1. Vineyard blocks and vine panels

The characteristics of the 10 commercial dryland vineyard blocks used for this study are shown in Table 1. The plots were selected to capture a wide range of soil and climatic conditions, with the aim of maximizing the range of rotundone concentrations found in the blocks. The 10 blocks were located within the Gaillac protected designation of origin, which covers a surface area of approximately 1000 km².

At each block, one 30-vine panel (comprising 10 consecutive vines from three adjacent rows) was selected for data collection and sampling. Vine management was performed by growers, except for fruit-zone leaf removal, which is a practice that is likely to affect rotundone concentration (Geffroy et al., 2014; Homich et al., 2017; Geffroy et al., 2018b). To avoid bias due to differences in the timing and extent of mechanical leaf removal between blocks when the practice had to be performed by growers, all the leaves were removed manually from the basal node to the first node above the top bunch on the morning-sun side of the canopy at mid-veraison, when rotundone starts to accumulate in berries (Caputi et al., 2011). Row orientation was determined using Google Earth Pro software version 7.3.0.3832 (Google Inc., Mountain View, CA, USA). Elevation and slope data were extracted from the RGE ALTI 5 m device, provided by the French National Institute of Geographic and Forest Information.

2. Phenology and vine measurements

For each block, the percentage of veraison was determined by randomly collecting 100 berries from both sides of the row and also several parts of the bunch, and by counting the number of colored berries every third day from the end of July to the end of August. Linear regression tests were performed using Excel software (Microsoft Corp., Redmond, WA, USA) to calculate the onset and end of veraison, the mid-veraison date, and the duration of veraison.

Exposed leaf area (m²/m² soil), which is the leaf area that can be exposed to direct solar radiation, was estimated at mid-veraison using the protocol proposed by Murisier (1996). Measurements were made for six consecutive vines on each of the three rows constituting the 30-vine panel. The ratio between canopy height and row spacing, which is an indicator of the shade of one vine row provided by an adjacent row (Smart, 1985), was also calculated.

For each sampled vine, the number of bunches per vine was counted and the yield at harvest (kg/vine) was monitored by weighing individual crop loads, using a Precia Molen C20 K balance (Precia SA, Privas, France). Mean bunch weights were also calculated for each block.

Leaf area to crop weight ratio, which is an indicator of vine balance and ability to produce high-quality fruit and wines (Kliewer and Dokoozlian, 2005) was calculated as follows:

leaf area to crop weight ratio = [exposed leaf area (m²/m² soil) × 10 000] / [yield at harvest (kg/vine) × plant density (vines/ha)]

In 2014 only, powdery mildew (Erysiphe necator) infections were detected on bunches from vines of one of the blocks. For this vintage, the severity of the infection was assessed at harvest by visually estimating the percentage of the bunch area colonized by the pathogen on 100 grape bunches (50 on each side of the vine). In 2013 and 2014, the severity of bunch rot (Botrytis cinerea) was assessed by determination of the gluconic acid content of the grapes (see the section Berry sampling and analyses), which correlates with the visually estimated percentage of rotten grapes (Bocquet et al., 1995).

Pruning wood weight (kg/vine) was estimated by weighing 1-year-old shoots of 10 out of 30 vines in January 2014 and 2015, using a Precia Molen C20 K balance.

3. Weather measurements and water-balance model

Daily high-resolution data provided by Météo France (Toulouse, France), the French weather service, were obtained using ANTILOPE (Champeaux et al., 2011) for rainfall and AROME (Brousseau et al., 2011) for air temperature and evapotranspiration. Resolutions were 1 km² and 4 km² for ANTILOPE and AROME, respectively. These data were used to calculate the average minimum, maximum and mean daily air temperature for each block between veraison and harvest; to calculate the Huglin index over the period from 1 April to 30 September (Tonietto and Carbonneau, 2002); to determine cumulative rainfall over different periods; and to run a water-balance model.

