Hierarchy of Factors Impacting Grape Berry Mass: Separation of Direct and Indirect Effects on Major Berry Metabolites

Hierarchy of Factors Impacting Grape Berry Mass: Separation of Direct and Indirect Effects on Major Berry Metabolites Roberta Triolo, 1,3 * Jean Philippe Roby, 1 Antonella Plaia, 2 Ghislaine Hilbert, 1 Simona Buscemi, 2 Rosario Di Lorenzo, 3 and Cornelis van Leeuwen 1 Abstract: Final berry mass, a major quality factor in wine production, is determined by the integrated effect of biotic and abiotic factors that can also influence berry composition. Under field conditions, interactions between these factors complicate study of the variability of berry mass and composition. Depending on the observation scale, the hierarchy of the impact degree of these factors can vary. The present work examines the simultaneous effects of the major factors influencing berry mass and composition to create a hierarchy by impact degree. A second objective was to separate the possible direct effects of factors on berry composition from an indirect effect mediated through their impact on berry mass. Vine water and nitrogen status of six blocks of Cabernet franc vines planted on sandy or gravelly soils were monitored over two years. Berries were analyzed from veraison to harvest. At each sampling date, fresh berry mass, berry seed mass and number, sugar content and concentration, and malic acid concentration were recorded. All studied factors significantly impacted final berry mass, but vine water status had the largest effect. The interaction between factors sometimes hid significant effects on berry compounds. Nevertheless, we showed by means of appropriate statistics that all factors had a direct impact on berry sugar and malic acid concentrations, although their order of impact varied with the metabolites considered. Conversely, the effect of factors significantly impacting berry sugar content is mediated through their impact on fresh berry mass. Key words: berry composition, berry mass, berry seed, soil, vine water status, virus, yeast assimilable nitrogen (YAN) Berry size is an important factor for quality assessment of grape berries used in wine production. Most grape growers agree that better wines, particularly red wines, are produced from small berries due to their higher ratio of skin to flesh (Matthews and Nuzzo 2007). However, the influence of berry size on grape composition is complex and far from being fully understood. The absence of a consensus among researchers could be attributed to the fact that final grape berry composition likely depends on factors affecting the growth of the various berry tissues rather than just differences in berry size (Roby et al. 2004, Walker et al. 2005, Poni et al. 2009). Several studies showed that final berry mass is the result of integrated 1 EGFV, Bordeaux Sciences Agro, INRA, University of Bordeaux, 33140 Villenave d Ornon, France; 2 University of Palermo, Scienze Economiche, Aziendali e Statistiche, Viale delle Scienze, 90128 Palermo, Italy; and 3 University of Palermo, Scienze Agrarie e Forestali (SAF), Viale delle Scienze, 90128 Palermo, Italy. *Corresponding author (roberta.triolo@hotmail.it; tel: +33 7 81 52 50 44) Acknowledgments: The authors thank the Association nationale de la recherche et de la technologie (ANRT), the Aquitaine Region, Vignobles Bardet and Amos Industrie, who funded this work. The authors are also grateful to Innovin for support in project management and to Vitinnov for technical and scientific assistance. Supplemental data is freely available with the online version of this article at www.ajevonline.org. Manuscript submitted Sept 2016, revised Feb 2017, Jul 2017, Oct 2017, accepted Oct 2017. Copyright 2018 by the American Society for Enology and Viticulture. All rights reserved. doi: 10.5344/ajev.2017.16087 effects of biotic and abiotic factors that affect cell number and/or cell volume. Some of these factors, such as berry seed content (Walker et al. 2005), are intrinsic, being related to the individual berry itself. Seeds influence berry cell division and expansion in pericarp through hormone production (Friend et al. 2009). Some virus infections, such as Grapevine Fan Leaf Virus (GFLV), can affect the quality of flower fertilization and induce poor fruit set (May 2004), on which the final berry seed content depends (Gillaspy et al. 1993). Water deficit is considered to be one of the major environmental factors limiting grapevine photosynthesis, plant growth, fruit size, and yield (Chaves et al. 2007). Vine water status, equally impacted by climate and soil, is highly variable inside and among vineyards because soil water availability varies with soil texture, percentage of stones, and rooting depth (van Leeuwen et al. 2009). Water deficit during berry growth reduces final berry mass, primarily when low water supply occurs before veraison (Ojeda et al. 2001). Late water deficit decreases vine vigor and shoot growth and consequently accelerates sugar accumulation because more sugar is directed to clusters (van Leeuwen et al. 2009). However, because the impact of water deficit on berry sugar concentration is cultivar-dependent (Deluc et al. 2009), this relationship is not always significant. Berries produced under water deficit conditions are also characterized by lower acid concentration (Esteban et al. 1999), probably due to increased malate breakdown (Matthews and Anderson 1988). Among nutrients the vine extracts from soil, nitrogen (N) has the most impact on vine growth, vigor, and grape composition. Beyond the addition of N fertilizer, vine N uptake 103

104 Triolo et al. varies considerably in relation to soil parameters such as soil organic matter content and mineralization rate. The latter, in turn, depends on soil temperature, aeration, ph, and moisture content (van Leeuwen et al. 2000). Moreover, because different nutrient ions dissolve into the soil solution at different concentrations in each soil layer, NO 3 - uptake also depends on water flow through the soil-root-shoot pathway and on rooting depth (Keller 2010). Low vine N status reduces vine vigor, berry mass, and malic acid concentration, and increases berry sugar (Choné et al. 2001a, Trégoat et al. 2002). Accumulation of total polyphenols in berries is increased under low N status (Soubeyrand et al. 2014). Hence, grape quality potential for red wine production is increased by limited soil N availability to the vines. Finally, berry tissue growth and final berry mass may also be affected by cultural practices such as cluster thinning (Guidoni et al. 2002), pruning severity (Walker et al. 2005), and leaf removal (Poni et al. 2009). Variability in berry mass can be observed at different scales, a combined effect of all impacting factors (Dai et al. 2011). Depending on the scale, the hierarchy of these factors may vary. At the parcel scale, grapevine cultivar is likely to be the dominant factor. At the intra-parcel scale, the variability may stem from variations in soil characteristics. Cluster mass variation has been linked to variation in berry mass (Pagay and Cheng 2010). Within a bunch and within a vine, the differences between berries may be related primarily to differences in seed number. However, this relationship may be affected by berry position within the cluster. This position effect possibly arises from sink competition (Prudent et al. 2014). Most studies published on berry mass and composition examine only a single impacting factor. Though this approach clarifies