The study was performed using 15 durum wheat genotypes (Triticum turgidum subsp. durum) including 12 landraces and three modern check cultivars originating from contrasting pedoclimatic regions (Table 1).

The durum wheat landraces were obtained from the U.S. National Plant Germplasm System (NPGS) which provides precise geographical coordinates of the sites of collection. Germplasm origin covers climatic conditions varying from arid B climates to more temperate C climates with both winter and summer rainfall as well as genotypes from continental D and tropical A climates (Kottek et al., 2006). Pedoclimatic characterization of the sites of origin was done based on data from the FAO weather database and using the New_LocClim software (Grieser et al., 2006) and the harmonized world soil database (Fischer et al., 2008). Table 1 also shows annual rainfall, reference evapotranspiration and their ratio (aridity index; UNEP, 1992) as an indicator of the climatic water balance deficit at the sites of origin of the tested accessions ranging from extremely arid high stress to less stressful sub-humid environments.

The genotypes were evaluated in a pot experiment at early vegetative stage for 4 weeks in the PhyTec Experimental Greenhouse at the Institute of Biosciences and Geosciences, Plant Sciences (IBG-2), Forschungszentrum Julich GmbH, Germany (50°54′36″N, 6°24′49″E). The experiment was set up in a factorial completely randomized design with 5 replications for each genotype and treatment. After each measurement the pots were automatically re-randomized via a laser positioning system and a robotic crane to avoid any systematic bias from position within the greenhouse. The fixed factors were genotype (12 landraces and 3 modern cultivars) and water regime (control and drought stress).

The experiment started on June 1st 2014 by visually selecting 20 medium-sized seeds of each genotype and sowing single seeds in plastic germination trays. Uniformly emerged seedlings at the one-leaf stage (BBCH = 11; Lancashire et al., 1991) were then individually transplanted into 5 L pots (23 × 17 cm). Soil substrate was a mixture of peat, sand and pumice (SoMi 513, Dachstauden; Hawita, Vechta, Germany).

For the first week after transplanting, soil moisture level was maintained at field capacity for both the control and stress treatment to ensure optimum establishment of seedlings. Afterwards, all pots were gradually dried down to the predefined moisture levels i.e., 75% (control) and 25% (drought) of plant available water (PAW). PAW was calculated as soil water content difference between field capacity (h = −0.01 MPa) and permanent wilting point (h = −1.5 MPa) derived from the water retention curve of the soil substrate (Figure S1). The analysis of the soil water retention curve was done in 2013 at the University of Kiel, Germany, Institute of Plant Nutrition and Soil Science. Water content at the respective water potential values was derived from the continuous Van Genuchten retention curve (van Genuchten, 1980). The water content levels of the single pots were then kept constant by automated irrigation after weighing every third day.

Five sensors collected environmental data of the greenhouse (relative humidity, global radiation, temperature) throughout the experiment. In addition to natural light, supplemental illumination was used for 14 h during the experiment. The resulting average light intensity at plant level during the experiment was 442 ± 210 μmol m⁻² s⁻¹.

Analysis of variance (ANOVA) was performed on all data using the MIXED procedure in SAS 9.2 (SAS Institute, Inc., Cary, NC). For repeated measures over time (projected area, evapotranspiration) an unstructured covariance model provided the best fit according to the AIC (Akaike Information Criterion). Genotypes and water regimes were treated as fixed effects. Comparison of means was performed using a Tukey post-hoc test. Differences among groups of genotypes (i.e., cultivars vs. landraces) and putative water-use strategies were tested by linear contrasts using the CONTRAST statement in the MIXED procedure. Relations among the data were assessed by regression analysis (PROC REG) with stepwise selection for maximum R². Slope comparison was performed following the procedure described by Sawand (2012).

The key variable for our analysis of distinctive plant water use among accessions was evapotranspiration rate, i.e., the daily water loss via plant transpiration and soil evaporation.

