A continuum of stomatal responses to water deficits among 17 wine grape cultivars (Vitis vinifera)

Additional keywords: genotypic variation, grapevine physiology, stomatal conductance, water potential.

Increasing demand for fresh water in the face of finite resources is intensifying pressure to improve water use efficiency in agriculture in general, and in crop production in particular. Because regulation of stomatal aperture plays a major role in plant control of water loss in dry environments, it has been the focus of intensive research directed towards improved drought tolerance and water use efficiency. For example, reduced maximum transpiration rate (Sinclair et al. 2005) and early stomatal closure (Martin-St. Paul et al. 2017) have been proposed as traits of drought avoidance. In both cases, the putative benefit in yield stability arises from postponing dehydration. Indeed, the sensitivity of stomata to water deficits could be viewed as a measure of drought avoidance (Jones 1974; Bates and Hall 1981) and is sometimes described in terms of ‘isohydric’ vs ‘anisohydric’ behaviour (Tardieu and Simonneau 1998).

In addition to species variation, it has long been known that genetic variation in stomatal response to water deficits is present within crop species (e.g. in sorghum, Henzell et al. 1976; e.g. in rice, Dingkuhn et al. 1989). Schultz (2003) provided some evidence of such genetic variation within grapevine, Vitis vinifera L., but there have since been contradictory reports of cultivar-specific responses (Chaves et al. 2010; Costa et al. 2012). Ambiguity over the precise definition of what constitutes an iso- vs anisohydric response (Hochberg et al. 2018) has caused difficulties in categorisation of either cultivars within a species (Lavoie-Lamoureux et al. 2017) or across many species (Klein 2014; Martinez-Vilalta and Garcia-Forner 2017). However, in general, isohydric species/cultivars will maintain water status similar between droughted and well-watered plants (Tardieu and Simonneau 1998) or that a measure of plant water status (such as leaf water potential) will not decrease below a minimum value on a diurnal or seasonal basis (Schultz 2003). Such a response by a crop species may minimise the usefulness of assessing plant water status via the measurement of leaf water potential. Cifre et al. (2005) concluded that if grapevine cultivars respond to soil water deficits in an isohydric manner, then midday leaf water potential cannot be used in an irrigation management program or as one would suspect the validation of other means to assess plant water status (Williams 2017).

Based on observed responses of plant water relations and gas exchange, some grapevine cultivars such as Grenache, have been classified as isohydric (Schultz 2003; Cifre et al. 2005) whereas others, such as Syrah and Thompson Seedless have been classified as anisohydric (Schultz 2003; Soar et al. 2006b; Williams and Baeza 2007; Williams 2012). However, the discrepancy among cultivars is not entirely clear. This has led some researchers to modify the binary categorisation (i.e. isohydric vs anisohydric) and propose that cultivars may modulate their stomatal response (Domec and Johnson 2012) based on changes in soil moisture (Collins et al. 2010), or differences in leaf to air vapour pressure deficit (VPD) (Soar et al. 2006b). In addition, acclimation of stomatal response to water deficits has been shown in grapevine through a modification of leaf hydraulic conductivity and osmotic adjustment (Martorell et al. 2015; Hochberg et al. 2017a). Accordingly, alternative classifications such as ‘near-isohydric’ and ‘isohydrodynamic’ (a relatively constant difference between predawn (Ψ_PD) and midday leaf water potential (Ψ_l) throughout seasonal moisture cycles) have been offered to account for differences in stomatal response to drought among cultivars and the resulting effects on plant water status (Schultz 2003; Franks et al. 2007; Chaves et al. 2010; Shellie and Bowen 2014).

Putative cultivar differences in stomatal response would be important for water-conserving vineyard management strategies because g_s is highly correlated with daily water use (Williams et al. 2012), and because moderate water deficits have positive effects on fruit quality (Roby et al. 2004). Furthermore, relative to vegetative growth, reproductive growth of irrigated grapevines is less affected when applied water amounts are reduced by up to 40% of crop evapotranspiration (Williams et al. 2010a, 2010b). As such, some water deficit is beneficial in production of horticultural crops like wine grapes in contrast to production of agronomic crops like rice or sorghum.

The relationship between g_s and Ψ_l in grapevine is affected by numerous factors; predominate among them are rootstock and cultivar, accounting for 19 and 16% of the variability in that relationship, respectively (Lavoie-Lamoureux et al. 2017). Presumably, a large proportion of the remaining variation in response would be accounted for by plant-environment interactions (Hochberg et al. 2018). As mentioned above, soil moisture, VPD and plant hydraulic properties have all been shown to influence the relationship between g_s and Ψ_l across grapevine cultivars. However, since the initial classification by Schultz (2003), and numerous subsequent attempts by others – as reviewed by Chaves et al. (2010) – there have been no studies that have directly compared a large set of cultivars grown under the same environment conditions (soil and atmosphere) grafted onto the same rootstock in a replicated field trial.

The goal of this study was 2-fold: to (1) determine if numerous V. vinifera cultivars separate into distinct iso- or anisohydric classes in response to soil water deficits from measurements of predawn (Ψ_PD) and midday leaf water potential (Ψ_l); and (2) explore the genetic variation in the relationship of g_s to Ψ_l among these cultivars over several seasons. To accomplish this, the study was designed to maximise the effect of cultivar on the relationships in question by controlling for numerous factors that have all been shown to be influential such as rootstock, environment, instrument operator, and sampling methodology. Several grapevine cultivars evaluated in this study have been previously classified as being isohydric (or near-isohydric) – e.g. Cabernet Sauvignon, Grenache, Tempranillo and Touriga Nacional – while others have been classified as anisohydric – e.g. Cabernet Sauvignon, Grenache, Montepulciano, Syrah (syn. Shiraz) and Touriga Nacional (Schultz 2003; Chouzouri and Schultz 2005; Soar et al. 2006b; Williams and Baeza 2007; Chaves et al. 2010; Lovisolo et al. 2010; Rodrigues et al. 2012). Finally, an additional 11, red wine grape cultivars were evaluated representing a broad range of plant material grown in warmer viticultural regions around the world.

