Eight commercial bottled Spanish red wines of different vintages (between 2009 and 2014) and made from four different grape varieties, Garnacha (Grenache), Tempranillo Merlot, and Cabernet-Sauvignon, were used in the study. Details of the samples together with some compositional parameters are shown in Table 1.

Wines were subjected to three different controlled oxygen exposure conditions (R1, R2, and R3) in duplicate following the procedure described in Marrufo-Curtido et al. (2018) (Figure 3). In such procedure, perfectly controlled volumes of sterile filtered (OIV, 2015) wine are enclosed with perfectly controlled volumes of air in air-tight tubes (60 mL nominal internal volume from WIT-France, Bordeaux) containing PSt3 oxygen sensors (Nomacorc S.A., Thimister-Clermont, Belgium). The accurate internal volume and weight of each one of the tubes were previously determined by standard calibration practices.

In order to avoid bottle effects, five bottles of each one of the eight wines used in the experiment were opened, mixed in a large beaker, filtered and redistributed back in the five bottles. The bottles were purged for 1 min with a 415 mL min⁻¹ flow of argon, were further resealed with a Nomacorc Select 300 CP closure and were stored in the fridge until the beginning of the experiment. One bottle was then used to measure wine density and total sulfur dioxide, parameters required in the procedure.

The experiment begins by opening a bottle, transferring a volume of wine to a graduated cylinder and carefully closing the bottle with a wine saver vacuvin closure (IIC Brands, The Netherlands). Then, a specific volume of wine was quickly transferred to the calibrated WIT tube. The tube was immediately weighted and closed. The weight of the filled tube is used to determine the exact volume of wine contained in the tube (V_l). The exact volume of headspace (V_g) was estimated by subtracting V_l from V_int, the truly internal volume of the tube. The time elapsed between uncorking and closing the WIT tube with the wine was less than 2 min. The tubes were then left in a thermostatic bath Grant OLS23 with orbital shaking (90 rpm) at 25 ± 0.1°C to ensure that liquid and gas phases were always in equilibrium. Oxygen was measured with an oxygen analyzer Fibox 3 LCD trace from Nomacorc SA. Oxygen level was monitored each 30 min during the first 3 h, and then once per day. Each experiment was considered complete when the wine consumed 95% of the initial oxygen. At that time, the WIT tubes were opened inside a glove box without oxygen (oxygen < 0.002%) to get samples for analysis avoiding the wine to be further oxidized.

The total amount of oxygen contained initially in the tube was estimated from calibrated tubes filled with volumes of synthetic wine following exactly the same procedures used for the wines. In these cases, the dissolved oxygen in the liquid phase was measured after half an hour inside the incubator shaker to ensure a perfect equilibration. Those concentrations in the liquid phase of equilibrated synthetic wines were R1: 5.0 ± 0.13, R2: 6.35 ± 0.04, and R3: 6.75 ± 0.02 mg L⁻¹ (means of four replicates each). Then, knowing that the saturation concentration of O₂ in our wine models completely equilibrated with air (21% v/v in Oxygen) was 7.73 ± 0.07 mg L⁻¹ (n = 4), the exact volume of headspace (V_g) and of liquid (V_l) within the tubes, and assuming that oxygen in gas and liquid phases are in equilibrium, it is possible to estimate the initial total amount of oxygen. In R1 experiments, the total amount of oxygen contained in the tube corresponded to 9–11 mg of oxygen per liter of wine; in R2 experiments, the total amount within the tubes ranged from 22 to 35 mg of oxygen per liter of wine. In this case, the amount delivered to each wine was 18 mg L⁻¹ plus the stoichiometrically required amount to oxidize all its total SO₂. Similarly, in R3 experiments, the oxygen given to each wine was 32 mg L⁻¹ plus the stoichiometrically required amount to oxidize all its total SO₂ so that total amounts of oxygen within the tubes ranged from 35 to 53 mg of oxygen per liter of wine.

