The study was conducted in 2023–2024 in an experimental vineyard of INRAE (Institut national de recherche pour l’agriculture, l’alimentation et l’environnement) Pech Rouge (Gruissan, France; 43.14° N, 3.14° W) in a hot, semi-arid Mediterranean climate. Three mildew-tolerant grapevine varieties planted in three different plots, no more than 385 m apart, were monitored. The varieties were: ARTABAN (a Vitis hybrid from the Resdur-1 programme), and 3159-B and G5 (3197-81-B), both bred by Alain Bouquet (two Vitis hybrids from a cross between Vitis vinifera and Muscadinia rotundifolia) (Salmon et al., 2018). In addition to its fungus-tolerance, G5 is characterised by reduced hexose accumulation in berries at harvest, leading to low-ethanol wines (Ojeda et al., 2017). The vines were grafted on drought-adapted rootstocks (110R for G5 and ARTABAN; 140RU for 3159-B) and planted in 2012 (3159-B) and 2015 (ARTABAN, G5) at a density of 4400 plants ha⁻¹ (2.5 × 0.9 m), with a southwest-northwest row orientation.

Eight treatments were applied at both plant and plot scale to modulate water, nitrogen, and carbon availability (Table S1). Plots composed of six plants were selected based on homogeneity criteria, including plant health, trunk circumference, vegetative growth, shoot weight, and bud number.

At the plot scale, irrigation and inter-row management differed among treatments. Guard vines separated the different treatments. One treatment involved complete inter-row vegetation removal, organic fertilisation at budburst (40 kg N ha⁻¹ in 2024 only, in order to enhance the difference in soil nitrogen content between the treatments), and weekly irrigation with total water applications of 87, 55, and 98 mm in 2023 and 72, 44, and 77 mm in 2024 for G5, 3159-B, and ARTABAN, respectively (BARE-I₂). These amounts were calculated by dividing the total water supplied per treatment by the soil surface area specific to the treatment. Another treatment comprised a cover crop sown in September and removed in mid-April, with irrigation applied every two weeks (68, 17, and 36 mm in 2023; 34, 16, and 50 mm in 2024 for G5, 3159-B, and ARTABAN, respectively) and no fertilisation to induce moderate water and nitrogen constraints (SOWN-I₁). The third treatment was applied only to G5 and ARTABAN, as the 3159-B plot was not big enough for all treatments with a sufficient number of replicates; it comprised spontaneous inter-row vegetation being maintained until June, without irrigation or fertilisation, resulting in high water and nitrogen constraints (SPO-I₀).

Canopy management strategies comprised vertical shoot positioning (VSP) with standard (75–105 cm) (VSPH) or reduced canopy height (30–60 % reduction, applied in late June) (VSPLw). For 3159-B, minimal pruning was also tested. These treatments aimed to alter the carbon source-sink balance at the plant scale. Three guard vines also separated the different canopy treatments when positioned alongside the same row.

The year 2023 was particularly dry, with 271 mm of rainfall, of which 116.5 mm between April and August. In 2024, precipitation nearly doubled (491 mm), but it remained similar to 2023 during the growing season (152 mm). Cumulated thermal time (base temperature 10 °C, Lebon et al. (2004)) was higher in 2023 (2555 °Cd vs. 2246 °Cd over the year), but the values over the growing seasons of both years were similar (1689 °Cd in 2023, 1595 °Cd in 2024).

Soil characteristics were studied in three parcels, with two to three soil pits per parcel, whose locations were chosen based on soil management treatments and surface heterogeneity. According to the World Reference Base for Soil Resources (WRB), the soil in the G5 plot was classified as Calcisol (leptic) in the upper two-thirds and as Cambisol (coluvic) in the lower part. Soils in the 3159-B and ARTABAN plots were classified as Calcisol (arenic). The 3159-B plot had deeper soil than G5, but its sandier texture resulted in lower retention capacities. The soil water holding capacity (AWC) was similar for both G5 (97 mm) and 3159-B (116 mm), while the ARTABAN plot had the lowest depth and an AWC of 70 mm.

