Durum wheat (Triticum turgidum L. subsp. durum) accounts for about 10% of the global wheat production (Kantety et al., 2005). Its cultivation is concentrated in latitudes corresponding mostly to the North America, the Middle East, Australia and  especially  the  Mediterranean Basin. The aforementioned represent around 60% of its total growing area. In the southern Mediterranean countries, it occupies a key place in agricultural production. In Morocco for instance, out of a total of 8.7 million hectares cultivated annually,  5.3  million  hectaresare devoted to cereals, and durum wheat cultivation (1.1 to 1.3 million ha) ranks third after bread wheat and barley (Belaid et al., 2003; MAPM, 2014).

Durum wheat is mainly grown under rainfed conditions (e.g., cultivated in “Bour” at 81% in Morocco) (MAPMDREF, 2016). Its cultivation under arid and semi-arid conditions is thus expected to face water deficit or drought (Chennafi et al., 2006), and its annual production is therefore highly dependent on unpredictable seasonal rainfalls and temperatures (Anderson, 2010; Royo et al., 2010). In the Mediterranean region, losses in durum wheat yield due to water deficit vary from 10 to 80% depending on the year (Nachit et al., 1998).

In durum wheat as in other crops, the comparison between optimal yields and mean yields observed in the field generally reveals considerable differences, especially in traditional farming systems. These differences are explained by the influence of agronomic, genetic and climatic factors (Ricroch et al., 2011). Water stress can be quantified as the ratio between the amount of water required for optimal growth of the plant and that available in its environment (Laberche, 2004). In wheat, the moisture deficit in the soil affects the three main components of yield, namely the number of ears, the number of grains per ear and the weight of 1000 grains (Assem et al., 2006). The effect on these components, and therefore on yield, depends on the stage at which the deficit occurs (Mongensen et al., 1985; Debaeke et al., 1996). For instance, drought, at the beginning of cultivation cycle is known to affect emergence and tillering and, at end of the cycle, to affect the filling of the grains (Fisher, 1973; Watts and El Mourid, 1988; Kobata, 1992).

Maintenance of the major physiological functions (photosynthesis, transpiration, growth) under water shortage represents an important challenge (Passioura, 1996; Tardieu, 2003, 2005; Amigues et al., 2006). In addition to the maintenance of growth of leaves and reproductive organs, delayed leaf senescence, by keeping the photosynthetic capacity, helps to feed the reproductive organs. This strategy can allow high yields but also increases the risk of total yield loss. It is favorable in conditions of moderate water deficit but may prove to have detrimental consequences in the case of more severe water deficit (Amigues et al., 2006). Tolerance to water deficit may therefore be considered rather as the ability of a genotype to produce an acceptable yield under conditions of water deficit according to the scenario of the constraint (degree, developmental stage, duration) (Yokota et al., 2006; Hamon, 2007; Tardieu and Tuberosa, 2010).

In cereals, leaf rolling is a characteristic response to water deficit (O’Toole and Cruz, 1980; Kadioglu et al., 2012). It is believed to occur when the evaporative demand is no longer balanced by water uptake by the roots. This movement of the limb is indeed caused by the loss of turgor of bulliform cells, which are large  epidermal cells of the upper epidermis (Begg et al., 1980; Willmer, 1983; Jane and Chiang, 1991). The degree of leaf rolling is often used as a marker of intensity of drought stress in cereal cultures (Riboldi et al., 2016), but leaf rolling can also be considered as a mechanism of avoidance of dehydration (Belhassen et al., 1995; Amokrane et al., 2002). It contributes, by decreasing the leaf surface exposed to sunlight, to the reduction of transpiration and to increase water use efficiency in water stressed conditions (O’Toole and Cruz, 1979; Monneveux and Belhassen, 1996). Furthermore, it allows the plant, by limiting the direct illumination of leaf surface, to avoid overheating of leaf tissues, harmful to cellular metabolism and is considered to play a significant role in the resistance to high temperatures and end-of-cycle water deficit (Ortiz et al., 1991). In rice, leaf rolling was reported to improve photosynthetic efficiency and to delay leaf senescence (Richards et al., 2002; Richards et al., 2004; Zhang et al., 2009). The available photosynthetic surface of rolled leaf is however reduced, which has a negative impact on plant growth (Li et al., 2016a).

The objective of this study was to characterize 16 durum wheat varieties for their leaf rolling ability under drought stress and their agronomic performance in order to investigate the relationship between leaf rolling and tolerance or sensitivity to water stress applied at tillering .