The percentage of soil covered by grass differed greatly between the blocks, and the water balance for intercropped systems (WaLIS) model was used to simulate the stem water potential (ψ_stem) of the vine (Celette et al., 2010; Dufourcq et al., 2013) at different time points between 15 days before mid-veraison and harvest for each block. The minimal and maximal values of ψ_stem over the mid-veraison to harvest period were also identified for each block. Calibration of the model was carried out in 2013 and 2014 based on three measurements of ψ_stem that were obtained between the beginning of July and the end of August for each block. Each measurement was taken from 3 of the 30 vines and from two mature exposed leaves per vine, according to the protocol proposed by Choné et al. (2001). These three vines were marked, and ψ_stem was always measured on these selected plants. On cloudless days, leaf blades were initially enclosed in plastic bags between 11:00 and 12:00 hours, and measurements were made between 14:00 and 15:00 hours.

The hours of sunshine, solar irradiation, and hygrometry data over the mid-veraison to harvest period, with a 9-km² spatial resolution, were obtained for each block from HelioClim-3, a satellite-based surface solar irradiation database (Blanc et al., 2011; Eissa et al., 2015).

In April 2015, just before budburst, trunk circumference, which is an index of vine vigor that correlates well with the rotundone concentration in vines within the same block (Geffroy et al., 2015b), was determined for every vine within each block. We used the average of measurements obtained at three different heights: just above the graft, 10 cm under the trellis wire, and halfway between the two first points of measurement.

4. Berry sampling and analyses

Previous modeling of the kinetics of rotundone accumulation in berries has shown that the concentration increases from veraison and reaches a plateau after a certain time (Zhang et al., 2015b). According to Zhang et al., this plateau is higher and is reached earlier during cool and wet seasons. In 2011 and 2012, two hot and dry vintages that did not promote high rotundone concentration in the grapes of vines grown in the south-west of France, this plateau was achieved for Duras wines 44 days after mid-veraison (Geffroy et al., 2014).

To collect grapes with the maximum rotundone concentration, each block was harvested in 2013 and 2014 exactly 44 days after mid-veraison. For each block and for each year, two samples, consisting of 200 berries from both sides of the row and several parts of the bunch (four berries per vine on each side of the row), were collected.

Before the start of the analysis, the mass of the 200 berries in each sample was measured. The berries were then crushed gently, and the musts obtained were centrifuged (14,000 g for 6 min) to enable conventional laboratory and δ¹³C analyses. Sugar concentration (°Brix) was determined with a PAL digital handheld pocket refractometer (Atago, Tokyo, Japan). Titratable acidity was measured according to the methods of the Organisation Internationale de la Vigne et du Vin (2009), and pH was measured with an HI 3221 pH meter (Hanna Instruments, Woonsocket, RI, USA).

A Konelab Arena 20 sequential analyzer (Thermo Fisher Scientific, Waltham, MA, USA), in conjunction with enzyme kits provided by several suppliers, was used to determine the concentration of amino acids, ammonium, gluconic acid (Megazyme, Bray, Ireland) and malic acid (Thermo Fisher Scientific). Potassium concentration was determined by flame photometry (Bio Arrow, France) according to the methods of the Organisation Internationale de la Vigne et du Vin (2009), and the concentration of tartaric acid was determined by colorimetric titration (Hill and Caputi, 1970). δ¹³C, a proxy for vine water status (van Leeuwen et al., 2009) was determined at harvest according to the protocol described by Geffroy et al. (2014). All the analytical determinations were carried out twice.

The berry skins that had not been used for the previous analyses were crushed for 2 min at low speed (600 rpm) using a Faciclic food blender (Moulinex, Ecully, France), in preparation for phenolic analyses. The concentration of anthocyanins and the total phenolic index of the grapes were quantified according to the method described by Cayla et al. (2002). In 2014 only, samples of 240 berries were also collected from the 10 blocks at mid-veraison to determine the weight of 200 berries, δ¹³C, sugar concentration, titratable acidity, pH, concentration of malic acid and amino acids, and ammonium content, according to previously described methods.

