Twenty-one T. crinita accessions from the GBNG, at IADIZA, were used. These materials were selected to maximize genetic diversity, from collections of representative plants from 48 natural populations of T. crinita dispersed throughout extensive arid and semi-arid regions (~350,000 km²) in Argentina. The selected accessions are genetically and morphologically diverse, and they constitute a representative sample of the broad phenotypic variation observed naturally in the Monte phytogeographical region (Cavagnaro et al., 2006). The geographic locations and main climatic and edaphic characteristics at the collection sites of the accessions are presented in Table 1 and Figure 1 . Noteworthy, these accessions are different from the plant materials used in previous studies by Quiroga et al. (2013) and Marinoni et al. (2020).

The field study consisted of a partially-replicated experiment, conducted during the 2017-2018 growing season (henceforth “2018”), in which 21 T. crinita accessions were evaluated. The experiment was repeated in the following season (2018-2019; henceforth “2019”) with the six most contrasting accessions with regards to their drought-tolerance performance in the previous year. The climatic conditions at the site of the experiment for 2018 and 2019 trials are presented in Supplementary Figure S1 . For implantation, individual seeds were sown in 250 cm³ pots with sterile soil, and plants were grown under greenhouse conditions until they had 5–6 leaves, after which the pots were placed outdoors for a period of rustication, and after one week they were transplanted to the experimental field of the Faculty of Agricultural Sciences, Nacional University of Cuyo (Mendoza, Argentina) (33° S, 68° 8’ O). In the field, the experimental design consisted of two plots, one for the irrigated (control) and one for the drought treatment, separated from each other by five meters, with each plot containing 252 T. crinita plants, for a total of 504 plants used in the experiment. All the plants in both treatment plots were drip-irrigated to field capacity during the first 80 days after the transplant, to ensure that the plants were fully established in the field. After the 80^th day, and coincidently with flowering initiation, the drought treatment was imposed in one of the plots by completely eliminating irrigation, while the control plot continued with the same irrigation regime throughout the experiment. A 21 × 2 factorial design with divided plots (subplots) was used, being the two plots the drought and control treatments, and the 21 subplots the accessions within each plot. Each subplot consisted of 12 plants of each T. crinita accession, with a completely randomized distribution in the field (within each plot) with adjacent plants separated 80 cm from each other. To prevent any water from precipitations to infiltrate the ground, the soil of the entire experimental plot in both treatments, drought and irrigated, was covered with a 200-µm thick polyethylene, and covered with soil on top. By this means, water availability for the plants was completely controlled, and provided only by the drip irrigation system. Supplementary Figure S2 shows the field trial for evaluating drought tolerance in T. crinita accessions.

Plants of all the accessions were harvested at four different sampling times during the vegetative cycle; at the beginning of the drought treatment (day 0), and 28, 56 and 84 days after initiation of the drought treatment (DAIDT). At each sampling date, three plants per accession were harvested in both the irrigated and drought plots. The aerial and underground parts of each plant were independently collected. The underground plant parts were obtained from a volume of soil of 0.064 m³ (0.4 m × 0.4 m × 0.4 m) by excavation around the plant to a depth of 40 cm, considering that the majority of the root system in this species is found within this soil volume (Greco and Cavagnaro, 2005). The roots were then separated from the soil by rinsing with water and sieving through a 0.6 mm mesh screen, and dried in an oven at 60°C until constant weight, to obtain their dry matter (DM) content. The aerial part of the plant, henceforth referred to as “foliage”, was subdivided into its components leaf blades (from the ligule to the leaf tip, henceforth referred to as “leaves”), ‘stems+culms’ (henceforth “stems”), and panicles, and oven-dried at 60°C until constant weight, to estimate their respective dry matter contents. Dry matter partitioning to each of the plant parts (i.e., leaves, stems, panicles, and roots) was calculated by dividing the DM of a particular plant part by the total DM of the plant, and expressed as percentage of the total plant biomass. The ratio of ‘foliage DM/root DM’, referred to as ‘foliage/root ratio’ (FRR), was calculated for each plant based on its DM contents in the foliage and roots. Total leaf area per plant was determined, prior to desiccation of the leaf samples in the oven, using a leaf area meter (LICOR, Model LI3000A. USA).

