Four maize (Zea mays L.) hybrids: B73xUH007, N22xUH007, P148xUH007, and PHT77xUH007 and their parental inbred lines B73, N22, P148, PHT77, and UH007 were used for the first case study. Four of these inbred lines, B73, N22, P148, and PHT77, and one additional inbred line, S052, were used for the second case study. Each line had nine replicates (individual pots/plants). All the lines are part of the EPPN/EPPN2020 reference maize panel and were selected for the present study according to results of a previous investigation on the genetics of shoot growth (Muraya et al., 2017). The lines B73, N22, and PHT77 were initially provided by Alain Charcosset and Cyril Bauland, INRA(E) Mulon, France (see Rincent et al., 2014) and lines S052, P148, and UH007 were made available by Albrecht Melchinger, University of Hohenheim, Germany, and were propagated at IPK Gatersleben. Seeds of the hybrids B73xUH007, N22xUH007, P148xUH007, and PHT77xUH007 were supplied by Claude Welcker, INRA(E), LEPSE, Montpellier, France.

Seeds were germinated on wet filter paper and, after 4 days, seedlings were transplanted into the custom-made ‘rhizo-pots’, one plant per pot. The special rhizo-pot was designed based on the root phenotyping concept, which was validated in our previous work (Shi et al., 2018) and the prototype was pre-tested in the system. The bottom and three sides of the box are made of black PVC in order to prevent the roots from being exposed to light. The front side is tilted 30 degrees allowing the roots to be visualized effectively. A NIR filter which allows the spectrum above 750 nm to pass through is inserted at this side. The plants were transplanted close to the NIR filter, and the root images were taken only from this side. The size of the box is 35 x 25 x 40 cm (LxBxH; Figure 1 ). The boxes were filled with about 7 kg of black peat soil (Graberde, Plantaflor, Germany). The top surface of the boxes were covered with a black mesh to improve image quality and reduce water evaporation. All the rhizo-pots were placed in carriers and entered the conveyor belt-based automated plant phenotyping system. The system is located at the IPK in a climatized greenhouse and plants were grown under controlled long-day conditions with 25/20 °C and 16/8 h day/night, as descripted by Junker et al. (2015).

All the hybrids and inbred lines used for the first case study were grown only under well-watered conditions (60% field capacity (FC)). The FC was determined on a gravimetric basis as described by Junker et al. (2015). Briefly, soil water content corresponding to 100% field capacity was determined by weighing soil-filled pots (0.3L) after watering to full saturation (100% FC) and weighing them after drying the soil completely (20 days at 70 °C). The weight corresponding to 60% or 35% field capacity was then calculated accordingly. For the second case study, the five inbred lines were grown under well-watered (WW) and drought (D) conditions within the same experiment. The plants of the well-watered inbred lines B73, PHT77, P148, and N22 were the same in both studies. Water supply was maintained by automated weighing and watering towards target weight and was stopped at 13 days after transplanting (DAT) to induce drought stress and was kept to 35% FC. Five soil moisture sensors (Decagon 5TE, UMS, Germany) were inserted in the soil at a depth of 10 cm for each genotype under different treatment to record the water content in the pot. All plants were fertilized with Hakaphos Blau (Compo Expert, Germany) 150 ml/pot at 11 DAT. The hybrids and WW plants were fertilized two times more at 21 DAT and 27 DAT.

In order to further evaluate the newly integrated root phenotyping system, additional plants of two inbred lines B73 and N22 (each with seven replicates) were cultivated simultaneously with the plants of the two case studies and sampled at 19 DAT. These two lines have contrasting root biomass as shown in our previous work (Shi et al., 2018): B73 has a relatively large root biomass while N22 has a small root biomass. At the end of the experiment, roots of B73 and N22 were dug out, washed, and placed in a 28 x 40 cm transparent tray filled with distilled water. The roots were scanned at 400 dpi on an Epson Expression 10000 XL scanner (Seiko Epson).

The V stage (the number of visible leaf collars) and leaf number were counted manually after 13 DAT on a weekly basis. At 40 DAT, shoots were cut from the base and the fresh and dry weight (oven-dry at 70˚C for one week) were determined.