For the following measurements, we selected mature and healthy leaves, located halfway up the canopy and fully exposed to sunlight (south-east side of the row). Predawn leaf water potential (Ψ_pd) is an indicator of plant and soil water status and is less sensitive to climatic conditions than midday stem potential. Here, it was measured during each leaf gas exchange measurement session using a Scholander chamber on one leaf from six plants per treatment between 1:00 and 5:00. From mid-June to mid-August, leaf gas exchange measurements were conducted every three weeks on sunny days, between 9:00 and 12:30, when Photosynthetically Active Radiation (PAR) was over 1500 µmol m^-2 s^-1 on six plants per treatment. The order in which the plots and the treatments inside each plot were measured was randomly changed between each measurement session. Chlorophyll fluorescence and stomatal conductance were measured using a porometer-fluorimeter (LI-600; LI-COR Biosciences) on one or two leaves per plant, depending on the weather conditions during the session. Simultaneously, maximum CO₂ assimilation rate (A_max) was assessed under ambient CO₂ (400 μmol mol^-1) and saturated light (1500 μmol m² s^-1) using an infrared gas analyser (LI-6800; LI-COR Biosciences) on three leaves per treatment, with optimal chamber conditions (27.5 °C, VPD 1.5–2 kPa, saturated light).

On each leaf measured for ecophysiological parameters, a leaf disc was sampled using a punch, then dried. For three leaves per treatment, a second disc was sampled per leaf and frozen for subsequent biochemical analysis of starch, glucose, fructose, sucrose, and nitrogen content (hereafter referred to as LS, LG, LF, LSc and LN, respectively). NIRS measurements were performed on dried samples using a laboratory NIR spectrometer (ASD LabSpec4) with fibre optics to estimate nitrogen and non-structural carbohydrate contents for samples not analysed by biochemical assays.

The carbon balance of aerial organs per plant was estimated using the Leaf-to-Fruit ratio (L:F, cm² berry^-1), defined as the ratio total leaf area (TLA) to total number of berries. L:F values were calculated at harvest. Measurements taken on plants from the VPSLw treatment before late June were excluded, as leaf area values prior to trimming at the end of June would have overestimated the true leaf area after pruning.

To estimate TLA for each of the six plants per treatment, 15 leaves from primary and secondary shoots were sampled, and their surface area was measured with a planimeter (LI-3100C; LI-COR Biosciences). Mean leaf areas for primary (LA_I) and secondary (LA_II) leaves were determined. The lengths of 15 primary shoots were measured for each treatment ($\stackrel{-}{{SLg}_I}$). The number of primary and secondary leaves (NL_I and NL_II, respectively) was also recorded to calculate the average number of leaves per centimetre of primary shoot (L_I and L_II, respectively). The number of shoots per plant (NS_I) was counted on each of the six plants per treatment. The total leaf area per plant was then estimated using the following equation:

$Leaf Total Area= {NS}_I\times \stackrel{-}{{SLg}_I}\times (\stackrel{-}{L_I}\times {LA}_I+\stackrel{-}{L_{II}}\times {LA}_{II})$

where $\stackrel{-}{{\mathrm{L}}_{\mathrm{I}}}$ and $\stackrel{-}{{\mathrm{L}}_{\mathrm{I}\mathrm{I}}}$ are the mean values of L_I and L_II in each treatment.

Partial Least Squares (PLS) regression models were developed to predict leaf nitrogen and carbon-related variables using NIRS spectra or pre-processed NIRS data (e.g., smoothed spectra, standard normal variate spectra, first and second derivatives). The models were built using the ‘pls’ package in R, based on 270 samples on which LS, LG, LF, LSc, and LN were measured by biochemical assays. Model performance was assessed on the basis of R², normalised root mean square error (nRMSE, i.e., RMSE divided by the range of observed values), and relative bias (i.e., bias relative to the mean of observed values). The best models showed R² values of 0.76 and nRMSE of 7.1 % for LN but with a tendency to underestimate the observed values (relative bias = –7.7 %) (Table 1). For the other variables, the bias was close to 0 and R² = 0.78 for LS with nRMSE of 9.6 %, and 0.82 for LG with nRMSE of 8.2 %. For LF and LSc, the R² values were lower (0.58 and 0.55, respectively) with higher nRMSE (12.2 % and 15.3 %, respectively). LG, LF, and LSc concentrations were summed to estimate total non-soluble sugar content (LSS).

