This study was carried out in the winegrowing area of Franciacorta DOCG, a famous Italian sparkling wine production area located in the Lombardy Region. The experiment was repeated for four consecutive years (from 2010 to 2013) in a vineyard (45.572850° Lat, 9.967622° Lon) belonging to Azienda Agricola Castello Bonomi Tenute in Franciacorta. This company is located in the southern part of the Franciacorta region (Supplementary Figure 1).

The soil of experimental vineyard is grass-covered and characterized by a loamy texture, high total organic carbon content and CEC (Cation Exchange Capacity). The vineyard, with row orientation from north to south, was planted in 2004 with two international Vitis vinifera L. cultivars, Chardonnay (clone ENTAV-INRA^® 96) and Pinot noir (clone 292), both grafted onto Kober 5BB rootstock. Both cultivars are cordon trained.

The experimental plan included five different treatments: a control test without defoliation and shading (ND), a test where about six basal leaves, equal to about 35% of total leaf area, were removed, east and west side (TD), and three different systems adopting shading nets applied along the bunch zone; two of the shaded treatments were defoliated as for TD and covered with one layer of shading net (TD1) or two layers of shading net (TD2), while a third treatment was covered by only one layer of shading net, but not defoliated (ND1L).

The treatments were replicated on both the Chardonnay and Pinot noir cultivars. All treatments were set in two replicates represented by two rows of the vineyard for each cultivar. Each treatment was extended for 25 vines for each of the two replicates. All treatments took place at about 20% veraison, corresponding to July 19th, July 14th, July 19th, July 29th, respectively, in 2010, 2011, 2012 and 2013. Both manual leaf removal and shading net application were carried out along the bunch zone; shading was realized through a UVstabilized polyethylene net of approximately 95 g m^-2 (shading net OF50N supplied by Retes srl). The transmittance of global solar radiation of the single layer and double layer nets was preliminarily tested in order to evaluate the percentage of global solar radiation passing through the nets. A reduction by 50% and 70% was detected for single- and double-layer shading nets, respectively.

During the four years, weather conditions were monitored by means of the weather station of Rovato (BS) of the Agrometeorological Network of the Province of Brescia, which is located at a distance of 5 km from the experimental vineyard.

To provide a more precise description of the thermal conditions in the experimental vineyard a four-channel data logger (Onset HOBO U12) equipped with an air temperature sensor (thermistor Onset HOBO S-THC), protected by a solar shield was placed in the vineyard.

Precipitation data recorded by the Rovato weather station and temperature data from the vineyard thermometer were used to characterize the general meteorological features of the four years of investigation and to calculate thermal indices, the grapevine water balance and the derived water stress indices, as described further below.

With the aim to compare musts and wine composition at a standard potential alcohol degree, for both cultivars and all treatments, the harvesting time was established on the basis of the strategy adopted by the wine company for the specific year and ranging from 10 to 10.7% of potential alcohol.

The evolution of ripening for both Chardonnay and Pinot noir was monitored weekly (maintaining an interval of 7-5 days between each sampling), in order to identify the relevant harvesting time for each treatment.

Experimental harvesting was organized by selecting 5 grapevines from the two 25 vine replicates for each treatment, for a total of 10 vines per treatment (ND, TD, TD1, TD2 and ND1L). For each grapevine, the average bunch weight (ABW) was calculated counting the number of bunches and weighing the entire grapevine production. A sample of three bunches was collected from each grapevine to check must quality. These samples were then crushed and the total soluble solids concentration (TSS), pH, titratable acidity (TA) and malic acid (MA) concentration were analyzed in the resulting musts. These measurements were determined using a traditional handheld refractometer for soluble solids concentration, a Crison compact titrator analyser for both pH and TA, and the enzymatic method (Hyperlab wine analyser) to determine MA concentration.

About 100 kg of grapes were harvested for each cultivar and treatment upon reaching the standard potential alcohol. Grapes were harvested in small boxes weighing maximum 15 kg, and stocked in a cold room at about 10°C. Microvinifications was then initiated according to a standardized protocol. The wine-making protocol was organized in different steps described in Supplementary Figure 2.

The wine so obtained was stocked in 15 l demijohns and periodically monitored for free sulfur content, using the WineScan™ SO2 (FOSS) instrument, to avoid undesired malic fermentation. After about five months the wine was stabilized by adding 10 ml hl^-1 silica sol and 1 ml hl^-1 gelatin. The wine was then bottled using 0.75 l bottles (for a total of 18 bottles); nine bottles were then stocked and 0.3 g l^-1 of yeast and 22 g l^-1 of sugar were added to the other nine bottles to induce refermentation, thus allowing to produce a final sparkling wine. After this last step, all bottles were stocked and manually disgorged, after a manual remuage to remove the residue of yeasts without any further addition. Disgorgement was performed for all wine samples at the time of the volatile aroma compounds analysis in 2016 and 3 different bottles for each treatment were used as replicates.