5. Microvinification and rotundone analysis

Rotundone content was analyzed indirectly in wines fermented under microvinification conditions. For each block and for each year, two grape samples, comprising 800 g of berries from both sides of the row and several parts of the bunch, were collected at harvest 44 days after mid-veraison. Microscale fermentation was carried out in a 1-L Erlenmeyer flask, according to the protocol described by Geffroy et al. (2014). After 8 days of fermentation, the amount of wine and of skins was measured to determine the skin-to-juice ratio at pressing (%). The wines were centrifuged (14,000 g for 6 min), and sulfite (80 mg/L) was added before bottling and storage at 4°C or less until the rotundone analysis.

The rotundone concentration of the wines was determined by solid-phase microextraction-multidimensional gas chromatography–mass spectrometry, according to the protocol described by Geffroy et al. (2014). The analyses were performed in two different years but during the same period of each year, to reduce potential variation associated with different post-bottling times.

6. Data treatment

Details of the variables for which data were collected are shown in Table 2. Rotundone concentration and data for the other variables, collected for the 10 blocks over the 2 years of the study, were subjected to analysis of variance, with the vintage as a factor and blocks as repetitions, using Xlstat software version 19.5 47062 (Addinsoft, Paris, France). Fisher’s least significant difference test was used as a post-hoc comparison of means at P ≤ 0.05.

Based on previous work conducted with the aim of modeling 3-isobutyl-2-methoxypyrazine concentration in Cabernet Franc grapes (Scheiner et al., 2011), a multivariate analysis using partial least squares regression (PLSR) was carried out to identify key factors affecting rotundone concentration in wine. The data were averaged over the replicates before PLSR was performed. For the multi-vintage model, variables that had been determined only in 2014 were discarded. A first set of PLSR for which the number of latent variables was determined by the lowest predicted residual sum of squares was conducted, using Xlstat software to model rotundone concentration in 2013, in 2014, and in both vintages (Wold et al., 1984). Variables with a variable importance in projection score below 1 were removed manually and the model was regenerated. Once the number of variables remaining in the model was lower than the number of observations (10 for 2013 and 2014, and 20 for the multi-vintage data set), PLSR was conducted with Unscrambler X software version 10.4 (Camo, Oslo, Norway).

The results were validated by a full cross-validation (leave-one-out) procedure. Variables that did not contribute sufficiently to the model (absolute values of the loading weights below 0.1) were removed manually and the model was regenerated. For the multivintage model, one outlier (block no. 6 in 2013) detected using principal component analysis was removed.

1. Main differences between the vintages

The 2013 vintage was characterized by a cold spring and regular and heavy rainfall events over the vine vegetative cycle that resulted in severe bunch rot (Botrytis cinerea) damage, ripening difficulties and delay of harvest. The mean harvest date for the 10 blocks was 6 October ± 4 days. In 2014, the winter was rainy and warm and summer was extremely rainy between mid-July and the end of August, whereas conditions during ripening were dry and hot. The mean harvest date was 22 September ± 3 days. Compared with 2014, the 2013 vintage was significantly wetter as well as cooler over the entire winegrowing and ripening periods, as reflected by the cumulative rainfall between 1 April and 30 September (423 ± 22 mm in 2013, 358 ± 19 mm in 2014) and between veraison and harvest (115 ± 18 mm in 2013, 63 ± 13 mm in 2014), the Huglin index (1862 ± 51 in 2013, 1969 ± 48 in 2014), and the mean air temperature over the veraison-to-harvest period (18.50 ± 0.56°C in 2013, 19.48 ± 0.42°C in 2014). The values of ψ_stem were significantly lower at mid-veraison during the first year of the study (–0.87 ± 0.20 MPa in 2013, –0.60 ± 0.12 MPa in 2014). Owing to the larger quantities of rainfall received in 2013 over the veraison-to-harvest period, the level of water deficit was higher at harvest in 2014 (–0.51 ± 0.17 MPa in 2013, –1.06 ± 0.14 MPa in 2014).

