene 40.60 0.015 1.56 0.051 1742 1744 204.4 C 15 H 24 0.78 0.78 (±0.03) Sweet, floral - Analytes not quantified           Hexane 3.58 0.135 0.45 0.049 600 600 86.2 C 6 H 14   Alkane  Octane 4.87 0.066 0.99 0.063 800 800 114.2 C 8 H 18   Solvent 940 [5] Acetone 5.05 0.017 0.45 0.049 850 845 58.1 C 3 H 6 O  Pungent - 2-Methyl butanal 7.08 0.140 0.64 0.073 924 926 86.1 C 5 H 10 O  Malty 10 [5] Ethanol 7.35 0.015 0.37 0.051 928 933 46.1 C 2 H 6 O  Ethanol-like - 3,4-Diethyl−1,5-hexadiene (RS/SR) 7.90 0.018 1.21 0.049 946 952 138.2 C 10 H 18   - - 3-Methylbutanal 8.00 0.133 0.72 0.066 939 930 86.1 C 5 H 10 O  Malty 13 [5] 3,4-Diethyl−1,5-hexadiene (meso) 8.45 0.145 1.40 0.083 956 956 138.2 C 10 H 18   - - Pentanal 8.76 0.027 0.75 0.050 975 974 86.1 C 5 H 10 O  Almond-like, pungent, malt 150 [4] (Z)−1-Methoxy−3-hexene 9.74 0.025 1.17 0.050 1007 997 114.2 C 7 H 14 O  - - (5Z)−3-Ethyl−1,5-octadiene 9.93 0.026 1.69 0.050 1012 1006 138.2 C 10 H 18   - - (5E)−3-Ethyl−1,5-octadiene 10.47 0.027 1.69 0.049 1026 1032 138.2 C 10 H 18   - - Toluene 10.73 0.050 0.91 0.057 1029 1036 92.1 C 7 H 8   Chemical-like - (E,Z)−3,7-Decadiene 12.52 0.028 1.92 0.051 1078 1068 138.2 C 10 H 18   - - (E,E)−3,7-Decadiene 12.85 0.016 1.89 0.049 1085 1082 138.2 C 10 H 18   - - (continued on next page) A. Caratti et al. Journal of Food Composition and Analysis 137 (2025) 106975 4 Table 2 (continued) Compound Name 1 D t R min SD 2 D t R sec SD Exp I T Lit. I T MW Formulae PRRF β(±SD) Odor quality OT (ng/g)* (Z)−2-Pentenal 13.49 0.017 0.74 0.049 1101 1105 84.1 C 5 H 8 O  Pungent, apple-like 300 [3] 3-Methylbutyl acetate 14.42 0.053 0.97 0.060 1123 1126 130.2 C 7 H 14 O 2   Banana-like, fruity  2-Heptanone 16.85 0.027 0.98 0.086 1178 1171 130.2 C 7 H 14 O  Sweet, fruity 470 [4] (Z)−2-Hexenal 17.66 0.022 0.82 0.048 1197 1193 98.1 C 6 H 10 O  Fruity - γ-Terpinene 19.59 0.024 1.71 0.058 1239 1247 98.1 C 6 H 10 O  Petrol-like - trans-β-Ocimene 19.95 0.006 1.44 0.050 1246 1249 136.2 C 10 H 16   Citrus-like, soapy, geranium-like - o-Cymene 20.72 0.027 1.42 0.050 1264 1268 134.2 C 10 H 14   - - 3-Hydroxy−2-butanone 21.45 0.021 0.46 0.059 1279 1283 88.1 C 4 H 8 O 2   Buttery - 2-Octanone 21.50 0.014 1.07 0.052 1280 1284 128.2 C 8 H 16 O  Mould, green 510 [4] (E)−3-Hexen−1-ol acetate 22.49 0.020 1.57 0.050 1303 1300 142.2 C 8 H 14 O 2   Fruity - Dioxa−1,6-spiro[4.5]decane 23.54 0.032 1.44 0.077 1326 - 142.2 C 8 H 14 O 2   - - 3-Methyl-cyclopentanol 23.68 0.034 0.65 0.064 1329 1342 100.1 C 6 H 12 O  - - (E)−3-Hexen−1-ol 24.78 0.031 0.47 0.044 1354 1352 100.2 C 6 H 12 O  green, grassy - Methoxymethyl-benzene 26.05 0.006 0.93 0.053 1382 1382 122.2 C 8 H 10 O  - - (Z)−2-Hexen−1-ol 26.62 0.035 0.43 0.029 1396 1405 101.2 C 6 H 12 O  Green, grass - Dimethyl Sulfoxide 32.60 0.015 0.49 0.051 1537 1549 78.1 C 2 H 6 OS   - - 3-Methyl−2-hexen−4-one 34.60 0.000 0.47 0.054 1587 1586 112.2 C 7 H 12 O  - - Benzoic acid, methyl ester 35.66 0.028 0.75 0.060 1613 1614 136.1 C 8 H 8 O  Starfruit-like, sweet - 1,4-Cyclohex−2-enedion 39.93 0.025 0.53 0.047 1725 - 110.1 C 6 H 6 O 2   - - Pentanoic acid 39.93 0.024 0.25 0.050 1725 1730 102.1 C 5 H 10 O 2   Sweaty, fruity 400 [1] 2,3-Dimethylbenzaldehyde 39.99 0.052 0.66 0.049 1725 1736 134.2 C 9 H 10 O  - - 5-ethyl−2(5 H)-Furanone 40.73 0.024 0.51 0.045 1746 1757 112.1 C 6 H 8 O 2   Sweet, spicy - Methyl salicylate 41.53 0.029 0.72 0.046 1768 1765 152.1 C 8 H 8 O 3   Fatty, tallowy, terpene-like - 2-(2-butoxyethoxy)Ethanol 42.09 0.024 0.55 0.054 1783 1786 162.2 C 8 H 18 O 3   - - Hexanoic acid 43.84 0.022 0.26 0.049 1831 1839 116.2 C 6 H 12 O 2   Goat-like, sweaty 460 [1] Benzyl alcohol 44.78 0.025 0.38 0.046 1858 1857 108.1 C 7 H 8 O  Sweet, fruity - Phenylethyl alcohol 46.02 0.029 0.44 0.053 1893 1904 122.2 C 8 H 10 O  Floral, honey-like - (E)−2-Hexenoic acid 47.89 0.022 0.30 0.047 1947 1941 114.1 C 6 H 10 O 2   - - Phenol 49.10 0.006 0.26 0.054 1983 1987 94.1 C 6 H 6 O  Ink-like, phenolic - Benzoic acid, 2-methoxy-, methyl ester 51.31 0.045 0.66 0.060 2049 2032 166.2 C 9 H 10 O 3   - - *Odor threshold references: [1] (Neugebauer et al., 2020); [2] (Van Gemert, 2003); [3] (Luna et al., 2006); [4] (Squara et al., 2023a); [5] (Purcaro et al., 2014) Retention times ( 1 t R min, 2 t R sec), standard deviation (SD), experimental and tabulated linear retention indexes (I T ), molecular weight (MW), predicted FID relative retention factor (RRF), decay function (β), odor threshold (OT) in ng/g A. Caratti et al. Journal of Food Composition and Analysis 137 (2025) 106975 5 reliable peaks is used for re-aligning chromatograms in the temporal domain. (c) Align chromatograms of representative samples to the average retention times of the reliable peaks, then sum/fuse them into a composite chromatogram. (d) Generate a comprehensive untargeted feature template from the composite chromatogram, including reliable peaks and peak regions defined by the footprint of all detected peaks (Reichenbach et al., 2013). This process was used to generate a composite image from a selection of sample chromatograms representative of the different expressions of functional variables. From the composite image, a feature template consisting of 190 UT features (i.e., peak regions corresponding to untargeted and targeted components) was defined and applied to single chromatogram images (i.e, samples and replicates) for reliable tracking and alignment of features. The list of targeted features is reported in Table 2 together with chemical names, 1 D and 2 D retention times ( 1 t R ; 2 t R ), experimental I T , tabulated I T , odor quality, and odor thresholds (OTs) in oil as from reference literature. 2.6.2. Data acquisition and statistical analysis Raw chromatographic data were acquired by MassHunter Workstation (Agilent Technologies). Raw data were processed by GC Image™ V2020 r1.2 suite (GC Image, LLC Lincoln, NE, USA). Statistical analysis and chemometrics were performed using GC Investigator™(GC Image), XLSTAT statistical and data analysis solution (Addinsoft 2020, New York, USA), and Microsoft Office Excel 2016 (Microsoft, Redmond, WA USA). Fig. 1. PCA on untargeted and targeted (UT) peak features (analytes) % responses; ellipses set at 95 % of confidence level, show cultivar natural clusters. In Fig. 1A all 190 UT features were computed for all samples; dotted lines indicate Maturity Stage evolution according to sample loadings. Fig. 1Bresults from Fisher ratio features reduction considering cultivar discrimination (Picual vs. Hojiblanca) the PCA refers to 92 UT peaks with a F >10. Fig. 1CFisher ratio features reduction considering cultivation type on Picual oils (organic vs. conventional) the PCA refers to 49 UT peaks with a F >10. Fig. 1DFisher ratio features reduction considering cultivation type on Hojiblanca oils (organic vs. conventional) the PCA refers to 49 UT peaks with a F >10. A. Caratti et al. Journal of Food Composition and Analysis 137 (2025) 106975 6 3. Results and discussion EVO oil volatilome encrypts information on many key-quality variables (e.g., olives cultivar and geographical origin, cultivation methodologies, olives harvesting stage, oil auto-oxidation level, and sensory profile (Luna et al., 2006; Melucci et al., 2016; Morales et al., 2005), by GC×GC it was possible to consistently monitor more than 190 reliable peak features (including targeted and untargeted compounds (Magagna et al., 2016)), of which 84 were identified according to 1 D retention index (I T ±10 units of tolerance) combined to EI-MS spectral similarity (Direct Match Factor threshold >900) with