Plant material and experimental site

The study concerns 16 varieties of durum wheat (Triticum turgidum L. var. durum) provided by the National Institute of Agronomic Research of Settat (Morroco): Kyperounda (“2777"), Amjad, Anouar, Irden, Isly, Jawhar, Korifla, Marjana, Marouane, Massa, Oum Rabia, Sebou, Tomouh, Vitron, Waha and Yassmine. These varieties displayed diverse ranges of agronomic adaptation, a number of them being adapted to semi- arid areas (Taghouti et al., 2010; Nsarellah et al., 2011; Zarkti et al., 2012). The field experimentation was conducted at Tamellalet, CMV 408, Marrakech region (31 "81'N, 7" 50'W), during three consecutive cropping seasons: 2013-2014, 2014-2015 and 2015-2016. The Supplementary Table 1 represents more information on durum wheat varieties studied.

Climatic conditions

The monthly precipitations recorded over the three crop years, quite variable from one year to another, are given in Supplementary Table 2. The rainfall received from sowing to maturity was 186 mm, 171 mm, and 95 mm for the first, the second and the third years, respectively. The end-of-cycle period of culture at the field trial site coincided with low rainfall associated with high temperatures. Irrigation water was added to overcome the water deficit during the whole cultivation cycle, to meet the needs of the crop, fully or only partially when water stress was applied.

Field trial stress application

After the preparation of the soil (tillage, then leveling of the plot  and limitation of the three subplots of 16 m² each) with a clay-loam soil, it was proceeded to the mineral fertilization consisting of a triple super phosphate feed (TSP-46%) as a baseline fertilizer applied before sowing (with a dose of 130 g / 16 m²) and ammonitrate 33% as a cover fertilizer (with a dose of 220 g / N). 16 m²⁾ applied to tillering. The application of fertilizers was done on the fly.

A fungal treatment (product offered by the INRA of Settat) was applied on March 20th for the second and the third year, then during the first year this product was applied late at the stage of the run. Fertilization during the first, second and third years of study took place on 30/11/2013, 29/12/2014 and 30/12/2015, respectively. Before the sem-grains durum wheat, the three subplots were first irrigated to saturation of the soil. After 24 h, the water was brought in daily to reach the lost ETM (Table 1).

Seeding was carried out the first year on 01/12/2013 and the two other years in early January (01/01/2014 for the second year and 01/01/2015 for the third year). The test was carried out under two irrigation treatments. The first treatment (T1) consisted of normal irrigation (control) providing  volumes   of  water  just  to  compensate  for  the ETM. The second treatment (T2) consisted of a week of water stress, applied 46 days after the semi, during tillering: the first year of irrigation was stopped on 16/01/2014, and 16 / 02/2014 for the second year and 16/02/2015 for the third year. The total water amounts received by the two irrigation treatments are given in Table 2.

Leaf rolling determination

In our experiment, the degree of winding of durum wheat varieties studied was evaluated according to the indices used at INRA to estimate the winding and the type of adaptation of each of the varieties having manifested a winding under the supervision of Dr N. Nasser Elhaq (Research director at INRA Settat Genetic improvement and plant genetic resources unit Specialized: Genetic improvement of cereals). There are other methods for estimating leaf roll in cereals, for example in rice, Leaf Roll Index (LRI) has been calculated at flag leaf level with the following formula: LRI = (Lw-Ln) / Lw (Li et al., 2016). Lw was the largest flag leaf width in extending leaf blade, Ln was the natural distance from flag leaf margins at the same Lw measurement area.

One week after the water stress application, the flag leaf rolled from noon to 4 pm (at the daytime of peak temperature), and beyond 4 pm, unrolled. The degree of leaf rolling was evaluated from noon to 2 pm on the 7th day of water stress, based on the indices established by the INRA of Settat (N. Nsarellah (Research director at INRA Settat Genetic Improvement and Plant Genetic Resources Unit and Specialized: Genetic improvement of cereals) 8 different rolling degrees from absence of rolling to a leaf rolled all over its length displaying a thorn shape (Table 4 and Supplementary Figure 2). The flag leaf is the most sensitive leaf to water stress, when the soil starts to dehydrate; this leaf rolls on itself and just 30 min after hydration of the soil the leaf begins to unfold. Its presence is obligatory for growth and normal production in wheat, because it feeds the ear (Table 4).