For photosynthesis (A_max), a calibration model was built using 322 A_max measurements and variables from LI-600 (e.g., Electron Transport Rate, Photosystem II activity, stomatal conductance, leaf temperature, and VPD). It was verified that the variations in leaf temperature, resulting from light intensity increase during the morning, were equally distributed for each treatment in the calibration set (Figure S2). The A_max model had an R² of 0.71 and nRMSE of 12.5 %, with a tendency to overestimate values.

*Max, Min, Maximum and Minimum of calibration dataset; SD, Standard deviation of calibration dataset; CV, coefficient of variation [(SD/mean)*100]

All statistical analyses and graphics were performed using R software. The analyses related to the genotypic effect were performed by assessing the differences between the genotypes for a given set of combined values of Ψ_pd, LN, and L:F, which reflect the water, nitrogen and carbon status of the plants. This approach was used because other interactive factors not considered in the present study (e.g., scion × rootstock interactions, nutritional status, pedo-climatic variables, and microclimatic variability) could potentially influence genotype effects.

A multiple linear regression model was built to examine the effects of Ψ_pd, LN, and L:F on A_max. The significance of each explanatory variable was assessed with a Type II ANOVA. The proportion of variance of A_max explained by each variable was computed as the ratio of its sum of squares to the total sum of squares. To meet ANOVA assumptions (normality and homoscedasticity of residuals), A_max was log-transformed.

A_max was classified into three groups: i) high (> 14 µmol m^-2 s^-1, 30 % of data), ii) low (< 8 µmol m^-2 s^-1, 10 % of data), and iii) intermediate (8–14 µmol m^-2 s^-1). A concentration ellipsoid was fitted to the 3D point cloud of Ψ_pd, LN, and L:F to enclose the maximal number of high-A_max values while minimising low-A_max values, using the ellipse3d function from the rgl package. The ellipsoid’s centre and its minimum and maximum Ψ_pd, LN, and L:F values were computed. The range of each variable within the ellipsoid was expressed as a fraction of its total observed range in the dataset.

Pearson correlation coefficients were computed among A_max, LN, L:F, Ψ_pd, LS, and LSS, both across and within genotypes, and their significance was assessed.

Structural equation modeling (path analysis, Shipley (2016)) was used to analyse the relationships between water, source-sink, and nitrogen-related variables (Ψ_pd, L:F, LN), maximal photosynthetic activity (A_max), and Non Structural Carbohydrate (NSC) contents (LSS and LS). The same model was applied across genotypes and measurement periods (before or after veraison) to assess differences in variable relationships based on standardised Lavaan coefficients. This approach accounts for all covariations among variables, estimating the independent effect of each one. The model was designed to integrate physiological knowledge about variable interactions. It assumed causal relationships: (1) between L:F, Ψ_pd, LN, and A_max, (2) between L:F, Ψ_pd, LN, and NSC content (LSS, LS), and (3) between A_max and LSS/LS, with a free correlation between LSS and LS. Potential causal links among water, source-sink, and nitrogen-related variables were also considered, including relationships between L:F and LN, Ψ_pd and LN, and a free correlation between L:F and Ψ_pd. The final model retained only significant relationships, leading to the removal of some pre-established links, such as the causal effects of Ψ_pd on LS and LSS.

The experimental design generated a broad range of predawn leaf water potential (Ψ_pd), leaf nitrogen surface concentration (LN), and leaf-to-fruit ratio (L:F) over the two years (Figure 1). Ψ_pd tended to decrease after veraison. Before veraison, its values were similar across genotypes, but after veraison, G5 reached the lowest values (-1.5 MPa); meanwhile, ARTABAN never dropped below -1 MPa. Additionally, 3159-B and G5 exhibited greater Ψ_pd variability than ARTABAN after veraison (Figure 1A).

Seasonal LN variations were less pronounced than those of Ψ_pd, with a slight decline after veraison for G5 and 3159-B (median values dropping from 0.13 and 0.17 to 0.12 and 0.14 mg cm^-2, respectively). G5 consistently exhibited lower LN than ARTABAN and 3159-B, particularly after veraison (Figure 1B). Before veraison, ARTABAN displayed the lowest LN variability, but after veraison, all genotypes showed similar levels of variation.