Sodium metabisulfite 99% (Na₂S₂O₅), tartaric acid (99%), glycerol (99.5%), 1,2-propanediol (99.5%), sodium hydroxide (98%), ortho phosphoric acid (85%), hydrogen peroxide 3% stabilized w/v VINIKIT, indicator 4,4, mixed (methyl red-methylene blue) VINIKIT, sodium hydroxide 0.01 mol L⁻¹ VINIKIT were from Panreac (Barcelona, Spain). Dichloromethane, ethanol and methanol for gas chromatography analyses were purchased from Merk (Darmstadt, Germany). Methanol and acetonitrile of HPLC quality were obtained from Fluka Analytical (Buchs, Switzerland). Hydrochloric acid 37%, formic acid and ammonium formate high purity grade were purchased from VWR Prolabo (Fontenay sous Bois, France). Phloroglucinol (≥99%), ascorbic acid (≥99%), acetaldehyde (≥99.5%), 2-chloroethanol (≥ 99.0%), methyl 2-methylbutyrate (≥ 99%), 2-butanol (≥ 99%), glyoxal 40% in water, Folin-Ciocalteu's phenol reagent, sodium carbonate (≥ 99%), gallic acid (≥ 99%), and (+)-catechin (≥ 99%) were supplied by Sigma-Aldrich (Madrid, Spain). Standards and reagents for aroma compounds and amino acids determination were purchased from Sigma-Aldrich, Fluka, Panreac, Lancaster (Eastgate, UK), PolyScience (Niles, IL, USA), ChemService (West Chester, PA, USA), and Firmenich (Switzerland), and details of the chemicals have been already reported (Singleton et al., 1998; Ortega et al., 2001; Hernández-Orte et al., 2003; Ribéreau-Gayon et al., 2006; OIV, 2009b; Herrero et al., 2011; Bueno et al., 2014; Grindlay et al., 2014; Vallverdu-Queralt et al., 2016; Carrascón et al., 2017). Water was purified in a Milli-Q system from Millipore (Bedford, Germany).

Initial wines (R0) were analyzed in duplicate for total acetaldehyde and total odor-active carbonyls, as well as for free and total SO₂, pH, color parameters, total polyphenol index and Folin-Ciocalteu index, phenolic, and tannin composition, major aroma compounds, metals, and amino acids.

Final samples after oxidation procedures were analyzed for total acetaldehyde, total odor-active carbonyls, as well as for free (only for the low exposure) and total SO₂, color parameters, total polyphenol, and major aroma compounds.

Levels of carbonyls obtained in the different samples were analyzed by repeated measures ANOVA using SPSS v15. (SPSS Inc., IBM Company, Chicago, IL, USA) for assessing the effects of oxygen level and wine sample. Correlation studies were directly carried out with Excel 2013 (Microsoft, Washington, USA). The uncertainty of each mean was calculated taking into account experimental uncertainty, measured via experimental replicates, and the analytical uncertainty, measured via analytical replicates.

Principal Component Analysis was performed using XLSTAT (Addinsoft, version 2015). Partial Least Square Regression analysis was carried out using The Unscrambler 9.7 (CAMO Software AS, Oslo, Norway). PLS modeling was carried out using cross-validation criteria. In this strategy, the model is built in iterations leaving out one of the samples in each iteration. The predicted result for the sample left out is us to compute the model error. The process is repeated until every sample has been left out once; then all prediction residuals are combined to compute the validation residual variance and the RMSEP (Root mean square prediction error).

The relationship between the total content in acetaldehyde found in the different samples of each wine and their levels of O₂ not SO₂, is given in Figure 4A. The figure splits wines attending to their age. Levels of acetaldehyde in samples derived from young wines (1-year-old bottled wines) were under 19 mg L⁻¹, regardless of the O₂ delivered to the sample. By contrast, levels in samples derived from aged wines increased with oxidation becoming in all cases above 25 mg L⁻¹ in R3 samples. Within each category there are also some differences. In young wines, acetaldehyde levels slightly increased up to 5.8 mg L⁻¹ with oxygen exposure only in the wine MF. In aged wines, increases were most noticeable at low O₂ exposure (R1) and only in the wine BL there was a strong increase at the highest exposure (R3).