commercial and in-house databases. Table 2 lists putatively identified compounds (i.e., targeted features) and potent odorants together with retention data, odor qualities, odor thresholds (OTs) in oil, and FID predicted RRFs calculated according to hexanal as reference internal standard. Table S1, provided as supplementary material, lists targeted and untargeted features comprehensively mapping the detectable volatilome of analyzed samples. As a first approach to understanding the existence of natural samples’clusters according to volatile patterns distribution, unsupervised exploration by Principal Component Analysis (PCA) was conducted. Fig. 1Aillustrates PCA results obtained by examining the data matrix including untargeted and targeted (UT) peak features (analytes) % responses for all Picual oils (blue indicators) and Hojiblanca oils (green indicators) (190 ×32 –features ×samples replicates). It represents a total explained variance of 41.62 % (PC1 and PC2) and, along the PC1 with 31.32 % of the total explained variance, the two cultivars form natural clusters as shown by ellipses, set at 95 % confidence. Besides some outliers, the cultivars’chemical signature is almost clear confirming that the volatile fraction in toto encrypts information on olives’ chemotype useful for their discrimination. This result agrees with other authors who observed that botanical traits have the most influence on the volatile profile of EVOOs (Ríos-Reina et al., 2021). Nevertheless, a certain degree of overlap between the two clusters suggests that for some samples the volatilome has similar composition, possibly due to the influence of the harvest region with its common pedoclimatic and soil characteristics. However, the simultaneous presence of many variables influencing the volatiles’qualitative and quantitative profiles, such as the ripening stage of the fruits, might have a confounding effect. According to previous investigations, the ripening stage of the fruit was, after the variety/cultivar, the most influential factor on the volatile fraction (Magagna et al., 2016; Ríos-Reina et al., 2021; Stilo et al., 2021b). A deeper observation of the samples’distribution on the Cartesian plane (Fig. 1A), indicates a clear trend along PC1 with oils obtained by olives harvested at Stage I characterized by lower loadings and those obtained by olives at Stage III with higher values. This is in accordance to the increasing ripening index as reported in Table 1. Analytes responsible for this distribution are, among the others: (E,E)- α -farnesene, (5Z)-3-ethyl-1,5-octadiene, octane, (E,Z)-3,7-decadiene, 3,4-diethyl-1,5-hexadiene (meso), 3,4-diethyl-1,5-hexadiene (RS+SR), (Z)-3-hexenal, (Z)-2-hexenal, phenylethyl alcohol, 6-methyl-5-hepten-2-one, (Z)-3-hexen-1-ol, (E,E)-3,7-decadiene, 1-penten-3-one, 1-penten-3-ol, and nonanal. Results are in keeping with previous studies on ripening indices (Angerosa et al., 1998; Magagna et al., 2016; Stilo et al., 2021b). To identify cultivar discriminating variables, the UT features % response data were filtered by the Fisher ratio (F) value criterion (Sch¨ oneich et al., 2022). Since F crit (1,15) with α =0.05 is 4.54, UT variables with a F calc >10 (arbitrarily set above the critical value) were retained (n=92 UT features), and by applying this features selection criterion, the resulting PCA shown in Fig. 1Bclearly clusters samples according to the cultivar type. The total explained variance raises 60.74 % (PC1 and PC2) and samples are now independently clustered and discriminated along the PC1 (49.65 % of the total explained variance). F values referred to UT features discriminating cultivars are reported in Table S1. The impact of cultivation methodology (i.e., conventional vs. organic) was then