The experimental site used in this study was an existing variety trial that was planted with 1103 Paulsen (1103P) (Vitis berlandieri Planch. × Vitis rupestris Scheele) rootstock in June 2003 at the University of California Kearney Agricultural Research and Extension (KARE) Center near Parlier, California (36°48′N, 119°30′W). The soil was a Hanford fine sandy loam. The total vineyard area was ~0.53 ha. Vine and row spacing was 1.83 and 3.05 m respectively (5.58 m² per vine).

The trellis system consisted of a 0.3 m cross-arm atop a stake, 1.7 m above the vineyard floor. Fixed catch wires were positioned at each end of the cross-arms. Vines were trained to bilateral cordons on a single fruiting wire at a height of 1.2 m, and were pruned to spurs. In early spring, shoots were thinned to a density of ~15 shoots per meter of row. The vineyard was drip-irrigated with two, 2 L h^–1 emitters per vine placed 0.5 m on either side of the trunk. The dripline was attached to a wire installed 0.5 m above the vineyard floor.

The vineyard layout consisted of 20 rows with 48 vines per row. In May 2004, 20 different Vitis vinifera L. cultivars were field-budded onto the rootstock. Foundation Plant Services (University of California, Davis, CA, USA) supplied budwood for the scion varieties. The cultivars, clonal selection, and sources of budwood are given in Table 1. Carmenere, Tinta Francisca and Tinta Cão were not included in this study due to foliar symptoms of viral infection.

Table 1.  Clone and source of budwood for the 20 cultivars planted at the Kearney Agricultural Research and Extension Centre

Three irrigation treatments were imposed beginning in the 2012 growing season as detailed below.

1. Late deficit (LD) – irrigated at 100% of estimated ET_c from berry set until the onset of ripening (veraison); no applied water from veraison until harvest.

2. Sustained deficit (SD) – irrigated at 50% of estimated ET_c throughout the entire growing season.

3. Early deficit (ED) – no applied water from berry set until veraison; irrigated at 50% of estimated ET_c from veraison until harvest.

Prior to berry set, all vines were irrigated at 100% of estimated ET_c to maintain midday leaf water potential (Ψ_l) at or above –1.0 MPa. After berry set, irrigation treatments were imposed once Ψ_l reached the threshold value of –1.0 MPa (averaged across cultivars). Drip emitters were plugged or unplugged to impose the irrigation treatments, and inline water meters were used to quantify applied water amounts. Veraison timing was averaged across replicates of a cultivar-treatment combination (e.g. separately quantified for Syrah-ED vs Syrah-LD). Therefore, treatment transition (e.g. in the ED and LD treatments) occurred over a period of several weeks across cultivars. After harvest, all vines were irrigated at 50% of estimated ET_c until leaf fall.

Vineyard ET_c was estimated using the following equation: ET_c = ET_o × K_c, where ET_o is reference ET and K_c is the crop coefficient. Prior to the initiation of the experiment, seasonal crop coefficients were developed in this vineyard across seven growing seasons during which vines were kept well watered, so as to not inhibit canopy development (LE Williams, unpubl. data). They were determined by measuring the shade cast on the ground beneath the canopy, and then using the relationship between the percentage of canopy shaded area and K_c according to Williams and Ayars (2005). Finally, a relationship between K_c and accumulated growing degree-days (GDD) was established. During the experiment, evolution of the K_c over the course of the growing season was calculated using accumulated GDD from average budbreak. GDD data were downloaded from the University of California Statewide Integrated Pest Management Project’s website (www.ipm.ucdavis.edu, accessed 12 September 2019). ET_o and GDD data were obtained from the CIMIS weather station at the KARE Center (Parlier #39, California Irrigation Management Information System). GDD were calculated using the single sine method with a lower threshold of 10°C.

Vine water status (water potential; Ψ) was measured according to (Williams and Araujo 2002) using a pressure chamber (Model 1000, PMS Instrument Co.). Specifically, predawn water potential (Ψ_PD) measurements began at ~0400 h Pacific daylight time (PDT) and were completed before sunrise, whereas midday leaf water potential (Ψ_l) measurements were generally taken between 12:30 and 13:30 hours PDT. Ψ_PD was only measured in 2012. Leaves chosen at predawn and at midday were fully expanded, mature leaves – and at midday exposed to direct solar radiation. Leaf blades chosen across all times of day were covered with a plastic bag, quickly sealed, and petioles then cut within 1 to 2 s (s). Time between leaf excision and pressurisation was generally between 10 to 15 s.

Stomatal conductance (g_s) was measured with a steady-state diffusion porometer (LI-1600, LI-COR Biosciences). The porometer was allowed to warm-up to ambient vineyard conditions (placed in the shade) at least 30 min (min) before the first measurement. The null point humidity was set to ambient after warm-up and before first measurements, and subsequently reset every 20 to 30 min. Measurements were recorded after steady-state conditions were reached (10–20 s). Care was taken to reduce sensor head exposure to direct sunlight in between measurements so as not to increase cuvette temperature. Sampled leaves were of same age and position in canopy as for Ψ_l measurements described above, thus ensuring light saturation.

During the 2012 growing season, the measurements of Ψ_l and g_s were made by LE Williams for all cultivar/irrigation treatment replicates and required two days. In 2013, measurement techniques were refined by AD Levin (data not reported). During the last 2 years of the study, g_s was measured on the same leaves used for the measurement of Ψ_l, just before leaf excision. Sample leaves were selected for g_s, which was measured by LE Williams, and Ψ_l was measured by AD Levin both years, thus minimising potential operator error (Goldhamer and Fereres 2001). Measurements began at 12:00 hours and were completed by 14:00 hours those 2 years. A single leaf from each three-vine irrigation treatment replicate was measured and used for data analysis (n = 3–4). There were two, 16, and four measurement dates for the 2012, 2014 and 2015 growing seasons respectively.

Stomatal response to Ψ_l was characterised with the following sigmoidal model:

where Ψ_l is midday leaf water potential, g_max is the asymptote that represents a theoretical maximum stomatal conductance, Ψ_l⁵⁰ is the inflection point of the curve that represents the value of Ψ_l at which stomatal conductance is half of the maximum, and b is a scaling parameter that represents the transition width of the function. From the fitted curves, vine water status values were extracted at fixed values of g_s relative to g_max (Ψ_l⁹⁵, Ψ_l⁷⁵, Ψ_l⁵⁰ and Ψ_l²⁵ represent the estimated Ψ_l at 95, 75, 50 and 25% of g_max respectively) for each cultivar, with Ψ_l⁹⁵ assumed to be the threshold for the beginning of stomatal closure. The value of b was determined to be equal to (ΔΨ_l^75–25 – 0.002) × 2.196^–1 by linear regression, where ΔΨ_l^75–25 is the difference in Ψ_l (MPa) between 75 and 25% of g_max (data not shown; R² = 0.99, P < 0.001). The instantaneous slope of the fitted sigmoidal curve was calculated as the first derivative of the function at the inflection point and was taken as a measure of the maximum sensitivity of stomatal closure (g_sensitivity; mol m^–2 s^–1 MPa^–1).

