*2.1. Sample Characterization*

Seventeen commercial wines were used, including white and red wines from di fferent vintages and worldwide winemaking regions. For the selection of the wines, a specialized press investigation was carried out in order to choose wines that would have been described as "mineral". For this, reviews sourced from specialized magazines (The Wine Advocate, Wines and Spirits, Wines & Vines, etc.) were used. Similarly, preference was given to those wine regions identified as "mineral" regions by specialized magazines from the sector (http://www.thewinesociety.com/society-news-and-viewsregular-features-opinion-mineral-wines).

Eleven white wines and six red wines were chosen. The description of the wines used in the study is detailed in Table 1.


**Table 1.** Identification of the samples of white and red wine used in the study.

 MLF: malolactic fermentation; \*\* OBA: Oak Barrel Aging.

## *2.2. Sensory Analysis*

\*

In order to carry out the sensory part of the study, two tasting panels were trained in accordance with the Asociación Española de Estandarización (UNE) 87024-2. One of them, located in Barcelona, formed by 10 professional judges from the wine export sector not specialized in winemaking, and a second panel located in La Rioja formed by 12 students of oenology. The panels were recruited in two ways: the panel from the University of La Rioja was trained for one year by following the program of the sensory analysis course that is part of the University's degree in Enology. The panel of winemakers from La Rioja was made up of sensory judges aged between 22 and 36 (61% women and 39% men). In the case of the panel of exporters in Barcelona, this was composed of sensory judges aged between 35 and 48 (72% men and 28% women).

The judges from the University of La Rioja were trained with chemical standards and real samples for one year. The second panel, from Barcelona, was made up of students who passed Level 2 of the Wine and Spirits Trust beverages program that the company Outlook Wine S.L. teaches in Barcelona.

Sensory descriptive analysis was performed according to the International Organization for Standardization (ISO) 11035. The wines were served in certified glasses according to ISO 3591-1977, with 50 mL in each glass, covered with a Petri dish so that the aromas reached equilibrium in the head space. The judges were first asked to evaluate the aromas orthonasally and record their intensity for each of the descriptors set out in the tasting sheet, on a scale of 0 to 5 for positive attributes and 0 to −5 for negative attributes or defects The judges were then asked to conduct the evaluation at the gustatory level with the same scale.

## *2.3. Statistical Analysis*

A set of 17 wines were tasted and, subsequently, a partial least squares regression analysis was conducted with data obtained by using XLSTAT 2017 Addinsoft statistical software (40, rue Damrémont 75018 Paris, France).

Partial least squares regression statistical methodology is an analysis that is related to the regression of major components: Instead of finding hyperplanes of maximum variance between the response variable and independent variables, there is a linear regression through the projection of forecast variables and the observable variables to a new space.

In a first step, and in order to reduce the variables that would be part of the predictive mathematical algorithms, various selection criteria were taken into account [4,5].

High level of positive or negative correlation between the chemical compound and the "minerality" attribute and likewise for organoleptic attributes [6].

Percentage of explanatory variance by the model and Q<sup>2</sup> value accumulated as high and as close to 1 as possible.

Subsequently, a dynamic search process for the best PLS model was performed. For this, those compounds and chemical families with a relationship greater than 2 in their concentration between the maximum and minimum value and with activity aroma values (OAV) that met the condition OAVmax/OAVmin >1 were taken into consideration, since these are compounds that can contribute with significant di fferences in sensory perception against the scores of the "minerality" attribute by the panel of tasters. To reduce the number of variables of the models, a partial least squares (PLS) regression analysis was performed, retaining those chemical compounds that showed a significant correlation with a confidence level of 90% with the olfactory and gustatory "minerality" attribute.

## *2.4. Chemical Characterization*

## 2.4.1. Volatile Composition

The minor compounds were analyzed by solid phase extraction (SPE) followed by analysis via Gas Chromatography and Mass Spectrometry (GC–MS) and major compounds by liquid–liquid extraction (LL) followed by Gas Chromatography (GC) with flame ionization detection (FID) [7]. To perform the analysis, 5 mL of wine, 0.8 mL of acetonitrile, 0.1 mL of chloroform and an internal standard containing the deuterated compounds 3-octanol, 4-methyl-pentanol and 4-hydroxy-4-methyl-2-pentanol were added to a centrifuge tube with a 15 mL thread. The tube was centrifuged at 4800 rpm for 5 min. Once the phases were separated, the organic phase was recovered and injected into an Agilent gas chromatograph model 7890 (G3440A) under the following conditions: an initial temperature of 40 ◦C, which was maintained for 5 min and then increased by 3 ◦C/min to 200 ◦C. The gas used was Helium injected at a flow of 3 mL/min. Three milliliters were injected with a divided mode flow of 30 mL/min. The column (50 m × 0.32 mm) and film thickness of 0.5 μm used was DB-FFAP (nitroterephthalic-acid-modified polyethylene glycol (PEG) column of high polarity) from J & W Scientific (Folsom, CA, USA) and the identification and quantification was carried out by means of a mass detector. To perform the quantification of the majority and minority compounds, calibration graphs for 70 compounds were previously prepared by analyzing synthetic samples containing known quantities of the odorants analyzed.
