In this greenhouse study, two spring canola cultivars previously screened for drought tolerance were utilized: the drought-tolerant cultivar “Czyzowska” and the drought-sensitive cultivar “BN-1.” These cultivars served as founder lines for developing a nested association mapping population (Ebersbach et al., 2022). For each cultivar, six tubs each with 60 L capacity were filled with peat moss + soil mix and watered to field capacity. The bottom was perforated (six holes) to allow water drainage. A slow-releasing fertilizer (Osmocote, Everris, USA) was added at 10.7 g L⁻¹ to avoid nutrient deficiencies. Nine canola seeds per tub were sown at an equal distance and thinned after emergence to five plants per tub. Plants were grown under a day/night temperature regime of 23°C/18°C, respectively, and relative humidity of 45% to 65%. The day/night photoperiod was set at 16 h/8 h with a minimum photosynthetic photon flux density (PPFD) of at least 400 μmol m⁻² s⁻¹ during the day supplemented by sunlight. Plants were regularly watered until bolting. From bolting to physiological maturity, the temperature was increased from 23°C to 28°C over a 60-min period starting at 09:30 h and stayed at 28°C till 15:30 h to ramp down to 23°C within a 60-min window.

Once plants started bolting, tubs were randomly assigned to one of the two treatments (N = 12: 2 treatments × 2 cultivars × 3 replications)—i) control treatment (WW): to mimic field conditions, plants were watered at 90% of maximum water holding capacity (WHC), and ii) drought treatment (D): plants were exposed to 40% of WHC. The water holding capacity (%) of the growth medium was determined as described in Elferjani and Soolanayakanahally (2018). Prior to physiological maturity, plants exposed to drought were watered optimally (90% of WHC) and allowed 48-h recovery before leaf sampling for amino acid analysis.

Leaf temperature was monitored with a thermal imaging camera (FLIR T530, FLIR Systems, Wilsonville, Oregon, USA) by taking images of the canopy at 08:00 h, 12:00 h, 14:00 h, 16:00 h, and 18:00 h. The normalized difference vegetation index (NDVI) was measured using a GreenSeeker handheld crop sensor (Trimble, Westminster, CO, USA). The sensor was held 80 cm above the plant canopy, as recommended by the manufacturer. The measurements were taken between 09:00 h and 11:00 h. Plants were harvested at maturity and pods were collected in brown paper bags and stored under ambient temperature for 3–5 days until threshing. Seeds were cleaned and weighed, and oil and protein contents were determined using the NIR method as described in Siemens and Daun (2005).

In a separate experiment, the canola cultivars (Czyzowska and BN-1) were grown in 2-L pots (1 plant per pot) filled with peat moss soil mix and regularly watered under 23°C/18°C and 18 h/6 h day/night temperature and photoperiod (N = 10: 5 replicates × 2 cultivars). At the four-leaf stage, water loss by transpiration of plants was monitored by weighing pots, using an electronic balance. First, plants were watered to field capacity, then medium surface and pot edges were covered with aluminum foil to prevent water evaporation from the substrate, and the initial weight of each pot was recorded at 19:00 h (i.e., greenhouse lights were off). Then, pots were weighed every 2 h for the next 36 h. After pot weighing, the second fully developed leaf from the top of each plant was sampled and petiole tips were covered with Vaseline to assess non-stomatal transpiration. Cut leaves were immediately transferred to a dark room at 23°C to minimize stomatal water loss, where their initial weight and surface area were measured. Then, their weights were recorded every 30 min from 10:00 h to 18:00 h.