The experimental design was a randomised complete block with a strip-plot factorial treatment structure and four replications. Experimental blocks were assigned across the rows (i.e. 20 rows per block), and scion cultivars (main plot) were field-budded down the rows in a pattern of 12 vines per cultivar per block (i.e. four cultivars per row). The three irrigation treatments (strip plots) within each block were arranged down the rows in a line-source design (Williams et al. 2010a, 2010b), alternating sequentially among blocks from LD-SD-ED to ED-SD-LD such that vines in adjacent blocks received the same irrigation treatment. One non-irrigated vine was included among irrigation treatments within a block, such that an experimental unit consisted of the three remaining vines in the irrigation treatment.

Statistical analyses and graphics were done using R statistical software (R Core Team 2018). Linear regression, non-linear regression, and ANOVA were conducted using the functions lm, nls and anova, respectively, from R base package stats. For the analyses of Ψ_l and g_s over time, a linear mixed-effects model was fit using the lmer function from the packages lme4 and lmerTest with cultivar, irrigation treatment, and sample date (and all interactions) as fixed factors, and block (and its interactions with the fixed factors) as a random factor (Bates et al. 2015; Kuznetsova et al. 2017). Calculation of least-squares means and multiple comparisons were conducted using the package lsmeans and the Tukey-Kramer adjustment method for multiplicity (Lenth 2016). Prior to non-linear regression analyses of g_s in response to Ψ_l, g_s values were pooled and means calculated across cultivars at each 0.1 MPa of Ψ_l. Smoothed cubic regression splines were fitted using the gam function from the package mgcv. (Wood 2017). Data were plotted using the package ggplot2 (Wickham 2016).

Predawn (Ψ_PD) and midday leaf water potential (Ψ_l) were measured on vines in each irrigation treatment on numerous dates during the 2012 growing season. Although there were significant differences among irrigation treatments in both measures of vine water status, there were never significant differences among cultivars within an irrigation treatment. On 24 July 2012, just before changing the irrigation treatments that year, mean Ψ_PD (±s.d.) of vines in the ED treatment averaged across cultivars was –1.18 ± 0.13 MPa whereas those in the LD treatment averaged –0.13 ± 0.02 MPa. Averaged across cultivars, midday Ψ_l of the ED and LD treatments were –1.78 ± 0.02 and –0.98 ± 0.02 MPa, respectively, whereas that of the SD treatment averaged –1.28 MPa. Ambient temperature and VPD in the vineyard at the time of measurement (midday) was 34.9°C and 4.63 kPa respectively. Notably, mean Ψ_l of Cabernet Sauvignon, Grenache and Syrah on that day was –1.76 ± 0.04, –1.86 ± 0.07 and –1.80 ± 0.05 MPa, respectively. Stomatal conductance for the ED and LD treatments on that day averaged 0.089 and 0.568 mol m^–2 s^–1 respectively.

Midday Ψ_l of the LD treatment averaged across cultivars on 29 August (more than 5 weeks after irrigation had ceased for that treatment) was –1.76 ± 0.05 MPa. Midday Ψ_l of two vines of each cultivar (located at the end of the row on the east side of the vineyard) still being irrigated at 100% of estimated ET_c averaged –0.92 ± 0.01 MPa. Midday Ψ_l of vines irrigated at 50% of ET_c averaged –1.45 MPa. Ambient temperature and VPD at the time of measurement were 33.9°C and 4.11 kPa, respectively. On the last measurement date in 2012 (5 and 6 September) mean Ψ_PD and midday Ψ_l across cultivars were –0.73 ± 0.01 and –1.79 ± 0.02 MPa respectively. Midday Ψ_l of Cabernet Sauvignon, Grenache and Syrah on those dates were –1.84, –1.78 and –1.80 MPa respectively. The slope of the linear relationship between midday Ψ_l and Ψ_PD using data from Cabernet Sauvignon, Grenache and Syrah across the 2012 growing season was 0.784 with an R² value of 0.73 (P = 0.001). Notably, the individual slopes of the three cultivars did not differ significantly from one another.

In general, seasonal responses of Ψ_l and g_s to irrigation treatments were sensitive and consistent over the course of each growing season. The seasonal pattern of applied water amounts and responses of both water status variables (Ψ_l and g_s) during 2014 is representative of those for the other growing seasons (Fig. 1). Prior to treatment imposition, all vines received 62 and 41 mm of applied water in 2014 and 2015 respectively. Following treatment imposition (12 and 11 May in 2014 and 2015 respectively), vines were irrigated at previously defined rates with applied water amounts illustrated by the curves of each treatment (Fig. 1a). Ψ_l and g_s were significantly affected by the irrigation treatments, with three levels of water stress established before veraison, and two levels of water stress established after veraison in each year of the study (Fig. 1b, c). While Ψ_l and g_s responded similarly to the irrigation treatments, there were small, but notable differences in each variable’s response to an irrigation treatment during each phenological period.

Fig. 1.  Seasonal time course of applied water (a), leaf water potential (Ψ_l) (b), and stomatal conductance (g_s) (c) for each irrigation treatment in 2014 (representative of all years in the study). (b, c) Data are means (±s.e.) calculated at each time point across all 17 cultivars. (a) Dot-dashed line without points is the estimated ET_c. (b) Dot-dashed line without points is a non-stressed baseline Ψ_l. (c) First group of points are from 2015. See ‘Materials and methods’ section for definition of irrigation treatments.