All the analyses were carried out following the method described in our previous works (Carlin et al., 2016; Carlin et al., 2022). In order to monitor instrument stability, pooled quality controls (QC) consisting of equal proportion of each sample were placed at the beginning of the run (n=5) and thereafter every 10th sample as a common practice in metabolomics studies (Arapitsas et al., 2016). For each sample, 2 ml of wine with 50 μl of internal standard solution (2-octanol in ethanol at a concentration of 1 mg l^-1) was mixed with 1 g NaCl in a 20 ml headspace vial. A Gerstel MultiPurpose Sampler autosampler (Gerstel GmbH & Co. KG Mülheim an der Ruhr Germany) with an agitator and SPME fiber 2 cm (50/30 DVB/CAR/PDMS), from Supelco Merk KGaA (Darmstadt, Germany), was used to extract the volatiles from the sample vial headspace. The sample was pre-incubated for 5 min at 35°C. Adsorption lasted 20 min, at the same temperature, Then, desorption took place in the injector in splitless mode (4 min) at 250°C. The fiber was then reconditioned for 10 min at 250°C. The GC×GC system was the Agilent 7890 A (Agilent Technologies, Santa Clara, CA). Injections were performed in splitless mode. Equipped columns were the VF-Wax column (100% polyethylene glycol; 30 m × 0.25 mm × 0.25 μm, Agilent J&W Scientific Inc., Folsom, CA) as the 1^st dimension and Rxi-17Sil MS 1.50 m × 0.15 mm × 0.15 μm, Restek Bellefonte, USA) as the 2^nd dimension. The GC system was equipped with a secondary column oven and non-moving quadjet dual-stage thermal modulator. The injector/transfer line was maintained at 250°C. Oven temperature program conditions were as follows: initial temperature of 40°C for 4 min, programmed at 6°C min⁻¹ at 250°C, where it remained for 5 min. The secondary oven was kept at 5°C above the primary oven throughout the chromatographic run. The modulator was offset by +15°C in relation to the secondary oven; the modulation time was 7 s and 1.4 s of hot pulse duration. Helium (99.9995% purity) was used as carrier gas at a constant flow of 1.2 ml min^-1. The MS signal was obtained with a Pegasus IV time-of flight mass spectrometer (Leco Corporation, St. Joseph, MI) with electron ionization at 70 eV and the ion source temperature at 230°C, detector voltage of 1317 V, mass range of m/z 35–450 and acquisition rate of 200 spectra s⁻¹.

For GC×GC-MS data, LECO ChromaTOF Version 4.22 software was used for all acquisition control, data processing and Fisher ratio calculations. Automated peak detection and spectral deconvolution with a baseline offset of 0.8 and signal-to-noise of 100 were used during data treatment. Before proceeding with the data analysis a quality control of the data sets, checking the distribution of the QC injections was carried out (Arapitsas et al., 2016). The identification of the wine volatile compounds was achieved by comparing the mass spectrometric information for each chromatographic peak with NIST 2.0, Wiley 8 and the FFNSC 2 mass spectral library (Chromaleont, Messina, Italy), with a library similarity match factor of 750 and comparing also the experimental linear temperature retention index (LTPRI) with retention indices reported in the literature for 1D-GC (VCF Volatile Compounds in Food 16.1 database). Retention data for a series of n-alkanes (C10–C30), under the same experimental conditions employed for chromatographic analysis of wine volatiles, were used for experimental LTPRI calculation. The content of norisoprenoids were expressed as peak areas.

In the statistical analysis, Chardonnay and Pinot noir were kept separated in relation to the different composition of their grapes and their different sensory characterization in wines (Herrero et al., 2016).

The effect of the year was initially investigated for variables related to ABW and musts quality through an analysis of variance (ANOVA) and Siegel-Tukey’s post-hoc test using the SPSS software (Statistical Package for Social Science) (IBM Corp, 2021).

The relationships between year and agrometeorological indices and AWB and must quality were also investigated through a principal component analysis (PCA) on autoscaled data (i.e., mean-centered and divided by the standard deviation of each variable), through R software (R Core Team, 2022). In the PCA analysis for each cultivar, five points for each year, corresponding to the average value of the single treatment in a specific year, were considered. This approach was used because one single value of agrometeorological indices was available for each year without having the possibility to differentiate between treatments.

Additional ANOVA and Siegel-Tukey’s post-hoc test were performed to determine the influence of the treatment on variables related to ABW and musts quality (TSS, TA, MA, pH). In this analysis, both the cultivars and the four years were considered separately. The decision to process the different years separately was determined by the meteorological differences in the individual years and the effect on the AWB and must quality revelead by the results obtained from the analysis of the relationship between year and these variables.