explored. The PCA was conducted on the data matrix including UT peak features (analytes) % responses with an F value >10 for organic vs. conventional classes. F values discriminating cultivation methods are listed in Table S1. PCA loadings plot based on the 35 features with F>10 is shown in Fig. 2. Results confirm the major role played by cultivar traits on the volatilome expression, with a total explained variance of 61.58 % the first two PCs do not show any independent clustering for organic and conventional cultivated olive tree samples. However, for Picual oils, a distinction of the two sub-groups along the PC1 appears (pink and light green indicators). Due to the concurrent effect of external variables, as cultivar and cultivation practices, the identification of marker volatiles for organic cultivation was conducted on samples belonging to the same cultivar with the aid of supervised strategies. 3.1. Picual olive oils: organic vs. conventional cultivation diagnostic volatiles signature Considering the Picual cultivar, the unsupervised statistics on the two cultivation methodologies (F calc >5 for organic vs. conventional classes) show two independent groups along PC1, with a total explained variance of 58.93 % (e.g., Fig. 1C). At the same time, heatmap visualization (e.g., Fig. S1A) captures diagnostic patterns of UT volatiles and by hierarchical clustering (HC) based on Pearson correlation of variables form two independent clusters corresponding to cultivation methodologies. Sub-clusters are coherent with biological and technical replicates. To select statistically relevant yet informative compounds related to cultivation methodologies on Picual oils, supervised analysis by partial least squares-discriminant analysis (PLS-DA) was adopted (Lee et al., 2018). The model was developed on an estimation set consisting of 70 % randomly selected measures/samples (20 samples over 28) and validated on the residual 30 % (8 samples over 28). Model performances, after 10 reiterations, were good referring of a 98 % of correctness. The variable importance in the projection scores (VIPs) were used to identify meaningful variables, resulting from the classification model on Picual EVO oils conventional vs. organic, they are visualized as histogram in Fig. 3 (orange bars). Table sS2 lists VIPs values together with relative standard deviation. Discriminant analytes between the cultivation methodologies belongs to the class of potent odorants with 2-pentanol, 1-hexanol, (Z)-3-hexen-1-ol likely discriminating the Hojiblanca cultivar from the Picual which has a characteristic pattern of odorants with (E)-2-penten-1-ol, heptanal, and (E)-2-pentenal. Variables showing Fig. 2. PCA on untargeted and targeted (UT) peak features (analytes) % responses after Fisher ratio reduction according to classes organic vs. conventional and F>10 (35 UT features). Red indicators are for Hojiblanca conventional, pink is for Hojiblanca organic, dark green is for Picual conventional, and light green is for Picual organic. A. Caratti et al. Journal of Food Composition and Analysis 137 (2025) 106975 7 a role in the discrimination are also validated by HC results. Of interest are carbonyls [2-heptanone, 2-octanone, heptanal, (E)-2-pentenal] and alkenes [(5E)-3-ethyl-1,5-octadiene, (E,E)-3,7-decadiene, 3,4-diethyl-1, 5-hexadiene (RS+SR), (5Z)-3-ethyl-1,5-octadiene, (E,Z)-3,7-decadiene]; this last class of chemicals known for their correlation with olives ripening (Angerosa et al., 1998). The relative distribution of the most informative variables is illustrated by box-plots in Fig. 4. Analytes with higher % response in conventional cultivation are methoxymethyl-benzene [ α -methylbenzyl ether already documented(da Silva et al., 2012)], (5E)-3-ethyl-1, 