Before veraison, Ψ_l and g_s responses were like one another in the ED and LD irrigation treatments, but they responded differently to the SD treatment. Both Ψ_l and g_s values in LD vines remained high early in the season, with Ψ_l closely following the calculated non-stressed baseline Ψ_l value, and g_s increasing slightly from berry set to a maximum near the end of June (just before veraison). Similarly, Ψ_l and g_s values in ED vines both declined immediately following treatment imposition and were significantly lower compared with SD and LD at every measurement date from berry set until veraison (Fig. 1b, c). In contrast, while Ψ_l for SD vines declined after treatment imposition and was significantly different than that of LD vines from the third measurement date after treatment imposition, g_s response was less sensitive compared with Ψ_l, and remained high in SD relative to LD for a longer period before veraison (Fig. 1b, c). Overall, preveraison g_s was less sensitive to the SD irrigation treatment and remained high relative to that of LD vines despite a constantly declining Ψ_l until veraison.

Following veraison, Ψ_l and g_s recovered in ED vines once irrigation water was reapplied, and both variables declined in LD vines when water was withheld (Fig. 1b, c). Ψ_l of LD vines was significantly lower than that of SD and ED beginning with the first measurement postveraison and remained so until harvest. Ψ_l and g_s of ED vines recovered to values like those of SD vines and were not significantly different from SD for the remainder of the season. In general, postveraison Ψ_l was less stable week-to-week compared with postveraison g_s, whereas preveraison Ψ_l was more stable week-to-week compared with preveraison g_s.

The response of g_s to Ψ_l is summarised for all cultivars analysed together and individually in Figs 2 and 3 respectively. Because year was not a significant factor in either analysis – for all cultivars together or for all cultivars individually – data from the 2014 and 2015 growing seasons were analysed together. When all cultivars were analysed together in a composite analysis, Ψ_l⁹⁵ and Ψ_l²⁵ were estimated to be –0.89, and –1.74 MPa respectively (Fig. 2a). Across all cultivars, the response of g_s to Ψ_l was well characterised by the sigmoidal function at Ψ_l values greater than Ψ_l²⁵, with relative residuals generally varying within ~30% from the fitted curve, and in most cases within ~10%. Below Ψ_l²⁵, the response of g_s to Ψ_l diverged from the fitted curve with a corresponding increase in residuals to values greater than 30% of the fitted values (Fig. 2b).

Fig. 2.  (a) Stomatal conductance (g_s) at various leaf water potentials (Ψ_l) averaged across all 17 cultivars with data collected in 2014 and 2015. Within a given year, g_s values were pooled and means calculated across cultivars at each 0.1 MPa of Ψ_l. Data are means ± 95% confidence intervals. The solid line is a non-linear regression line fit with a sigmoidal function (adjusted R² = 0.96). (b) The plot of residuals as percent deviation from the fitted curve. Curve is fit using a smoothed cubic regression spline. Vertical dotted lines represent Ψ_l values at Ψ_l²⁵ and Ψ_l⁹⁵ according to the sigmoidal function in the top panel.

Fig. 3.  Response of g_s to Ψ_l for all 17 cultivars used in this study, arranged from top-left to bottom-right in order of decreasing g_sensitivity (see ‘Materials and methods’ section for definition). Data are means ± 95% confidence intervals of g_s for samples that were pooled at each 0.1 MPa of Ψ_l in 2014 and 2015. Vertical dotted lines represent Ψ_l values at Ψ_l²⁵ and Ψ_l⁹⁵ according to the fitted curves.

The three-parameter logistic function was well fit to all cultivars individually as well (Fig. 3), with adjusted R² values ranging from 0.72 (Durif) to 0.94 (Aglianico) (Table 2). From the fit sigmoidal functions, parameter estimates and modelled values at various levels of vine water status were compared across cultivars (Tables 2, 3). Excluding Syrah, the coefficient of variation (CV) in estimates of g_max was less than 7% across cultivars, ranging from 0.398 (Grenache) to 0.489 mol m^–2 s^–1 (Refosco). The estimate of g_max (±s.e.) for Syrah (0.595 ± 0.130 mol m^–2 s^–1) was 1.2 times larger than the next greatest estimate (Refosco; g_max = 0.489 ± 0.029 mol m^–2 s^–1), but with a large standard error (Table 2). Interestingly, the range and standard deviation (s.d.) of measured g_max values among cultivars – measured range = 0.151 mol m^–2 s^–1; measured s.d. = 0.0428 mol m^–2 s^–1 – was smaller than those of estimated g_max values among cultivars – estimated range = 0.196 mol m^–2 s^–1; estimated s.d. = 0.0461 mol m^–2 s^–1. In either case – when comparing estimated or measured values – there were no significant differences in g_max among cultivars.

Table 2.  Parameter estimates (Est.), associated standard errors (s.e.), and regression fit statistics (Adj. R²) from non-linear regression analyses of g_s to Ψ_l
The parameter g_max represents the curve asymptote, i.e. a theoretical maximum g_s. The parameter Ψ_l⁵⁰ represents the inflection point of the curve, i.e. the Ψ_l (MPa) at which g_s is 50% of g_max. The parameter b characterises the transition width of the function, i.e. the difference in Ψ_l from 25 to 75% of g_max and is equal to (b × 2.196) + 0.002. Adj. R² is a bias-corrected R² accounting for the number of parameters in the non-linear regression model

Table 3.  Maximum slope and vine water status values modelling the relation of g_s to Ψ_l for each cultivar
Vine water status values (Ψ_l⁹⁵, Ψ_l⁵⁰ and Ψ_l²⁵) were extracted from the fitted curves using parameters from Table 2. Maximum slope (g_sensitivity) was determined from the first derivative of each fitted curve

For several cultivars, the estimated asymptotic g_max value from the sigmoidal fit was greater than the highest actual measured mean g_max value (e.g. Syrah and Tinta Madeira,) or there were few data points near the estimated g_max (e.g. Tinta Amarela) (Fig. 3). One consequence was that the derived Ψ_l⁹⁵ values fell outside of the measured values for Syrah and Tinta Madeira, or that the estimated g_max value was a poor representation of a cultivar’s actual g_max. Since subsequent analyses depended upon correct determination of g_max, the estimated g_max values were regressed on the highest mean measured g_max (taken at the highest Ψ_l). There was a statistically significant relationship between estimated and measured g_max (P < 0.05), though there were some notable outliers (e.g. Syrah and Durif) that weakened the relationship (R² = 0.31). Nevertheless, the estimated slope of the relationship was not significantly different than unity (slope ± 95% confidence interval = 0.599 ± 0.493). Thus, it was assumed that the estimated g_max value from the sigmoidal fit was representative of actual g_max for all subsequent analyses.

