Wine Lees as Source of Antioxidant Molecules: Green Extraction Procedure and Biological Activity

An ultrasound-assisted extraction method, employing ethanol and water as solvents at low temperature (30 °C) and reduced time (15 min), was proposed to extract bioactive molecules from different cultivars (Magliocco Canino, Magliocco Rosato, Gaglioppo, and Nocera Rosso) of wine lees. All the extract yields were evaluated and their contents of phenolic acids, flavonoids, and total polyphenols were determined by means of colorimetric assays and high-performance liquid chromatography coupled with diode-array detection (HPLC-DAD) and Fourier transform infrared (FTIR) techniques. Radical scavenging assays were performed and the Magliocco Canino extracted with a hydroalcoholic mixture returned the best results both against ABTS (0.451 mg mL−1) and DPPH (0.395 mg mL−1) radicals. The chemometric algorithms principal component analysis (PCA) and partial least square regression (PLS) were used to process the data obtained from all qualitative–quantitative sample determinations with the aim of highlighting data patterns and finding possible correlations between composition and antioxidant features of the different wine lees cultivars and the extraction procedures. Wine lees from Magliocco Canino and Magliocco Rosato were found to be the best vegetable matrices in terms of metabolite content and antioxidant properties. The components extracted with alcoholic or hydroalcoholic solvents, specifically (−)-epigallocatechin gallate, chlorogenic acid, and trans-caftaric acid, were found to be correlated with the antioxidant capacity of the extracts. Multivariate data processing was able to identify the compounds related to the antioxidant features. Two PLS models were optimized by using their concentration levels to predict the IC50 values of the extracts in terms of DPPH and ABTS with high values of correlation coefficient R2, 0.932 and 0.824, respectively, and a prediction error lower than 0.07. Finally, cellular (SH-SY5Y cells) antioxidant assays were performed on the best extract (the hydroalcoholic extract of Magliocco Canino cv) to confirm its biological performance against radical species. All these recorded data strongly outline the aptness of valorizing wine lees as a valuable source of antioxidants.


Introduction
Wine is produced in all parts of the world, but France, Italy, Spain, and some emerging countries such as Australia, Chile, South Africa, and the United States represent the main world producers [1]. The market data highlight the extreme complexity of the wine sector, often based on thousands of small farms that produce small quantities of wine, together with other large companies with high wine production [2]. Italy, together with France, ranks at the top of the classification of wine-producing countries. Italy is still the leading wineproducing country with 44,500 hectoliters, while European Union countries reached 145,000 hectoliters in 2021, with a substantial decrease of 13% (−21,000 hectoliters) compared to 2020, almost achieving 60% of world wine production [3].

Sample Collection and Preparation
A total of four lees samples of autochthonous Calabria Vitis vinifera red grape varieties (Nocera Rosso (1), Magliocco Rosato (2), Magliocco Canino (3), and Gaglioppo (4)) were collected in September 2021 from a local producer (Azienda Agricola Donna Fidelia, Belvedere Marittimo, Italy). The grapes were harvested at technological maturation, and the agronomic practices as well as the winemaking process were the same for all analyzed samples. The removal of the ethanol from the lees samples was accomplished by rotary evaporation (ALC Multispeed Centrifuge Thermo Electron Corporation, Waltham, MA, USA ) for 15 min at 35 • C and the concentrated solution was frozen at −18 • C and then freeze-dried at a pressure of 0.45 mbar (Micro Modulyo, Edwards). Freeze-dried lees samples were crushed and kept in hermetically sealed containers at −18 • C until further analysis.