Taking into account that once SO₂ has been depleted, H₂O₂ oxidizes mostly ethanol to yield acetaldehyde (Danilewicz, 2013) (Figure 1), it is possible to estimate the amount of acetaldehyde theoretically formed by wine in each sample. The comparison of this expected amount with the real amount of acetaldehyde accumulated in each case will provide an estimation of the fraction of acetaldehyde reacted with polyphenols or further oxidized to acetic acid. As an example, the wine BL consumed 32.4 mg L⁻¹ of O₂ not SO₂ in R3. This amount should have produced 32.4/32 = 1.01 mM of H₂O₂, which in turn will have oxidized 1.01 mM of ethanol to produce the same molar amount of acetaldehyde: 44.6 mg L⁻¹. Since the measured increase of acetaldehyde is 17.7 mg L⁻¹ and the increment of acetic acid was null, it can be inferred that 60.4% of the acetaldehyde formed reacted most likely by forming ethyl bridges with polyphenols and that 39.6% of acetaldehyde remained unreacted in the wine.

These calculations are summarized in Table 3. As can be seen, differences between aged and young wines are very high, even taking into account that in some cases the small changes observed in levels of acetaldehyde and the large changes in total SO₂ cause a large imprecision. The unreacted fraction in old wines was in all cases above 80% in the low oxygen exposure experiment, while in three of the young wines the unreacted fraction was close to zero. A negative percentage means that more acetaldehyde is reacting than it is being formed. A second observation is that the unreacted fraction decreases with the extent of the O₂ exposure, so that, in the high O₂ exposure experiment, the unreacted fraction in aged wines was smaller than 25% in three of the wines.

The effect of the level of O₂ on the unreacted fraction of acetaldehyde becomes more evident by assuming that the oxidation process can be segmented into three “virtual” steps (first 2.8–10 mg L⁻¹; from 15 to 19 mg L⁻¹; from 29 to 34 mg L⁻¹ of O₂ not SO₂, second column of Table 2), neglecting the fact that the experiments were not consecutive but independent. For instance, the amount of acetaldehyde produced in the second virtual step was estimated by subtracting the levels of acetaldehyde found in R1 from those found in R2, while differences in O₂ not SO₂ between both experiments were used as estimators of the amount of O₂ not SO₂ consumed in such virtual second step. The unreacted fraction of acetaldehyde in those three steps is represented in Figure 5A for aged wines, showing that the unreacted fraction decrease was progressive for SL, TS and CH (quite dramatic in the case of TS) while BL showed a unique increase in the unreacted fraction in the 3rd step. Regarding young wines, shown in Figure 5B, the unreacted fraction of TP was close to 0 in all cases, while for HV and BS became negative in the second period. MF followed a completely different pattern, halfway to those of aged wines.

The previous patterns can be explained attending to three concepts:

the ability of acetaldehyde to form strong and stable unreactive complexes with SO₂

Attending to those concepts, the parameters which should determine the evolution of acetaldehyde in a specific wine are:

the wine content of SO₂ and of SO₂ binders

The higher accumulation of acetaldehyde in aged wines can be explained attending to the much lesser amounts of ARPs which these wines should contain, because of their potentially larger previous exposure to O₂. Similarly, the higher accumulation of acetaldehyde in the 1st oxidation period observed in Figure 5A can be explained attending to the presence of some remaining SO₂ in the media. As complexes SO₂-acetaldehyde are the strongest amongst carbonyls (de Azevedo et al., 2007), newly formed acetaldehyde could displace weak SO₂-binders from their associations, becoming itself protected from reaction with the limited fraction of ARPs present in aged wines. In the second and third period, however, SO₂ is progressively depleted, so that newly formed acetaldehyde cannot be further protected from ARPs. The wine-specific patterns seen in Figure 5A should be attributed to their differences in initial SO₂ and in SO₂ binders. For instance, the highest fraction of unreacted acetaldehyde of TS in 1st period corresponds to its highest content in total and free SO₂ (see Table 1). Similarly, the highest fraction observed in the 2nd period for SL is in agreement with its highest content in total SO₂ and highest level in SO₂ binders (manifested in its low free SO₂). The unique behavior of BL wine may be attributed to a possible exhaustion of ARPs. In this context it may be noteworthy that this wine had the lowest TPI and Folin indexes amongst aged wines. In the case of young wines (Figure 5B), the negative fraction of unreacted acetaldehyde of HV in the 2nd period may be related to its lowest level of SO₂ and may be also to its highest TPI and Folin amongst young wines.