L:F varied widely among genotypes, with G5 showing the highest values (median of 53 cm² berry^-1), followed by 3159-B (30 cm² berry^-1) and ARTABAN (12 cm² berry^-1). ARTABAN had the smallest within-genotype variability, with a range of 4.2–33.3 cm² berry^-1 (Figure 1C).

ARTABAN exhibited higher maximal photosynthesis (A_max), with a median value of 14.3 μmol m^-2 s^-1, compared to 11.4 and 12.3 μmol m^-2 s^-1 for G5 and 3159-B, respectively (Figure 1D). Median A_max values declined after veraison for all genotypes. Variability remained comparable for G5 and 3159-B before and after veraison, whereas ARTABAN showed more variability after veraison, with a larger number of very high values (A_max > 25 μmol m^-2 s^-1).

Boxplot representation of the variables used to determine water and nitrogen leaf status, balance between sink and source organs for carbon, and carbon assimilation rate in G5, 3159-B and ARTABAN in 2023 and 2024: a) predawn leaf potential Ψ_pd for water status, b) nitrogen estimated surface leaf concentration LN for nitrogen status, c) Leaf-to-fruit ratio L:F for balance between sink and source organs for carbon, and d) photosynthetic maximal net activity (A_max) for carbon assimilation rate. Each point represents the value for one leaf at a given measurement period (n = 950).

Figure 1. Boxplot representation of the variables used to determine water and nitrogen leaf status, balance between sink and source organs for carbon, and carbon assimilation rate in G5, 3159-B and ARTABAN in 2023 and 2024: a) predawn leaf potential Ψ_pd for water status, b) nitrogen estimated surface leaf concentration LN for nitrogen status, c) Leaf-to-fruit ratio L:F for balance between sink and source organs for carbon, and d) photosynthetic maximal net activity (Amax) for carbon assimilation rate. Each point represents the value for one leaf at a given measurement period (n = 950).

The variation of A_max as a function of Ψ_pd, LN, and log(L:F) was analysed considering the whole dataset (Figure 2) and separately by measurement period (Table 2). An ellipsoid was fitted to capture the highest number of high A_max values (> 14 µmol m^-2 s^-1) while minimising low A_max values (< 8 µmol m^-2 s^-1), thus grouping 72.2 % of high A_max and only 6.4 % of low A_max (Table 2). The ellipsoid's coordinates ranged from -0.12 to -0.75 MPa for Ψ_pd, 0.09 to 0.21 mg cm^-2 for LN, and 1.66 to 4.59 cm² berry^-1 for log(L:F), encompassing 45 %, 59 %, and 80 % of total dataset variation in these variables, respectively. This suggests a stronger influence of Ψ_pd on A_max than the other variables. The ellipsoid centre (Ψ_pd = -0.44 MPa; LN = 0.15 mg cm²; log(L:F) = 3.12 cm² berry^-1) indicates that high A_max values are associated with higher Ψ_pd and LN but lower L:F ratios.

When the dataset by phenological stage was split before veraison, the ellipsoid captured 85.5 % of high A_max values, with only 3.6 % of low A_max. After veraison, 72 % of high A_max values were captured, but low A_max values increased to 10.6 %. The impact of each variable (Ψ_pd, LN and L:F) on A_max variations within the ellipsoid also shifted over time. Before veraison, Ψ_pd had the strongest impact, the ellipsoid covering only 45.9 % of its total variation, compared to 87.7 % for LN and 88.4 % for L:F. After veraison, the contributions of Ψ_pd and LN to discriminating A_max values were similar (around 50 % of their total variations), while L:F discriminated A_max values to a lower extent (78 % of its total variation).

Figure 2. 3D representation of photosynthetic maximal net activity (A_max) depending on predawn leaf potential Ψ_pd, nitrogen estimated surface leaf concentration LN and Leaf-to-fruit ratio log(L:F), respectively.

^a A_max values in the ellipsoid are classified as high A_max (superior to 14 µmol m^-2 s^-1 yellow dots), intermediate A_max values (between 8 and 14 µmol m^-2 s^-1, light blue dots), low A_max (inferior to 8 µmol m^-2 s^-1, dark blue dots).