Considering the limited number of biological replicates used for the untargeted volatile compounds analyses (i.e., 3 replicates for each cultivar, treatment, and year) a different approach was outlined. In this case, it was checked whether there were any statistically significant differences associated with each treatment compared to the untreated thesis (ND) by means of a linear model on log-scaled data using R software (R Core Team, 2022).

Table 1 shows the monthly variation of temperature and precipitation by season, respectively, while the agrometeorological indices are shown in Table 2.

The year 2010 was characterized by optimal water conditions. The scarce precipitation of the winter period (Jan-Mar) was preceded by good precipitations in the 2009 Oct-Dec period and followed by abundant spring precipitations. From a thermal point of view, 2010 showed lower maximum temperatures during spring than 2011 and 2012, resulting in the highest NHH value in the series. However, the indices for this year were strongly affected by a longer interval between fruit set and harvest, than in the other three years. Low temperature stress (LHH) was the second highest over the four years and high temperature stress (HHH) was the highest. The 33°C limit was exceeded for 65 days during the season.

The year 2011 was still characterized by very low water stress, thanks to the refilling of the water reservoir during winter and the good distribution of precipitation over the months. As compared to other years, low temperatures caused a high level of NHH (second highest), the highest value of low temperature stress (LHH) and the lowest value of high temperature stress (HHH), confirmed by only 2 days with maximum temperatures above 33°C.

The year 2012 was characterized by the highest water stress value over the four years (as shown by WSTR and ΔET values), mainly due to the limited precipitation during June, July and August that led to a long period of a limited water soil content. From a thermal point of view, the year 2012 featured the lowest levels in terms of NHH accumulation (though very close to 2013), and LHH and the highest level in HHH (thermal excess, with 93 days of maximum temperature above 33°C).

The year 2013 was very similar to 2012, with a certain amount of water stress (second highest in both indices), the second lowest in NHH and LHH, and the second highest in HHH and the first in the number of days with a maximum temperature above 33°C (99 days).

Considering the thermal behavior in the four seasons in relative terms, in 2010 the percentages of NHH, LHH and HHH versus their total (NHH+LHH+HHH) were 63.85, 18.23 and 17.92 respectively. 2012 and 2013 showed a similar behavior with 62.85, 17.78, 19.38 and 62.81, 18.41, 18,77 respectively. The year 2011, with 65.41, 26.62, 7.97 showed a strongly different repartition, again emphasizing its unique environmental conditions.

The results obtained in this study highlight the influence of temperature and water availability on grapevine yield, musts quality and on the composition of aromas in wine. This confirms the results obtained by other authors that underlined the effect of meteorological conditions on grape and wine quality (Mariani et al., 2009). The average weight of the bunches was higher when water needs were satisfied; 2010 and 2011 were, indeed, characterised by lower water stress indices (WSTR and ΔET) and the highest values of ABW. The grapevine yield positive response to higher water availability was confirmed by other authors (Ramos and Mulligan, 2005; Ramos and Martínez-Casasnovas, 2006; Intrigliolo et al., 2016; Pérez-Álvarez et al., 2021; Ohana-Levi et al., 2022). On the other hand, water scarcity in 2012 lead to a reduction of the bunch weight, causing in turn the typical phenomena of sugar and acidity concentration (Intrigliolo et al., 2016; Pérez-Álvarez et al., 2021) and pH level increase (Reynolds and Naylor, 1994).

Previous studies reported how the ripening of grapes for sparkling wine production was positively influenced by a cool climate (Jones et al., 2014). The optimum temperature for the accumulation of malic acid during grape ripening was estimated to be around 20°C (Lakso and Kliewer, 1975), while the negative correlation between high temperatures and MA content after véraison was reported by many studies (Rienth et al., 2016; Blank et al., 2019). Preserving malic acid degradation made it possible to contrast the decrease of the acidic level in must (Volschenk et al., 2006), which is a desirable condition for sparkling wine production (Jones et al., 2014). In the present study, the positive effect of cooler temperatures on must quality was revealed in 2011, when high values of indicators of suboptimal temperature and optimal temperature were recorded. This led in 2011 to an early ripening (1 August – 12 August), while preserving the must sugar-acid balance, and a low pH. On the contrary in 2010, where similar values for meteorological indicators were reported, the reference sugar level was reached later (24 August - 6 September), leading to lower acidic levels, and higher pH levels. Moreover, the year 2010 reported the highest total amount of cumulated precipitation between berry set and harvest. The low level of titratable acidity and the high value of pH in musts can therefore be related to the effect of dilution of acidic components (Keller, 2006) and the increase of K absorption (Marais et al., 1992) caused by water absorption. The negative effect of high temperatures on the must acidic content was evident in 2013, when a value of the stress caused by heat waves (33STR) was recorded. This was particularly evident in the case of malic acid in Chardonnay musts that showed low values during this vintage. Many previous studies reported the increase of malic acid degradation kinetics when high temperatures were recorded (Michelini et al., 2021).