A first insight about the nature of ARPs can be obtained by modeling the unreacted fraction of acetaldehyde measured in the different experiments vs. the initial wine chemical composition. Models (summarized in Table 4) were obtained for the different oxidation doses and were in most cases highly explicative, with explained variances by cross-correlation above 94%, except for the high oxygen dose.

In general, models give a very large positive coefficient to combined SO₂ and negative coefficients for nearly all the other components, which suggests that the accumulation of acetaldehyde is the result of the balance between its ability to displace SO₂ from its adducts with weaker SO₂-binders and the rate at which it reacts with ARPs.

Attending to the first model, derived from R1 samples, ARPs are mainly anthocyanins: malvidin 3-O-glucoside, the vinylguaiacol derivative of peonidin, and the catechin-ethyl-malvidin dimmer. The positive coefficient for epigallocatechin may be related to recent evidences indicating that epigallocatechin is oxidized with a concomitant high SO₂ consumption (Carrascón et al., 2018) which would cause a shortage of free SO₂, a concomitant increase of H₂O₂ and a peak in acetaldehyde production which could not completely react with ARPs, and instead would displace other SO₂-binders to form 2-hydroxyethylsulfonate.

In the medium O₂ exposure level or in the virtual second step, which takes place under small levels of free SO₂, malvidin 3-O-glucoside, galloylated tannins, and epicatechin-3-O-gallate have negative coefficients in the model, suggesting that they are the main ARPs for the newly formed acetaldehyde in this stage of oxidation.

Modeling the high exposure level has been more complicated because of the unique behavior of sample BL. Additionally, the virtual step could not be modeled because the predictive ability of the initial chemical composition is limited since wines already have suffered a deep chemical change. Nevertheless, the model is consistent with previous observations, and suggests that under large O₂ exposure conditions, the weight of SO₂ is null and that the accumulation of acetaldehyde is basically the result of the balance between its formation and its reaction to form ethyl-bridged structures. In this oxidation step, tannins do not seem to be kinetically critical, and only mDP (medium Degree of Polymerization) has a positive coefficient with the unreacted fraction of acetaldehyde. This would suggest that large tannins are less efficient at forming ethyl bridged structures, and that kinetically, small tannins would be favored.

The previous discussion suggests that when SO₂ levels become very low, as will inevitably happen in some moment of wine aging, the effects of oxidation will be the result of the relative rates of some key chemical processes, as illustrated in Figure 6.

In the first place, the rate at which ethanol is oxidized to form acetaldehyde depends on the following balance

In the second place, the fate of the acetaldehyde formed will depend on:

The rates at which acetaldehyde react to wine ARPs, critically dependent on the nature and concentration of those ARPs

The figure also highlights the relationships between aldehydes, ARPs and SO₂. The exhaustion of free SO₂ will have as consequence first the cleavage of the relatively weak associations between anthocyanins and SO₂, and these anthocyanins will further react with acetaldehyde. Once weak SO₂ binders have been consumed, the other SO₂ binders will be released from weakest to strongest. Some of the binders are Strecker aldehydes which once released will change wine aroma (Bueno et al., 2016). Some others, such as diacetyl, are reactive compounds which can then promote different chemical changes.

The evolution of diacetyl with the progress of oxidation is presented in Figure 4B and again, there is a neat difference between young and aged wines. Levels of diacetyl in aged wines were low, strictly inversely proportional to wine age, and quite stable. This result already suggests that diacetyl slowly but continuously reacts with different wine components during aging in processes more related to time than to the presence of oxygen. Among aged wines, only in TS there is a significant decrease (p ≤ 0.05) during oxidation (Table 2). In strong contrast, diacetyl in young wines followed a clear inverse sigmoid characterized by a flat initial period, a strong decrease in the medium O₂ exposure samples and a last flat step. The decrease was proportional to the initial level and in all cases final diacetyl contents of the wine seemed to stabilize at around 40% of its initial content.

