The study areas are in two hilly environments of southern Italy at different latitudes and elevations, Molise (41° 42’; 606 m a.s.l.), and Sicily (37° 40’; 400-700 m a.s.l.), in farms oriented to the production of high-quality wines. The parent material of soil is different between the experimental sites, clayey marls in Molise, and volcanic material in Sicily that led to the development of Vertisols and Andosols, respectively. The studied grapevines were Aglianico (AGL) and Cabernet Sauvignon (CS) cultivars planted between 2008 and 2010 on 140 Ruggieri rootstocks. In Molise, vines were planted in north-south rows, trained according to espalier Guyot, placed in a flat area. In Sicily, both vineyards were bush trained at 0.5 m, with two to six main branches (each branch was spur-pruned to one spur, with two buds per spur), and placed in anthropic terraces.

In each experimental site, the representative soil profile has been described and analyzed. Soil profiles were described according to IUSS (IUSS Working Group WRB, 2014). Chemical analyses were performed according to the official methods of the Italian Ministry of Agriculture and Forestry (Colombo and Miano, 2015). Disturbed and undisturbed soil samples (volume 750 cm³) were collected from each soil horizon to determine chemical and physical soil characteristics. The hydraulic properties were determined in the laboratory according to the procedures reported by Basile et al. (2012) and Bonfante et al. (2010). Climatic characterization of each site was realized on the reference climate dataset 1971-2005 of Italy (spatial resolution about 8 km) realized and optimized on regional high-resolution observational data sets by Bucchignani et al. (2015). This choice has been necessary because there was no available local long-weather dataset reliable near the sites. Weather information (temperature, rainfall, solar radiations, wind speed, air humidity, etc.) was collected using an in-situ weather station (netsense.com) placed in each experimental vineyard within the canopy. Daily reference evapotranspiration (ET₀) was estimated according to the equation of Penman-Monteith equation (Monteith, 1965) and the bioclimatic index of Amerine and Winkler (A&W; Amerine and Winkler, 1944) applied to compare the different thermal regimes.

Berries were collected at the full ripening stage (BBCH stage 89, Lorenz et al., 1995) during the 2020 growing season for biochemical, microbiological, and transcriptional analyses, all at the same time. Thirty clusters were harvested from different positions of the vineyard and from random positions on the plant to ensure the representation of the entire vineyard. At least ten berries were randomly selected from different parts of the cluster, avoiding those with visible damage and/or signs of pathogen infection, and pooled with berries from the other plants. For subsequent analyses, three independent pools (biological replicates) of 50 (for transcriptional analysis) or 100 (for biochemical and microbiological analyses) whole berries were selected, immediately frozen in liquid nitrogen, and stored at -80°C for subsequent analyses. For each analysis, four experimental samples were considered: Aglianico Molise (AM), Aglianico Sicily (AS), Cabernet Sauvignon Molise (CM), and Cabernet Sauvignon Sicily (CS).

For each cultivar/terroir combination and time of measurement, 21 fully exposed primary leaves, and 21 lateral leaves were collected. All the measurements have been conducted at the end of flowering, at the berry’s pea-sized stage, and at berry softening (vèraison). The total leaf area (TLA) was estimated through the measurement of leaf area (LA) of the main and lateral leaves realised in the lab with the area meter LI-3100 (Licor, Inc., Lincoln, Nebraska). Plant water status was determined in each experimental site using midday leaf water potential (LWP) measurement realised with a Schöelander pressure chamber (Soil Moisture Equipment Corp., Sta. Barbara, CA, USA; Scholander et al., 1965). Bunches were collected from 5 vines per treatment on each block for yield assessment. A sample of five bunches per treatment on each block was used to determine the bunch weight and both the number and weight of the berry. A 100-berry sample per experimental block was divided into three sub-samples, crushed with a manual press. The free-run juice was utilised to determine the total soluble solids (TSS) measured by a digital refractometer with temperature correction (RX-5000 Atago Co., Ltd., Bellevue, WA, USA). Must pH and titratable acidity (TA) were determined using an automatic titrator (Titrino model 798, Metrohm, Riverview, FL, USA) on 5.0 mL juice samples titrated against 0.1 M NaOH up to pH 8.2. TA was expressed as g/L of tartaric acid equivalents. The maturity index was estimated at harvest in 2020 (soluble solids -Brix° and pH), using the following formula (Lana et al., 2021):