Evapotranspiration rate responds to both changing plant properties as well as environmental conditions. Figure 1A shows the daily mean air temperature and reference evapotranspiration (ET₀, _{Penman-Monteith}, Allen et al., 1998) as two key atmospheric variable driving plant growth and development (temperature) as well as water losses (ET₀).

With increasing leaf development, water loss per day steadily increased. Evapotranspiration rate varied significantly between genotypes and watering treatments, both showing significant interaction with time (p_GENxDAS < 0.001; p_TRTxDAS < 0.001). The induced water stress resulted in a decrease of average water loss from 80.2 ml d⁻¹ in the control treatment to 44.6 ml d⁻¹ in the stress treatment with differences between treatments becoming larger over time.

Figure 1B reveals the significant interaction between genotype × time. Generally evapotranspiration rate over time could be approximated by a linear increase with moderate scattering around this overall trend. Some genotypes like Neda (lowest water-use) and 96 (highest water-use) kept a constant rank over time, while others such as BPR vs. 9127 changed their ranks. Significant differences between Neda and 96 were registered from DAS 25 onwards. BPR with an intermediate water use kept a lower evapotranspiration rate compared to 96 from DAS 25 onwards, while higher rates compared to Neda were found from DAS 32 onwards.

The influence of evaporative demand in the greenhouse was reflected by the scattering of evapotranspiration rate around a linear trend (Figure 1C): the residuals from the linear regression of evapotranspiration rate vs. time were significantly explained (R² = 0.72) by reference evapotranspiration (ET_{0, Penman-Monteith}) calculated from atmospheric data (temperature, radiation, relative humidity) following Allen et al. (1998). As expected the atmospheric demand, approximated by ET₀, was an important driver of vapor losses from leaf and soil surfaces.

In this experiment no significant interaction between genotype and stress treatment on evapotranspiration rate was found. On average the modern cultivars Neda, Levante, and Floradur had low evapotranspiration rates over time under both control and stress conditions. Similarly, the genotypes ELS 63, 96, and S44 consumed the highest amount of water throughout the experiment, independent of the watering regime (Table 2).

Multiple linear regression with leaf area duration and stomata conductance as dependent variables explained 89% of the observed variation in transpiration rate. Both variables significantly contributed to the model.

Plotting the values of these traits standardized by their treatment means allocates the genotypes to four groups. This visualization identifies four different water use types (Figure 5) which can be hypothesized to distinguish the single genotypes according to their specific trait combinations.

Genotypes with leaf area duration and stomatal conductance higher than the average represent water-spending types (ELS114, ELS63, B1). Genotypes with high leaf area duration but lower conductance than average were defined as area types (96, BPR, IWA86081, S44, Ziraat). Few genotypes had both lower than average leaf area duration and stomatal conductance, constituting a group of water-saving types (9923, Da Terra). All modern cultivars (Floradur, Levante, Neda) as well as two landraces (6924, 9127) were characterized by lower leaf area duration and higher stomatal conductance than average. We denominated this group as conductance types.

The groups identified from this analysis (Figure 5) were then analyzed by linear contrasts in a one-way ANOVA with treatment (TRT; control, stress), trait (TRAIT; leaf area duration, stomata conductance) and water use type (TYPE; spender, saver, area, conductance) as fixed effects. TYPE (p < 0.001) and the interaction of TYPE × TRAIT (p < 0.001) were significant, while all other effects were not significant. Thus the water use types emerging from Figure 6 were stable over the two moisture treatments. Comparison of means confirmed the differences determining the four water use types (Table 3).

Average transpiration rate differed significantly among water-spenders and area-types vs. water-savers and conductance types. Thus, transpiration rate alone would have suggested only two distinctive groups, while consideration of underlying morphological and physiological traits identified four distinctive water use strategies among genotypes.