The CV was similarly low (6%) across cultivars for estimates of Ψ_l⁵⁰, which ranged 0.32 MPa from –1.27 (Touriga Nacional) to –1.59 MPa (Montepulciano) with no differences among cultivars. In contrast, the unitless scalar parameter b – the transition width of the function related to the slope – varied broadly among cultivars (CV = 29%). Accordingly, there was broad variation in estimates of g_sensitivity (CV = 32%) among cultivars, with values ranging 3-fold from 0.428 (Petit Verdot) to 1.283 mol m^–2 s^–1 MPa^–1 (Tannat). Further, g_sensitivity was 0.562 and 0.525 mol m^–2 s^–1 MPa^–1 for Grenache and Syrah respectively (Table 3).

When comparing modelled Ψ_l and g_s values among cultivars at each extracted vine water status value (i.e. Ψ_l⁹⁵, Ψ_l⁵⁰ and Ψ_l²⁵), the across-cultivar CV for modelled Ψ_l decreased with vine water status (i.e. Ψ_l⁹⁵ > Ψ_l⁵⁰ > Ψ_l²⁵), whereas across-cultivar CV for modelled g_s remained constant at each value (i.e. Ψ_l⁹⁵ = Ψ_l⁵⁰ = Ψ_l²⁵) (Table 3). This was not surprising given that the model was fit to minimise variation in g_s. The across-cultivar mean modelled g_s and Ψ_l at high water status (i.e. Ψ_l⁹⁵) was more than 0.4 mol m^–2 s^–1 and –0.99 MPa, respectively, whereas the across-cultivar mean modelled g_s and Ψ_l at low water status (i.e. Ψ_l²⁵) was near 0.1 mol m^–2 s^–1 and –1.68 MPa. Furthermore, the cultivars with high g_s at high vine water status (Ψ_l⁹⁵) tended to have similarly high g_s at low vine water status. For example, Syrah had the highest g_s among all cultivars at Ψ_l⁹⁵ and at Ψ_l²⁵, whereas Grenache had the lowest g_s at both Ψ_l⁹⁵and Ψ_l²⁵ (Table 3). However, their g_sensitivity values were similar (as stated above).

There was a significant negative relationship between g_sensitivity and Ψ_l⁹⁵ across all cultivars (Fig. 4). As the water status threshold for the onset of stomatal closure (Ψ_l⁹⁵) decreased, sensitivity to further closure increased across the entire range of Ψ_l⁹⁵ values. Moreover, all 17 cultivars were broadly distributed along the curve and did not cluster together. Notably, Grenache and Syrah were distributed similarly on the vertical axis (g_sensitivity) and were separated only on the horizontal axis (Ψ_l⁹⁵).

Fig. 4.  Relationship of maximum stomatal sensitivity (g_sensitivity) to onset of stomatal closure (Ψ_l⁹⁵) for all 17 cultivars. A non-linear regression was fit through the data (y = 0.11e^–1.80x; adj. R² = 0.73).

Overall cultivar mean Ψ_l (averaged across treatments and years) was positively related to Ψ_l²⁵ both pre- and post-veraison (Fig. 5). Petit Verdot was removed from the analysis as an outlier due to its large absolute residual value in both phenological periods. There were no significant differences in response between phenological periods (slopes not significantly different), except that cultivar mean Ψ_l values were significantly lower post-veraison by 0.21 MPa (likely due to the lack of a fully irrigated post-veraison treatment). Nevertheless, in both phenological periods a higher Ψ_l²⁵ resulted in a higher cultivar mean Ψ_l with nearly a 1 : 1 ratio based on the slopes of the linear regression lines. The linear regression model was highly significant, and Ψ_l²⁵ explained 83% of the variation in cultivar mean Ψ_l when the model included both phenological periods.

Fig. 5.  Cultivar mean Ψ_l and its corresponding Ψ_l²⁵ for each of the 17 cultivars from set to veraison and from veraison to harvest. Cultivar mean Ψ_l values were computed across 2014 and 2015 and across irrigation treatments. Cultivar Ψ_l²⁵ values were calculated from fitted curves (Fig. 4). Petit Verdot was excluded from the regression because it was an outlier. Equations for each phenological period were: y = 0.07 + 0.80x for set to veraison, and y = –0.14 + 0.80x for veraison to harvest. Fit and significance of regression model was R² = 0.83 and P < 0.001.

Since the paper published by Schultz (2003) suggesting that V. vinifera cultivars may differ in their stomatal behaviour in response to water deficits, there has been great interest among researchers in unravelling the causes and underlying mechanisms of the an/isohydric phenomenon across cultivars (Soar et al. 2006b; Collins et al. 2010; Gerzon et al. 2015; Hochberg et al. 2015; Coupel-Ledru et al. 2017; Charrier et al. 2018). However, the obvious experimental difficulty in conducting large, field-scale experiments across many cultivars has limited previous investigations to only a few cultivars at a time or to the comparison of various cultivars grown in different vineyards or locations. As such, much work has been done on potted vines whose responses may not be representative of field performance – particularly in terms of plant water relations (Passioura 2006) – and on cultivars selected based on a priori classification as either iso- or anisohydric from few data – notably Grenache and Syrah. Several reviews have been published on the topics of irrigation management and/or water relations of grapevine that have attempted to synthesise the existing literature on across-cultivar stomatal behaviour (Cifre et al. 2005; Chaves et al. 2010; Lovisolo et al. 2010; Costa et al. 2012; Domec and Johnson 2012). Unfortunately, differences in plant material, environmental conditions, and experimental methodology and techniques have limited the ability of past reviews to characterise various cultivars’ responses and have ultimately resulted in inconsistent classification. In an attempt to resolve these problems, Lavoie-Lamoureux et al. (2017) conducted a meta-analysis of 40 publications across 28 V. vinifera cultivars and hypothesised that rather than separating into two distinct groups, cultivars’ stomatal behaviours were distributed along a continuum. A similar result was found by Klein (2014) in a meta-analysis across 70 plant species.