On a weekly basis and until the 84 DAIDT, the number of panicles per plant was recorded; and stomatal conductance, chlorophyll index, and maximum quantum efficiency of photosystem II (PSII) were determined on the fully expanded penultimate leaf of each plant, between 10 am and 2 pm. For each accession and treatment, three plants (i.e., three biological replicates) were measured at each time point. Stomatal conductance (g_s) was measured with a leaf porometer (Decagon Devices, Model SC-1, USA) on the surface of the abaxial side of the leaves, avoiding the midrib. Mean values of three readings per plant, for three plants per accession and treatment, at each time point, were considered to calculate mean g_s values. The chlorophyll index was determined using a chlorophyll meter (Minolta, SPAD 502 Plus, Japan). For each plant, the mean of three readings was obtained, and three plants per accession and treatment were analyzed at each time point. Maximum quantum efficiency of PSII (Fm/Fv) was determined by measurements of chlorophyll a fluorescence using a Plant Efficiency Analyzer (Pocket PEA; Hansatech Instruments, England). Prior to the measurements, the leaves were fully dark-adapted for 30 min to achieve the complete oxidation of the primary electron carriers. Chlorophyll fluorescence induction was prompted by a 3-s pulse of red light (peak wavelength of 627 nm) emitted from a LED lamp filtered by a NIR filter. This pulse was emitted at maximal saturation irradiance of 3500 μmol m⁻² s⁻¹. The basal non-variable chlorophyll fluorescence (F₀) and the maximum fluorescence induction (Fm) were determined, and the variable fluorescence (Fv) was calculated as follows: Fv = Fm - F₀. Then, the maximum quantum efficiency of PSII (Fv/Fm) was estimated according to Maxwell and Johnson (2000).

In order to reflect the degree of change in the plant parameters analyzed due to the drought stress, the data for all the variables were expressed as percentage (%) relative to the values in the respective control treatments, according to the formula (Value_Drought/Value_Control) × 100, unless otherwise stated.

The data were analyzed using analysis of variance (ANOVA) by mixed linear models with a factorial structure, treating drought treatment, accession, and their interactions, as fixed effects, while biological replicates were treated as random effects. Different structures of residual variance were evaluated, and the best models were selected using the Akaike (AIC) and Schwarz (BIC) information criteria (Di Rienzo et al., 2017). Prior to the analysis, percentage values were transformed by the square root of the bow-sine function. All the statistical and graphic analyses were performed with the software InfoStat version 2020 (Di Rienzo et al., 2020) and R^® version 3.5.3 (R Core Development Team). Means comparisons were performed using the DGC test (Di Rienzo et al., 2002). For all the variables, the data were expressed as mean value ± standard deviation and p-values ​​< 0.05 were considered significant. Principal component analysis (PCA) was implemented in the InfoStat software to classify the accessions based on drought-response variables using a data matrix of 21 × 9, where the rows represent the 21 T. crinita accessions and the columns comprised the data for nine morpho-physiological traits.