Wine Lees Extraction
Lees samples were extracted as described by Carullo et al. [13]. Two hundred mL of ethanol (E), HCl solution at pH = 2.0 (W), or hydroalcoholic solution (50% v/v) acidified at pH 2.0 with HCl (WE) were added to 1.0 g of a freeze-dried sample of each cultivar, and the samples were extracted using an ultrasound-bath Branson model 3800-CPXH (Milan, Italy) at 30 • C (10 cycles/sec) for 15 min at an ultrasonic frequency of 40 kHz. The supernatants were decanted and the solid was removed, while the solution was evaporated, frozen, and freeze-dried to provide a vaporized solid. Wine lees and sample preparation were performed in triplicate.
The tentative identification of the compounds in the extracts was achieved by matching the positions of absorption maxima (λ max ), absorption spectra profiles, and retention times (RT) with those from commercially available pure standards and studies already reported in the literature for comparable matrices [14,15]. Moreover, the quantification was performed using the calibration curves in the concentration range of 100-0.2 µg mL −1 of malvidin-3Oglucoside (R 2 = 0.9983), caffeic acid (R 2 = 0.9988), epicatechin (R 2 = 0.9944), and gallic acid (R 2 = 0.9991).

Fourier Transform Infrared Analysis of Lyophilized Wine Lees
The infrared fingerprints of the lyophilized wine lees samples were acquired by using the Spectrum Two infrared spectrometer (Perkin Elmer Italia Spa, Milan, Italy) supplied with an attenuated total reflection (ATR) accessory with a ZnSe crystal on the sampling flat plate. The ATR detector was cleaned with ethanol after each spectral collection and Antioxidants 2023, 12, 622 4 of 18 then dried. The infrared spectrum of the laboratory air was considered as background to check the instrumental status and H 2 O/CO 2 interferences. The FTIR spectra of the samples (3 samples for each extraction procedure), placed on the ATR detector, were recorded in triplicate in the range 4000-450 cm −1 , with 32 scans and a resolution of 4 cm −1 . The fingerprint analysis of samples was focused on a specific wavenumber range, the variables selection was carried out by taking out the spectral window 4000-1850 cm −1 [16,17]. The interval 4000-2500 cm −1 contained mainly signals of the stretching vibrations due to -OH functions and aromatic bonds (C-H), and information about phenolic compounds was relatively poor. The 2500-1850 cm −1 interval also contained no relevant spectral information besides the band corresponding to CO 2 . The remaining 1800-450 cm −1 interval was set in matrix form and subjected to chemometric modeling after opportune pre-treatment and signal transformation from reflectance to absorbance data.

Polyphenols Total Content
The total phenol content (TPC) of wine lees extracts was determined as described by Spizzirri et al. [18]. In a volumetric flask, 6.0 mL of an aqueous solution of each lees extract, 1.0 mL of Folin-Ciocalteu reagent, and 2.0 mL of Na 2 CO 3 solution (2% w/v) were added. After two hours at room temperature, the absorbance was spectrophotometrically measured at 720 nm by a Jasco V-530 UV/Vis spectrometer (Jasco, Tokyo, Japan) against a control. This analysis was performed in triplicate and the TPC value was determined by a standard curve constructed using gallic acid (GA) in the range of 8-40 µM (R 2 = 0.9988). The TPC for each extract was expressed as mg GA equivalents per gram of dry matrix (mg GA/g).

Phenolic Acid Content
The phenolic acid content (PAC) of each dry matrix sample was evaluated by a literature procedure with some changes [19]. In a volumetric flask (10.0 mL) 1.0 mL of an aqueous solution of each extract, 1.0 mL of HCl (0.5 mol L −1 ), 1.0 mL of Arnov's reagent (Na 2 MoO 4 and NaNO 3 0.1 mg mL −1 ), 1.0 mL of NaOH (1.0 mol L −1 ), and distilled water were added. The absorbance of the solutions was spectrophotometrically determined at 490 nm. This analysis was performed in triplicate and the PAC value was determined by a standard curve constructed using GA in the range of 10-80 µM (R 2 = 0.9973). The PAC for each extract was expressed as mg GA equivalents per gram of dry matrix (mg GA/g).