This study explored genetic variation in vine water status and stomatal behaviour across 17 V. vinifera cultivars (clone wood and source known) grafted onto the same rootstock (1103P), grown in a replicated field trial in a high VPD area (San Joaquin Valley of California), and subjected to three season-long deficit irrigation treatments over the course of multiple seasons. Therefore, soil type, weather conditions and cultural practices were the same for all cultivars. Midday Ψ_l was measured by the same individual on each date in 2014 and 2015 using the technique whereby leaf blades were covered with a plastic bag just before leaf excision therefore minimising potential technique (Williams and Araujo 2002; Williams 2017) and operator errors (Goldhamer and Fereres 2001). Unfortunately, this was not one of the factors considered in the meta-analysis by Lavoie-Lamoureux et al. (2017) and even now the exact method used or a reference provided to describe how Ψ_l was measured is not given (Charrier et al. 2018). The timing interval between the two measurements only had to be within two hours in the meta-analysis of Lavoie-Lamoureux et al. (2017), and it is unknown to what degree this would affect their results. Stomatal conductance was measured with the same LI-COR porometer in each year on the same leaf used for the measurement of Ψ_l just before its excision. Thus, the design and techniques used in this study should have maximised the effect of cultivar on the parameters in question. The hypothesis that cultivars do not separate into clear iso- or anisohydric groups, but rather are distributed along a continuum was tested by (1) establishing a range of plant water deficits using variable amounts and timings of applied water, (2) measuring Ψ_PD and Ψ_l the first year and (3) measuring g_s and Ψ_l on the same leaf samples the third and fourth years, and then comparing response curves among cultivars.

In general, an isohydric plant will maintain its water potential value similar between water stressed and well-watered cohorts (Tardieu and Simonneau 1998). The results from the first year of the study clearly demonstrate that none of the cultivars in this study responded to soil water deficits in an isohydric manner but were more anisohydric-like, in agreement with that of (Soar et al. 2006b). Midday Ψ_l of vines in the three irrigation treatments differed significantly from one another – midday Ψ_l of well-watered vines across cultivars were always much greater than the other irrigation treatments and as the water in the soil profile was depleted, midday Ψ_l decreased for all cultivars including that of Grenache. None of the cultivars examined in this study maintained a minimal midday Ψ_l plateau (> –1.5 MPa) as reported previously for near-isohydric grape cultivars (Schultz 2003; Prieto et al. 2010). In addition, the slope of the relationship between midday Ψ_l and Ψ_PD reported here did not differ among cultivars, similar to that observed by Charrier et al. (2018) for Grenache and Syrah. Lastly, the cultivars did not appear to respond to soil water deficits in an isohydrodynamic manner (Franks et al. 2007; Shellie and Bowen 2014).

A sigmoidal g_s relation to Ψ_l was present in all 17 cultivars and in a composite analysis. A similar result was found by Charrier et al. (2018). Linear relationships between g_s and Ψ_l (and Ψ_stem) have been commonly reported in grape (Williams and Araujo 2002; Shackel 2007; Williams 2012; Williams et al. 2012). However, nonlinearity may have been overlooked in those studies. The reported ranges of Ψ_l or Ψ_stem were narrower (~0.5 to 1.0 MPa) than that reported herein (1.4 MPa), thus may have only represented the linear portion of the entire response curve. Furthermore, there are few data in the aforementioned studies at low vine water status (Ψ_l or Ψ_stem < –1.5 MPa), and the reported data at high vine water status (Ψ_l or Ψ_stem > –1.0 MPa) show increasing vertical scatter. Indeed, when a broader range of vine water status was evaluated, non-linear relationships between g_s and Ψ_l (and Ψ_stem) have been shown for V. vinifera cv. Sauvignon blanc (Naor et al. 1994, 1997), and between g_s and Ψ_PD across five cultivars (Prieto et al. 2010). Recently, sigmoidal functions were also well fit to data relating grapevine leaf petiole percent loss of conductivity to xylem tension in potted grapevine (cv. Merlot) (Hochberg et al. 2017a). Finally, sigmoidal relations of g_s to Ψ_l has been reported for woody and non-woody crops alike, including: Populus spp. (Silim et al. 2009), Olea europea (Ennajeh et al. 2008), Oryza sativa (Dingkuhn et al. 1989), Sorghum bicolor (Henzell et al. 1976); and across a broad range of tree species (Klein 2014). Nevertheless, in this study the nature of the sigmoidal relationship between g_s and Ψ_l depended greatly upon cultivar.

Under well-watered conditions, there were no cultivar differences in g_max. While both high g_max (Henzell et al. 1976) and low g_max (Sinclair et al. 2005) have been proposed means to improve productivity under dry conditions, there was surprisingly little variation in g_max under well-watered conditions among cultivars over two years. Measured and estimated g_max values in this study (~0.4 to 0.6 mol m^–2 s^–1 across cultivars) were similar to other reported values for well-watered grapevines (Williams and Baeza 2007; Bota et al. 2016), and there were no significant differences among cultivars. Interestingly, measured g_max values among cultivars were more similar than were estimated g_max values, further supporting the point that cultivars behave similarly under well-watered conditions. Thus, no prospects were revealed for improved efficiency at either high water status (Price et al. 2002) or at low water status (Silim et al. 2009).

The lone g_max outlier – Syrah, who’s estimated g_max was more than 0.1 mol m^–2 s^–1 greater than measured g_max – was likely due a lack of data density in the well-watered range (> –1.0 MPa) that ultimately resulted in a poor estimate of the asymptote by the sigmoidal fit. This was also true for Tinta Madeira, though the difference between measured and estimated g_max was smaller. The question remains whether or not higher g_s values would be measured for those cultivars – and a more accurate estimate for g_max would be obtained – at higher Ψ_l values. Although it is not biologically impossible – Soar et al. (2009) reported g_s values for Shiraz (syn = Syrah) near 0.8 mol m^–2 s^–1 – it is unlikely under field conditions, primarily because those values were obtained through artificial heating and under high relative humidity. Also, significantly higher mid-season Ψ_l values (> –0.7 MPa) would also be difficult to reach under field conditions (Williams and Baeza 2007).