Having identified distinct water use strategies, the main question in a crop production context is their implication for growth potential. In case of the ScreenHouse phenotyping platform where genotypes are observed in their early stages only, this refers to vegetative (from germination to pre-flowering stage) dry matter accumulation.

Dry matter showed significant genotype and treatment main effects without significant interaction. Exposure of genotypes to drought stress led to a strong decrease in accumulated dry matter (6.3 vs. 3.6 g; Table 2) which was a common response for all genotypes.

Differences in treatment averaged dry matter accumulation of the four water use types resembled their differences in transpiration rate: Water-spending types (5.3 g) and area types (5.8 g) significantly contrasted with water-saving types (4.6 g) and conductance types (4.0 g).

We notice here that the tested accessions also differed in phenology (BBCH, phyllochron) and tillering (cf. Table 2). However we did not find strong influence of these traits on both transpiration (control: RBBCH2 < 0.01, p = 0.929; Rtiller2 = 0.09, p = 0.269; stress: RBBCH2 = 0.02, p = 0.611; Rtiller2 < 0.01, p = 0.917) and dry matter (control: RBBCH2 = 0.05, p = 0.442; Rtiller2 = 0.27, p = 0.045; stress: RBBCH2 = 0.18, p = 0.117; Rtiller2 = 0.13, p = 0.179). Still an indirect influence of these traits on dry matter accumulation can be assumed via their clear relation to final leaf area (control: RBBCH2 = 0.68, p < 0.001; Rtiller2 = 0.28, p = 0.041; stress: RBBCH2 = 0.72, p < 0.001; Rtiller2 = 0.66, p < 0.001).

The objective of phenotyping is to support the selection process for superior cultivars by specific trait information (Passioura, 2012). Here used image based shoot phenotyping combined with physiological measurements to identify the diverse mechanisms of drought response within a pre-breeding sample of durum wheat genetic resources compared to modern cultivars. Different frameworks have been used to reveal relevant traits for drought resistance. For example Passioura (1977) defined yield formation under drought as the product of water use, water use efficiency and harvest index. Reynolds and Tuberosa (2008) identified traits to be selected for in order to improve each of these single components. Another framework of drought resistance frequently used in breeding studies follows Levitt (1980), distinguishing between drought escape, dehydration tolerance, and dehydration avoidance. Relevant traits for each of the single strategies have been elaborated by Farooq et al. (2009) and Blum (2011).

ScreenHouse captures shoot traits of plants during the early vegetative stage until increasing leaf overlap impedes accurate inference on canopy growth via image analysis only. Our observations demonstrate that canopy growth and architecture essentially drive the two functions of interest in drought resistance: dry matter accumulation and transpiration water losses. Water saving vs. spending as two distinct strategies of dehydration avoidance is first of all related to differences in the transpiring leaf area over time. Drought escape mostly refers to the timing of stress sensitive stages in relation to stress occurrence in a given environment, i.e., addressing stress effects on reproductive processes beyond the phenological development of plants in this experiment. However, also early vigor influences the dynamics of canopy closure and thereby light and water use efficiency. Loel et al. (2014) for example associate historic yield increases in sugar beet with more vigorous early growth resulting in quicker canopy closure and thereby improved light interception. Furthermore, early vigor is also relevant for lower soil evaporation losses by shading (Richards, 1996) and, in some climates, growth under less water demanding conditions due to lower saturation deficit of the atmosphere during a longer part of the crop cycle. These consequences of phenological differentiation can be readily captured by the ScreenHouse imaging platform, thereby providing important information to infer on potential drought escape advantages.