The portable photosynthesis LI-6400XT system equipped with a 6400-08 chamber attached to a 6400-02B LED light source (LICOR Inc., Lincoln, NE, USA) was used to measure leaf gas exchanges on the 12th to 15th days following the start of the stress treatment. Measurements were made on the 4th fully developed leaf from the top (N = 12: 2 treatments × 2 cultivars × 3 replicates) between 10:30 h and 12:30 h. The response of the net photosynthesis (A, μmol m⁻² s⁻¹) to the changing C _i was measured under saturated photon flux density, PPFD = 1,000 μmol m⁻² s⁻¹. The leaf was first exposed to the ambient atmospheric CO₂ concentration, C _a (400 μmol CO₂ mol⁻¹) in the chamber using CO₂ cartridges to reach a steady state. Next, C _a was changed in the following order: 400, 300, 200, 100, 50, 400, 500, 600, 800, 1,000 and 1,200 μmol mol⁻¹. At each step, we ascertained that the net photosynthetic assimilation rate (A), water vapor, and CO₂ fractions reached steady values at each step before moving to the next step. During the measurement periods, the leaf chamber temperature was set to the ambient temperature (28°C), airflow at 500 μmol s⁻¹, relative humidity at 55–65%, and vapor pressure deficit (VPD) at 1.4 ± 0.2 kPa. The order of the measurements was randomized among the treatments and the cultivars throughout the measuring period. A (μmol CO₂ m⁻² s⁻¹) and g _s (mol CO₂ m⁻² s⁻¹) values were extracted from A–C _i response measurements for C _a = 400 μmol CO₂ mol⁻¹. The intrinsic water-use efficiency was then deduced (A/g _s). The maximum rate of RuBisCO carboxylation (V _cmax, µmol m⁻² s⁻¹), the rate of photochemical electron transport (J, µmol e⁻ m⁻² s⁻¹), and the rate of CO₂ diffusion through the leaf mesophyll conductance (g _m, µmol m⁻² s⁻¹) were estimated by A–C _i curve fitting, according to Ethier and Livingston (2004); Ethier et al. (2006), and the biochemical model of C₃ photosynthesis developed by Farquhar et al. (1980) as detailed in Elferjani and Soolanayakanahally (2018).

The same leaves were subjected to gas exchange measurements at night 16 days after the stress started, using the same apparatus used to measure photosynthetic activity during the day. Leaf chamber parameters that were changed were as follows: PPFD = 0 μmol m⁻² s⁻¹, leaf temperature = 18°C, and airflow = 200 μmol s⁻¹. While the C _a was maintained at 400 μmol CO₂ mol⁻¹ as well as A, VPD and CO₂ fractions were allowed to stabilize before records could be taken. Leaf gas exchanges were monitored twice during the night: the first measurement at night (between 21:45 h and 22:45 h) and the second measurement at predawn (between 03:00 h and 04:00 h). Daytime gas exchanges were also measured early in the morning (between 05:00 h and 06:00 h) with PPFD = 1,000 μmol m⁻² s⁻¹, airflow = 500 μmol s⁻¹, and leaf temperature set at 23°C.

One fully expanded leaf was sampled from both cultivars (N = 24: 2 cultivars × 2 treatments × 3 replicates × 2 leaf surfaces) immediately after gas exchange measurements. Freshly cut leaves were kept in a cool and dry container and stored in a refrigerator at 4°C before being analyzed the next day at the Canadian Light Source Facility (https://www.lightsource.ca/Saskatoon, SK, Canada). Mid-infrared (mid-IR) spectroscopy was used to determine the total wax load and to identify the different functional groups of the epicuticular wax on canola leaves according to the protocol of Willick et al. (2018). Briefly, the mid-IR attenuated total internal reflection (ATR) spectra of fresh leaves were collected using the Cary 600 series FTIR spectrometer (Agilent Technologies, Santa Clara, CA, USA). The ATR crystal used was germanium (45 degrees). Mid-IR data in the spectral range between 4,000 and 600 cm⁻¹ (wave numbers) at a resolution of 4 cm⁻¹ were recorded at 256 scans per sample on average. Three parameters were considered to characterize the variation of functional groups of waxes following the application of treatments—i) aromatic carbon groups (C=C functional group, 1,650–1,500 cm⁻¹): spectral region with variable intensity depending on the plant species (Heredia-Guerrero et al., 2014); ii) methylene/methyl ratio (CH₂/CH₃, 3,000–2,800 cm⁻¹): related to the length of the aliphatic chain and to the branching of the side groups. A higher value of this ratio indicates longer and straighter (less branched) chains; iii) carbonyl (C=O, 1,800–1,600 cm⁻¹): the relative contribution of carbonyl/carboxyl group to a group containing oxygen + aromatic carbon (C=C) structures.

Samples were prepared for scanning electron microscopy (SEM) by sampling a 5 × 5-mm portion of the leaf, taped to aluminum foil, and allowed to air dry for 1 week in a desiccator. Samples were then mounted on 9 mm diameter aluminum specimen holders with double-sided carbon tape. Mounted samples were Au sputter-coated in a Denton Vacuum Desk IV Sputtering unit. Sputtering utilized air for plasma gas at 50 mTorr; 38% power applied to the target resulted in 10 mA of current achieving 26 nm film thickness at a 4.3 nm/s deposition rate. SEM was conducted using a Zeiss EVO 60 at 10.19 kV and 8 mm working distance. Three images at different magnifications were collected at ×125, ×2,000 and ×5,000 with 895.2 nm, 55.84 nm, and 22.33 nm spatial resolutions, respectively.