Diacetyl is a quite reactive molecule which can react with a broad range of organic molecules. It has been reported that diacetyl can react with ARPs, although at a much slower rate than acetaldehyde (Blanco-Vega et al., 2011). It can also react to thiols, such as cysteine, producing a wide range of subproducts (Marchand et al., 2000, 2011). Remarkably, one of the reactions suffered by diacetyl and other α-dicarbonyls is Maillard reaction with amino acids, one of whose possible outcomes is the Strecker degradation to form the corresponding Strecker aldehydes (Oliveira et al., 2011a). The possible implication of diacetyl in the formation of these components will be discussed later. The flat initial period observed in Figure 4B can be explained because in the low oxygen exposure conditions, diacetyl would be mostly complexed with SO₂ (Bueno et al., 2016), and would be protected. At higher O₂ exposure levels, however, unprotected diacetyl would react with available nucleophiles. The last flat period may be attributed to the exhaustion of such nucleophiles.

The five Strecker aldehydes accumulate during wine oxidation in all wines, as can be seen in Table 2. As previously observed, the accumulation takes place at relatively high consumptions of oxygen (Bueno et al., 2016), and in most wines it was observed only when the wine was exposed at least to medium amounts of oxygen (R2). The relationship between levels of Strecker aldehydes in the different wines and levels of oxygen “not invested in the oxidation of sulfur dioxide” can be seen in the Principal Component plot shown in Figure 7. The plot, which retains more than 95% of the original variance, shows that non-aromatic aldehydes (methional, isobutyraldehyde, 2-methylbutanal, and isovaleraldehyde) follow a quite similar and correlated behavior, while phenylacetaldehyde follows a quite distinct pattern.

Regarding samples, the PCA plot reveals that samples are largely segregated by its age: older wines are placed in the central and right south parts of the plot, while young wines are in the left and upper part of the plot. Older wines displace strongly to the right along the first component and just slightly up along the second when they oxidize, indicating that these compounds form comparatively more methional, isobutyraldehyde, 2-methylbutanal, and isovaleraldehyde than phenylacetaldehyde. The plot also shows that the magnitude of the change suffered by aged wines is related to the position of the original wine in the plot: the further right the original samples are, the further right are displaced the samples exposed to higher levels of oxygen. This implies that in these older wines the amount of aldehyde formed is proportional to the amount of aldehyde already present in the wine. This makes sense, since the history of previous exposure to oxygen of older wines is large, indicating that their initial content maybe strongly influenced by their specific ability to form aldehydes.

Quite differently, younger wines when oxidize displace north and slightly right, indicating that these wines tend to form higher levels of phenylacetaldehyde than of the other Strecker aldehydes. Moreover, the plot basically separates initial and low O₂ exposure samples (R0 and R1) from those of medium and high exposure (R2 and R3), indicating that phenylacetaldehyde formation takes part preferably at medium O₂ exposure. Additionally, the magnitude of the change in these cases does not seem to be related to the initial position of the wine in the plot, meaning that the aldehyde content of the initial wine seems to be a poor predictor of its ability to produce aldehydes upon oxidation.

The more specific differences in oxidation patterns between samples can be appreciated with the plots in Figure 4. Isobutyraldehyde, in Figure 4C, exemplifies the patterns followed by methional, 2-methylbutanal, and isovaleraldehyde, and shows that levels of aldehyde in older samples steadily increases with oxygen exposure and that increases are proportional to the original content of the wine. By contrast, levels in young wines increase just slightly and only become significant at high O₂ exposure levels. It is worth mentioning that samples in the plot are nearly perfectly arranged by age. This, as was previously mentioned for diacetyl, could be a clue indicating that time, and not only oxygen, plays a role in the accumulation of these components. Phenylacetaldehyde, in Figure 4D, follows a quite different pattern. The four young wines follow a clear sigmoidal pattern of accumulation, indicating that the formation of phenylacetaldehyde took specifically place when the wine was exposed to medium levels of oxygen. Higher or lower levels did not have any additional effect on the levels of this aldehyde. The resemblance with the plot seen for diacetyl in Figure 4B is remarkable. Older wines, by contrast, follow a pattern quite similar to the one observed for isobutyraldehyde in Figure 4C; levels increase steadily with level of oxygen, although, less evidently for SL, the formation was yet more intense when samples were exposed to medium levels of oxygen.