Pot experiments require a sound understanding of experimental conditions (Poorter et al., 2012). The imposed hydrology in pots is the basis for a sound interpretation of phenotyping observations toward plant breeding. In the current study a constantly lower water content compared to a well-watered control was used. Passioura (2012) showed that this might discriminate against water savers using less water during irrigation intervals, while spenders with high water consumption are favored by receiving more water: e.g., the total irrigation amount required to maintain the pre-set moisture level in the stress treatment (25% PAW) was 28% less for the saving type Da Terra compared to the spending type ELS63. The saving types on the contrary would have profited from their lower water demand in case of an experimental setup stopping irrigation at a certain point. In such a case lower transpiration by reduced leaf area could be expected to result in higher stress tolerance compared with water spenders (i.e., less reduction of dry matter) because of delayed exhaustion of soil moisture and therefore later stomata closure and wilting.

We did not observe interactions between genotype and watering treatment in most traits; i.e., superior genotypes under well-watered conditions were also superior under drought, although the imposed stress regime clearly reduced growth (−42% average dry matter reduction). Using dry matter accumulation as selection criteria, water spenders generally outperformed all genotypes with a small leaf area. The lower transpiring surface and the related limitation of water demand did not provide any advantage under the regular rewetting regime in terms of balancing water availability over longer time. The smaller leaf area however limited light interception and thereby the potential for dry matter accumulation under both moisture treatments. In natural environments with severe and prolonged drought, dehydration avoidance via traits limiting water demand (saving types Da Terra and 9923) still may be beneficial. They ensure better water availability for grain filling and yield formation (Condon et al., 2004; Monneveux et al., 2005; Mori et al., 2011). On the other hand water savers might tend to suboptimal use of available soil water under mild to moderate water deficits, thereby limiting the capacity for photosynthesis and biomass accumulation (Rebetzke et al., 2002). In such conditions water spenders (accessions ELS114, ELS63, B1) would profit from their high leaf area and stomatal conductance, implying high carbon gains and growth rates such as observed in our phenotyping experiment. In rain fed agro-ecosystems the yield advantage of spenders however frequently depends on root system traits that confer optimized water uptake to buffer intermittent stress periods (Peleg et al., 2005). Therefore, root system information is of particular interest to study possible allometric relations between shoot and root traits (e.g., Siddique et al., 1990) that sustain the productivity advantage of landraces with extensive leaf area we would expect from early stage phenotyping.

A frequently used breeding target is transpiration efficiency, i.e., the amount of dry matter per unit water transpired. Still the outcomes of targeting improved transpiration efficiency are contradictory (Araus et al., 2002; Condon et al., 2002; Blum, 2005). Transpiration efficiency is a function of whole plant morphology, i.e., the leaf area driving transpiration and light interception, and leaf scale physiology, i.e., gas exchange at the stomata level and photosynthetic capacity (Steduto et al., 2007). Our data showed an increase in transpiration efficiency under dry conditions. This is explained by the stronger reduction of transpiration compared to assimilation when stomata close: assimilation decreased by only −2.9% with a −15.9% reduction in stomatal conductance under stress. For transpiration stomata are the main resistance, while CO₂ transport is also dependent on mesophyll resistance; thus the relative response of assimilation to stomata conductance is less compared to the response of transpiration (e.g., Farquhar and Sharkey, 1982; Barbour et al., 2010). It is worth mentioning that both relevant components of transpiration efficiency, leaf area, and stomata conductance, are related also to distinct phenology among accessions. Similar to our results, Condon et al. (2002) found that early genotypes with high crop growth rate are associated with higher stomata conductance, while their leaf area expansion is constraint by the shorter duration of growth stages.