Leaf removal and artificial shading effects were differentiated by year and cultivar. In general, years characterized by satisfying water availability and cooler temperatures (i.e., 2010 and 2011) emphasized the difference in ripening delay between the non-artificial shaded and the artificial shaded treatments. This is in agreement with previous papers that reported how shading makes it possible to slow down grape maturation (Caravia et al., 2016; Ghiglieno et al., 2020). The effect of no defoliation and artificial shading on acidity preservation was more evident in years characterized by high temperatures and high-water stress (i.e., 2012 and 2013). This can be related to the positive relationships between malic acid degradation and high temperature and water scarcity (Blank et al., 2019) that suggest the need for shading during warm and dry vintages. In the case of ABW and pH there was no unique response.

The analysis of the main norisoprenoids showed how the canopy treatments greatly affect their quantity (boxplots in Supplementary Figures 3, 4). A general increasing effect of total defoliation can be highlighted, especially concerning Chardonnay. Even though Oliveira et al. (2004) reported that carotenoid contents were consistently higher in grapes in the shade than in those exposed to direct sunlight, the positive effect of exposure on norisoprenoids was underlined by many previous studies (Marais et al., 1991; Marais et al., 1992; Asproudi et al., 2016; Asproudi et al., 2020). The results obtained in the present study, in terms of TDN content in wines, showed that this aromatic compound, even if not always in a statistically significant way, was produced in larger quantities in TD while the shaded theses TD1 and TD2 contained less TDN; this behavior was clearly visible in Chardonnay and in two vintages of Pinot noir, and would suggest that it is the light that most influences the biosynthesis of this compound. TDN with a very low olfactory threshold (2 µg l^-1), can sometimes represent a problem, especially in sparkling wines intended for a long period of aging, because it increases over time. It has been seen that the olfactory threshold of TDN in sparkling wine is slightly higher and goes from 2.1 µg l^-1 in still Riesling to 4.0 µg l^-1 in sparkling Riesling (Ziegler et al., 2019); however, this is a compound to pay attention to when deciding on agronomic practices. Several glycosylated precursors have also been reported to originate these compounds during fermentation and wine ageing, through acid-catalysed reactions, and these reactions are certainly promoted in a wine with a low pH, such as sparkling wine (Schneider et al., 2001; Janusz et al., 2003).

The content in vitispirane was always much higher in TD theses and also in this case, although in Pinot noir there was no statistical proof, the defoliated and shaded theses presented a lower content of this compound. The olfactory descriptor of vitispirane is “camphor” or “eucalyptus”. Some studies (Silva Ferreira and Guedes de Pinho, 2004) showed that in aged wines the presence of vitispirane can reach the threshold of perception (800 µg l^-1) and, consequently, that it participates in the aroma of wine. As with the TDN, pH and temperature are the two factors that have the greatest influence on its formation over time.

(E)-1-(2,3,6-trimethylphenyl) buta-1,3-diene (TPB-1) was tendentially higher in the TD and less present in the shady theses TD1-TD2; this compound, like TDN at high concentrations, could lead to an unpleasant chemical note at higher content (40 ng l^-1 sensory threshold) (Janusz et al., 2003). The behavior of actinidols and their ethoxy forms was very similar to that seen previously with a higher content in TD, a little lower in the shaded theses and even lower in the not defoliated ones. For this compound the situation was very clear in both Chardonnay and Pinot Noir. Actinidols have an odor that has been described as camphoraceous or as woody and resinous, but their contribution to wine aroma is considered limited at best, as their concentrations are usually much lower than their detection threshold. Perhaps the corresponding ethyl ethers that have been found with the GC-O technique by Schneider et al. (2001), with fruity, citric and eucalyptus notes offer a greater contribution to the aroma.

Another compound that has been increasingly found in TD treatments and less in the other not defoliated and shaded ones was safranal. This compound is responsible for the characteristic smell of saffron and can in some cases exceed the olfactory threshold and be considered too intense, leading to possible depreciation in sparkling wine. It is known that the principal monoterpene glycoside precursor of safranal in saffron is picrocrocin (Zougagh et al., 2006). We don’t know if picrocrocin is a precursor also in wine or if safranal is formed starting from some other precursor or by the rearrangement of some other molecules, but previous studies reported that the content in this compound tends to increase over time, especially at low pH and if the temperature during the storage of the wine is high. Very similar trends in both Chardonnay and Pinot noir, with high levels in the TD thesis and less content in other thesis, were also observed for an unidentified norisoprenoid. On the other hand, we did not observe any clear treatment effects for β- damascenone.