Total plant biomass (i.e., DM of the four plant parts analyzed combined), expressed as percentage relative to the total DM content in control plants, henceforth referred as “RTDM”, was influenced by the accession, year of cultivation, sampling time, and their interactions ( Table 2 ). In 2018, ample and significant differences (p<0.001) were found among T. crinita accessions and among sampling times (i.e., 0, 28, 56 and 84 DAIDT) for RTDM ( Table 3 ). As expected, before initiation of the drought treatment (0 DAIDT), all the accessions had values close to 100% and they were statistically comparable, indicating no substantial differences between the two treatment plots due to factors unrelated to the drought treatment. After the drought treatment was imposed, all the accessions were significantly and differentially affected by the drought stress at the second sampling date (28 DAIDT), as evidenced by the broad variation found for RTDM values, ranging from 37.4% (in acc. 5) to 87.6% (acc. 21). As the drought treatment progressed, RTDM values were further reduced, with mean values at the end of the drought treatment in the range of 21.6% (acc. 5) to 83.3% (acc. 3). For the last two sampling dates, eight statistically different groups of accessions were revealed for this trait ( Table 3 ). Together, the nearly four-fold difference in RTDM between the least and the most affected accessions, and the large number of statistically different groups found for this trait, suggest broad genetic variation for drought tolerance in the T. crinita germplasm.

In general, RTDM was most-severely reduced in accessions 5, 8, 12, and 18, showing significantly lowest mean RTDM values at 56, and 84 DAIDT. In contrast, accessions 3, 1, 23, 9, 11, 21, 24, 14, and 17 exhibited the greatest mean RTDM values for these sampling times ( Table 3 ). Although RTDM was significantly reduced in all the accessions at the first sampling time after the drought stress was imposed (i.e., 28 DAIDT), further variations for this trait differed, in rate and direction, for the different T. crinita accessions, as the drought treatment progressed. Thus, eleven accessions evidenced additional significant decreases in RTDM levels from 28 to 84 DAIDT, seven accessions did not vary significantly, and three accessions (accs. 1, 3, and 4) revealed significant increases in this time-frame ( Table 3 ).

Based on these RTDM data, and results from the other drought-response variables analyzed in 2018 (described in sections below), the six most contrasting accessions with regard to drought tolerance were selected, considering accessions 1, 3, and 9 as the most drought-tolerant; and accessions 5, 18, and 22 as drought-sensitive. Accession 22, although its performance for RTDM was rather intermediate ( Table 3 ), generally ranked among the most sensitive accessions for the other variables analyzed, along with accessions 5 and 18, therefore justifying its inclusion in the drought-sensitive group. These six selected accessions were re-evaluated in 2019, using the same methods and experimental procedures as in 2018.

Results for RTDM content in 2019 confirmed the data from the previous season. Thus, broad and significant variation (p<0.001) was found for this trait among the accessions, with accessions 1, 3, and 9 (considered as drought-tolerant) presenting significantly greater mean RTDM values than accessions 5, 18, and 22 (considered sensitive) for the last two time points of the drought treatment ( Table 3 ). For these sampling times, and coincidently with results from 2018 using the complete set of 21 accessions, accessions 5 and 3 exhibited the lowest (19.5%) and greatest mean RTDM (90.1%) values, respectively.

The foliage/root ratio of T. crinita plants under drought stress, relative to this ratio in irrigated controls, referred to as ‘relative foliage/root ratio’ (RFRR), reflects whether the plant allocates more resources to the roots (RFRR > 100%) or the foliage (RFRR < 100%) under drought conditions, in comparison to controls. As depicted in Figure 2A , in 2018, highly variable and divergent responses were observed early in the drought treatment (28 DAIDT) among the accessions, with mean RFRR values varying from ~49% (indicating greater biomass allocation in the roots) to 168% (greater allocation in the foliage). In general, after this initial response, most of the accessions leveled-off or slightly inverted their trends as the drought stress continued but, in most cases, their RFRR values at 84 DAIDT did not vary much from the initial response at 28 DAIDT (i.e., the accessions that initially allocated more biomass to the roots finished the drought treatment with less than 100% RFRR, and the opposite occurred with accessions that initially allocated more biomass to the foliage). A group of six accessions (14, 6, 10, 11, 17, and 21) presented statistically greater RFRR values than the rest of the accessions at 84 DAIDT, and three of them (11, 21, 14) also ranked among the plant materials with greatest relative foliage biomass ( Figure 3 ). Figure 2B presents RFRR data for year 2019 in the six selected contrasting accession. Accession 5 consistently allocated the greatest proportion of biomass to its roots, as indicated by the significantly lowest RFRR values found in this material for both years and all the time-points. On the drought-tolerant extreme, in 2019 accession 1 revealed the greatest mean RFRR value at the end of the experiment, but without reaching statistical differentiation with the rest of the accessions.