Agronomic parameter measurements

(i) The number of tillers per plant was counted in each plant.

(ii) The height of the plant at maturity was measured from the visible base to the average top of the ears.

(iii) The number of ears per plant was counted in each plant.

(iv) The number of grains per ear was counted in each ear in the different varieties.

(v) The weight of 1000 grains (WTG) was determined per variety using a balance, with 1000 grains counted.

(vi)The grain yield (in g) per plant was calculated by multiplying the number of ears per plant by the number of grains per ear and by 10^-3 times the weight of 1000 grains.

Statistical analyses

Statistical analysis of the data was performed using the Statistical Product and Service Solutions, SPSS software version 20.0, an IBM product since 2009 (Hejase and Hejase, 2013: 58; https://www-01.ibm.com /support/docview.wss?uid=swg24031901). For each irrigation treatment, we performed variance analysis to determine the effect of the year, the variety and the irrigation treatment and that of their interaction. When the effect of the variety was significant, we performed the Newman and Keuls test to identify groups of homogeneous varieties. The Newam and Keuls test is a multiple comparison procedure that  allows  sample  means significantly different from each other to be identified. This procedure is often used as a post-hoc test whenever an analysis of variance (ANOVA) revealed a significant difference between three or more sample means.

Similar conclusions as those relative to the grain yield could be drawn when analysing the different yield components (the number of ears per plant, the number of grains per ear and the weight of 1000 grains), and the number of tillers and height of plants at maturity (Table 3). The variance analysis of examined agronomic parameters and yield indicated both highly significant inter-varietal and inter-annual variations (Tables 3 and 6). The variability between varieties was higher in water stressed parcels than in control ones for all the agronomic parameters. In water stressed parcels, the variability between varieties was also higher than the variability between years, except for the weight of grains. The variability between years was the highest in control conditions, except for the number of grains per ear and the weight of grains (Table 3).  The  parameters  showing the highest variability between varieties were the height of plants at maturity and the number of grains per ear in both control and water stress conditions (Table 3).

From agronomic performances measured in the control and water stressed parcels, the performance decrease due to water stress (or conversely preservation under water stress) was also examined for all selected agronomic parameters (Supplementary Figure 3). An important variability in the sensitivity to water stress was observed within the panel of wheat varieties (mean grain yield decrease under water stress from 21 to 93%). Slight differences were noticed for the different agronomic parameters: the weight of grains showed the largest range of variability in stress sensitivity between species (mean decrease under water stress varying by a factor of 8.3), and the number of ears per plant showed the lowest range of variability (mean decrease under water stress varying by a factor of 5). Analysis of inter-annual variation indicated that the preservation of agronomic performances under water stress was more strongly reproducible (slope  of  inter-annual  correlation  between 1.02 and 1.07, 0.79 <R <0.89 for grain yield; Supplementary Figure 3) than the agronomic performances in control conditions or under water stress (Figure 1 and Table 5). Thus, the sixteen varieties of durum wheat of our study displayed a large range of variability in their agronomic performance, especially when subjected to the one-week water stress. Low inter-annual variability was noticed when the decrease in performance due to water stress was considered.

Effect of water stress on leaf rolling in the different durum wheat varieties

Plants of the water-stressed parcels displayed flag leaf rolling at the end of the water stress treatment during the hottest hours of the day, in contrast to those of the well-watered control parcel where no leaf rolling was observed at that time. A large variability among the 16 durum wheat varieties in the degree of leaf rolling was noticed (Figure 2). Indeed, observed rolling in the different varieties ranged from concerning at most the very tip (mean score of the variety between 0 and 1, e.g., in Marjana and Waha) to reaching at least two third of the leaf (mean score between 6 and 7, e.g., in Irden and 2777). Thus, the rolling behavior in the different varieties almost ranged from absence of rolling (score 0) to maximal degree (score 7), all the rolling degrees being represented. The same range of variation among varieties was observed in each of the three crop years (Table 4 and Figure 2A-C).