After transplanting, visible light (RGB) and static fluorescence (FLUOR) top view and side view images of shoots, as well as near infra-red (NIR) images of roots, were taken on a daily basis from 4 to 40 DAT (4 to 39 DAT for roots). Due to technical failures, time points were missing at 9 (only for root), 30, and 35 to 39 DAT. Root images were taken by a CMO Mono sensor with 12 Mp resolution (UI-5200SE-M-GL Rev.4, iDS, Germany) which was integrated into the system. During imaging, an LED (LZ4-00R408 peak: 850 nm, range 835-875 nm) panel was used for NIR illumination.

To extract image-based shoot traits, the Integrated Analysis Platform (IAP) software (version 2.0; Klukas et al., 2014) was used with a customized analysis pipeline. From the very elaborate output of the image analysis with 445 phenotypic traits, the shoot traits ‘estimated shoot biovolume (px³)’ (ESV; biomass-related), ‘projected leaf area (px²)’ (PLA; biomass-related), ‘estimated plant height (px)’ (EPH; architecture-related), and the ‘brown to green ratio’ (colour-related) were selected and presented. They are based on visible light and from a combined view, top view, side view, and side view, respectively.

The scanned roots from B73 and N22 were analysed by WinRhizo Pro ver. 2013c (Regent Instruments). Root images derived from the phenotyping facility were pre-processed and analysed by the semi-automated Root Image Analysis (saRIA) software (Narisetti et al., 2019). It supports efficient image segmentation on soil-root images, while user input for selecting the best combination of algorithmic parameters is required. Noisy regions could be manually removed as well (Narisetti et al., 2019). The root traits ‘total root length (px)’ (TRL), ‘total root surface area (px²)’ (TRSA), ‘total root volume (px³)’ (TRV), and ‘average root diameter (px)’ (RD) were used for further analysis. The NIR root images from B73 and N22 at 19 DAT, which were used for validation, were additionally analysed by the SmartRoot software (Lobet et al., 2011). The root trait values extracted from images acquired after 29 DAT were not considered for statistical analysis as they were regarded as unreliable due to the increasing density of the root system and the progressive merging and overlapping of roots. Nevertheless, values derived from images taken at two time points, 34 and 39 DAT, were included in the figures, but only for illustrative purposes (shaded grey in the figures).

For the first case study, mid-parent heterosis (MPH in %) was calculated as the difference between hybrid performance (F1) and the mean value of the two parents [MP=(P1+P2)/2] for each trait at all time points as Eq.1.

To evaluate the drought tolerance of the lines, the biomass ratio was assessed by comparing the biomass under drought (D) with the biomass under well-watered (WW) conditions. The calculation was done as follows (Eq.2-Eq.5; Fischer and Maurer, 1978; Correia et al., 2022). Shoot DW ratio was calculated based on the data at the end of the experiment, while the estimated shoot biovolume (ESV), total root length (TRL), and total root volume (TRV) were derived from the daily acquired images. Mean values were calculated for each day of the growth period after drought was imposed at 13 DAT.

The manually measured traits were analysed by an analysis of variance (ANOVA) or t-test using GENSTAT software ver. 16.0. Correlations between traits were analysed using the Pearson product moment correlation. The data visualization for phenotyping data was performed using the R software (R Core Team, 2019). Significant differences between the treatments for each day and trait were determined by one-way ANOVA at a significance level of 0.05.

Two genotypes contrasting in root biomass, B73 and N22, each analysed with seven replicates, were sampled at 19 days after transplanting (DAT) and used to validate the root phenotyping setup in our phenotyping platform. To this end, roots were dug out manually, washed, scanned, and root morphological traits were analysed using the WinRhizo (SC) software to generate ground truth data. The high-throughput phenotyping images were processed with ‘SmartRoot’ (SR) and ‘Semi-automated Root Image Analysis’ (saRIA).

The correlations between root dry weight (RDW) and root traits obtained by different software tools are shown in Table 1 . There were high positive correlations among all three root traits: total root length (TRL), total root surface area (TRSA), and total root volume (TRV). The TRV obtained by scanning the root system displayed the highest correlation with RDW (r = 0.99). TRV reflected RDW better than TRL, regardless which software was used. Notably, the traits obtained by the saRIA software showed higher correlation with RDW than the traits obtained by the SmartRoot software which required manually tracing the roots. Therefore, the saRIA software was used to analyse root traits in the two case studies performed.