5-octadiene, (5Z)-3-ethyl-1,5-octadiene, 3,4-diethyl-1,5-hexadiene (RS+SR), (E,E)-3,7-decadiene, (E,Z)-3,7-decadiene, (Z)-3-hexenyl acetate, (E)-3-hexen-1-ol acetate, and α -copaene. Compounds with an opposite trend are 2-heptanone, 2-octanone, and butanoic acid. According to the literature, (5E)-3-ethyl-1,5-octadiene, (5Z)- 3-ethyl1,5-octadiene, 3,4-diethyl-1,5-hexadiene (RS or SR), (E,E)-3,7-decadiene and (E,Z)-3,7-decadiene have been found in Picual EVOOs and were considered as markers of early ripening stages, independently of the variety/cultivar and geographical origin (Stilo et al., 2021b). In addition, 3-ethyl-1,5-octadiene, a compound deriving from the lipoxygenase pathway, recorded significant variations according to the cultivar and/or environmental conditions on olive tree growing (Kosma et al., 2020). (Z)-3-Hexenyl acetate is a compound responsible for sensory attributes as green leaves, and it was shown to be relevant in the differentiation between non-defective (EVO) and defective (non-EVO) olive oil samples, being present in higher concentration in EVOO samples (Ríos-Reina et al., 2021). Non-defective (EVO) olive oil is of the highest quality and free of sensory defects, while defective (non-EVO) olive oil does not meet these strict standards and is of lower quality. In addition, aligned with the current evidence, (Z)-3-hexenyl acetate was differentially distributed in organic and conventional samples considered in a previous study (Jurado-Campos et al., 2021). α -Copaene is a mono-unsaturated sesquiterpene that has already been detected in Spanish oils obtained from olives of the Hojiblanca cultivar and Picual (Bortolomeazzi et al., 2001; Guinda et al., 1996), and together with α -muurolene and α -farnesene were the terpenes that aided the discrimination between extra virgin olive oil according to the cultivar and geographical origin (Bubola et al., 2014; Luki´ c et al., 2018; Zunin et al., 2005) and they have been suggested as markers of olive oil differentiation (Kosma et al., 2020). According to the literature, butanoic acid has been associated with vinegary,musty, and rancid defects (Angerosa, 2002; Cecchi et al., 2019). 3.2. Hojiblanca olive oils: organic vs. conventional cultivation diagnostic volatiles signature A first exploratory approach was conducted on Hojiblanca samples considering the cultivation practice. Fig. 1Dshows the PCA resulting from UT features filtered by Fisher ratio value (F calc >10). Samples, as expected, are clearly clustered according to cultivation practice and the total explained variance for PC1-PC3 achieves 61.72 %. At the same time, heatmap visualization (provided as supplementary material Fig. S1) captures diagnostic patterns of UT volatiles and by hierarchical clustering (HC) based on Pearson correlation of variables form two independent clusters corresponding to cultivation methodologies. Subclusters are coherent with biological and technical replicates. Chemical variables highlighted by PLS-DA as distinctive for the cultivation Fig. 3. : Histogram showing the VIPs values derived by the PLS-DA classification models conventional vs. organic classes. Blue bars are for Hojiblanca and orange bars for Picual. The Venn diagram visualizes the number of targeted analytes (features) in common (n=6) or unique (n=10 for Picual and n=6 for Hojiblanca) for the two cultivars. A. Caratti et al. Journal of Food Composition and Analysis 137 (2025) 106975 8 Fig. 4. Box plots of discriminant compounds (from PLS-DA modeling) between organic and conventional cultivation on Picual cultivar samples. Green boxes represent data from organic farming (mean/median) while pink boxes are for conventional farming (mean/median). A. Caratti et al. Journal of Food Composition and Analysis 137 (2025) 106975 9