The correlation analyses revealed numerous associations between variables, but the high number of correlations prevented us from drawing clear conclusions about causal relationships. To disentangle the specific contributions of each variable, we applied structural equation modelling to the dataset. The final model retained only significant relationships between the variables, leading to the removal of pre-established causal relationships, such as those between Ψ_pd, LS, and LSS. This analysis revealed that A_max was directly influenced by Ψ_pd across both periods and all genotypes (Figure 5A and 5B). Before veraison, L:F was positively associated with Ψ_pd for the whole dataset, as well as for 3159-B and ARTABAN. However, after veraison, this relationship became negative for the whole dataset and G5. Additionally, the analysis showed that the positive correlation between A_max and LN observed after veraison (Figure 4) mainly resulted from an indirect effect via the negative relationship between Ψ_pd and LN, as no clear effect of LN on A_max was observed, except for a slight positive effect before veraison.

The path analysis further showed that A_max was positively associated with LS, and that L:F also directly influenced LS independently of A_max. This direct effect was positive before veraison for the whole dataset, G5, and ARTABAN, but became negative after veraison for the whole dataset and G5 while remaining positive for ARTABAN. Although Ψ_pd and LS were positively correlated for the whole dataset and by genotype, the path analysis showed that this relationship was mainly an indirect effect via A_max, as no clear direct relationship between Ψ_pd and LS was indicated, except in G5 after veraison, where Ψ_pd had a small direct negative effect on LSS.

Furthermore, the positive correlation between L:F and LSS for the whole dataset was mainly attributed to a direct effect of L:F, with no indirect influence through A_max. Finally, after veraison, LS and LSS showed a small but significant relationship, whose nature (positive or negative) varied depending on the genotype.

Figure 5. Results of the path analysis performed on photosynthetic maximal net activity A_max (with logarithm transformation), leaf water, nitrogen, carbon status, and balance between sink and source organs for carbon, assessed on the basis of predawn leaf potential Ψ_pd, estimated leaf surface nitrogen concentration (LN), starch, estimated leaf soluble sugars concentrations (LS and LSS), and Leaf-to-fruit ratio (L:F), respectively for all the genotypes or considering G5, 3159-B and ARTABAN, separately before (5A) and after (5B) veraison.

^a Arrow thickness is associated with the strength of the relationships.

The distribution of A_max as a function of Ψ_pd, LN and log(L:F) was analysed over the whole season for the three genotypes (Figure 2). The fitted ellipsoid captured a major part of high A_max values (> 14 µmol m^-2 s^-1), while limiting the inclusion of low values (< 8 µmol m^-2 s^-1). The ‘optimal’ ellipsoid centre for high A_max values corresponded to intermediate Ψ_pd (-0.44 MPa) and LN (0.15 mg cm²), with L:F at around 3.12 cm² berry^-1, (i.e., around 15 cm² g of fruit), slightly above the threshold known to reach ‘optimal’ maturity (Howell, 2001). Beyond this centre, the ellipsoid contains high A_max values which were associated with quite a wide range of Ψ_pd values, including very low values (-0.75 MPa) associated with severe stress (Deloire et al., 2004). The ranges of LN and L:F within the ellipsoid were even higher than those of Ψ_pd, indicating the limited contribution of these two factors to reaching high A_max.

Unsurprisingly, except for ARTABAN, which displayed high A_max for non-irrigated treatment (SPO-I₀), the ellipsoid mostly includes the irrigated treatments, primarily the most irrigated one (BARE-I₂), and then the intermediate one (SOWN-I₁) (Figure 3). Interestingly, high A_max were mainly observed for VSP with a low canopy in both I₂ and I₁ for G5, and to a lesser extent for I₁ in ARTABAN (no low VSP was tested for 3159-B, I₁). However, high A_max was achieved for these treatments despite the low Ψ_pd values after veraison (< -0.5 MPa) (data not shown). For these treatments (i.e., VSPLw-I₂, VSPLw-I₁), we hypothesized that photosynthesis was stimulated by the low source-sink ratio to achieve berry maturation, as mentioned previously (Poni et al., 2006; Zufferey, 2000). On the other hand, in some situations, reduced total leaf area may have mitigated the daily decline in water potential by reducing transpiration at the plant level, which can in turn limit the drop in photosynthesis, as previously observed under high evaporative demand or water deficit conditions (Abad et al., 2019; Pascual et al., 2015; Mirás-Avalos et al., 2017). Nevertheless, depending on the genotypic characteristics, the trimming practices that are used to reduce leaf area development can induce a compensation of leaf area reduction by the development of lateral axes, thus limiting the expected effects of such practices (Poni et al., 2023). Ultimately, as well as the three variables that were measured to explain A_max variations (Ψ_pd, LN, L:F), additional factors influencing overall plant functioning (e.g., transpiration dynamics driven by microclimatic fluctuations or differences in aerodynamic conductance at the leaf level) may have also affected the photosynthesis of the targeted leaf (Villalobos-González et al., 2019; Albasha et al., 2019).