Generally speaking, the role of g_max in terms of drought tolerance has been evaluated almost exclusively in the context of crop species that are grown for agronomic purposes (yield/biomass production) – e.g. Populus spp., O. sativa, and S. bicolor. Considering that grapevine water use is highly correlated with g_s (Williams et al. 2012) and canopy percent shaded area/canopy coverage (Williams and Ayars 2005), similar g_max across cultivars would suggest that canopy percent shaded area alone – ultimately determined by relative cultivar vigour and canopy training/trellising system – determines total consumptive water use in grapevine. However, understanding the importance of high or low g_max in wine grape production is further complicated by the common horticultural purposes (fruit quality) in which water deficits are often beneficial (Roby et al. 2004).

As water deficits increased, differences among cultivars in the onset of stomatal closure (Ψ_l⁹⁵) became apparent, ranging from Syrah at –0.76 MPa to Tannat at –1.26 MPa. There are few reported data regarding the Ψ_l at the onset of closure for grapevines, primarily due to the predominant practice of fitting linear curves through the data (as discussed above), or simply not conducting a regression analysis. Williams and Baeza (2007) suggest a Ψ_l value of –1.2 MPa as signifying when grapevines are no longer responsive to changes in VPD, but rather to changes in soil moisture. In their recent meta-analysis of the literature, Lavoie-Lamoureux et al. (2017) suggest using –1.214 MPa as the water status level at which to compare cultivars. It is likely that this value – Ψ_l = –1.2 MPa – varies by cultivar, and indeed falls in the lower end of the across-cultivar Ψ_l⁹⁵ range obtained herein. The data of Hochberg et al. (2017b) show that Ψ_l⁹⁵ for Chardonnay was closer to Ψ_l ≈ –0.8 MPa, and Hochberg et al. (2017a) show a 12% loss of conductivity (i.e. Ψ_l⁸⁸) in Merlot occurring at Ψ_stem ≈ –0.66 to –0.91 MPa – which would correspond to a Ψ_l ≈ –0.97 to –1.20 MPa (Williams and Araujo 2002). Finally, a similar range in the onset of stomatal closure has been reported among Poplar clones (Silim et al. 2009) and across 70 other tree species (Klein 2014). Therefore, the cultivar-specific water potential values for the onset of stomatal closure in this study are within the expected range for V. vinifera, and moreover seem to be highly conserved across other woody perennial plant species as well.

Like the onset of stomatal closure, g_sensitivity was cultivar-specific, revealing a 3-fold difference in slopes (Fig. 4). Prevailing definitions of iso- and anisohydry state that isohydric plants maintain Ψ_l almost constant as soil dries due to more sensitive stomata, whereas anisohydric plants maintain g_s and allow Ψ_l to decline with soil drying (Tardieu and Simonneau 1998; Jones 2014). Using this definition, cultivars in this study with high g_sensitivity (e.g. Tannat, Durif and Montepulciano) could be categorised as isohydric/near-isohydric, whereas those with low g_sensitivity (e.g. Petit Verdot, Tinta Madeira, Freisa) could be categorised as anisohydric/near-anisohydric.

In the present study, Grenache (often considered the archetypal isohydric cultivar) and Syrah (often considered the archetypal anisohydric cultivar) had almost the same g_sensitivity (maximum slope of the linear part of the curves), and did not segregate based on this criterion. However, simple observation of the sigmoidal functions fit to Grenache and Syrah clearly shows that they behave differently at moderate water status – Grenache is more sigmoidal with a clear asymptotic levelling off of g_s at high Ψ_l values, whereas Syrah is less sigmoidal without a clear asymptotic levelling off of g_s at high Ψ_l. If g_sensitivity were calculated at moderate water status values (e.g. Ψ_l⁹⁵ > Ψ_l > Ψ_l⁵⁰) instead of in the linear portion (e.g. at Ψl⁵⁰), it is possible that Grenache and Syrah would segregate clearly based on this criterion. Simply looking at this constrained dataset supports some previous classification of these two cultivars. Yet the fact that their behaviour – and ultimate classification – depends on how the stomatal sensitivity parameter is defined calls into question its usefulness for classification.

In their comparison of five cultivars, Prieto et al. (2010) reported that a stomatal sensitivity factor, k – calculated using the Ball, Woodrow, and Berry model – significantly differed between Grenache and Syrah, but their level of significance for this parameter was α = 0.10. Furthermore, they show regressions of k to predawn water potential for Syrah and Ekigaina (as examples of extreme cases), but do not show the comparison data for Grenache. Recent work has also shown that there are no statistically significant differences between Grenache and Syrah with respect to sensitivity of transpiration rate to declining soil water availability under both field and greenhouse conditions (Charrier et al. 2018). Therefore, other factors affecting stomatal behaviour such as leaf hydraulics or environment may be more important to consider for cultivar classification going forward.

It is notable that Grenache and Syrah – the two cultivars first compared by Schultz (2003) – differed in this study only in their Ψ_l⁹⁵ value, but had almost the same g_sensitivity (Fig. 4). Although widely cited as the foundational evidence for the an/isohydric phenomenon within V. vinifera cultivars, it is seldom discussed that Schultz (2003) actually used Grenache and Syrah scions grafted onto different rootstocks, 140 Ruggeri and 110 Richter respectively (Schultz 1996). Vitis spp. rootstocks differ in their effects on vine water relations and gas exchange (Padgett-Johnson et al. 2003), and have also been shown to significantly affect scion gas exchange (Soar et al. 2006a) as well as scion growth and productivity (Williams 2010). Lavoie-Lamoureux et al. (2017) reported that rootstock generally accounts for 19% of the variability in scion g_s response to water deficits. When grafted onto the same rootstock, similar rates of stomatal closure between Grenache and Syrah in response to Ψ_stem (ranging from –0.5 to –1.5 MPa) have been reported in a recent three-year field study (Bota et al. 2016). It did appear that Grenache and Syrah behaved differently at moderate water status in this study, in contrast to Schultz (2003) who showed a greater divergence between the two cultivars at lower water status. Well-watered conditions have been shown to increase variation in stomatal response across Grenache-Syrah progeny (Coupel-Ledru et al. 2017). Indeed, it should be noted that Syrah maintained a higher g_s compared with Grenache at all water status levels in this study. Nevertheless, whether considering Grenache and Syrah separately, or all the cultivars together in this study, there were more differences among them in stomatal behaviour at moderate water status relative to low water status.