the specific impact of each considered factor, it does not lead to a greater understanding of berry mass and composition variability under field conditions, where environmental and plant-related factors interact with each other. Hence, the present work examined the simultaneous effect of vine water uptake conditions, vine N status, and berry seed content (seed number per berry and total seed mass per berry) on berry mass and composition to determine the hierarchy of their intra-parcel impact. Materials and Methods Location and experimental design. This study was carried out over the 2014 and 2015 seasons, on two parcels of a commercial vineyard located in the Saint Emilion region in southwestern France (Bordeaux area). The selected parcels were dry-farmed and planted on two different soil types, classified according to the French Réferentiel Pédologique (Baize and Girard 1995): a sandy soil (ARENOSOL Rédoxique), characterized by a water table within reach of the roots, and a gravelly soil (PEYROSOL) (Supplemental Table 1). Inside each parcel, three experimental blocks were chosen for their different water potentials measured in previous years (data not shown). The blocks S5, S6, and S7, located on the sandy soil, were characterized by similar high water availability. This homogeneity in soil composition was not observed in the gravelly soil. In this parcel, G7 and G8 blocks were characterized by a low water availability, related to the high gravel content of the soil layers (~25% in the topsoil [0 to 65 cm] and ~80% in the following soil layers [65 to 160 cm]). The third block, G1, showed a higher water availability, related to the presence of a layer without gravel at ~120 cm. The experimental design of this study was based on this large range of water availability, possibly impacting berry mass and juice composition. Plant material. All experimental blocks, composed of nine adjacent plants, were planted with Vitis vinifera L. cv Cabernet franc grafted onto 101-14 rootstock. Grapevines were trained to a double Guyot with vertical shoot-positioning and pruned to eight to 10 buds per vine. To minimize the effect of non-environmental factors, soil and canopy management practices were identical in both parcels. To determine the sanitary status of the plants, in both seasons, leaf and wood samples were collected from selected vines in each block and analyzed using ELISA and RT-PCR (Beuve et al. 2007, 2013), to check for possible presence of GFLV (genus Nepovirus), Arabis Mosaic Virus (ArMV, genus Nepovirus), and Grapevine Leaf Roll-associated Virus (GLRaV-1, -2, and -3, genus Ampelovirus). All plants were healthy except for vines belonging to the S6 blocks, which were infected with viruses involved in fanleaf degeneration (GFLV and ArMV). Because of the possible impact of these viruses on berry mass and composition, we included this block to collect information on possible effects of these viral diseases on berry mass. Weather conditions. Temperature and rainfall data were recorded by an automatic weather station based on one of the parcels studied. Because the parcels are located in a flat area at exactly the same altitude (8 m asl) and within ~500 m of each other, the climate for each vintage was homogenous among blocks. Vine water status assessment. Stem water potential. Dynamic evolution of vine water status during the growing season was monitored using a pressure chamber (Scholander et al. 1965) equipped with a digital LDC manometer. In 2014 and 2015, midday stem water potential (Ψ stem ) was measured 10 and seven times, respectively, from early July until the end of September. At each date, Ψ stem was measured on four adjacent vines per block. Measurements were taken in the early afternoon on fully expanded leaves from primary stems. Leaves were enclosed in a reflective plastic envelope for at least one hour before measurement. Carbon isotope discrimination. The 12 C isotope is preferentially used by the enzymes involved in photosynthesis for production of hexoses over the 13 C of ambient CO 2 (Gaudillère et al. 2002). This process, called isotope discrimination, is reduced when plants face water deficit conditions because of stomatal closure. Sugars produced under these conditions contain more 13 C than those produced when plant water status is not limiting. Therefore, the 12 C/ 13 C ratio (so-called δ 13 C) measured on products of photosynthesis at ripeness is an integrative indicator of vine water uptake conditions during grape ripening. The results of this measure can range from -20 (severe water deficit stress) to -27 (no water deficit stress)

Hierarchy of Factors Impacting Grape Berry Mass and Metabolites 105 (Gaudillère et al. 2002). In this study, δ 13 C measurements were carried out by isotope mass spectrometry on grape juice extracted from 54 individual berries per block. Berries were collected from the basal cluster of the central primary shoot of only three individual vines per block. To span a large range of berry mass variability, within each bunch, berries of different mass were sampled (small, medium, and large, with respect to the mean value for each bunch). Vine N state assessment. In this study, yeast assimilable nitrogen (YAN) content in grape juice was chosen to assess the dynamic evolution of vine N status during the ripening period (van Leeuwen et al. 2000). YAN was measured weekly in grape juice extracted from 200 fresh berries per block, using Sørensen formol titration (Ribéreau-Gayon et al. 2012). In 2014 and 2015, YAN was measured five and six times, respectively. At harvest, using the same method, the YAN was measured berry by berry on the same sample of berries used to measure δ 13 C. Assessment of grape ripening. To follow the seasonal dynamics of berry mass and composition, samples of 400 berries were collected weekly from vines in each experimental block. Measurements started at veraison, when over 50% of berries per bunch had developed color (83 to 85 on the BBCH scale) until harvest. Berries were analyzed five times in 2014 and six times in 2015. Each sample of berries from each block was weighed to determine mean berry mass. Then berries were pressed and the must, after gentle centrifugation, was analyzed for soluble solids and malic acid using Fourier transform infrared spectroscopy (FTIR) (Destrac Irvine et al. 2015). Berry analysis at harvest. To determine a highly representative value for berry mass, one basal cluster located on a central primary shoot was sampled from each individual vine per block 44 days after veraison (DAV). The mass of each individual berry was recorded. Berry seed content and the concentrations of sugar and malic acid, which are among major berry compounds, were measured berry by berry on the same subsample of 54 berries per block, corresponding to the sample in which δ 13 C and YAN had been estimated. Total soluble solids (Brix) was measured using a hand-held refractometer. The quantity of sugar per berry (mg/berry) was calculated as described (Deloire 2011). Finally, malic acid concentration was estimated by a colorimetric method using a Bran and Luebbe TRAACS 800 autoanalyzer. Statistical data analysis. Data were analyzed using R software (R development Core Team 2015, version 3.2.3). Block effect. Effects of blocks on vine water, N status, and berry characteristics (berry mass, berry sugar, and malic acid concentration) were tested using a one-way analysis of variance (ANOVA), considering berries as replicates. Tukey s honest significant