Fully developed leaves were sampled to assess the stomatal density on the abaxial and the adaxial sides using the clear varnish and tape technique. Leaves were coated with a clear varnish and allowed to dry for approximately 1 min. Then, the clear tape was applied on the varnished surface, then peeled off and stuck onto a glass slide. The number of stomata of the epidermal impression on each slide was counted using a fluorescence microscope (Axio Imager Z1, ZEISS, Oberkochen, Germany) and ZEN software (2.3 Pro, ZEISS). Stomatal density, pore length, and pore width were calculated using the scale tool of the ZEN software.

Fully expanded leaves (third or fourth leaf from the top) were harvested and immediately packed in plastic tubes and frozen in liquid nitrogen before being stored in a −80°C freezer until processed for analyses. Abscisic acid content was determined as previously described (Yan et al., 2016). Briefly, the freeze-dried and powdered samples were suspended in methanol containing 1% acetic acid, and d6-ABA was added as an internal standard and placed in the fridge overnight. The samples were centrifuged to remove debris, and the pellet was washed twice. The supernatant was evaporated in a SpeedVac and reconstituted in 1 mL of 1% (v/v) acetic acid. ABA was purified by solid-phase extraction using Oasis HLB, MCX, and WAX cartridge columns (Waters). The solvent was removed under vacuum and subjected to LC-ESI-MS/MS analysis (Agilent 6,410 Triple Quad LC/MS system). An LC (Agilent 1200 series) equipped with a 50 × 2.1-mm, 1.8-μm Zorbax SB-Phenyl column (Agilent) was used with a binary solvent system comprising 0.01% (v/v) acetic acid in water (solvent A) and 0.05% (v/v) acetic acid in acetonitrile (solvent B). Separations were performed using a gradient of increasing acetonitrile content with a flow rate of 0.2 mL min⁻¹. The gradient was increased linearly from 3% B to 50% B over 15 min. The retention time of ABA was 14.0 min. MS/MS transitions 269/159 (d6-ABA) and 263/153 (ABA) were used to quantify ABA.

Total amino acids were extracted from 10 mg powder of freeze-dried tissue samples following Inaba et al. (1994) with some modifications. Briefly, 1.5 mL of 80% (v/v) ethanol solution was added to each sample and shaken for 30 min at 40°C, and the supernatant was recovered by centrifugation (4,000 rpm for 10 min) at 4°C. Amino acids were derivatized following the Waters AccQ-Tag Reagent Kit (Waters, Milford, Massachusetts, USA; Cohen and Michaud, 1993). Briefly; 10 µL aliquot of the sample was mixed with 70 µL of borate buffer and 20 µL of AccQ-Fluor reagent which was reconstituted in acetonitrile. The AccQ-Fluor reagent was reconstituted as follows: 1 mL of AccQ-Fluor reagent diluent was transferred to a vial containing AccQ-Fluor reagent powder and vortexed for 10 s before heating at 55°C for a maximum of 10 min or until dissolved. The derivatized mixture was transferred to an autosampler vial and incubated at 55°C for 10 min. The derivatized samples were subjected to high-performance liquid chromatography (HPLC) as described in the Waters AccQ-Tag Chemistry Package Instruction manual, with an excitation wavelength of 285 nm and emission wavelength of 320 nm on a Waters Amino Acid Column—3.9 × 150 mm using 2475 scanning fluorescence detector (Waters, Milford, Massachusetts, USA). The column was set at 37°C with 5 µL of injection volume. Waters AccQ-Tag buffer (100 mL of AccQ-Tag Buffer concentrate + 1,000 mL of Super-Q water), acetonitrile, and Super-Q water were used as mobile phase A, mobile phase B, and mobile phase C, respectively. The concentration of amino acids (pmol/µL) from a sample was calculated using peak area values of the chromatogram against the calibration curve of serial dilution (10, 25, 50, 100, 150 pmol/µL) of known amino acid calibration standards (WAT 088122, Waters, Milford, Massachusetts, USA) with α-aminobutyric acid as an internal standard. The values were then converted to µmol/mg using the extraction volume and weight of the initial sample.