Vintage had a weak effect on crop and grape analytical characteristics, because no significant differences in yield (1.75 ± 1.01 kg/vine in 2013, 2.65 ± 1.22 kg/vine in 2014), sugar concentration (21.8 ± 0.9°Brix in 2013, 21.8 ± 0.6°Brix in 2014), pH (3.06 ± 0.12 in 2013, 3.05 ± 0.07 in 2014), titratable acidity (9.20 ± 1.13 g/L as tartaric acid in 2013, 9.10 ± 1.22 g/L as tartaric acid in 2014), amino acid concentration (120 ± 33 mg/L in 2013, 136 ± 48 mg/L in 2014) or total phenolic index (79.6 ± 10.8 in 2013, 88.5 ± 15.2 in 2014) were observed between 2013 and 2014. Concentrations of anthocyanins (872 ± 143 mg/L in 2013, 1048 ± 194 mg/L in 2014) and ammonium (40.9 ± 20.3 mg/L in 2013, 63.8 ± 20.4 mg/L in 2014) were the only fruit compound variables that differed significantly between the two vintages.

Rotundone concentration was significantly higher (P < 0.001) in the cooler and wetter 2013 growing season (Figure 1). This result is consistent with those of previous studies (Caputi et al., 2011). Surprisingly, some blocks that had high rotundone concentration in 2013 (i.e. blocks no. 2 and no. 8), with values among the highest ever reported for Duras, had low to moderate concentrations in 2014. In the same way, some blocks that had high concentrations in 2014 (i.e. blocks no. 3 and no. 5) had low to moderate rotundone concentrations in 2013. This observation runs counter to the belief that some vineyards within a given wine region may always be prone to yielding wines with substantial amounts of rotundone. This led us to the notion that qualitative or quantitative fixed variables such as year of plantation, elevation, training system, clone or rootstock do not make a substantial contribution to the rotundone model.

Figure 1. Rotundone concentration in wines made from grapes from the 10 vineyard blocks of Duras N in 2013 and 2014.

Save View full size Expand inline Collapse inline

2. Partial least squares regression rotundone models

Basic statistics of the models, regression coefficients, means and standard deviations of the variables included in the best models, are shown in Table 3. Regression plots of predicted rotundone concentration versus observed rotundone concentration in 2013, in 2014, and in both vintages are presented in Figure 2. The performance of the 2013 vintage model, reflected by its values of R² and root mean square error, was better than that of the other models. However, all the models appear to be appropriate for predicting rotundone concentration using a limited number of input variables for blocks with low or high rotundone concentrations.

Figure 2. Predicted versus observed rotundone concentration.

Save View full size Expand inline Collapse inline

Predicted versus observed rotundone concentration, using data from the models described in Table 3: A, 2013 (R² = 0.95); B, 2014 (R² = 0.78); and C, multi-vintage (R² = 0.81).

The loadings of the latent variables used in the regression models are shown in Table 4. The first latent variable in the 2013 model was mainly characterized by gluconic acid concentration. For 2013, this variable was significantly negatively correlated with rotundone concentration (P < 0.01, R² = 0.69) (Figure 3). It also contributed to the multi-vintage model. This observation was somewhat unexpected, because laccase, the polyphenol oxidase produced by Botrytis cinerea, has the ability to convert α-guaiene to rotundone via a non-specific oxidation reaction in the presence of chemical mediators (Schilling et al., 2013). However, it has also been documented that Botrytis cinerea can withstand the toxic effects of plant compounds such as sesquiterpenes by laccase-mediated oxidation and detoxification (Mayer and Staples, 2002). Further studies will be necessary to confirm this hypothesis and to better understand the mechanisms involved in rotundone degradation by the fungus.

Figure 3. Relation between gluconic acid concentration and rotundone concentration.

Save View full size Expand inline Collapse inline

Relation between gluconic acid content and rotundone concentration in wines in 2013 (n = 10). Linear regression model: y = –0.6667x + 204.47, P < 0.01, R² = 0.69.