difference (HSD) was used as post-hoc test for multiple mean comparison. Factor hierarchy on berry mass and its composition. The relationships between the potential impacting factors and berry mass or sugar levels and malic acid concentrations were investigated on the whole data set, independently from vintage or block. When the effects of water status, N status, and berry seed content on berry composition were examined, berry mass was included in the statistical model as a covariate, given its possible impact on berry sugar and malic acid concentration. A starting multiple linear regression model, considering all possible impacting factors and all possible interactions among them, was reduced by applying a stepwise backward procedure. This procedure drops out, step by step (according to the reduction in Akaike Information Criterion [AIC]), all nonsignificant regressors, finding a final model composed exclusively by significant covariates (Venables and Ripley 2002). On the final model, an ANOVA Type III was performed to compute the contribution of each component to the variance of berry mass and composition. Most of the previous studies performed the so-called Type I Sum of Squares (SS), determining the weight of each factor by the ratio of individual SS to the total SS. However, this approach can introduce bias when covariates are correlated, because individual SS depends on the order of the model terms. The Type III SS does not depend on the order of the model terms, but the individual SS do not sum to the total SS. Consequently, the incremental contribution of each covariate cannot be calculated as the ratio of the sum of squares due to this covariate to the total sum of squares of the model (Fox 2016). Therefore, in this study, in which correlations among covariates were observed during the statistical analysis, it was necessary to perform a Type III SS. The relative importance of each covariate (berry seed content, vine water status, and vine N status) to the variability of berry mass and berry compounds was calculated as the ratio of the SS due to this covariate to the sum of squares due to another covariate (e.g., SS of vine water status versus SS of berry seed content). Direct and indirect effect of factors impacting berry sugar and malic acid concentrations. A second statistical approach was used to separate the possible direct effect of a specific factor from that mediated through its possible impact on berry mass (or indirect effect). Three different models (one for each factor) were performed. In each model, the sugar content, sugar concentration, or malic acid concentration was the dependent variable. Berry mass was always the first covariate and vine water status, vine N status, or berry seed content was the second covariate (e.g., sugar concentration = berry mass + vine water status). An ANOVA Type I was then performed to test the significance of the additive effect (or direct effect) of the second covariate with respect to berry mass. Results Weather conditions. Weather conditions varied during the two vintages (Figure 1). Total rainfall and cumulative degreedays, calculated from 1 April through the end of October, were 463 mm and 1757 growing degree days (GDD) in 2014, and 299 mm and 1733 GDD in 2015. In 2014, the trend of monthly temperatures was close to the long-term mean and slightly warmer in September and October, while the growing season rainfall was above average except in April, May, and October. The 2015 growing season was warmer than average, except for September and October, and exceptionally dry, especially in the first part of the growing season. Only August was rainy, but most of this rainfall accumulated during one rainstorm in the middle of the month.

106 Triolo et al. Vine water status. According to the observed seasonal dynamic of Ψ stem, vines planted in the sandy soil with a shallow water table (blocks S5, S6, and S7) did not face a water deficit. This was observed during the two experimental seasons (Figure 2A and 2B). In contrast, a vintage effect on seasonal dynamics of Ψ stem was clearly evident on the gravelly soil. On 1 July 2014 (day of year 182), Ψ stem values were close to -0.4 MPa in all blocks (Figure 2A), showing no limitation in vine water uptake. Significant differences in water status among blocks started to develop around veraison in the beginning of August (day of year 216, Table 1). Vine water deficit continued to increase in August only in G7 and G8 and on the final measurement day (day of year 255), Ψ stem values recorded in these blocks indicated a severe water deficit. Conversely, no water deficit was recorded in G1 during the whole season, showing values similar to those observed in the blocks planted on sandy soil. In 2015, an exceptionally dry vintage, Ψ stem values recorded in G7 and G8 were significantly different from other blocks, starting with the first measurement day (Figure 2B), in which vines were subjected to moderate to severe water deficit. These water deficits continued to increase slightly until veraison, but disappeared after a rainfall event in mid-august (day of year 229). Moderate to severe water deficits reappeared at the end of the same month. At harvest, Ψ stem values were significantly different compared to remaining blocks (Table 1). In contrast, vines in block G1 faced no water deficit at the beginning of the 2015 season, showing Ψ stem values similar to those recorded in blocks S5, S6, and S7 (Figure 2B). However, Ψ stem values of G1 became progressively more negative over the season. At the beginning of August (day of year 215), measured Ψ stem values were intermediate compared to the remaining blocks, indicating a weak water deficit. Vine water deficit continued to increase during the following months and, on the final measurement day (day of year 251), this block had a moderate water deficit (Figure 2B and Table 1). Vintage and soil effects on vine water status were also reflected in δ 13 C values, measured in grape sugars at ripeness (Table 1). Results were consistent with vine water status measured as Ψ stem, and correlations between the two vine water status indicators were highly significant (Table 2). Vine N status. Because no N fertilizer was added in 2014 and 2015, or during the preceding years, we assumed that YAN level depended on soil characteristics. YAN values followed a similar tendency in both experimental parcels. Differences between soil types were noticeable during the growing season in both vintages (Figure 3A and 3B). At harvest, vines of blocks planted in sandy soil (S5, S6, and S7) had a similar N status, which was significantly lower than that of vines planted in gravelly soil, indicating an N deficit (Table 1). In contrast, fruit from block G7 had more YAN in both years, indicating moderate to high vine N status. Blocks G1 and G8 had slightly lower YAN in both years. In these blocks, vine N status was moderate. Combined vine water and N status. The combination of soil type and climatic conditions of the two vintages resulted Figure 1 Seasonal trends (1 April to 31 Oct) in monthly temperature and rainfall recorded in 2014 and 2015; the comparison is with average values (station Cheval Blanc, 33330, Saint-Emilion, average from 1996 to 2015). Figure 2 Seasonal dynamics of stem water potential (Ψ stem ) measured in the experimental blocks in the summers of 2014 (A) and 2015 (B). Observed values are presented as means (points) of four replicates with standard errors (bars). Arrows indicate 50% veraison.