Flavonoid Content
Flavonoid content (FC) in the lees extracts was spectrophotometrically determined by a literature procedure with some changes and expressing the results as the mass of catechin (CT) per gram of dry sample (mg CT g −1 ) [20]. Briefly, in a 5.0 mL volumetric flask, 0.5 mL of an aqueous solution of each extract was added to 0.15 mL of NaNO 2 (5% w/v). After 6 min, 0.3 mL of AlCl 3 (6% w/v), and after 5 min, 1.0 mL of NaOH 1.0 M were added to the mixture by measuring the absorbance at 510 nm against a control. A calibration curve of CT was built in the concentration range of 10.0-100.0 µM (R 2 = 0.9975).

Antioxidant Properties
The antioxidant features of the lees extracts were evaluated by specific measurements able to quantify their scavenger activities in organic and aqueous media against hydrophilic and lipophilic radical species, respectively.
The scavenger capacity of the extracts in an organic environment was evaluated by analyzing the concentration decrease in DPPH radical, slightly modifying a literature procedure [21]. Briefly, in a volumetric flask (10 mL) 1.0 mL of hydroalcoholic solutions (50:50 v/v) of each extract, 4.0 mL of a hydroalcoholic mixture (50:50 v/v), and 5.0 mL of a 200 µM ethanolic solution of DPPH were mixed. The mixture was kept at room temperature for 30 min and the residual concentration of the radical species was spectrophotometrically Antioxidants 2023, 12, 622 5 of 18 measured at 517 nm. The scavenging capacity of each lees extract against DPPH was expressed in terms of IC 50 . Ascorbic acid was used as a positive control.

Reactive Oxygen Species Detection
Intracellular reactive oxygen species (ROS) formation was investigated by using 2 ,7dichlorofluorescein diacetate (DCF-DA). SH-SY5Y cells were seeded (2 × 10 4 cells/well) into a 96-well plate, grown for 24 h under standard conditions, treated with LWE3 (10 and 25 mg mL −1 ) for 5 h at 37 • C, washed with PBS, and loaded with 10 µM DCF-DA for 30 min at 37 • C. The cells, rinsed twice with PBS, were treated with 2 mM H 2 O 2 in DMEM without phenol red and the intracellular fluorescence was read immediately for 60 min with the Thermo Labsystems Synergy HTX reader (BioTek, Winooski, VT, USA) (485 nm excitation, 528 nm emission). The Area Under Curve (0-60min) of the curve describing changes in fluorescence monitored every 5 min was calculated.

Statistical and Chemometric Analysis
HPLC-DAD analyses and antioxidant tests of the samples were performed in triplicate. Data were expressed as means ± SD and analyzed using the Wilcoxon test. A value of p < 0.05 was considered statistically significant. Biological experiment data are shown as the mean ± S.E.M. and analyzed by using ANOVA followed by the Tukey post-test with GraphPad Prism version 5.04 (GraphPad Software Inc., San Diego, CA, USA). In all comparisons, p < 0.05 was considered significant.
Principal component analysis (PCA) is an exploratory technique to extract the information contained in complex data systems, such as the spectral fingerprint of samples subjected to instrumental analysis. Samples and spectral variables are projected into the multidimensional space of the new principal components (PCs). PCA was performed to assess the composition differences in terms of the phenolic and flavonoid compounds of the wine lees extracts that can be detected by infrared spectral fingerprinting. The original data consisting of the extracts' analysis were organized into a matrix, in which 108 samples were described by their FTIR-ATR spectra. PCA analysis was elaborated by the Singular Value Decomposition (SVD) algorithm and the full cross-validation (FCV) procedure allowed for the determination of the appropriate number of PCs [23].
Partial least square regression (PLS) is a factor analysis method for the elaboration of the multivariate model with predictive ability, where descriptor variables X and vector y containing the response variables are decomposed in orthogonal factors. Consecutive factors are evaluated until the covariance between descriptors and responses is maximized. PLS modeling is achieved when the factors that explain most of the covariation between both data sets are found. The number of factors is chosen by calculating the figures of the merit root mean square error of validation (RMSEV) and correlation coefficient R 2 [24].
The multivariate analysis of the experimental data was carried out using the software Unscrambler X version 10.5 (CAMO, Trondheim, Norway).