A path analysis was conducted to explore the causal relationships in a highly interdependent system in which most factors influencing photosynthesis and carbohydrate concentration are strongly correlated. A_max was directly influenced by Ψ_pd, reflecting the plant’s rapid adaptation to water constraint not only through stomatal control (Medrano et al., 2002), but also through non-stomatal limitations, affecting light harvesting efficiency and biochemical processes or mesophyll conductance under severe water deficit (Lawlor & Tezara, 2009; Villalobos-González et al., 2022); these limitations are known to be coordinated (Flexas et al., 2016; Hochberg et al., 2013).

Higher water status promotes vegetative growth, increasing L:F (Lebon et al., 2006). Accordingly, Ψ_pd and L:F were positively correlated before veraison for 3159-B and ARTABAN. After veraison, the negative L:F and Ψ_pd relationship observed for G5 likely means that increased leaf area (source organs) (Figure S1) led to greater transpiration at the plant scale, reducing plant water status (Bravdo et al., 1985; Intrigliolo & Castel, 2011; Santesteban et al., 2011). This corroborates the indirect and negative impact of L:F on A_max through Ψ_pd after veraison for G5. Furthermore, L:F showed a direct negative effect on A_max for 3159-B and ARTABAN, consistent with photosynthesis regulation by sink demand (Naor et al., 1997; Poupard et al., 2024; Zufferey et al., 2009). The direct and independent effects of Ψ_pd and L:F on A_max are consistent with the findings of Poupard et al. (2024) and Dayer et al. (2016).

The positive relationship between Ψ_pd and LN can be explained by the fact that nitrogen root assimilation improves under sufficient soil water availability (Verdenal et al., 2021). While nitrogen availability can directly influence plant growth (Metay et al., 2014) and photosynthesis (Prieto et al., 2012) under well-watered conditions, this direct effect was not observed in our study, likely due to the greater impact of water deficit. Future studies using foliar nitrogen application, would allow leaf nitrogen content to be controlled more precisely without interference from soil water availability.

A positive effect of A_max on both LS and LSS was expected, as carbon assimilation depends on photosynthesis. Higher photosynthetic rates promote carbohydrate accumulation, primarily as starch (Gordon et al., 1986). Before and after veraison, a direct positive impact of L:F on LSS was also observed, as in previous studies (Zufferey et al., 2009). In high L:F conditions, sugar accumulation in leaves was observed due to the low carbon demand for sink growth compared to the amount of carbohydrate production. As such, L:F had a direct positive impact on LS before veraison. After veraison, this effect was less clear: it remained positive for ARTABAN (Faralli et al., 2022) but turned negative for G5, likely due to increased sink demand from grape growth, preventing transient sugar storage in leaves (Dayer et al., 2016).

In many species, a direct feedback effect of sugar accumulation in leaves has been observed (Pallas et al., 2018). However, in our study, no negative relationship between starch and photosynthesis was found. This suggests that the source-sink ratio likely affects photosynthesis through the regulation of stomatal conductance, rather than through sugar accumulation. This is supported by the literature, which highlights two main mechanisms for regulating photosynthesis: stomatal closure and end-product saturation, both influenced by fruit or seed carbon demand (Wünsche et al., 2005; Andrade et al., 2019).

While direct effects of water status on LS and LSS have been reported, albeit with contradictory results, in our study no clear effect was observed. Previous studies have found that water stress increases LSS while decreasing LS, with total LS+LSS remaining constant (Dayer et al., 2016; Pellegrino et al., 2014). This is likely due to the conversion of LS to LSS to maintain turgor pressure via osmotic adjustments and respiratory metabolism under prolonged water deficit (Düring et al., 1984; McDowell, 2011).