Hierarchy of Factors Impacting Grape Berry Mass and Metabolites 107 Table 1 Influence of block on vine water and nitrogen conditions, measured with different physiological indicators in 2014 and 2015. a 2014 2015 Block Ψ stem Ver Ψ stem Harv δ 13 C ( ) YAN Harv (mg/l) Ψ stem Ver Ψ stem Harv δ 13 C ( ) YAN Harv (mg/l) G1-0.55 ± 0.04 b b -0.74 ± 0.1 b -27.31 ± 0.45 a 121 ± 37 c -0.94 ± 0.07 c -1.31 ± 0.22 c -24.81 ± 0.46 b 147 ± 21 c G7-1.01 ± 0.11 a -1.68 ± 0.06 a -23.43 ± 0.88 c 140 ± 43 d -1.69 ± 0.07 a -1.77 ± 0.05 a -23.19 ± 0.23 d 167 ± 54 c G8-0.99 ± 0.11 a -1.61 ± 0.15 a -24.85 ± 0.77 b 101 ± 30 b -1.55 ± 0.04 b -1.57 ± 0.01 b -23.93 ± 1.04 c 118 ± 12 b S5-0.57 ± 0.06 b -0.69 ± 0.06 b -26.85 ± 0.76 a 64 ± 25 a -0.77 ± 0.02 d -0.75 ± 0.12 d -26.78 ± 0.34 a 92 ± 8 ab S6-0.6 ± 0.07 b -0.71 ± 0.04 b -26.73 ± 0.21 a 66 ± 22 a -0.6 ± 0.03 e -0.62 ± 0.11 d -26.98 ± 0.17 a 86 ± 4 ab S7-0.57 ± 0.05 b -0.67 ± 0.02 b -27.12 ± 0.35 a 58 ± 13 a -0.65 ± 0.07 e -0.42 ± 0.08 e -27.44 ± 0.2 a 69 ± 3 a a Ψ stem Ver: stem water potential at veraison; Ψ stem Harv: stem water potential at harvest; δ 13 C: carbon isotope discrimination; YAN Harv: yeast assimilable nitrogen at harvest. The difference between blocks was tested with a one-way analysis of variance. G, gravelly soil; S, sandy soil. b Different letters in the same column denote statistically significant differences between blocks at p < 0.05 by Tukey s honest significant difference test. Table 2 Correlations between indicators of vine water and nitrogen status used during the two years of experimentation. a Ψ stem Ver Ψ stem Harv δ 13 C ( ) YAN Harv (mg/l) Ψ stem Ver 1 Ψ stem Harv 0.81 *** b 1 δ 13 C ( ) 0.81 *** 0.89 *** 1 YAN Harv (mg/l) 0.60 *** 0.70 *** 0.64 *** 1 a Ψ stem Ver: stem water potential at veraison; Ψ stem Harv: stem water potential at harvest; δ 13 C: carbon isotope discrimination; YAN Harv: yeast assimilable nitrogen at harvest; r and p-value were computed using a linear regression, n = 108. b Significance: ***: 0.001. in a wide range of combinations of vine water and N status (Table 3). In blocks S5, S6, and S7, vine water status was high and vine N status was low in both vintages. Vine water and N status were both high in block G1 in 2014, but water deficit was moderate and N status high in 2015. In blocks G7 and G8, vines faced moderate to severe water deficit in both vintages, associated with non-limiting N conditions. Of these blocks, G7 had slightly higher N status and lower water status than G8. Berry seed content. The average seeds per berry of the six blocks varied over the two years of the experiment (Table 4). Most berries contained one or two seeds. This was observed in all blocks and in both years. However, the distribution of berries into seed number classes varied slightly among blocks. The most striking difference is a lower number of seeds per berry in block S6 (most berries with one seed). A vintage effect was observed when berry seed content was expressed as total seed mass per berry. The values were lower in 2015 than in 2014. There were no significant differences in total seed mass per berry among blocks in 2014, while in 2015, berries from block G7 had lower mass than berries from S5. Fresh berry mass at ripeness. Fresh berry mass followed a similar trend from veraison through harvest in 2014 and 2015 (data not shown). Differences among blocks at harvest were greater in 2015 than in 2014. In 2014, G1 berries had higher mass than other blocks, although differences were significant only with blocks G7, G8, and S6 (Table 5). Similarly, G1 produced heavier berries in 2015, especially compared to Figure 3 Seasonal dynamics of yeast assimilable nitrogen (YAN) in the experimental parcels in summer 2014 (A) and 2015 (B), from veraison through ripeness. Observed values are presented as means (points) of three replicates and standard errors (bars). G7 and G8. Berry mass at harvest in these two blocks was significantly less than in other blocks. Berry composition during ripening and at harvest. Dynamics of sugar accumulation and malic acid degradation