The experimentation was conducted on a set of durum wheat varieties maintained at the National Gene Bank of Morocco (INRA, Settat). This set gathered varieties originating from different world areas, although essentially Mediterranean ones (and mostly Morocco), and were issued from breeding programs at work at different periods (http://wheatatlas.org/varieties) over the 20th century (oldest variety: Kyperounda “2777" released in 1956 in Cyprus; most recent varieties: Irden and Marouane released in 2003 by INRA Mococco). Most of the selected varieties have been released in the late 1980s and the 1990s, as a result of programs in which the objective of productivity increase and yield stabilization between years was associated with a reduction of the plant cycle duration and plant size (Nsarellah et al., 2011; Taghouti et al., 2017). These objectives led to varieties with strongly homogenized plant development and, owing to the reduction in plant cycle and biomass, tended to improve the tolerance to drought. A disparity in drought adaptation however remained in the durum wheat cultivars released during that period (Nsarellah et al., 2011), a specific objective of tolerance to abiotic stresses having been introduced as one of the most recent steps in breeding programs with released varieties more systematically tolerant to drought only since the  beginning  of  the  21th  century (Nsarellah et al., 2011). The selected panel of 16 varieties of the present study displayed in our trial a large range of sensitivity to water stress with mean grain yield preservation over the three campaigns varying by a factor

of 12 between varieties (Figure 1 and Supplementary Figure 3). The grain yield under water stress in our panel of varieties appeared strongly correlated to that in control conditions (Figure 5; R = 0.92). This is in line with the assumption that selection for high yield over the years tend to associate characters of general dependability, the yield improvement being then based on the ability of the plant to overcome varying small or bigger stresses (Duvick, 2005; Tardieu, 2012).

The application of the one-week water stress widely affected the agronomic performances of the durum wheat varieties studied (Supplementary Table 3). All studied parameters (number of tillers, plant height, number of ears, number of grains, weight of grains) were affected, although slight variations in the level of sensitivity between the different parameters were noticed (e.g., decrease under water stress between 4.5 and 10% for the different parameters in the tolerant variety 2777, between 51 and 64% in the sensitive variety Marjana). The parameters that were the most affected (e.g., number of ears in 2777, number of grain per ear in Isly, and weight of grains in Marjana) depended on varieties. The water stress was applied at tillering. Grain yield is set up all over the plant cycle. At tillering is set up the number of tillers. Moreover, from the middle of tillering when the apex makes its floral transition and elaborates sketches of spikelets, the fertility of the ears starts being developed. Thus, a stress at tillering is expected to be able to affect different parameters of yield in relation to ear number and features. All the varieties from our panel were early or semi-early varieties, except 2777 which was classified as semi-late variety. A significant difference in the stage of embryonic ear development upon drought stress application in 2777 as compared to the other varieties may explain at least in part the low sensitivity of this variety to the applied water stress. In the other varieties, the quite similar cycles make it unlikely that differences in the developmental stage upon drought stress application would explain the strong differences observed between varieties in terms of subsequent agronomic performances. The observed strong differences between varieties in the sensitivity to the applied drought stress may denote differences in efficiency in water absorption of the water available for the crop or in its use (Royo et al., 2014). The response of the studied varieties to the imposed water stress appeared strongly linked to the ability of these varieties to manifest leaf rolling under water stress conditions (Figures 3 and 4). Leaf rolling in cereals is known as an adaptive response to water deficit in leaf tissues (O’Toole and Cruz, 1980; Kadioglu et al., 2012). Leaf rolling is believed to occur when the evaporative demand is no longer balanced by water uptake by theroots and the extent of rolling is often used as a marker of intensity of drought stress in cereal cultures (Riboldi et al., 2016). When leaf tissue hydration is restored, the leaf unrolls. Upon rolling, by decreasing the leaf surface exposed to sunlight, the water loss by transpiration is reduced (O’Toole and Cruz, 1979). This also allows the plant, by limiting the direct illumination of leaf surface, to limit the heating of leaf tissues, harmful to cellular metabolism. In rice, it was as such, reported to improve photosynthetic efficiency and to delay leaf senescence (Richards et al., 2002; Richards et al., 2004; Zhang et al., 2009). It however reduces the available photosynthetic surface, which negatively affects the plant growth (Li et al., 2016 b).

Leaf shape (angle, width/area), which influences light capture and gas exchange capacity, belongs to the agronomic traits highly considered for potential in grain yield improvement (Yuan, 1997; Wang and Li, 2005; Moon et al., 2011). Leaf rolling, which controls the photosynthetic surface and the water status of the leaf, is one of the traits commonly considered in modern programs of cereal selection (Turner, 1982; Richards et al., 2002; Hu et al., 2009). In rice, a number of QTL of leaf rolling have been mapped (Price et al., 1997, Singh and Mackill 2008, Zhang et al., 2016). Little assessment of the impact of leaf rolling on productivity under water stress conditions is yet available in other cereals (wheat, sorghum, etc.) (Peleg et al., 2009; Bogale et al., 2011). However, for example, five leaf-rolling QTLs co-localizing with QTLs associated with productivity in a population of durum wheat crossed with emmer wheat were recently reported (Peleg et al., 2009).