The segregation between aged and young wines shown in the PCA plot in Figure 7 is consistent with the assumption that Strecker aldehydes, except phenylacetaldehyde, are also subject to reaction with ARPs. Aged wines, which were found to contain less ARPs are the ones with higher increases in aldehydes. The odd pattern followed by phenylacetaldehyde, which strongly accumulates in young wines, could be simply related to a reduced reactivity toward ARPs because of steric reasons. The reactivity of the different Strecker aldehydes toward ARPs is not known and should be the subject of further research.

The accumulation of Strecker aldehydes in older wines is strongly and significantly correlated to the levels of precursor amino acids in wine: (R = 0.98; 0.96; 0.99; 0.99, and 0.96 for the pairs valine-isobutyraldehyde, leucine-isovaleraldehyde, isoleucine-2-methylbutanal, methionine-methional, and phenylalanine-phenylacetaldehyde). In younger wines, however, only the correlations between the pairs isoleucine-2-methylbutanal and phenylalanine-phenylacetaldehyde, are significant (R = 0.99 in both cases). The lack of significance in the other cases may be attributed to the low variability of their corresponding amino acids (<4 mg L⁻¹, with RSD inferior to 16% in all cases) and to the low levels of aldehyde accumulated.

The amount of aldehyde accumulated by unit of amino acid originally present in wine is in all cases strongly linked to wine age, being the effect of age in phenylacetaldehyde opposite to that observed in the other aldehydes. Young wines accumulated 9.2 μg L⁻¹ of phenylacetaldehyde per mg L⁻¹ of phenylalanine present, while just 3.0 μg L⁻¹ per mg L⁻¹ were observed in aged wines. Conversely, for 2-methylbutanal, aged wines accumulated 1.3 μg L⁻¹ of aldehyde, while young wines just 0.17 μg L⁻¹ per mg L⁻¹ of amino acid. If such difference was exclusively due to the effect of ARPs, that would mean that in young wines, more than 85% of aldehyde formed has been removed by reaction.

Levels of diacetyl are negatively and significantly (P < 0.05 in all cases) correlated to the accumulation of non-aromatic Strecker aldehydes. This may suggest that diacetyl reacts with the amino acid precursor to form different species. i.e., although diacetyl can induce the Strecker degradation of amino acids (Rizzi, 2006), it can also form a wide range of heterocyclic products belonging to chemical classes such as oxazoles, pyrazines, pyrroles, or pyridines (Piloty and Baltes, 1979; Pripis-Nicolau et al., 2000).

Further clues about the nature and relative importance of the chemicals involved in the accumulation of Strecker aldehydes can be extracted from the PLS models summarized in Table 5. The eight models for non-aromatic aldehydes closely follow the same structure, and in all cases attribute a positive weight to the amino acid, higher at high levels of oxygen (Table 5A), and negative roles to anthocyanins, which would act as ARPs. Some other elements in the models are worth mentioning. First, in the case of isovaleraldehyde in Table 5A, the negative role of diacetyl, which would compete for the amino acid, as previously suggested. Second, iron has positive coefficients in some cases, suggesting that it may play a role as catalyst. Third, in the medium oxygen exposure (Table 5B), the negative coefficients for procyanidins and catechin in tannins suggest that they act also as ARPs.

Models for phenylacetaldehyde are completely different. First, because anthocyanins do not have any weight, and second, because it is the single aldehyde for which tannins have positive weight. This may suggest that quinones derived from prodelphinidins and catechin in tannins would be particularly reactive toward phenylalanine yielding phenylacetaldehyde. This is consistent with the observation that o-quinones at pH 7 with K₃Fe(CN)₆ favor Strecker production of phenylacetaldehyde over methional (Rizzi, 2006). The higher levels of these quinones present in young wines, together with the negligible reactivity toward ARPs, would explain the higher formation of this aldehyde in young wines. A third specificity of phenylacetaldehyde is the large and positive coefficient taken by diacetyl in Table 5B, suggesting that in this particular case, diacetyl is one of the α-dicarbonyls inducing the Strecker degradation of phenylalanine. The distinct pattern followed by phenylacetaldehyde has been already observed (Rizzi, 2006; Grant-Preece et al., 2013).