108 Triolo et al. were monitored in grape juice in 2014 and 2015 (data not shown). From veraison to maturity, must from G7 and G8 had the most sugar and the least malic acid concentrations at most sampling dates in both years. In 2014, the dynamics of sugar accumulation and malic acid degradation in block G1 was similar to that observed in S5, S6, and S7. G1 was more similar to blocks G7 and G8 in 2015. Some significant differences were observed at harvest between blocks (Table 5). In 2014, S7 berries were low in sugar and G1 berries high in malic acid. In 2015, G1 berries were high in sugar, whereas S6 and S7 berries were high in malic acid. A clear vintage effect was noted for berry composition. Sugar concentration was greater in 2014, and malic acid concentration was markedly reduced in 2015. The effect of water, N, and berry seed content on fresh berry mass. The output of the R-code for the stepwise procedure showed that the final model was the following: fresh berry mass = vine water status + vine N status + berry seed content + vine water and N status interaction (Supplemental Table 2). The plant health status effect on berry mass was not Table 3 Vine water and nitrogen status of six experimental blocks in 2014 and 2015. a 2014 2015 Block Water deficit Nitrogen deficit Water deficit Nitrogen deficit G1 No water deficit No nitrogen deficit Moderate to weak No nitrogen deficit G7 Moderate to severe No nitrogen deficit Severe No nitrogen deficit G8 Moderate to severe No nitrogen deficit Severe No nitrogen deficit S5 No water deficit Moderate to severe No water deficit Moderate to severe S6 No water deficit Moderate to severe No water deficit Moderate to severe S7 No water deficit Moderate to severe No water deficit Moderate to severe a Thresholds for vine water and nitrogen deficit class definitions are those proposed in van Leeuwen et al. (2009). G, gravelly soil; S, sandy soil. Table 4 Effect of block on seed content per berry. Block a Seed number G1 G7 G8 S5 S6 S7 2014 Berries with 1 seed (%) 44 26 31 48 59 41 Berries with 2 seed (%) 37 50 45 37 31 31 Berries with 3 seed (%) 17 22 22 15 8 24 Berries with 4 seed (%) 2 2 2 0 2 4 Average seed number per berry 1.76 ± 0.80 ab b 2.00 ± 0.75 b 1.94 ± 0.79 ab 1.67 ± 0.73 ab 1.52 ± 0.72 a 1.91 ± 0.90 ab Total seed mass per berry (g) 0.079 ± 0.032 a 0.086 ± 0.029 a 0.0860.033 a 0.0710.027 a 0.0710.030 a 0.074 ± 0.033 a 2015 Berries with 1 seed (%) 38 50 43 45 69 63 Berries with 2 seed (%) 40 33 46 36 30 33 Berries with 3 seed (%) 21 17 9 19 1 4 Berries with 4 seed (%) 1 0 2 0 0 0 Average seed number per berry 1.83 ± 0.76 c 1.67 ± 0.75 ac 1.70 ± 0.72 ac 1.74 ± 0.76 bc 1.33 ± 0.51 a 1.41 ± 0.57 ab Total seed mass per berry (g) 0.059 ± 0.028 ab 0.052 ± 0.023 a 0.065 ± 0.024 ab 0.068 ± 0.023 b 0.055 ± 0.018 ab 0.057 ± 0.022 ab a The difference between blocks was tested with a one-way analysis of variance. G, gravelly soil; S, sandy soil. b Different letters in the same row denote statistically significant differences between blocks at p < 0.05 by Tukey s honest significant difference test. 2014 Table 5 Effect of block on berry mass, sugar, and malic acid measured at harvest in 2014 and 2015. a Block a G1 G7 G8 S5 S6 S7 Berry mass (g) 1.31 ± 0.14 c b 1.10 ± 0.08 ab 1.11 ± 0.10 ab 1.19 ± 0.11 bc 1.00 ± 0.15 a 1.20 ± 0.6 bc Sugars (Brix) 22.1 ± 0.9 bc 22.0 ± 0.7 bc 22.3 ± 1.5 bc 22.5 ± 1 c 21.7 ± 1.1 bc 20.9 ± 1.2 a Malic acid (g/l) 3.23 ± 0.9 b 3.04 ± 0.7 ab 2.75 ± 0.8 a 2.7 ± 0.5 a 2.91 ± 0.6 ab 2.75 ± 0.5 a 2015 Berry mass (g) 1.29 ± 0.10 c 0.79 ± 0.10 a 0.98 ± 0.14 b 1.29 ± 0.12 c 1.13 ± 0.12 bc 1.23 ± 0.13 c Sugars (Brix) 21.8 ± 1.02 b 20.7 ± 0.87 a 21.2 ± 1.35 ab 21.1 ± 0.85 a 20.7 ± 0.82 a 20.7 ± 0.95 a Malic acid (g/l) 1.05 ± 0.2 a 0.91 ± 0.3 a 0.88 ± 0.37 a 0.96 ± 0.32 a 1.35 ± 0.4 b 1.64 ± 0.6 c a The differences between blocks were tested with a one-way analysis of variance. G, gravelly soil; S, sandy soil. b Different letters in the same row denote statistically significant differences between blocks at p < 0.05 by Tukey s honest significant difference test.