Dynamic shoot and root growth of hybrid and inbred maize genotypes.

At the end of the experiment, most hybrids (B73xUH007, N22xUH007, P148xUH007, and PHT77xUH007) had a significantly higher plant height (PH) and shoot dry weight (DW) compared with the corresponding female parental inbred lines (B73, N22, P148, and PHT77) or the male tester UH007 ( Figure 2 and Supplementary Figure 1 ). The only exception was the DW difference between N22xUH007 and UH007. The manually measured traits V stage and leaf number, both representing the development of the maize plants, increased over time. At 13 DAT, the hybrids were further developed, displaying a higher V stage and leaf number than the female inbred lines. Significant differences in V stage between UH007 and the hybrids were detectable starting at 26 DAT (P148xUH007). For leaf number, B73xUH007 and P148xUH007 had significantly more leaves than UH007 at 13 DAT and all the hybrids had a higher leaf number than UH007 at 40 DAT, similar to the other female inbred lines. Among the inbred lines, N22 displayed significantly lower shoot DW and a lower plant height compared with B73, PHT77, and UH007. During the whole growth period, N22 developed fewer leaves than the other inbred lines, which could be observed as early as 13 DAT.

Similar tendencies of the manually measured traits were also observed for the image-derived shoot traits. Estimated shoot biovolume (ESV), estimated plant height (EPH), and projected leaf area (PLA) from 4 to 40 DAT are shown in Figures 3A, B and Supplementary Figure 2 , respectively. Due to a technical failure at 30 DAT, no imaging and watering was performed on this particular day. There were also no images between 35-39 DAT, but watering was carried out every second day during this period. Both hybrid and inbred lines showed continuous ESV, PLA, and PH increases over time. There was more variation at the later stages, most likely caused by losing some old leaves and/or greater overlap of leaves. As mentioned by Scharr et al. (2016), with the growth of plants, leaves tend to overlap which could result in less accurate estimates of the leaf area due to partial occlusion.

The traits TRL, TRV, and TRSA were extracted from the root images acquired by the NIR camera and their changes over time (from 4-29 DAT) were analysed ( Figures 4A, B and Supplementary Figure 3 , respectively). In concordance with shoot growth, the hybrid lines displayed higher TRL values compared to the inbred lines almost throughout the whole growth period. This tendency could be observed already during the early growth phase at 4 DAT ( Table 2 ). At 29 DAT, the hybrids B73xUH007, N22xUH007, P148xUH007, and PHT77xUH007 had a 1.6, 4.2, 2.5, and 1.6 times greater TRL than their respective parental inbred lines B73, N22, P148, and PHT77. Among the inbred lines, B73, PHT77, and UH007 had similar TRL, and they were substantially higher compared with N22 ( Figure 4A ). Roots of N22 were not only shorter, but also displayed a limited volume, which reflected a smaller root system. TRV showed a similar tendency as TRL regarding the growth pattern of hybrids and inbred lines, except that UH007 had the highest TRV among all inbred lines, which was even higher than the hybrid N22xUH007. This mainly resulted from the bigger RD, as show in Supplementary Figure 3 .

The mean mid-parent heterosis (MPH) for the hybrids was calculated based on the imaging-derived traits of both shoots and roots ( Figure 5 ). Generally, the MPH of all traits varied dynamically over time with higher MPH values at the early stage. TRL, TRV, and TRSA especially displayed high MPH at 4 DAT, though with quite high variation. The highest MPH for EPH was found at 5 DAT which reached 144%. The MPH of ESV and PLA showed the highest value several days later, in the period of 9 to 10 DAT, and both reached more than 200%. It seems that heterosis was manifested earlier in roots than in shoots, while the degree of heterosis was lower in roots compared to shoots. The range of MPH for TRL and TRV (4-29 DAT) was 70-146%, and 28-144%, while MPH for ESV and PLA (4-40 DAT) it ranged between 73-217% and 81-224%, respectively.

The inbred lines B73, N22, P148, PHT77, and S052 were evaluated for their phenotypic response to drought. Drought stress was induced by stopping the water supply starting from 13 DAT. Due to the large soil volume in the rhizo-pot, the soil moisture level dropped down progressively as shown either by soil VWC measured with soil sensors ( Figure 6 ) or by the calculated field capacity (FC; Supplementary Figure 4 ). Both methods showed the same tendency that only at a late stage, after 33 DAT, did the FC of the soil in the pots of B73 and PHT77 decrease to 35%. For P148 and S052, the FC dropped to 40% during the last two days, while the pots of N22 still had about 45% FC at the end of the experiment ( Supplementary Figure 4 ).