Separation and extraction of berry components was carried out in duplicate simulating the maceration process necessary to produce red wines (Mattivi et al., 2002). Monomeric anthocyanins were measured according to the OIV Compendium of International Methods of Analysis of Wine and Musts (https://www.oiv.int/OIV-MA-AS315-11). Total flavanols were determined as vanillin-reactive flavans, according to Gambuti et al. (2015). Total phenolic compounds were dosed as iron-reactive phenolics (Harbertson, 2003). A large-scale extraction (300 g of fresh product) was carried out for each collected sample of red grapes. Destemmed grapes were crushed and extracted twice overnight Skin with 500 mL of H₂O:EtOH 2:8 (v/v) mixture. The hydroethanolic extracts from each sample were concentrated under a vacuum and partitioned first against Ethyl Acetate (EtOAc) and then n-butanol 2-mL aliquots of each obtained extract were lyophilized and investigated by NMR. Likewise, 1-mL aliquots from each extract were lyophilized, solubilized in methanol, and subjected to MS analysis. NMR experiments were run on a Varian Unity Inova 700 spectrometer equipped with a 13C Enhanced HCN Cold Probe and by using a Shigemi 5 mm MR tubes. CD3OD (δH 3.31; δC 49.0) was selected as deuterated solvent. Standard Varian pulse sequences were employed for the respective classes of spectra. All NMR data reported in the text were derived from 1D 1H spectra and from 2D experiments, including COSY and TOCSY. Relative quantitation of each identified compound was conducted by selecting representative NMR signals and their areas determined by integration. For Mass Spectrometry analysis, a linear ion trap LTQ Orbitrap XL hybrid Fourier Transform MS (FTMS) instrument equipped with an ESI ION MAX source (Thermo-Fisher, Waltham, MA, USA) and coupled to an Agilent 1100 LC binary system was used. All the employed solvents were of laboratory grade. Following the NMR and High-Resolution MS data for the identified compounds: Catechin: HRESIMS m/z 291.0800 [M+H]⁺ corresponding to C₁₅H₁₅O₆. Epicatechin: HRESIMS m/z 291.0857 [M+H]⁺ corresponding to C₁₅H₁₅O₆. Gallic acid: HRESIMS m/z 169.0169 [M-H]^- corresponding to C₇H₅O₅. Syringic acid: HRESIMS m/z 197.0501 [M-H]^- corresponding to C₉H₉O₅. Quercetin: HRESIMS m/z 303.1858 [M+H]⁺ corresponding to C₁₅H₁₁O₇. NMR data for the above compounds were consistent with those reported in the literature (Forino et al., 2019; Gambuti et al., 2020).

Analyses of monomeric anthocyanins were performed by an HPLC Shimadzu LC10 ADVP apparatus (Shimadzu Italy, Milan, Italy) equipped with an SCL-10AVP system controller, two LC-10ADVP pumps to create the needed solvent gradient, an SPD-M 10 AVP detector and an injection system full Rheodyne model 7725 (Rheodyne, Cotati, CA, USA). The HPLC solvents were the following: solvent A: water milli- Q (Sigma-Aldrich, Milan, Italy)/formic acid (Sigma- Aldrich ≥ 95%)/acetonitrile (Sigma-Aldrich≥ 99.9%) (87:10:3) v/v; solvent B: water/formic acid/acetonitrile (40:10:50) v/v. The gradient was: zero-time conditions 94% A and 6% B; after 15 min, the pumps were adjusted to 70% A and 30% B, at 30 min to 50% A and 50%B, at 35 min to 40% A and 60% B, at 41 min, end of the analysis, to 94% A and 6% B. 5 minutes re-equilibration time were applied before the subsequent analysis. The column used for the analyses was a water spherisorb column (C 18, Silica particle substrate, ODS2 250 x 4.6 mm, 5 μm particles diameter, 80 Å pore size) with a pre-column was used. 50 µL of calibration standards or wine was injected into the column. The absorbance signals at 520 nm was detected. Detector sensitivity was 0.01 Absorbance units full scale (AUFS). All the samples were filtered through 0.45 µm Durapore membrane filters (Millipore-Ireland) into glass vials and immediately injected into the HPLC system. The calibration curve was obtained by injecting 5 solutions (in triplicate) containing increasing concentrations of malvidin-3-monoglucoside (Extrasynthese, Lyon, France).