Hierarchy of Factors Impacting Grape Berry Mass and Metabolites 109 significant. Therefore, this factor was deleted automatically. Hence, fresh berry mass was significantly and simultaneously influenced by vine water status, vine N status, and berry seed content. However, the extent to which these factors influenced berry mass varied. Table 6 reports the Type III ANOVA output for this final model. The incremental impact of vine water status on fresh berry mass (calculated as the ratio of Table 6 Effects of vine water, nitrogen status, and berry seed content on fresh berry mass. a Vine water status 6.9284 1 310.0228 <2.2E-16 *** c Vine nitrogen status 0.3144 1 14.067 0.00029 *** Berry seed content 0.0881 1 3.9412 0.04975 * Water/nitrogen 0.3337 1 14.9315 0.000194 *** interaction Residuals 2.3242 104 a Results were obtained from an analysis of variance Type III performed on the following model: berry fresh mass = vine water status + vine nitrogen status + berry seed content + vine water and nitrogen status interaction. b Sum Sq: sum of squares; Df: degrees of freedom. c Significance: ***: 0.001; *: 0.05. the SS due to this factor to the SS due to another factor: e.g., vine water status versus berry seed content = 6.9284/0.0881 = 78.64) was ~80 times the impact of berry seed content and ~22 times more important than vine N content. The impact of vine N status was around four times greater than the impact of berry seed content. The effect of water, N, and berry seed content on sugar content and sugar and malic acid concentrations. ANOVA Type III showed that the effect of vine N status on berry sugar levels (expressed as berry content and concentration) was not significant (Tables 7 and 8). Conversely, both sugar levels were significantly impacted by all remaining factors (berry mass, vine water status, and berry seed content), even though the effect of berry mass on sugar content is marginally significant. Moreover, several interactions between covariates were observed (data not shown). In contrast, all factors except berry mass significantly influenced the malic acid concentration of berries (Table 9). Results reported in Tables 7-9 allowed us to construct a hierarchy of the impact of covariates on metabolite concentrations. Vine water status had the greatest impact on both sugar and malic acid concentrations (Tables 7 and 9, respectively). The incremental impact of vine Table 7 Effects of berry mass, vine water, and nitrogen status and berry seed content on berry sugar concentration. a Berry mass 81.52 1 61.9145 3.931E-12 *** c Vine water status 479.58 1 364.2234 2.2E-16 *** Vine nitrogen status 2.17 1 1.6468 0.20231 ns Berry seed content 10.52 1 7.9869 0.00657 ** Residuals 134.30 102 a Results were obtained from an analysis of variance Type III. b Sum Sq: sum of squares; Df: degrees of freedom. c Significance: ***: 0.001; **: 0.01; ns: not significant. Table 8 Effects of berry mass, vine water, and nitrogen status and berry seed content on berry sugar content. a Berry mass 12545 1 3.6796 0.058024. c Vine water status 22572 1 6.6205 0.011596 * Vine nitrogen status 6426 1 1.8848 0.172951 ns Berry seed content 32397 1 9.5025 0.002672 ** Residuals 330705 97 a Results were obtained from an analysis of variance Type III. b Sum Sq: sum of squares; Df: degrees of freedom. c Significance: '.' = 0.1; **: 0.01; *: 0.05; ns: not significant. Table 9 Effects of berry mass, vine water, and nitrogen status and berry seed content on berry malate concentration. a Berry mass 1.891 1 2.2783 0.1343211 ns c Vine water status 13.085 1 15.7629 0.0001346 *** Vine nitrogen status 11.896 1 14.3301 0.0002605 *** Berry seed content 8.165 1 9.8354 0.0022426 ** Residuals 83.844 101 a Results were obtained from an analysis of variance Type III. b Sum Sq: sum of squares; Df: degrees of freedom. c Significance: ***: 0.001; **: 0.01; ns: not significant.

110 Triolo et al. water status on sugar concentration was around five times and ~45 times more important than those of berry mass and berry seed content, respectively. The impact of berry mass was around eight times greater than berry seed content. The incremental impacts of vine water status and vine N status on malic acid were similar at ~1.5 times greater than berry seed content. In contrast, berry sugar content was primarily impacted by berry seed content (Table 8). The incremental impact of this factor was ~2.5 times greater than that of berry mass and 1.5 times greater than that of vine water status. Direct and indirect effect of considered factors on berry composition. Since the effect of berry fresh mass on berry malic acid concentration was not significant (Table 9), the significant effect of all remaining covariates are direct effects and their additive effect on berry sugar concentration and content, with respect to berry mass, can be calculated (Tables 10 and 11, respectively). The results in these tables derive from three different models. The complete outputs of each model are reported in Supplemental Tables 3 and 4. In each model, the impact of berry mass on sugar concentration was significant, and the effect of each second covariate was also significant (Table 10). Therefore, vine water status, vine N status, and berry seed content all have significant indirect effects (through berry mass) and significant direct effects on berry sugar concentration (independent from berry mass). This was not the case for berry sugar content (Table 11). In each model, the impact of berry mass on sugar content was significant. Effectively, the linear regression between berry mass and sugar content, determined berry by berry (n = 648), was highly significant (R² = 0.96). Conversely, the effect of each second covariate (vine water status and berry seed content) was not significant. Therefore, vine water status and berry seed content only have a significant indirect effect (through berry mass) on berry sugar content, independent of berry mass. Discussion This study was conducted under field conditions on experimental blocks that were neither irrigated nor fertilized, so water and N availability to the vines depended only on variability in environmental conditions. Vine water status varied from year to year, during the season and among blocks. Because weather was homogenous among blocks for a given vintage, observed differences in intra-parcel vine water status reflect water availability of the soil, caused by variations in soil composition. Our results clearly show the impact of soil on vine water and N uptake, as observed previously in the Bordeaux area (Choné et al. 2001). Vines in the three blocks on sandy soil, characterized by a water table accessible to the roots, faced no water deficit during the two experimental years. On the gravelly soil, vine water status varied within the parcel due to variations in rooting depth and gravel content. Hence, only two of the three blocks (G7 and G8) faced severe water deficit during the two years, although the period during which this occurred varied among vintages. The third block with gravelly soil (G1) showed a similar stem water potential pattern as the blocks on sandy soil during the whole 2014 season and at the beginning of the 2015 season, but vines on this block faced greater water deficit at the end of the season. Vintage effects on vine N status were smaller than those of vine water status. YAN levels were highly variable among experimental parcels. Without added N fertilizer, vine N status depends on soil organic matter content and its mineralization rate that, in turn, increases with soil temperature and soil aeration (Choné et al. 2001). Therefore, in this study, the characteristics of the warm and well-aerated gravelly soil, favored turnover of organic matter. Conversely, in the sandy soil, which was cooler and less well aerated due to water logging in the spring (van Leeuwen et al. 1998), the mineralization rate was limited. These differences explain the higher vine N status on the gravelly soil compared to the sandy soil in our experiment. Similar results were obtained by Peyrot des Gachons et al. (2005). It also explains the significant relationship between vine water and N status, tested on the complete data set (Table 2). However, as the N availability was estimated by a physiological indicator, a dilution effect Table 10 Direct effect of vine water and nitrogen status and berry seed content on berry sugar concentration. a Vine water status 993 1 705.57 <2.2E-16 *** c Vine nitrogen status 227 1 61.451 3.935E-12 *** Berry seed content 147 1 43.375 1.837E-9 *** a Results were obtained from an analysis of variance Type I. b Sum Sq: sum of squares; Df: degrees of freedom. c Significance: ***: 0.001. Table 11 Direct effect of vine water and nitrogen status and berry seed content on berry sugar content. a Vine water status 181 1 0.0524 0.819528 ns c Vine nitrogen status 3438 1 0.9959 0.320930 ns Berry seed content 6410 1 1.8568 0.176324 ns a Results were obtained from an analysis of variance Type I. b Sum Sq: sum of squares; Df: degrees of freedom. c Significance: ns: not significant.