Drought stress significantly decreased PH and shoot DW of B73, PHT77, and S052. Although P148 displayed a decreased PH, the DW did not differ significantly between well-watered and drought-stressed plants. Both PH and shoot DW were unaffected by the drought treatment in N22, likely due to a lower stress intensity with a relatively high FC (higher than the intended 35%) even at the late growth stages ( Figure 7 ). The V stage and leaf number of B73 and PHT77 were affected by the drought from 33 DAT, while only at 40 DAT, a substantial effect was observed on P148 and S052. Similar to PH and shoot DW results, the V stage and leaf number did not differ between the two treatments in N22 ( Figure 7 ).

It took several days for plants to show detectable phenotypic changes after the drought stress was imposed. Starting from 29 DAT, a significant difference in ESV appeared between well-watered and drought-treated plants ( Table 2 ). Drought tended to reduce PH for all the tested inbred lines over time, although significant differences were not detectable until 32 DAT. The ESV and EPH at 40 DAT were 2.3 and 1.35 times higher, respectively, under well-watered conditions compared with drought treatment ( Figures 8A, B ). B73 and PHT77 were most severely affected by the drought stress. Their ESV and PLA decreased about ten days after drought imposition, while there was no significant difference detectable in N22 at that time ( Figure 8A ; Supplementary Figure 5 ). Most colour-related shoot traits, such as the brown to green ratio ( Supplementary Figure 6 ), showed no obvious changes due to the drought stress.

Root growth of all tested lines was substantially affected by the drought stress ( Figures 9A, B and Supplementary Figure 7 ). Water deficit significantly decreased TRV compared to well-watered plants, and this was evident from 24 DAT, about 10 days after the watering was stopped ( Figure 9B and Table 2 ). TRL showed a trend toward higher TRL in the drought-treated plants, although the differences between 18 and 23 DAT were not significant ( Figure 9A ). Subsequently, the TRL of most drought-treated plants stagnated or decreased over time and was on average 10% lower than that of the well-watered plants at 29 DAT. RD of all the lines were decreased under drought stress ( Supplementary Figure 7 ).

Drought stress reduced the shoot DW for all tested inbred lines ( Figure 10A ). Among the lines, B73 was most severely affected by the drought treatment and gained only 45.3% of shoot biomass of the corresponding well-watered plants, while N22 was least affected (79.5%). After one week of drought stress, the ESV ratio between drought and well-watered plants dropped below a value of 1 for all lines. The value continuously decreased over time and reached on average about 0.5 at 40 DAT ( Figure 10B ). There was variation in the degree of reduction in the five inbred lines. Starting from 32 DAT, the observed tendency became stable. Consistent with the end-point results at 40 DAT, the ESV ratios for B73, PHT77, and S052 were lower than for P148 and N22, indicating a stronger effect on those lines. Root growth was affected by the water deficit as well, as shown in Figures 10C, D . The TRV ratio decreased over time, and it appears there were genotypic differences between the lines. After ten days of water deficit, the TRV of S052 was much less reduced than the TRV of the other lines ( Figure 10C ). The ratio of TRL showed a different pattern. The ratio was above 1 for all lines during more than half of the drought stress period from 15 to 25 DAT. This increased root length (ratio >1) lasted for about ten days, and then decreased to an average value of 0.83 at 29 DAT (except for S052). Notably, for S052 the TRL increased for a period of 14 days (15 to 29 DAT), while for N22 it only increased for 7 days (15 to 22 DAT; Figure 10D ).

The non-invasive high-throughput shoot phenotyping platform at IPK has been utilized in many research studies for diverse models and crop plants (Junker et al., 2015; Muraya et al., 2017; Knoch et al., 2020; Dodig et al., 2021). As realized by many researchers, crop species with optimised root systems are essential for future food security and key to improving agricultural productivity and sustainability (Li A. et al., 2022). In order to enhance our understanding of the root system, in particularly the dynamics of root growth and development, a root phenotyping concept was developed (Shi et al., 2018) and root phenotyping units were established and integrated in the existing phenotyping platform for large plants. In this work, we present the upgraded facility and the use of the root phenotyping units to evaluate jointly shoot and root growth in two case studies.