Results and Discussion
An eco-friendly procedure was proposed to isolate antioxidant compounds from four different autochthones cultivars (Magliocco Canino, Magliocco Rosato, Gaglioppo, and Nocera Rosso) wine lees (Table 1). Table 1. Ultrasound-assisted extractions of wine lees from different cultivars.  Specifically, water (pH = 2.0), ethanol, and a water (pH = 2.0)/ethanol (50/50 v/v) mixture were employed, while the extractions were performed at low temperature (30 • C) and reduced time (15 min) to preserve the structural integrity of the bioactive compounds. For all the analyzed cultivars, the extraction procedure performed by employing acidic water returned the best results in terms of yield (range 27.7-35.8%), while the hydroalcoholic mixture performance displayed values statistically (p < 0.05) reduced (13.2-20.2%). Finally, pure ethanol represented the worst extraction solvent for all the analyzed cultivars (2.4-6.3%). The recorded analytical results clearly highlight that winery by-products are rich in highly hydrophilic compounds, poorly interacting with the organic solvents. In all cases, the cultivar Nocera Rosso guarantees the highest yields. Focusing our attention on the phenolic molecules, the wine lees represent a vegetable matrix rich in high-added-value compounds, such as flavonoids, tannins, and phenolic acids [25]. The literature has established that the chemical integrity and biological characteristics of the active molecules can be successfully preserved by employing eco-friendly procedures, considering low temperatures, reduced extraction times, and green solvents [26]. Specifically, bioactive molecules were effectively extracted from the vegetal raw material by employing an ultrasoundassisted methodology, able to decrease both extraction time and temperature, avoiding the loss of the biological properties of the active compounds [27]. The chemical compositions and yields of the extracts are strictly related to the hydrophilicity of the solvent and the solid-solvent ratio. Usually, an increase in the solvent volume ensures a better swelling of the vegetal raw material, making easy the transfer of the polyphenols and enhancing the yield of the extraction procedure [28]. Additionally, the employment of a binary solvent system based on water and eco-friendly organic solvents was suggested by the literature and can guarantee high performance for the extraction of polyphenols from the vegetal matrix [29]. Specifically, the effect of ethanol concentration on the extraction yield was reported in the literature and the highest extraction yields of phenolic molecules were obtained when hydroalcoholic mixtures were used [30] due to the presence of a more polar medium increasing the polyphenol extraction efficiency [31]. In this regard, Kalia et al.

Sample Extraction Conditions Yield
found a 50% (v/v) ethanol solution in water to be the most effective for the extraction of phenolics from Potentilla atrosanguinea [32].
A detailed characterization of the extracts was performed by HPLC-DAD and FTIR spectroscopy.
Regarding their content, as expected [14], anthocyanins were more concentrated in lees from red wines (Nocera Rosso cv) and particularly in the water extracts due to the high polarity of these compounds; indeed, either the mono-glucoside anthocyanins or the two coumaroyl conjugates (POCG and MCG) were significantly more concentrated in LW1 than LWE1 and LE1 ( Table 2).

Characterization of Wine Lees Extracts Via FTIR and PCA
The FTIR fingerprints of the wine lees are shown in Figure 2, where the variation in sample composition due to different cultivars and extraction strategies can be seen. The wavenumber window 1850-450 cm −1 presented the characteristic bands of phenolic compounds for all samples. The bands between 1470 and 950 cm −1 were due to the methoxy group of some phenolic acids. In the range of 1755-1630 cm −1 , phenolic acids can be distinguished from flavonoids by the typical bands of the carboxyl groups, while flavonoid moieties are mainly distinguished by bands associated with benzo-γ-pyrone and benzopyrylium signals in the regions 1650-1400 and 1200-450 cm −1 [16].
and Magliocco Canino, whereas the more hydrophilic GA and EG were more concentrated in water extracts, specifically in LW1 (650 mg g −1 ) and LW2 (220 mg g −1 ), respectively. It is worth pointing out that the behavior of EGG was an exception because its values were prevalent in all LE extracts ranging from 820 to 2900 mg g −1 (Table 2). Finally, just quercetin-3O-glucoside (QG) was identified as a flavonol compound ( Figure 1B) at higher levels in LWE1 and LWE3.