Hierarchy of Factors Impacting Grape Berry Mass and Metabolites 111 under non-limiting water status conditions, related to vine vigor and berry mass, cannot be excluded. Differences in berry mass among blocks are the result of the combined impact of several factors. An ANOVA Type III test confirmed that vine water availability, vine N status, and berry seed content significantly impacted final berry mass. Moreover, an interaction between vine water and N status was found. This combined effect of vine water and N status on fresh berry mass was particularly evident in block G1, where N and water uptake conditions were rarely limited. Hence, berries produced under these conditions were bigger than those produced under limited water and/or N conditions. The major objective of this work was to create a hierarchy of factors impacting final berry mass. At the intra-parcel scale used in this study, berry seed content was the least-impacting factor considered, and vine water status was most important. Its impact on fresh berry mass was much greater than the effect of N availability. Therefore, berries produced by vines subjected to severe water deficit were smaller than those produced under limited N uptake conditions. The exception in S6 berries in 2014 was probably related to the health status of the vines in this block. Berries produced by vines affected by GFLV in S6 were significantly smaller, confirming previous studies (Martelli and Savino 1990). Severe incidence of the fanleaf virus is generally accompanied by poor setting (May 2004). Our results support this observation. The proportion of berries containing only one seed was bigger in S6 than in the other blocks. Several studies showed the existence of a relationship between berry mass and berry composition (Roby et al. 2004, Walker et al. 2005, Barbagallo et al. 2011). Nevertheless, the origin of this relationship is still a subject of debate. In this study, berry sugar and malate concentration were considered among important compounds representing berry composition. We investigated the relations between these compounds and vine water status, vine N status, berry seed content, and berry mass. At harvest, differences among blocks were small, likely due to interactions among impacting factors. However, the combined effects of the soil and climatic conditions (vintage effect) created an interesting range of water and N availability, and influenced the patterns of sugar accumulation and malate degradation during ripening. Berries produced under water deficit conditions were characterized by higher sugar levels and lower malate levels. Similar results were found in Bordeaux, France (Trégoat et al. 2002). Conversely, berries produced with an unlimited water supply contained less sugar and more malic acid, as observed previously in Sangiovese grapevines (Storchi et al. 2005). In this study, statistical models applied to the whole data set showed that the hierarchy of impacting factors differed among metabolites. The overall effect of the covariates on malic acid was consistently less than their effect on sugar. Although not all covariates significantly affected sugar and malate concentration simultaneously, vine water status was, once again, the most important factor. Conversely, berry sugar content was primarily affected by berry seed content. Berry fresh mass had a nonsignificant effect on malic acid concentration. As a consequence, the significant impact of factors on malic acid concentration was exclusively direct and not mediated through berry mass. Conversely, berry mass had a significant influence on berry sugar content and concentration, while sugar levels did not change with vine N status. Indeed, the effect of vine N status on berry sugar concentration was hidden by the interactions among other covariates. This hypothesis was confirmed using a model chain where the effect of each covariate was separated from that of fresh berry mass. This approach showed that all factors considered in this study, including vine N status, affected berry sugar concentration. Conversely, when sugar levels were expressed as content (mg/berry), the direct effect of each factor was not significant. This means that vine water status and berry seed content have a significant but indirect effect (through berry mass) on berry sugar content. Conclusion This study produced a hierarchy of factors impacting grape berry mass under field conditions for the first time. Our statistical approaches also separated the direct effect of these factors on major berry compounds from a possible indirect effect through their influence on fresh berry mass. All studied factors significantly impacted the final fresh berry mass. Vine water status was the most important factor among those considered and berry seed content, the least. Vine water status also had the most impact on berry sugar and malic acid concentrations, while berry sugar content was primarily impacted by berry seed content. Nevertheless, not all remaining covariates (including fresh berry mass) significantly affected the concentrations of these two major compounds. This was probably due to correlation and/or interaction among factors that sometimes hid their real significant effect on berry composition. This was verified using a series of statistical models. No factor impacted berry sugar content directly. Conversely, all factors had a direct impact on berry sugar and malate concentrations, independently of their impact on fresh berry mass. Malic acid concentration was impacted directly by berry seed content and vine N status, while the additive effect of vine water status was only marginally significant (p = 0.1). In contrast, grape sugar concentration was driven primarily by vine water status, but also depended indirectly on these factors through their effect on berry mass. Hence, the statistical approaches used here clarified the importance of each factor responsible for variability in berry mass and berry composition, under field conditions where multiple factors act simultaneously. Literature Cited Baize D and Girard MC. 1995. Référentiel pédologique. INRA Editions, France. Barbagallo MG, Guidoni S and Hunter JJ. 2011. Berry size and qualitative characteristics of Vitis vinifera L. cv. Syrah. S Afr J Enol Vitic 32:129-136. Beuve M, Sempé L and Lemaire O. 2007. A sensitive one-step realtime RT-PCR method for detecting Grapevine leafroll-associated virus 2 variants in grapevine. J Virol Methods 141:117-24.