Fungal community composition was analysed by a culture-independent approach using next-generation sequencing (NGS). The grape samples from every vineyard were collected in duplicate, immediately transported to the laboratory and processed. Total genomic DNA was extracted using the Stool DNA Isolation Kit (Norgen, Biotek Corp., Thorold, ON, Canada) according to the manufacturer’s instructions. The concentration and purity of the extracted nucleic acids were visualised and quantified, after agarose gel electrophoresis, by Nanodrop (NanoDropTM 2000/2000c; Thermo Fisher Scientific, Italy). DNA quantity was standardized to a concentration of 10 ng/μl. The ITS2 (internal transcribed spacer) region of the rRNA was amplified using primers 2024F and 2409R. Amplicons were sequenced using the Illumina MiSeq PE300 platform (Illumina, San Diego, CA, USA) at Biodiversa s.r.l. (Rovereto, Italy). Adapter sequences were removed using Cutadapt (Martin, 2011). Read quality was assessed using DADA2 (Callahan et al., 2016). The taxonomic references were assigned using trained OTUs at 99% from UNITE database version 8.2 (https://unite.ut.ee) within Qiime2 tools version 2020.2 (https://qiime2.org) (Bolyen et al., 2019). The raw data have been deposited in Mendeley Data with the accession number DOI: 10.17632/tp2rrdz8wp.2.

Total RNA was isolated from 40 mg of ground whole berries, excluding seeds. The Spectrum™ Plant Total RNA kit (Sigma-Aldrich, St. Louis, MO, USA) was used following the manufacturer’s protocol with some modifications (400 mg of fresh tissue powder and an additional Column Wash step). The quantity and quality of the isolated RNA was measured using the NanoDrop ND-1000 spectrophotometer (Thermo Scientific, Wilmington, DE, USA) and Qubit 2.0 fluorometer (Life Technologies, Carlsbad, CA). For cDNA synthesis, 100 ng of each RNA sample was reverse transcribed using the SuperScript® III cDNA Synthesis Kit (Life Technologies), following the manufacturer’s protocol. Expression analysis was conducted by Real-Time quantitative PCR (RT-qPCR) using an SYBR Green method on a 7900HT Fast Real-Time PCR System (Applied Biosystems, Foster City, CA, USA). Each 15 μl PCR reaction contained 330 nM of each primer, 2 μl of 5-fold diluted cDNA and 7.5 μl of SYBR Green Mix (Applied Biosystems, Foster City, CA, USA). The SDS 2.3 and RQ Manager 1.2 software (both Applied Biosystems, Foster City, CA, USA) were used for data elaboration. Primer pairs used in the quantitative analysis were divided into five groups based on their roles: glutathione S-transferase as precursors of volatile thiols, terpene synthase enzymes for terpenoid biosynthesis, O-methyltransferases for the biosynthesis of methoxypyrazines, anthocyanin biosynthetic pathway genes and peroxidase for anthocyanin degradation (Sparvoli et al., 1994; Kobayashi et al., 2002; Dunlevy et al., 2010; Kobayashi et al., 2011; Battilana et al., 2017; Sicilia et al., 2020; Ji et al., 2021; Maritim et al., 2021; Wang et al., 2021). The relative expression value was calculated through the ΔΔCt method using the Sicilian berries as calibrators (Livak and Schmittgen, 2001; Villano et al., 2016). Three biological and three technical replicates were considered.

To evaluate the environmental difference between the experimental sites, the well-known t-test and f-test methods were applied to the weather information collected in 2020: mean temperature, day-night temperature excursion, ET₀, maximum wind, the Amerine and Winkler index (A&W), and rainfall. Quantitative grapevine data were compared using Tukey’s least significant differences procedure; all the variance resulted homogeneous. When the variances were not homogeneous, data were analyzed using Kruskall–Wallis test. When results of the Kruskal–Wallis test were significant (p< 0.05), the significance of between-group differences was determined by the Bonferroni–Dunn test (5% significance level). These analyses were performed using XLSTAT (version 2013.6.04; Addinsoft, Paris, France). All the data are expressed as means ± standard deviation of four replicates (two experimental replicates x two analytical replicates).

In the two experimental sites, different soils were identified and described as representative of selected vineyards ( Figure 1 and Table 1 ). In particular, a deep clay soil in Molise site (Gleyic Calcic Vertisols) for both AGL and CS, and two shallow volcanic sandy soils (Vitric Andosols) in the Sicilian site. The available water capacity (AWC) of the first 100 cm differed among the soils, with the highest value expressed in Molise (205 mm) and the lowest in Sicily (127-132 mm). The topsoil layer cation exchange capacity (CEC) was high in both sites except for the Sicilian AGL vineyard. This data agrees with the young nature of the latter soil, less evolved compared to the others (96.1% of volcanic sand), and less enriched in organic matter compared to the Sicilian CS soil. Moreover, a rising of water from a suspended groundwater table was found in the Molise site.