The high positive correlations between root DW and image-derived traits such as TRL and TRV indicate that our root phenotyping facility is well suited to monitor root growth ( Table 1 ). The root analysis software saRIA (Narisetti et al., 2019) was utilized to analyse the root images obtained by the automated high-throughput phenotyping system. This upgraded system now enables simultaneous non-invasive analysis of root and shoot traits of the same plants in a particular phenotyping experiment.

Hybrids often display superior phenotypes due to their vigorous nature (van Dijk et al., 2016), a phenomenon widely known as heterosis. The superior performance of F1 hybrids compared to their parental inbreds has been known for decades, although the underlying genetic and regulatory mechanisms remain largely unclear (Paschold et al., 2010). Heterosis studies rely on morphological and physiological analyses of inbred lines and corresponding hybrids. So far, most related research has focused either on shoot or root traits independently using separate approaches.

In the present study, the dynamics of vegetative shoot and root growth were investigated in a selection of maize hybrid and inbred lines. Growth-related image-derived shoot and root traits were analysed during the early vegetative growth phase, from 4 to 40 DAT. The phenotyping results indicated an early establishment of heterosis in the tested hybrids. EPH, TRL, and TRV were higher in hybrids than in their corresponding parental inbred lines as early as 4 DAT, the first day when shoot and root images were taken ( Table 2 ). At the end of the growth period, EPH and shoot DW in hybrids were on average 20% and 55% higher than in their parental inbred lines, respectively. These findings are consistent with the results of previous studies, which showed the superiority of hybrids (Hauck et al., 2014; Zhou et al., 2018). The range of MPH for ESV (73-217%) and PLA (81-224%) was similar to that reported by Tollenaar et al. (2004). In their study, they reported MPH values from 138% to 214% for dry matter accumulation and 92% to 204% for leaf area at the 14-leaf stage, respectively. Thanks to the better performance with superior biomass and higher seed yield, hybrid maize varieties have been predominantly grown worldwide since the 1960s (Li C. et al., 2022). Further increasing yield potential and yield stability through heterosis remains a major goal of maize breeding (Duvick, 2005; Li C. et al., 2022). The advances in high-throughput phenotyping facilities will assist and fasten this process by supporting the gain of fundamental genetic and mechanistic knowledge.

In addition to the aboveground parts, heterosis is also observable in the belowground organs of plants. In accordance with Hoecker et al. (2006), the primary root length, number of seminal roots, and the lateral root density can display substantial heterosis. They demonstrated that heterosis manifests in the very early stages of root development a few days after germination. We observed in our study that hybrids displayed high MPH values as early as 4 DAT (8 days after germination), with on average 67% and 70% higher TRL and TRV values than the inbred lines, respectively ( Figure 5 ). Compared to conventional methods of quantifying trait expression at one particular time point (usually using destructive techniques), heterosis in shoots and roots can be assessed here in a dynamic manner by a non-invasive, fast, and high-throughput procedure. We observed that heterosis occurred earlier in roots than in shoots and decreased over time in both organs.

The ability to compare heterosis for both shoot and root simultaneously, even under various conditions, will help researchers to further explore the developmental and physiological mechanisms associated with heterosis and to jointly study the genetic basis of heterosis for both organ systems, which are highly interdependent.

Drought stress is one of the most serious adverse environmental factors limiting crop productivity and a major threat to world food security (Boyer et al., 2013; Berdugo et al., 2020). Due to global climate change, the frequency and duration of drought periods will most likely increase (Trenberth et al., 2014). Therefore, it is of the utmost importance to understand how plants respond and adapt to water deficit in order to support solutions to this problem and enhance the sustainability of agricultural production.