Characterization of Wine Lees Extracts Via FTIR and PCA
The FTIR fingerprints of the wine lees are shown in Figure 2, where the variation in sample composition due to different cultivars and extraction strategies can be seen. The wavenumber window 1850-450 cm −1 presented the characteristic bands of phenolic compounds for all samples. The bands between 1470 and 950 cm −1 were due to the methoxy group of some phenolic acids. In the range of 1755-1630 cm −1 , phenolic acids can be distinguished from flavonoids by the typical bands of the carboxyl groups, while flavonoid moieties are mainly distinguished by bands associated with benzo-γ-pyrone and benzopyrylium signals in the regions 1650-1400 and 1200-450 cm −1 [16]. Considering the intense overlapping of the signals from different compound families, a PCA was performed to investigate more specific differences in spectral fingerprints. Figure 3 describes the 3D plot scores by considering the principal components 1, 2, and 3 with a total explained variance of 93%. The clustering of the samples was immediately Considering the intense overlapping of the signals from different compound families, a PCA was performed to investigate more specific differences in spectral fingerprints. Figure 3 describes the 3D plot scores by considering the principal components 1, 2, and 3 with a total explained variance of 93%. The clustering of the samples was immediately evident, and it was possible to distinguish the samples according to the type of solvent used to extract the phenolic compounds. The only exception was the sample LWE3, which stood out from the others by forming a distinct group. Analyzing the x-loading values, calculated for the processed wave variables with respect to the first three PCs, some characteristic peaks for each extraction technique or cultivar were highlighted. boxylic acid; epigallocatechin gallate (EGG) is characterized by bands in the range 1120-1050 cm −1 and showed a band around 800 cm −1 [16]. Chromatographic analysis showed a relative abundance in the content of metabolites TCA, CA, CLA, and EGG for samples extracted in ethanol (LE) or hydroalcoholic (LWE) solvents, which was confirmed by evaluating the infrared fingerprint of the sample. The higher levels of these metabolites in the LWE3 sample were also evident in the infrared signals and were able to push the samples out of the LWE cluster by forming a distinct cluster.

Antioxidant Characterization of the Wine Lees Extracts
The total polyphenol content of wine lees extracts was accomplished according to the Folin-Ciocalteu methodology and the obtained results are reported in Table 3. The PCA processing was able to handle and highlight a relative abundance of useful information in four spectral regions: 1736, 1720, 1390, and 1075 cm −1 . The first band at 1736 cm −1 was apparently not responsible for the clustering of samples in the 3D-scores space; however, these characteristic bands could be attributed to the anthocyanins such as POCG and MCG, which have been detected for samples belonging to Nocera Rosso cultivar (L1) (Table 2 and Figure 2). Signals at 1720, 1120, and 1075 cm −1 were able to give enough information to allow the grouping of the samples and especially the clusters consisting of wine lees extracted by ethanol (LE) and hydroalcoholic (LWE) solvents. Chlorogenic acid and esters of caffeic acid showed bands around 1720 cm −1 , which could be associated with the vibration of unsaturated aliphatic ester or the carbonyl stretching of protonated carboxylic acid; epigallocatechin gallate (EGG) is characterized by bands in the range 1120-1050 cm −1 and showed a band around 800 cm −1 [16]. Chromatographic analysis showed a relative abundance in the content of metabolites TCA, CA, CLA, and EGG for samples extracted in ethanol (LE) or hydroalcoholic (LWE) solvents, which was confirmed by evaluating the infrared fingerprint of the sample. The higher levels of these metabolites in the LWE3 sample were also evident in the infrared signals and were able to push the samples out of the LWE cluster by forming a distinct cluster.