The results of climate analysis (RC 1971-2005 dataset) explained the effects of latitude in a Mediterranean climate (cold winters and mild summers), with the high latitude site (Molise) being the coldest compared to the Sicilian site (low latitude). Looking at the grapevine growing season (April-October 2020), the mean temperature was about 7.6% higher in Sicilian than in the Molise site. The behavior of the A&W index reflected the temperature differences, with a gap of 15% between Sicily and Molise. The ET₀ for the two sites was quite similar, with a small difference of 9%, while the cumulative rainfall was lower for Sicily compared to Molise (average difference of -8%). Daily mean temperatures and rainfall collected in each experimental site during the monitored growing season (April-October) are reported in Figure 2 . The mean temperature over the selected time was 17.2°C ( ± 0.5) for the Molise site, and 19.6°C ( ± 0.4) for the Sicilian one, and the ET₀ was 748.4 mm ( ± 1.4) and 714.3 mm ( ± 1.8). The accumulated rainfall was 219.5 mm and 333.4 mm for Molise and Sicily, respectively. The thermal regime showed a higher value of the A&W index in Sicily (2056 GDD compared to 1600 GDD of Molise) with a different trend in the accumulation of GDD during the growing season ( Figure 3 ). Statistical analysis of weather conditions among the experimental sites showed a significant difference between the two sites with a p< 0.05 (p mean temperature = 9,6E-27, p day-night temperature excursion = 0,0036, p ET₀ = 0,0325, p maximum wind = 1,96E-51 and p A&W index = 4,29E-55), with an average difference of mean temperature = 2,4 C°, day-night temperature excursion = -0,9 C°, ET₀ = -0,2 mm, maximum wind = -2,3 ms^-1 and A&W index = 191,4 GDD. Regarding rainfall, there is no statistical difference between the two sites, but from the biological point of view, the difference of 100 mm over the growing season between Molise and Sicily can be considered significant in terms of the effect on plant responses.

During early summer, midday LWP was about less than -0.8 MPa in both environments, which is considered the negative threshold for well-irrigated vines (Olivo et al., 2009; Intrigliolo et al., 2012). Indeed, the ψ ≤ -1.0 MPa recorded from midsummer to the end of the season, are considered drought-like conditions occurring from the fruit set throughout the ripening stage. During the latter period in Sicily, the registered values of -1,79 MPa for Aglianico and -1.73 MPa for Cabernet Sauvignon were significantly below -1.2 MPa or even that is considered a condition of severe stress (Nicolosi et al., 2022). The LWP was measured at the end of flowering, at berry’s pea-sized stage and at berry softening. The values were significantly different between the first and third stages for Aglianico. LWP values were -1.31MPa (AS) and -1.79 MPa (CS) in Sicily, whereas in Molise they were higher. During the intermediate stage (berry’s pea-sized stage) no differences were detected. As regards Cabernet Sauvignon, the most stressed vines were observed in Molise during both the end of flowering and berry’s pea-sized stage (ψ= -1,12 MPa vs -0,95 MPa, and ψ= 1,32 MPa vs -1,15 MPa, respectively). At berry softening the most stressed vines were in Sicily (-1,73 MPa vs -1,17 MPa) (Nicolosi et al., 2022).

The total leaf area per vine (TLA) is reported in Table 2 . For both cultivars and each phenological stage, the Molise site showed significantly higher values than the Sicilian site. However, in the latter, for Cabernet Sauvignon the TLA slightly increased during the growing season, while in Molise because of the shoot topping (midsummer) a decrease was recorded. For Cabernet Sauvignon, the above-mentioned decrease was observed at the berry softening stage due to the manual late basal leaf removal. The main and lateral shoot behaviour influenced the TLA. In fact, for Aglianico in Sicily, the TLA/main shoot ratio increased during the experimental period (from 0.23 m² to 0.34 m²), whereas in Molise, a decrease in TLA/main shoot ratio was registered between the end of flowering stage and the berry pea-size stage. Concerning the lateral shoots in Molise, a substantial increase occurred at the berry softening stage. For Cabernet Sauvignon, significant differences between the two experimental sites were observed at the end of flowering and at the softening of berries, both for main and lateral shoots. Regarding the leaf area index, a significant difference was recorded at the berry pea-size stage for Aglianico and Cabernet Sauvignon and the highest value was observed in Sicily. On the contrary, the leaf area index of Aglianico in Molise showed a significant increase at the berry softening stage.