Other variables contributing to the latent variables of the models were associated with abiotic factors, which indicates that mesoscale climate is the key factor driving rotundone production. Notably, the Huglin index, a thermal index based on a degree-day approach from 1 April to 30 September, was a strong contributor to the 2014 model. This is consistent with the finding that cool vintages promote rotundone accumulation in grapevine berries (Caputi et al., 2011). A contribution to the models from mean air temperature between veraison and harvest would have been expected, because post-veraison berry surface temperature has been identified as the main determinant of rotundone concentration in grape berries (Zhang et al., 2015a).

Bunch temperature does not fully reflect air temperature, because it is also determined by absorbed radiation and convective heat loss (Dry, 2009). The extent to which bunch temperature exceeds air temperature depends on degree of exposure, radiation load, wind velocity, berry or bunch size, berry color and bunch compactness. With the exception of wind velocity, which was not assessed, most of the data for these variables were collected directly or depend on data collected for other variables, such as bunch weight, mass of 200 berries, training system, rootstock, row orientation and exposed leaf area. If bunch temperature were the key variable driving rotundone concentration, the best rotundone model would include air temperature and other variables affecting bunch exposure and penetration of solar radiation through the canopy. It has recently been reported that the percentage of degree hours above 25°C (Dh₂₅) in the fruit zone from veraison to harvest correlates with rotundone concentration in berries (Zhang et al., 2015a). Our study was designed and conducted 2 years before publication of these findings. Dh₂₅ could not be calculated a posteriori, because collection of the data for this variable would have required the use of temperature loggers located within the canopy. We cannot completely exclude the possibility that Dh₂₅ would have been included in the models.

Mean daily irradiation between veraison and harvest had a substantial contribution to the multi-vintage model. Hours of sunshine was also included in the best rotundone models for 2013 and 2013–2014. These two variables, reflecting the quantity of ultraviolet radiation received by the blocks, were positively correlated with rotundone concentration. Several studies have investigated the light dependency of sesquiterpene emission, with evidence that emission of some sesquiterpenes is solely temperature-controlled whereas emission of others is also affected by light (Duhl et al., 2008). Although additional studies would be needed to confirm this hypothesis, our data suggest that ultraviolet rays may stimulate the production of rotundone. This conclusion is consistent with recent research conducted in cool-climate vineyards on Noiret and Fer cultivars, indicating that grape exposure by leaf removal could promote rotundone accumulation (Homich et al., 2017; Geffroy et al., 2018b).

The amount of rainfall between 1 January and 31 December and during the period of vine vegetative growth (between 1 April and 30 September) had a positive contribution to the 2014 and multi-vintage models, whereas rainfall during the late stage of development (i.e. between veraison and harvest) was negatively correlated with rotundone concentration in 2013. These results are somewhat unexpected, because irrigation around veraison, when rotundone starts to accumulate in berries, stimulates its production (Geffroy et al., 2014; Geffroy et al., 2016b). For the 2013 rainy vintage, even if skin-to-juice ratio and mass of 200 berries were not included in the model, we cannot exclude the possibility that the large excess of water may have resulted in a dilution effect through an increase in berry size.

The contribution to the models from cumulative rainfall between 1 April and 30 September and between 1 January and 31 December deserves further comment. Rotundone is found at elevated concentrations in the early berry stages, notably in flower caps (Zhang et al., 2016b). Rotundone concentration then decreases to reach a minimal concentration at veraison, before increasing again from veraison to harvest (Zhang et al., 2016b). It could be assumed that the precipitation regimen over the entire berry development period, not only during the last stage of maturation, may have affected the final concentration of rotundone through degradation or emission before veraison. Another hypothesis is that increased water availability over the period of vegetative growth and between 1 January and 31 December would have led to higher-vigor vines, resulting in greater bunch shading and lower bunch-zone air temperature. However, this hypothesis is unlikely, because the level of water deficit experienced by the vines at mid-veraison, when shoot growth stops, was minor (ψ_stem = –0.87 ± 0.20 MPa in 2013, ψ_stem = –0.60 ± 0.12 MPa in 2014) and insufficient to affect vine vegetative expression (van Leeuwen et al., 2009). Additionally, no significant correlation at P < 0.05 could be established for the whole data set between cumulative rainfall and either pruning wood weight or exposed leaf area.