Antioxidant Characterization of the Wine Lees Extracts
The total polyphenol content of wine lees extracts was accomplished according to the Folin-Ciocalteu methodology and the obtained results are reported in Table 3. The literature data reported that the phenolic profile in lees is strictly related to the type of crushed grapes and other factors that are present during vinification [8].  The recorded data displayed as the cultivar is an important parameter to be considered and deeply influences the performance of the extraction mixture. Ethanolic extracts returned the highest TPC values for Magliocco Rosato and Magliocco Canino cultivars, while the acidic water extraction of Nocera Rosso significantly guaranteed the best performance. Finally, the hydroalcoholic extract of Magliocco Canino displayed the highest TPC value (49.56 mg GA per g of extract) according to the data recovered by HPLC-DAD analysis. Recorded TPC value is in the same order of magnitude reported in the literature and related to wine lees deriving from different grape varieties and performing the extraction by several analytical methodologies. In particular, solvents with different polarities were also evaluated in the analyses of Tempranillo wines lees, returning TPC values in the range of 26 (acetone)-254 (ethanol/water 25:75 v/v) mg GA per gram of dry matter [35], according to another study performed a conventional extraction procedure and reporting a phenolic concentration equal to 547 mg GA per gram of dry matter [7]. Additionally, the TPC values of different extracts deriving from various red wines (Cabernet Sauvignon, Cabernet Franc, Merlot) by ultrasound were in the range of 44.02-58.71 mg GA per gram of dry extract [30], while other studies displayed decreased values of 23.2-30.9 mg GA per gram of dry extract [34,36]. On the other hand, these partial discrepancies can be largely due to the wine lees types, strictly related to the vinification process and grape variety as well as the extraction procedure.
LWE3 also showed the greatest concentration of flavonoid (20.15 mg GA per g of extract) and phenolic acid (22.67 mg CT per g of extract) compounds. Concerning the phenolic acids, p-coumaric and caffeic acids as well as their derivatives such as transcoutaric acid and trans-caftaric acid, respectively, have been found in wine lees samples, according to the literature data reporting trans-caftaric acid as the predominant compounds in both Vitis vinifera and non-vinifera types [36].
Polyphenol moieties display numerous beneficial effects such as a remarkable scavenger ability against various radical species [37]. As antioxidant bio-compounds explain their action through different mechanisms, usually, various tests should be explored to fully estimate the antioxidant capacity of a compound mixture [38]. For this reason, the antioxidant profile of the extracts was evaluated in terms of scavenger activity both in aqueous and organic environments against DPPH and ABTS radicals, respectively. The capacity of the extracts to inhibit the lipophilic DPPH radical, expressed in terms of IC 50 (mg mL −1 ), is reported in Table 3. The recorded results are in the range of 0.395-1.360 mg mL −1 and LWE3 appears as the more active extract. This trend was also established in the hydrophilic environment against the ABTS radical (IC 50 for LWE3 equal to 0.451 mg mL −1 ), highlighting a greater scavenger effect of the extract in the organic environment with respect to the aqueous one. In general, scavenger activity measurements confirmed the TPC, PAC, and FC values. Specifically, a direct correlation between the antioxidant capacity of the extracts was carried out, verifying the existence of a direct dependence between TPC value and the scavenger activity, considering the sum of the contributions of the various antioxidant compounds in the sample and their possible synergistic effects.