During early growing stages morphological traits are the main limiting factor, i.e., the low leaf surface acts as the predominant constraint to assimilation (via light interception) and transpiration. Only at later stages, when leaf area is fully developed and light interception is at maximum, other traits related to physiological efficiency at leaf level can become predominant constraints (Richards, 2000). Morphological differences in plant canopies are therefore the principal driver during an early stage phenotyping experiment. This implies that the two components of transpiration efficiency, dry matter and transpiration, are reduced proportionally. The importance of the physiological response, resulting in a proportionally lower decrease of dry matter compared to transpiration, is still subordinated. During later stages of plant growth, however, when leaf area is sufficient for maximum light interception, physiological efficiency at the single leaf scale can become essential for enhanced plant productivity. Udayakumar et al. (1998) argue that a yield advantage might be achieved from higher transpiration efficiency when the underlying reason is superior photosynthetic capacity rather than lower stomata conductance. Steduto et al. (2007) however consider the variability in photosynthetic capacity as limited under comparative nutrient and water status of photosynthetic tissues within a given species and during vegetative growth. Therefore, Blum (2009) argued against transpiration efficiency as a selection target, except for high stress conditions where stomata mediated water saving ensures reproductive success at relatively low yield levels.

Particularly for early stage phenotyping, transpiration efficiency does not provide the best selection criterion. Differentiation among accessions is clearly less compared to dry matter and transpiration alone: coefficients of variation under well-watered conditions were 10.7% for transpiration efficiency, 15.9% for transpiration, and 18.4% for dry matter respectively, while under stress they were 9.0, 18.7, and 21.1%. The common leaf area dependence of both parameters underlying transpiration efficiency hides existing variability among accessions for each single process.

In order to go beyond the expected dominance of the leaf area influence on both dry matter and transpiration during early stage phenotyping, it is necessary to assess traits and processes per unit of leaf area. This reveals elements of plant productivity others than those mediated by extensive canopies, particularly those related to physiological regulation via stomata conductance. Thereby a more differentiated picture of water use and dry matter accumulation strategies among accessions can be obtained. Figure 7 resumes the distinction among accessions when combining morphological phenotyping data (leaf area) and physiological measurements (stomata conductance).

Two thirds of landraces belonged to types characterized by extensive canopies (area and spending types), while all three modern cultivars were in the group of capacitance types with compact canopy and high stomatal conductance. Only two landraces where identified as water savers with low leaf area and restricted stomata conductance. This suggests that landraces tend to produce an extensive vegetative canopy to ensure successful establishment, while probably investing excessively in vegetative biomass at the cost of grains. The important role of grain sink limitation for cereal yield advance has been shown by Sadras and Lawson (2011). Huge vegetative canopies are also considered inadequate for environments with late season drought due to imbalanced water use. Exceptions might be for accessions where large stems provide a source for carbohydrate translocation to grains. The saving types with restricted water loss and low dry matter accumulation might provide adaptive advantages under high stress environments. Still considering landraces as accessions selected for productivity under low input conditions, a constitutive restriction of growth due to conservative resource use could be a selection disadvantage. This might have reduced the presence of saving types among landraces. The allocation of cultivars toward the capacitance type confirms findings of Nakhforoosh et al. (2015) from field trials: cultivars have been selected for compact canopies that ensure sufficient light interception for photosynthesis in field stands, while high stomata opening allows an optimization of CO₂ assimilation. Nakhforoosh et al. (2014) also revealed that such types often show effective water uptake to sustain high stomata conductance.

Also several retrospective studies on yield progress of wheat cultivars released over the last decades concluded that the ideotype of modern cultivars has short stature (Slafer and Araus, 2007), small and erect leaves (Austin et al., 1980; Feil, 1992; Fischer, 2001), restricted tillering capacity (Richards et al., 2010), and enhanced stomata functioning compensating reduced leaf area in terms of assimilation (Fischer et al., 1998; Sadras and Lawson, 2011). These characteristics confer high assimilation potential as well as high transpiration efficiency to modern cultivars such as Neda (Shearman et al., 2005). Already during early stage phenotyping modern cultivars showed distinctive canopy traits and water use type compared to most landraces. Within a breeding process for better drought resistance, the early stage phenotyping based distinction among water use strategies can contribute to narrowing a screening population while keeping diversity of the sample. Properly interpreted phenotyping data thereby provide a gate from an early stage of the breeding process toward subsequent field validation of the trait implications for yield formation in a given target drought environment.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.