Models in Table 5 suggest that fusel alcohols are less relevant than amino acids as precursors for Strecker aldehydes, in agreement with previous studies (Grant-Preece et al., 2013; Ferreira et al., 2014; Bueno et al., 2016). The potential relevance of fusel alcohols can be further assessed by estimating the proportion of alcohol oxidized, assuming that they will be oxidized at extents similar to those of ethyl alcohol. The molar fractions of ethanol oxidized were estimated from the expected levels of H₂O₂ produced and are given in Table 6. As can be seen, ethanol oxidizes by percentages which range between 0.003 and 0.045%, depending on the alcoholic degree of the wine and on the oxygen exposure. If equivalent amounts of isobutanol, 2-methylbutanol, isoamyl alcohol, methionol, and β-phenylethanol have been oxidized to form the corresponding aldehydes, data show that only in the cases of isobutyraldehyde, 2-methylbutanal and isovaleraldehyde, the direct oxidation of the precursor alcohol can be a significant source of those aldehydes. Taking into account, however, the large fraction of aldehyde removed by reaction with ARPs and the fact that levels of alcohols tend to be less variable than those of amino acids, the contribution of the alcohols to the specific ability of a sample to form these aldehydes should be quite limited. In the case of methional, the oxidation of methionol may account just for a marginal amount of the aldehyde accumulated, in accordance with previous observations (Bueno et al., 2016). In the case of phenylacetaldehyde and taking into account its low reactivity toward ARPs, the oxidation of the alcohol as source of aldehyde should not be discarded, particularly in wines containing high amounts of this alcohol.

Attending to the previous discussion, the patterns of accumulation of Strecker aldehydes (Figures 4C,D) can be explained attending to the availability of reactive quinones able to induce the Strecker degradation of the amino acid precursors and to the presence of ARPs. In the first stage of the oxidation, quinones surely will be poorly available because they will be reacting fast with SO₂ or with other competing nucleophiles, so that Strecker degradation of amino acids will not be intense. In addition, the few molecules of aldehyde formed will react with ARPs, so that non-aromatic Strecker aldehydes do not accumulate, as seen in Table 2 and Figure 4C. In contrast, in the case of phenylacetaldehyde (Figure 4D), which is the least reactive to ARPs, clear increments of the aldehyde in older wines can be observed even in the low O₂ exposure conditions. The reasons why those increments are not observed in young wines may be related to the highest levels of ARPs of these wines, which would mean that reactivity of phenylacetaldehyde toward ARPs contained in young wines is not null. At medium O₂ exposure levels, in which SO₂ becomes poorly available, both plots suggest that there is an intense degradation of amino acids through the Strecker pathway. In the case of non-aromatic aldehydes increases are limited because of the presence of ARPs and because in these cases diacetyl, which is now released from its SO₂ adducts, would be competing for the amino acids. On the contrary, in the case of phenylacetaldehyde, the huge increases observed in young wines in this specific range of oxygen, should be attributed to the lower fraction consumed by ARPs, to the potential extra contribution of diacetyl as reactive α-dicarbonyl and to the possible specific Strecker degradation of phenylalanine by quinones derived from catechin in tannins. Finally, the flat part of the plot in Figure 4D may be due to the exhaustion of the amino acid precursors.

A quite important corollary of all the previous observations and hypotheses, is that regarding the temporal pattern of accumulation of aldehydes, Strecker aldehydes accumulate long before acetaldehyde does. This can be seen in Figure 8, which compares the evolution during oxidation of acetaldehyde, methional and phenylacetaldehyde in a young (HV) and an aged wine (BL). The highest difference between acetaldehyde and Strecker aldehydes is found in the second oxidation stage and is particularly evident in the young wine (filled lines). It can be clearly appreciated that while acetaldehyde decreases, the levels of Strecker aldehydes peak up. In aged wines, even if in the low and high O₂ exposure levels the three aldehydes increase in parallel, increases of Strecker aldehydes in the medium exposure levels are much higher than those of acetaldehyde. What this implies in sensory terms can be best assessed by estimation of the corresponding Odor Activity Values, as seen in Figure 9. It can be seen than in methional and phenylacetaldehyde reach much higher OAVs than acetaldehyde, and more important, even at low oxygen exposure they reach OAVs above 20, when acetaldehyde is barely detectable.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The handling Editor declared a past co-authorship with one of the authors, VF.