Effect of LWE3 on H 2 O 2 -Induced ROS Production in SH-SY5Y Cells
In SH-SY5Y cells, treatment with H 2 O 2 markedly increased the intracellular level of ROS ( Figure 4). However, pre-treatment with LWE3 significantly prevented ROS production in a concentration-dependent manner, suggesting that the extract protected neuroblastoma cells against oxidative stress damage.
to the literature data reporting trans-caftaric acid as the predominant compounds in both Vitis vinifera and non-vinifera types [36].
Polyphenol moieties display numerous beneficial effects such as a remarkable scavenger ability against various radical species [37]. As antioxidant bio-compounds explain their action through different mechanisms, usually, various tests should be explored to fully estimate the antioxidant capacity of a compound mixture [38]. For this reason, the antioxidant profile of the extracts was evaluated in terms of scavenger activity both in aqueous and organic environments against DPPH and ABTS radicals, respectively. The capacity of the extracts to inhibit the lipophilic DPPH radical, expressed in terms of IC50 (mg mL −1 ), is reported in Table 3. The recorded results are in the range of 0.395-1.360 mg mL −1 and LWE3 appears as the more active extract. This trend was also established in the hydrophilic environment against the ABTS radical (IC50 for LWE3 equal to 0.451 mg mL −1 ), highlighting a greater scavenger effect of the extract in the organic environment with respect to the aqueous one. In general, scavenger activity measurements confirmed the TPC, PAC, and FC values. Specifically, a direct correlation between the antioxidant capacity of the extracts was carried out, verifying the existence of a direct dependence between TPC value and the scavenger activity, considering the sum of the contributions of the various antioxidant compounds in the sample and their possible synergistic effects.

Effect of LWE3 on H2O2-Induced ROS Production in SH-SY5Y Cells
In SH-SY5Y cells, treatment with H2O2 markedly increased the intracellular level of ROS ( Figure 4). However, pre-treatment with LWE3 significantly prevented ROS production in a concentration-dependent manner, suggesting that the extract protected neuroblastoma cells against oxidative stress damage.

Relationship between Antioxidant Properties and Metabolite Composition
In the present study, the PLS algorithm was applied to study the relationship between antioxidant properties and the metabolite composition of the extracts. The root mean square error of validation (RMSEV) and the correlation coefficient R 2 were used to evaluate the fitting of the multivariate models with the experimental results. Usually, an R 2 value higher than 0.5 is indicative of an acceptable model and a value over 0.8 proves an excellent correlation [39].
The chemical composition detected by HPLC analysis was used as a descriptor variable (matrix X) for each sample, and two different response vectors, y DPPH and y ABTS , consisting of the respective IC 50 values evaluated for each sample, were selected. All variables used in the building of the model were re-scaled and standardized prior to elaboration; they were normalized by dividing by the standard deviation value in the original scale and in logarithmic form.
First, multivariate calibration built two PLS models with very low predictive performance with R 2 below 0.6 for both y DPPH and y ABTS . These models were used as an initial screening of the correlation between the compound concentrations and the antioxidant power of the extracts. After evaluating the regression coefficients of each variable, it was possible to determine which of them provided the most information in the calibration of the models. Figure 5 shows the weighted regression coefficients (Bw) plot. Regression coefficients describe the relationship between all the descriptors (x-variables) and a given response (y-variable). Bw coefficients provide information about the importance of the x-variables: a high Bw value means an important role in the building of the models, while x-variables with small coefficients may be negligible; a positive coefficient shows a direct proportionality with the response, and a negative coefficient shows an indirect proportionality. The most important variables were TCA, CLA, and EGG for the prediction of y DPPH and TCA and CLA for the prediction of y ABTS .