Total sugars were extracted from 10 mg powder of freeze-dried leaf tissue samples. One milliliter of 75% (v/v) methanol solution containing 0.1% formic acid was added to each sample and mixed by vortexing for 10 s followed by sonication in a water bath at room temperature for 15 min. The supernatant was obtained by centrifugation (20,000 rpm for 15 min) at room temperature. The resulting supernatants were filtered through a 0.2-μm PVDF filter syringe onto HPLC slit vials and stored at −20°C until use. Sugar analysis was performed on a Waters Acquity UPLC system with evaporative light scattering (ELS) and photodiode array (PDA) detection. A Waters UPLC BEH amide 1.7 μm 2.1 × 100 mm column at 70°C was used with solvents: A—95% ACN/5% water + 0.1% TEA and B—30% ACN/70% water + 0.1% TEA with a 60-min runtime. The ELS detection settings used included gain: 400, mode: cooling, gas pressure: 60.0 psi, and drift tube: 60°C. Sucrose was identified and quantified by comparison with known standards (25 µg/mL) prepared in 80% acetonitrile.

A two-way analysis of variance (ANOVA) was used to test the effects of the cultivar, the water status, and the interaction between them on the measured traits. The means were compared using Tukey’s honest significant difference (HSD) at a p <0.05 significance level. The coefficients and the p-values of the correlations were calculated using Pearson’s correlation coefficient. All the statistical analyses were performed with R software version 4.2.1 (R Core Team, 2021). A correlation bubble plot was generated using the corrplot function (Taiyun and Viliam, 2017), and bar plots and line plots were generated using the ggplot2 function (Wickham, 2016) in R software. Principal component analysis (PCA) was computed using the online clustVis tool (Metsalu and Vilo, 2015). The multitrait genotype-ideotype distance index (MGIDI) was used to rank the amino acids based on information of multiple traits as proposed by Olivoto and Nardino (2021). Lavaan software package with SEM function (Rosseel, 2012) within the R statistical software (R Core Team, 2021) was used to produce the hypothetical pathway model for observed traits.

At physiological maturity, plants of the drought-tolerant (DT) canola cultivar were taller and had higher NDVI values than those of the drought-sensitive (DS) cultivar under well-watered (WW) conditions ( Supplementary Figures S1A, B ). The DT cultivar flowered 7 days later (46 days after sowing, DAS) compared to the DS cultivar (39 DAS) under WW conditions ( Supplementary Figure S1C ). Plants grown under drought conditions were shorter in height and flowered earlier (43 DAS and 37 DAS, respectively, for the DT and DS cultivars). Similarly, seed yield and the number of seeds per pod were significantly lower in the DS cultivar compared to the DT cultivar ( Supplementary Figure S1E ). Drought treatment further reduced the yield for both cultivars. The number of branches, however, was higher in the DS cultivar, both of which were reduced under drought conditions ( Supplementary Figure S1D ). On the other hand, the seeds per pod were higher in the DT cultivar ( Supplementary Figure S1F ).

Leaf temperature was recorded from 08:00 h to 18:00 h every 2 h ( Supplementary Figure S2 ). The lowest leaf temperature was recorded at 08:00 h and the highest at 10:00 h when the greenhouse reached the 28°C target temperature. For both the DT and DS canola cultivars, leaf temperature was significantly higher under drought than under WW conditions at every time point. Water loss was measured by weighing every 2 h for two nights and one full day ( Figure 1A ). Based on daylight time and greenhouse conditions, the nighttime period starts at 21:00 h and ends at 05:00 h the next morning. The amount of water lost every 2 h decreased during the night until 03:00, after which it started to rise in the morning, continued to rise throughout the afternoon, and then began to decrease once more during the night. A subtle difference between the two cultivars was observed during the night, but the differences became significant during the day at 15:00 h and 17:00 h ( Figure 1A ). Average water loss per hour was significantly different between DT and DS during the day, but not during the night ( Figure 1B ).

Leaf stomatal conductance (g _s) was recorded at night, predawn, and in the morning for both the DT and DS canola cultivars under WW and drought conditions. At each of the three time points and under both conditions, g _s was higher in the DS cultivar ( Figures 2A, B ). Among the time points recorded, there was no significant increase from night to predawn; however, there was a drastic increase during the early morning ( Figures 2A, B ). The early morning stomatal conductance rates under the WW conditions for the DS (0.19 mol m⁻² s⁻¹) and the DT (0.12 mol m⁻² s⁻¹) cultivars were higher than those under drought treatments (0.042 and 0.025 mol m⁻² s⁻¹, respectively, for DS and DT). Stomatal conductance was negatively correlated to leaf cysteine and ABA content (R = −0.68, p = 0.015 and R = −0.78, p = 0.0027, respectively) ( Figures 3A, B ). Although ABA content was positively correlated to cysteine (R = +0.49, data not shown), the correlation was not significant (p = 0.1). Furthermore, sucrose showed a significant negative correlation with net assimilation rate (R = −0.83, p = 0.00084) and dark respiration rate (R = −0.60, p = 0.04) ( Figures 3C, D ).