Figure 3 depicts time-course variation for relative biomass content in the foliage and roots of all the accessions grown under drought conditions, and expressed as percentage of the DM content in the foliage (RFDM) and roots (RRDM) of control plants. As expected, at 0 DAIDT, all the accessions had RFDM and RRDM values close to 100%, being not statistically different from each other (data not presented). After the drought stress was imposed, significant variation among the accessions was found for RFDM and RRDM at all the time points analyzed, with mean values for both traits generally decreasing in a genotype-dependent fashion as the drought conditions progressed. Thus, after 56 and 84 DAIDT, mean RFDM, which includes all the above-ground organs (stems, leaves, and panicles), varied from 19.6% to 87.4% across the accessions, whereas RRDM ranged from 26.2% to 89.7%. These data indicate broad genetic variation in the T. crinita germplasm with regard to the accessions ability to maintain root and foliage growth under drought stress.

For RFDM, accession 5 was the most affected plant material, and this was evidenced early during the drought stress (at 28 DAIDT), followed by accessions 8, 18, 12, and 4. Conversely, accessions 11, 3, and 1 were the least affected by drought at 56 DAIDT, with accessions 3 and 11 being also the least affected at 84 DAIDT. While most of the accessions either decreased or maintained statistically unaffected their mean RFDM values as the drought progressed, two accessions, namely accs. 3 and 9, significantly increased their RFDM values at 56 and 84 DAIDT compared to 28 DAIDT.

Variation for RRDM partially coincided with RFDM, as shown in Figure 3 , and as indicated by the low to moderate -yet significant- correlation coefficient (r) values obtained between these two variables at 56 DAIDT (r=0.30, p=0.0178) and 84 DAIDT (r=0.55, p<0.0001), suggesting some level of independence between the two traits. Under prolonged drought conditions (56-84 DAIDT), mean RRDM was significantly lowest in accession 5; and greatest in accession 1 (at 84 DAIDT) and accessions 9, 13, 4, 24, and 23 (at 56 DAIDT). The majority of the accessions revealed significant decreases in –or maintained statistically unaffected- their mean RRDM values as the drought treatment progressed, with the exception of accessions 1 and 3, which increased their RRDM values at 84 DAIDT.

Results for the six contrasting accessions reevaluated in 2019 fully agreed with data from the previous season, clearly separating the drought-tolerant from the sensitive accessions ( Supplementary Figure S3 ). Briefly, broad and significant variation (p<0.001) was found among the accessions for both RFDM and RRDM, with the range of variation increasing as the drought treatment progressed For the last two sampling times, all the tolerant accessions (accs. 1, 3, and 9) had significantly greater RFDM and RRDM values than the sensitive accessions (accs. 5, 18, and 22). At these time points, accessions 3 and 1 were the least affected, and accession 5 the most affected, for both variables.

Further analysis investigated DM partitioning to individual organs of the plant under irrigated and drought conditions. ANOVA for the percentage of the total DM allocated in roots, stems, leaves, and panicles revealed significant effects for the accession, the drought/irrigation treatment, the year of cultivation, the sampling time, and many of their interactions ( Supplementary Tables S1, S2 ). Figure 4 depicts the variation found among the accessions for photoassimilates partitioning to these organs under irrigated and drought conditions. Nine accessions (accs. 6, 7, 10, 11, 14, 17, 19, 20, 22) consistently revealed significant decreases in DM partitioning to the roots for all the time points considered after the drought stress was imposed (i.e., 28, 56, and 84 DAIDT). Conversely, three other accessions (accs. 4, 5, and 13) consistently exhibited significantly greater biomass partitioning to the roots during the entire drought treatment. For the remaining eight accessions, no consistent variation was observed in photoassimilates partitioning to the roots. Overall, under drought conditions, biomass partitioning to the roots was greatest in accession 5, presenting 25-32% of the total plant DM allocated in its roots, whereas accession 21 had the lowest partitioning to this organ, accounting for 14-18% of the total DM.