Relationship between Antioxidant Properties and Metabolite Composition
In the present study, the PLS algorithm was applied to study the relationship between antioxidant properties and the metabolite composition of the extracts. The root mean square error of validation (RMSEV) and the correlation coefficient R 2 were used to evaluate the fitting of the multivariate models with the experimental results. Usually, an R 2 value higher than 0.5 is indicative of an acceptable model and a value over 0.8 proves an excellent correlation [39].
The chemical composition detected by HPLC analysis was used as a descriptor variable (matrix X) for each sample, and two different response vectors, yDPPH and yABTS, consisting of the respective IC50 values evaluated for each sample, were selected. All variables used in the building of the model were re-scaled and standardized prior to elaboration; they were normalized by dividing by the standard deviation value in the original scale and in logarithmic form.
First, multivariate calibration built two PLS models with very low predictive performance with R 2 below 0.6 for both yDPPH and yABTS. These models were used as an initial screening of the correlation between the compound concentrations and the antioxidant power of the extracts. After evaluating the regression coefficients of each variable, it was possible to determine which of them provided the most information in the calibration of the models. Figure 5 shows the weighted regression coefficients (Bw) plot. Regression coefficients describe the relationship between all the descriptors (x-variables) and a given response (y-variable). Bw coefficients provide information about the importance of the xvariables: a high Bw value means an important role in the building of the models, while x-variables with small coefficients may be negligible; a positive coefficient shows a direct proportionality with the response, and a negative coefficient shows an indirect proportionality. The most important variables were TCA, CLA, and EGG for the prediction of yDPPH and TCA and CLA for the prediction of yABTS. A new calibration procedure was thus optimized by considering just the variables TCA, CLA, and EGG. A significant improvement in the statistical parameters was observed when the concentration levels of phenolic acids and flavonoids were considered on the logarithmic scale. The PLS model built to predict the DPPH IC50 gave good statistical parameters with a correlation coefficient R 2 value of 0.932 and a prediction error A new calibration procedure was thus optimized by considering just the variables TCA, CLA, and EGG. A significant improvement in the statistical parameters was observed when the concentration levels of phenolic acids and flavonoids were considered on the logarithmic scale. The PLS model built to predict the DPPH IC 50 gave good statistical parameters with a correlation coefficient R 2 value of 0.932 and a prediction error RMSEV of 0.023 by using three factors, while ABTS IC 50 was correlated with R 2 of 0.824 an RMSEV of 0.064 by using two factors.

Conclusions
In this study, wine lees from different cultivars (Magliocco Canino, Magliocco Rosato, Gaglioppo, and Nocera Rosso) were extracted with an ultrasound-assisted method using an eco-friendly procedure employing low temperature and reduced time and ethanol and water as solvents to obtain polyphenolic fractions deeply analyzed by HPLC-DAD and FTIR techniques. Specifically, anthocyanins were more concentrated in lees from red winemaking (Nocera Rosso cv) and particularly in the water extracts due to the high polarity of these compounds, while the main hydroxycinnamic acid present in the extracts was trans-caftaric acid, reaching values of 2100 mg g −1 in LWE3 samples. Additionally, colorimetric tests allowed for the evaluation of total polyphenol, phenolic acid, and flavonoid contents, while radical scavenging assays were performed to quantify the antioxidant performances of the extracts, both in organic and aqueous media, against DPPH and ABTS radicals, respectively. The combined use of multivariate tools and FTIR spectral analysis proved very effective in characterizing samples from different cultivars and extraction procedures, grouping samples with respect to their spectral fingerprint, and highlighting the correspondence between spectral bands and metabolite abundance in extracts. The collected data confirmed the spectroscopic recovery, highlighting LWE3 as representing the higher-performing extract. Cellular (SH-SY5Y cells) antioxidant assays were performed on the best extract (LWE3) to confirm its biological performance against radical species.
The chemometric processing of the qualitative-quantitative experimental data allowed the correlations between chemical composition and antioxidant features of the different wine lees extracts to be found. PLS optimization identified the metabolites responsible for the detected antioxidant effects, and two PLS models were built to predict the IC 50 values of DPPH and ABTS radicals with good predictive ability by using the concentration values of the compounds TCA, CLA, and EGG.
Further studies will employ the best extract to prepare innovative high-value functional foods. Alternative antioxidant extracts will be involved in a radical conjugation process to synthesize more stable macromolecular systems able to be used as an additive in the food industry.