Consistent with the fact that the hierarchy of blocks for rotundone concentration can differ greatly from one vintage to the other, none of the rotundone models included any fixed variables. This is somewhat unexpected for the clone, because higher rotundone concentrations have been reported in Duras wines from clones 554 and 654 than from clone 555 (Geffroy et al., 2015a). In the same way, the rootstock could have affected the concentration of rotundone in wine. The rootstock is likely to affect some key variables driving rotundone production, such as scion water status and bunch surface, through its effect on vine vigor, leaf area and penetration of solar radiation (Koundouras et al., 2008). However, it may be difficult to draw firm conclusions, due to the small number of observations for each level of these two factors and the fact that not all clone–rootstock combinations were available.

It has been suggested that the grapevine defense response to powdery mildew could enhance rotundone production in berries (Geffroy et al., 2015a). This hypothesis could not be confirmed under in the present study, because severity of powdery mildew infection was not included in the 2014 model. Period of infection by Erysiphe necator was not assessed in that previous study nor in the present study. On the one hand, as rotundone accumulates around veraison, differences in timing of infection, which usually occurs when bunches are receptive (from blooming to veraison), may play a role in defense mechanisms leading to production of rotundone. On the other hand, powdery mildew damage was noticed only in 2014 at just one block, and it remains difficult to draw firm conclusions regarding the contribution of this biotic factor.

Consistent with the findings that rotundone is not translocated from vegetative tissues (Geffroy et al., 2016b; Zhang et al., 2016a) and that its accumulation does not depend on source–sink relations (Geffroy et al., 2014), it should be pointed out that the leaf area to crop weight ratio was not included in any of the rotundone models.

The absence of contribution to the models from sugar concentration at harvest is somewhat unexpected. Although recent studies have shown that an increased alcohol content of wine improves rotundone extraction rate from grapes into wine (Zhang et al., 2017), our data suggest that ethanol concentration in the resulting wines had no effect on rotundone concentration. This may be explained by the relatively small amplitude of difference between the lowest and highest sugar concentrations (20.1°Brix and 22.8°Brix, respectively) recorded between blocks for the 2 years of the study.

We cannot rule out the possibility that some blocks were harvested in 2013 and 2014 before rotundone concentration reached its maximum. This would mean that the proposed models would be valid for prediction of rotundone concentration only in wines from grapes harvested exactly 44 days after mid-veraison and may not be applicable to grapes harvested beyond that time. However, this hypothesis is unlikely, because 2013 and 2014 were characterized by substantial concentrations of rotundone in Duras wines (Geffroy et al., 2016b). Therefore, considering previous findings (Zhang et al., 2016a), the rotundone plateau and the maximum rotundone concentration were expected to have been reached earlier than in 2011 and 2012, and before 44 days after mid-veraison.

With only 2 years of data, we cannot completely disregard the possibility that our model may not be generalizable. This is particularly true because the two vintages of the study were relatively cool and wet for the area, as illustrated by the data for Huglin index and cumulative rainfall between 1 April and 30 September. The 2015 vintage, which was hotter and dryer, was characterized by a different climatic pattern than that of 2013 and 2014. The equation of the multi-vintage model was used to predict rotundone concentration in wine fermented under the same microvinification conditions from grapes collected in 2015 on a Duras plot 44 days after mid-veraison. Based on measurements made on this block for mean daily irradiation between veraison and harvest (1930 W/m²), hours of sunshine between veraison and harvest (419), cumulative rainfall between 1 April and 30 September (348 mm), cumulative rainfall between 1 January and 31 December (516 mm) and gluconic acid concentration (12 mg/L), the predicted rotundone concentration was 30.8 ng/L; the observed value was 26.0 ng/L. In light of this prediction, we believe that the model may be sufficiently robust to also allow estimation of rotundone concentration during hot vintages.