Analysis of g_s at Ψ_l²⁵ and the similarity of Ψ_l among cultivars at Ψ_l²⁵ suggests a common water status limit to low g_s among grapevine cultivars. At Ψ_l²⁵, g_s was already down to 0.08 to 0.1 mol m^–2 s^–1 and declined gradually thereafter. Grapevines have been previously classified as severely stressed when g_s <0.05 to 0.1 mol m^–2 s^–1 (Cifre et al. 2005; Lovisolo et al. 2010). The lack of apparent cultivar differences at low water status may have been due to the relatively poor fit of the sigmoidal functions at these levels (shown by increasing homoscedasticity in residuals in Fig. 2b). However, all extracted Ψ_l²⁵ values were above these poorly fit regions of the dataset. Brodribb and Holbrook (2003) regarded Ψ_l²⁰ as the point of stomatal closure in their analysis of stomatal behaviour across four tropical tree species, and Klein (2014) used Ψ_l²⁵ as a functional parameter by which to compare stomatal behaviour of 70 woody perennial plant species. Hence, it is reasonable to regard the value of Ψ_l²⁵ as the functional lower limit of stomatal contribution to leaf gas exchange and overall vine performance.

Stomata function in part to protect the hydraulic integrity of the xylem (Tyree and Sperry 1988; Brodribb and Holbrook 2003), but also close in response to reduction in leaf turgor (Shackel et al. 1987). Thus, a reduction in g_s is both an impact of and a protective response to declining plant water status. This inherently complicates behavioural interpretation. Choat et al. (2010) showed that xylem cavitation commences at much lower water status than previously accepted, and Hochberg et al. (2017b) recently showed that stomata close well in advance of xylem cavitation in grapevine. In the latter work, g_s was less than 10% of g_max before the first embolisms were detected in the xylem (–1.46 > Ψ_l > –1.75 MPa). Those Ψ_l values at the appearance of the first embolisms closely correspond to the Ψ_l²⁵ values found among cultivars in this study (-1.5 > Ψ_l > –1.8 MPa; Fig. 5). Therefore, our data indicate most stomata were closed for all cultivars at or below their Ψ_l²⁵ value, and that the cultivar-specific Ψ_l²⁵ value may be related to the appearance of first embolisms in the xylem. Thus, each cultivar-specific Ψ_l²⁵ value may indicate a hydraulically determined water status limit, which may be governed by cultivar differences in vascular anatomy (Gerzon et al. 2015; Hochberg et al. 2015).

It is likely that the different stomatal behaviours observed among cultivars were mediated by hydraulic or turgor-dependent signals, which are closely associated with incipient stomatal closure (Brodribb and Holbrook 2003). In their modelling work on partitioning stomatal responses to drought, Rodriguez-Dominguez et al. (2016) found that leaf turgor-dependent mechanisms accounted for a large majority of the decline in g_s to soil drought across three woody perennial crop species, including grapevine. Reduced hydraulic conductivity and increased osmotic adjustment made up a large proportion of the overall stomatal response in that study.

In general, differences in stomatal closure among plant species may arise in part from differences in osmotic adjustment (Hsiao 1973), leading to acclimation of stomatal response and leaf hydraulic conductivity to low water status (Matthews and Boyer 1984; Martorell et al. 2015). Differences in osmotic adjustment have also been closely associated with differences in stomatal behaviours between tree species (Nolan et al. 2017). Grape leaves also go through significant osmotic adjustment (Schultz and Matthews 1993), and there is recent evidence for differences in osmotic adjustment among V. vinifera L. cultivars (Martorell et al. 2015). Differences in osmotic adjustment have also been previously reported among cultivars in wheat, sorghum, and cotton (Morgan 1984), and in O. europea (Ennajeh et al. 2008). Thus, genetic variation in the osmotic adjustment response may play an important role in the wide range of cultivar responses found in this study.

The results of this study provide clear evidence in support of an across-cultivar continuum of stomatal behaviour, and call into question previous classifications of V. vinifera cultivars as having either exclusively iso- or anisohydric stomatal responses. First, cultivars did not separate into distinct iso- or anisohydric groups based on their Ψ_l v. Ψ_PD relationships. Moreover, a deeper exploration into genetic variation among cultivars revealed that instead of separating into distinct groups, cultivars were distributed along a continuum based upon the negative relationship between the sensitivity (g_sensitivity) and the onset (Ψ_l⁹⁵) of stomatal closure in the face of declining vine water status. Notably, this continuum was well defined at moderate status values, but cultivars behaved similarly at high and low water status.

At high vine water status, cultivars had similar g_max values, suggesting that overall vine water use would simply be a function of canopy size under well watered and saturating light conditions. As Ψ_l decreased from well-watered conditions, all cultivars responded similarly to one another by maintaining a high g_s – an anisohydric characteristic. This was despite several of the cultivars having been assigned to the isohydric/near-isohydric category in other studies. However, closer examination of the g_s vs Ψ_l relationship showed that as vine water status declined from well-watered conditions, stomata of various cultivars responded differently to minimise water loss – an isohydric characteristic. Finally, at low water status, g_s was the similar for all cultivars (<0.1 mol m^–2 s^–1), but the water status value at the end of stomatal closure (Ψ_l²⁵) differed among them, indicating that there is a cultivar-dependent water status limit to stomatal functioning.

The large and significant differences in midday Ψ_l values measured as a function of irrigation treatment across all cultivars in this study would indicate that midday Ψ_l provides an accurate assessment of vine water status, and would therefore be useful to validate other measures of monitoring vine water status, or as a tool in an irrigation management program. In terms of cultivar-specific stomatal behaviour, all cultivars examined herein responded in an anisohydric manner at high water status by keeping stomata open as Ψ_l decreased. This was despite several of the cultivars having been assigned to the isohydric/near isohydric category in other studies. However, closer examination of the g_s vs Ψ_l relationship indicates there may be times – at moderate water deficits – in which stomata of various cultivars may respond differently to minimise water loss, an isohydric response. This would indicate that V. vinifera cultivars have both iso- and anisohydric stomatal responses to water deficits, and that any future classification should incorporate a holistic evaluation of the entire response curve. Finally, differences in osmotic adjustment among cultivars may play the primary role in determining the observed differences in stomatal behaviour.

Larry E Williams, Mark A Matthews and Alexander D Levin conceived of and designed the research. ADL conducted experiments, analysed data, and wrote the manuscript. All authors read and approved the manuscript. The authors declare no conflicts of interest.