The berries and wine characteristics are reported in Table 3 . Aglianico and Cabernet Sauvignon had the highest yield, bunch weight, and berry number in Molise. As regards the berry weight, a significantly high value was detected in Molise for Aglianico, while the berry weight of Cabernet Sauvignon was similar in each region. The rachis weight was always higher in Sicily than in the Molise site. The technological maturity of grapes was evaluated by determining the content of soluble solids, titratable acidity, their ratio, and the pH in 2020. The main classes of phenolic compounds extractable from skins and seeds were also determined ( Table 3 ). Total iron reactive phenolics, total flavanols and anthocyanins are higher in samples grown in Molise, while the content of flavanols strongly differs Aglianico from Cabernet. Data on alcohol content of wines were correlated to the content of sugars in grapes ( Table 3 ). The lower values of titratable acidity of all wines with respect to respective grapes are expected considering tartaric precipitation during fermentation and post-fermentative aging of wines. Comparing data on total phenolics in grapes and wines ( Table 3 ) in AS and AM phenolics were transferred from grapes to wine during winemaking but, especially for AM, a slight loss, probably due to adsorption on grape marcs, was detected. In contrast, in CS higher contents of phenolics compounds in wines with respect to berries were observed. This is easily explained since the method used to determine total phenolic compounds involved the reaction with iron and, thus, it is likely that during winemaking, for this grape variety, the formation of new phenolic structures highly reactive towards iron occurred. This is not surprising given the high reactivity of phenolic compounds among themselves and with fermentation metabolites such as acetaldehyde during winemaking. The lower values of total flavanols and anthocyanins detected in wines with respect to grapes is due to the great number of reactions of oxidation, polymerization, and precipitation that native phenolic compounds underwent. To identify the major polyphenols occurring in the analyzed samples, an NMR-based investigation was conducted. To this aim, both Aglianico and Cabernet Sauvignon grapes were extracted and partitioned as reported in the experimental section. Polyphenols were accumulated in the EtOAc extracts. Gallic acid and syringic acid were identified as phenolic acids, (+)-catechin and (-)-epicatechin and detected as the most abundant flavan-3-ols derivatives, while quercetin emerged as the most abundant flavonol compound. In Figure 4 , ¹H-NMR spectra of CM and CS EtOAc extracts display all the above proton resonances attributed to the identified polyphenols. Typical NMR signals attributable to hydroxycinnamic acids were identified, but due to their low intensities and occurrence in quite crowded spectral regions, unambiguous identification of specific molecules was prevented. It is to be underlined that anthocyanins were not analyzed by either NMR and MS techniques, as they were investigated by means of DAD-HPLC and spectrophotometric analyses, as reported above. All the above NMR-based structural hypotheses were corroborated by High Resolution Mass Spectrometry analysis (See Materials and methods).

The data on the fungal community present in the harvested grapes is shown graphically in Figure 5 . Overall, at harvest time the fungal populations, present in the vineyards of the examined sites, at the phylum level are very similar ( Figure 5A ). Ascomycota was the most abundant phylum followed by Basidiomycota. Other phylums such as Chytridiomycota are present only in low abundance. The analysis of the fungal community by high-throughput sequencing allowed to identify a total of 13 families and 12 genera. At the family level ( Figure 5B ), mean values of relative abundance revealed that the population was mainly represented by the Cladosporiaceae family with relative percentages between 11.2 (AS) and 48.6% (CS). The Saccotheciaceae family was widely present on the grapes analyzed with relative percentages between 11.4% (AM) and 33.5% (AS). Saccharomycodaceae was among the predominant family on AS grapes with a relative abundance of 43.9% while on the other grapes sampled this family was little detected. Widespread presence of Pleosporaceae with relative abundance between 2.5% (AS) and 18.9% (CM). At the genus level ( Figure 5C ), the main contaminants of grapes were fungi belonging to the genera Cladosporium, Aureobasidium, Alternaria and Sporobolomyces. A significant presence of the genus Hanseniaspora was recorded only on AS grapes (43.9%). The diffuse presence of Cladosporium was recorded on all the grapes analyzed, with relative abundances between 11.2% (AS) and 48.4% (CS).

The transcript abundance of representative key genes of phenylpropanoid metabolism is shown in Figure 6 as log₂ of the fold change of the samples from Molise normalized to the samples from Sicily. Overall, both varieties showed an overexpression of F3H, LDOX, and UFGT, while DFR and PRX21 were downexpressed. MYBA and CHS were overexpressed in Cabernet Sauvignon and downexpressed in Aglianico. As regards volatile thiol, methoxypyrazine, and terpenoid metabolism in Molise we observed, with respect to Sicilian grapes, an overexpression of GST2, GST3 and GGT for volatile thiols, OMT1 and OMT2 for methoxypyrazine, and TPS56 for terpenoids, in both Aglianico and Cabernet Sauvignon. On the counterpart, OMT3 showed downregulation in both varieties ( Figure S1 ).