In the studied panel of durum wheat varieties, leaf rolling positively correlated with the maintaining of growth (height at maturity) and with that of the grain yield parameters (Figures 3 and 4). Only the preservation of grain weight showed lower correlation with leaf rolling index (Figure 4, Supplementary Figure 4), which is in agreement with the period of application of the drought stress which induced the leaf rolling at early stage of plant and ear development. The strong positive correlation observed between the leaf rolling index and the maintaining of grain yield in plants subjected to water stress, may be explained by better preservation of the water in leaf tissues when rolling is stronger, thanks to the reduction of the surface of heated tissue and transpiration. This would help maintaining photosynthetic activity and metabolism during the critical period of water stress, allowing to preserve agronomic performance. Epidermal bull-shaped cells (so-called bulliform cells) are involved in the regulation of leaf rolling (Itoh et al., 2005; Li et al., 2010; Zou et al., 2011; Xiang et al., 2012). According to Willmer (1983), these cells serve as water storage. The loss of turgor pressure in these cells due to the lack of water causes a winding of the leaves. During drought, loss of moisture due to vacuoles causes bulliform cells to roll the leaves of many grass species as the two edges of the leaf fold together (Hsiao et al. 1984, Moulia 1994, Price et al 1997). Once enough water is available,  these  cells  expand  and   the   leaves   unfold again. Folded leaves provide less sun exposure and, as a result, they receive less heat, which reduces transpiration and helps retain the remaining water in the plant. In addition, they also play a role in the development of developing leaves (Moore and Clarke, 1998). Thus, bulliform cells can be considered as a primary target in molecular selection, to modulate leaf rolling and unwinding in response to alternating wet and dry periods.

During water deficit, it may be possible that the water, coming from bulliform cells, contributes to the maintenance of the mesophyll cell functioning, thus allowing them to store photosynthetic assimilates (Bois et al., 1987). This could explain that varieties displaying strong rolling maintain at best agronomic performance. On the other hand, in varieties with low or no rolling under stress conditions, a decrease in transpiration via stomatal closure would cause tissue heating and slow down photosynthesis. In this case, both the water stress, the reduction of the photosynthetic capacity of the plant during this period of tillers developments and formation of the ears will penalize the yield. Based on the present results, leaf rolling is concluded to be one of the morphological parameters that can be used as powerful indicator of strong productive performance in semi-arid conditions, for pertinent selection of adapted wheat varieties.

Leaf rolling belongs to the traits commonly considered in modern programs of cereal selection for drought tolerance. Very few, and contradictory, results however exist on its impact on yield. Here, a water stress was applied at tillering and resulted in a decrease in height and number of stems and a reduction in the yield components in the sixteen varieties of durum wheat studied, thus affecting the final grain yield. These negative impacts of the water stress were much reduced in the varieties with strong rolling behavior. This provides evidence that leaf rolling can mitigate the effects of water deficit.

The agronomic parameters studied in our sixteen varieties of durum wheat revealed a significant variability according to the rolling group. The high-rolling varieties (Amjad, Vitron, 2777, and Irden) showed the best performances in terms of yield (number of ears per plant, number of grains per ear, and weight of grains) in both water stress and control conditions. This suggests that these varieties with high leaf rolling behavior are likely to display a general tolerance to environmental conditions. These varieties are thus promising for cultivation in arid and semi-arid zones. In these areas, high temperatures may also be a limiting factor in durum wheat production, which justifies the choice of both high-rolling and short-cycle varieties that escape drought and high temperatures at the end of the cycle (Karrou, 2003).

Future physiological and genetical investigations will allow to better understand the mechanisms underpinning leaf rolling and their contribution to drought resistance.

The authors have not declared any conflict of interests.

The authors are grateful to Mr P. Dorchies for organizing exchanges between France and Morocco. We thank all those who participated in the execution of the fieldwork and special thank goes to Dr. Nasser Elhaq Nsarellah (Research director at INRA Settat, Genetic Improvement and Plant Genetic Resources Unit, Specialized: Genetic improvement of cereals), who provided us with durum wheat seeds, and for all his advice.