112 Triolo et al. Beuve M, Moury B, Spilmont AS, Sempé-Ignatovic L, Hemmer C and Lemaire O. 2013. Viral sanitary status of declining grapevine Syrah clones and genetic diversity of Grapevine Rupestris stem pittingassociated virus. Eur J Plant Pathol 135:439-452. Chaves MM, Santos TP, Souza CR, Ortuño MF, Rodrigues ML, Lopes CM, Maroco JP and Pereira JS. 2007. Deficit irrigation in grapevine improves water-use efficiency while controlling vigour and production quality. Ann Appl Biol 150:237-252. Choné X, van Leeuwen C, Chéry P and Ribéreau-Gayon P. 2001. Terroir influence on water status and nitrogen status of non-irrigated Cabernet Sauvignon (Vitis vinifera). Vegetative development, must and wine composition (example of a Medoc Top Estate Vineyard, Saint Julien Area, Bordeaux, 1997). S Afr J Enol Vitic 22:8-15. Dai ZW et al. 2011. Ecophysiological, genetic, and molecular causes of variation in grape berry weight and composition: A review. Am J Enol Vitic 62:413-425. Deloire A. 2011. The concept of berry sugar loading. Wineland Wynboer 257:93-95. Deluc LG, Quilici DR, Decendit A, Grimplet J, Wheatley MD, Schlauch KA, Mérillon JM, Cushman JC and Cramer GR. 2009. Water deficit alters differentially metabolic pathways affecting important flavor and quality traits in grape berries of Cabernet Sauvignon and Chardonnay. BMC Genomics 10:212. Destrac Irvine A et al. 2015. Harmonizing grapevine phenology recording for budburst and flowering as recommended by the PER- PHECLIM project. In Proceedings of the Nineteenth International Giesco Meeting, pp. 641-645. Pech Rouge - Montpellier, France. Esteban MA, Villanueva MJ and Lissarrague JR. 1999. Effect of irrigation on changes in berry composition of Tempranillo during maturation. Sugars, organic acids, and mineral elements. Am J Enol Vitic 50:418-434. Fox J. 2016. Applied Regression Analysis and Generalized Linear Models. 3rd ed. Sage Publications, Inc. Friend AP, Trought MCT and Creasy GL. 2009. The influence of seed weight on the development and growth of berries and live green ovaries in Vitis vinifera L. cvs. Pinot noir and Cabernet Sauvignon. Aust J Grape Wine Res 15:166-174. Gaudillère JP, van Leeuwen C and Ollat N. 2002. Carbon isotope composition of sugars in grapevine, an integrated indicator of vineyard water status. J Exp Bot 53:757-763. Gillaspy G, Ben-David H and Gruissem W. 1993. Fruits: A developmental perspective. Plant Cell 5:1439-1451. Guidoni S, Allara P and Schubert A. 2002. Effect of cluster thinning on berry skin anthocyanin composition of Vitis vinifera cv. Nebbiolo. Am J Enol Vitic 53:224-226. Keller M. 2010. Managing grapevines to optimise fruit development in a challenging environment: A climate change primer for viticulturists. Aust J Grape Wine Res 16:56-69. Martelli GP and Savino V. 1990. Fanleaf degeneration. In Compendium of Grape Diseases. Pearson RC and Goheen AC (eds.), pp. 48-49. APS Press, St. Paul, MN. Matthews MA and Anderson MM. 1988. Fruit ripening in Vitis vinifera L.: Responses to seasonal water deficits. Am J Enol Vitic 39:313-320. Matthews MA and Nuzzo V. 2007. Berry size and yield paradigms on grapes and wines quality. Acta Hortic 754:423-436. May P. 2004. Flowering and Fruit Set in Grapevines. Lythrum Press, Adelaide, Australia. Ojeda H, Deloire A and Carbonneau A. 2001. Influence of water deficits on grape berry growth. Vitis 40:141-145. Pagay V and Cheng L. 2010. Variability in berry maturation of Concord and Cabernet franc in a cool climate. Am J Enol Vitic 61:61-67. Peyrot des Gachons C, van Leeuwen C, Tominaga T, Soyer JP, Gaudillère JP and Dubourdieu D. 2005. Influence of water and nitrogen deficit on fruit ripening and aroma potential of Vitis vinifera L cv. Sauvignon blanc in field conditions. J Sci Food Agric 85:73-85. Poni S, Bernizzoni F, Civardi S and Libelli N. 2009. Effects of prebloom leaf removal on growth of berry tissues and must composition in two red Vitis vinifera L. cultivars. Aust J Grape Wine Res 15:185-193. Prudent M, Dai ZW, Génard M, Bertin N, Causse M and Vivin P. 2014. Resource competition modulates the seed number fruit size relationship in a genotype-dependent manner: A modeling approach in grape and tomato. Ecol Model 290:54-64. Ribéreau-Gayon P, Dubourdieu D, Donèche B and Lonvaud A. 2012. Traité d œnologie. Dunod, Paris. Roby G, Harbertson JF, Adams DA and Matthews MA. 2004. Berry size and vine water deficits as factors in winegrape composition: Anthocyanins and tannins. Aust J Grape Wine Res 10:100-107. Scholander PF, Bradstreet ED, Hemmingsen EA and Hammel HT. 1965. Sap pressure in vascular plants. Science 148:339-346. Soubeyrand E, Basteau C, Hilbert G, van Leeuwen C, Delrot S and Gomès E. 2014. Nitrogen supply affects anthocyanin biosynthetic and regulatory genes in grapevine cv. Cabernet Sauvignon berries. Phytochemistry 103:38-49. Storchi P, Costantini EAC and Bucelli P. 2005. The influence of climate and soil on viticultural and enological parameters of Sangiovese grapevines under non-irrigated conditions. Acta Hortic 689:333-340. Trégoat O, van Leeuwen C, Choné X and Gaudillère JP. 2002. Etude du régime hydrique et de la nutrition azotée de la vigne par des indicateurs physiologiques. Influence sur le comportement de la vigne et la maturation du raisin. J Int Sci Vigne Vin 36:133-142. van Leeuwen C, Renard R, Leriche O, Molot C and Soyer JP. 1998. Le fonctionnement de trois sols viticoles du Bordelais: Conséquences sur la croissance de la vigne et sur le potentiel œnologique du raisin en 1997. Revue Française d Œnologie 170:28-32. van Leeuwen C, Friant P, Soyer JP, Molot C, Choné X and Dubourdieu D. 2000. Measurement of total nitrogen and assimilable nitrogen in grape juice to assess vine nitrogen status. J Int Sci Vigne Vin 34:75-82. van Leeuwen C, Tregoat O, Chone X, Bois B, Pernet D and Gaudillere JP. 2009. Vine water status is a key factor in grape ripening and vintage quality for red Bordeaux wine. How can it be assessed for vineyard management purposes? J Int Sci Vigne Vin 43:121-134. Venables WN and Ripley BD. 2002. Modern Applied Statistics with S. Springer-Verlag, Berlin. Walker RR, Blackmore DH, Clingeleffer PR, Kerridge GH, Rühl EH and Nicholas PR. 2005. Shiraz berry size in relation to seed number and implications for juice and wine composition. Aust J Grape Wine Res 11:2-8.