Under the WW conditions, leaf ABA content for the DT and DS canola cultivars was similar (323.8 and 273.1 ng g⁻¹, respectively). Under drought, ABA content increased significantly to 1,322.7 and 869.7 ng g⁻¹ for DT and DS, respectively ( Supplementary Figure S3A ). Sucrose content under the WW conditions was higher in DS (53.4 µg mg⁻¹) compared to DT (20.3 µg mg⁻¹) and increased in both cultivars under drought (by 3-fold and 4.5-fold, respectively; Supplementary Figure S3B ). Similarly, cysteine content was low in both cultivars under the WW conditions and then increased significantly under drought ( Supplementary Figure S3C ).

To analyze the physiological impact of the drought treatments, we measured the maximum carboxylation rate (V _cmax), electron transport rate (J), and mesophyll conductance (g _m) in both cultivars. Mesophyll conductance decreased under drought stress and was not significantly different between the DS and DT cultivars ( Supplementary Figure S4B ). Meanwhile, V _cmax and J differed among the treatments but not between cultivars ( Supplementary Figures S4E, F ). However, the photosynthetic rates differed among the treatments and cultivars ( Supplementary Figure S4C ), but not in intrinsic WUE ( Supplementary Figure S4D ).

SEM images were taken for both the abaxial and adaxial surfaces of the leaves from plants under the WW and drought conditions. Wax morphology was different between treatments and to a lesser extent between the two cultivars ( Figures 4A, B ). The adaxial surface appeared to have more wax platelets, while the abaxial surface contained more wax tubules. Under drought, the DT canola cultivar developed more wax (platelets and β-diketone tubules).

Lipid unsaturation of the epicuticular waxes, measured by C=C content, decreased by 22.8% and 37.7%, for DT and DS, respectively, when plants were exposed to drought stress. Under this treatment, no significant difference was observed between the two cultivars (data not shown). The CH₂/CH₃ ratio had the same variation pattern as C=C with a higher value in the DS cultivar under the same treatment ( Figure 4C ). Drought treatment significantly reduced the CH₂/CH₃ ratio in the DS cultivar ( Figure 4C ). The C=O functional group increased by 30% under drought stress for the DS cultivar. However, there was no significant difference between the two cultivars (data not shown). We also measured the stomata pore length/breath (l/b) ratio of adaxial and abaxial surfaces in both cultivars under the WW and drought conditions. On both surfaces, the ratio increased significantly under drought treatment in the DT cultivar but not in the DS cultivar ( Figures 4D, E ).

Oil content (% seed dry matter) was similar between the DT and DS cultivars under the WW conditions (35.3% and 34.5%, respectively) ( Supplementary Table S1 ). Under drought treatment, oil content decreased significantly to 23.8% and 23.7% for DT and DS, respectively. Similarly, seed protein content was not significantly different between DT and DS under the WW treatment ( Supplementary Table S1 ). However, drought conditions significantly increased protein content in both cultivars.

Principal component analysis (PCA) of sucrose and organic acids was performed to provide a preliminary understanding of the metabolic differences between the DT and DS cultivars under the WW and drought conditions ( Supplementary Figure S5 ). The PCA showed cultivar as the main factor for variance (37.5%) along PC1, while PC2 explained 30.1% of the variance.

Next, we looked at how WW, drought, and drought recovery affected amino acids in both cultivars. We found that out of the 17 essential and non-essential amino acids analyzed, 12 of them had genotype and/or treatment × genotype effect. These 12 amino acids were grouped into five families based on their biosynthetic pathways from the intermediates of the carbon metabolism pathway ( Figure 5 ). When we ranked these amino acids, MGIDI based on the WW and drought conditions selected histidine, serine, and aspartic acid ( Figure 6A ). Similarly, proline, histidine, and aspartic acid were selected when the drought and recovery selection criteria were applied ( Figure 6B ). A correlation matrix was plotted for physiological and metabolic variables that showed significant treatment or genotype effects ( Figure 7 ). Lastly, we used the semPath function from semPlot for confirmatory factor analysis to demonstrate that leaf sulfur and sucrose content directly influenced oil content ( Figure 8 ). The pathway diagram also revealed a clear connection between the production of cysteine, which is produced by diverting leaf sulfur, and the increase in ABA levels that results in stomatal closure.