In the present study, drought stress was induced at 13 DAT and drought symptoms such as leaf wilting and rolling could be observed in B73 and PHT77 at the growth stages V6-V10. The effect of drought was reflected by decreases in PLA and ESV, as well as in the final shoot DW. Plants generally decline the number and area of leaves in response to drought stress. This was confirmed by our manually measured parameters and image-derived shoot phenotypic traits ( Figures 7 , 8 ). To cope with drought, plants induce protective mechanisms against water deficit. In addition to stomatal closure, assimilates are often re-allocated from the shoot to the root, thereby enhancing root growth and extension into deeper soil layers (Rich and Watt, 2013; Xu et al., 2015). Roots have the ability to plasticly change their spatial distribution in the soil in response to drought stress (Orman-Ligeza et al., 2018; Orosa-Puente et al., 2018; von Wangenheim et al., 2020). The degree of plasticity, however, is trait and genotype dependent. As mentioned by Tracy et al. (2020), phenotypes in roots and shoots are expressed differently depending on environmental conditions. Our results show that roots respond faster to drought than shoots as significant differences between drought and WW treatments occurred 10 days after drought imposition in roots, while in shoots significant differences could only be observed 4 days later ( Table 2 ). Drought affected not only root biomass, which was represented by TRV, but also modified the morphology of root with changes in other traits, such as TRL, and also the root diameter.

Notably, there was a tendency for TRL to increase shortly after the onset of drought stress, although the difference was not significant. Sharp and Davies (1979) showed that a mild degree of water stress resulted in a higher root elongation rate compared to well-watered maize. The rate of root growth probably depends on the degree and the duration of the stress. Under extreme water deficit, root growth will be inhibited in many species (Rich and Watt, 2013; Kou et al., 2022). It is therefore highly important to assess the dynamics of root growth changes over time and under different conditions to gain deeper insights into the responsible mechanisms. Addressing only single time points will be inappropriate. Our high-throughput phenotyping platforms supports such measurements and thus the elucidation of the responses of plant organs to environmental cues and the adaptation to stress conditions.

The biomass ratio between well-watered and drought-treated plants provides a parameter to compare different genotypes with respect to their response to drought (Harb et al., 2010; Correia et al., 2022). Among the five inbred lines tested here, B73 exhibited the lowest ESV ratio at most time points. This suggests that B73 was most severely affected by drought, which is in line with the study of Chen et al. (2012), who classified B73 as a drought-sensitive line. The estimated TRL and TRV ratios suggested that S052 produced more roots, likely through enhanced lateral root growth and this effect lasted also longer than in the other lines. This might partly result from the relatively small shoot biomass (and transpiration) and the lower water consumption of the plants of this line ( Figure 10 ), which caused a longer time period to reach water deficit than for other genotypes.

In future, some features of the phenotyping installation need to be taken into account when it is used to assess genotypic difference in responses to drought: due to the relatively large volume of the rhizo-pots compared to regular growth pots, soil moisture levels will decrease more slowly. This is closer to the natural drought scenarios in the field, where gradual changes in water availability occur rather than abrupt changes (Poorter et al., 2012). Soil drying in the rhizo-pots occurs due to the water consumption by the plants and the evaporation from the soil surface. While the latter is rather uniform (also due to the use of the soil cover), the former is affected by the size of the plants and by their physiological states, in particular their rates of transpiration. The impact of small plant size was evident for the line N22, which did not suffer from the drought stress as much as the other lines, since the FC was still 40% at the end of the experiment ( Supplementary Figure 4 ). This is probably the main reason why the reduction of shoot mass was least compared to the other genotypes. A similar tendency could be observed for many other measured traits. Also, most colour-related traits did not differ after water deprivation, which might also be caused by the weak stress imposed in the system. If the different water consumption of plant lines under investigation (mainly due to different plant sizes) cannot be avoided by an appropriate selection of the population under investigation, we suggest adjusting the drought regime imposed to all plants to that of the plants with the lowest water consumption. This could be achieved by programming a gradual decrease of the watering target weight, which is gauged to the weakest water consumer, rather than a complete stop of watering. To avoid a too long period to reach considerable water stress levels, the soil moisture level in the rhizo-pot used in the initial phase of the experiment should be carefully adjusted.

Plant productivity is the results of integrating processes occurring in both the root and the shoot systems. Therefore, a deeper understanding of the effects of different stresses such as water deficit or nutrient limitation on roots and shoots is of great value for cultivar selection, the improvement of crop models, and low-input agriculture (Hadir et al., 2020). Our observations confirm that the root phenotyping upgrade of the platform supports experiments to assess the dynamic phenotypic changes of both shoots and roots (and thus the relations of these two important organ systems) caused by genetic variation and/or induced by environmental triggers such as drought stress.