Partitioning to individual above-ground organs of the plant varied significantly among the accessions, and between the drought and irrigated treatments within each accession ( Figure 4 ). In some accessions, drought stress was associated with reduced biomass partitioning to the leaves while increasing partitioning to the stems and panicles. This was more evident at the end of the drought treatment (84 DAIDT), when ten accessions exhibited significantly greater DM allocation in the stems and panicles, and nine accessions had reduced DM in leaves. Some exceptions to this trend were accessions 4, 5, and 18, which exhibited significantly lower partitioning to the stems consistently during all the drought-stress period (i.e., at 28, 56, and 84 DAIDT), and accession 7, showing the same performance for the last two time points. For the rest of the accessions, no consistent variations in DM allocation to specific above-ground organs were observed during the drought treatment.

Considering the six selected accessions, no direct association was found between the patterns of photoassimilates partitioning to different plant organs under drought conditions with the two sub-classes of accessions (i.e., tolerant vs. sensitive). For example, within the drought-sensitive group, accession 5 increased partitioning to the roots and decreased DM in the leaves, whereas accession 22 decreased DM allocation in the roots while increasing its content in panicles, and their performances were consistent throughout the drought treatment ( Figure 4 ). In the ‘tolerant’ subgroup, accession 1 consistently revealed reduced DM in leaves while it was increased in stems, whereas no clear trend was observed for accessions 3 and 9. In 2019, data for biomass partitioning to different plant parts generally agreed with results from the previous year ( Supplementary Figure S4 ). Again, no clear association was found between DM partitioning patterns under drought conditions with the two subgroups of accessions. Coincidently with results from 2018, under drought stress, accession 5 consistently showed increased DM allocation in the roots and decreased DM in the stems, relative to its control, and presented the greatest percentage of DM allocation in the roots for all the accessions,

ANOVA for relative total leaf area (RLA) revealed significant effects for accession and sampling time (but not for year), and for all the interactions among accession, sampling time, and year ( Table 2 ). Figure 5 depicts time-course variation for RLA for both years of experiments. In 2018, the drought treatment significantly reduced RLA in all of the accessions, as evidenced by the sudden and significant drop in mean RLA values from 28 DAIDT to the end of the experiment ( Figure 5A ). For most of the accessions, such decay in RLA further progressed during the drought treatment, reaching minimum values at 84 DAIDT. However, for a few accessions (namely, accs. 1, 3, 7, 9, 17, and 20), mean RLA values rapidly dropped at the beginning of the drought stress (28 DAIDT), along with most other accessions, but then they remained statistically invariable throughout the rest of the drought treatment. As result, these accessions had the greatest RLA at the end of the experiment (84 DAIDT), together with accession 23. Similar results were obtained in 2019 for the subset of six selected accessions, with the ‘tolerant’ accessions 1, 3, and 9 exhibiting significantly greater RLA than the ‘sensitive’ accessions 5, 18, and 22 at the end of the drought treatment ( Figure 5B ).

In order to integrate and compare the relative leaf area of the accessions during the entire drought treatment, the RLA data for the four time points analyzed were plotted in a graph (the one shown in Figure 5 ) and the ‘area under the curve’ (AUC) was calculated for each accession. Comparisons of the resulting mean AUCs revealed significant variation among the accessions (p<0.05), with accessions 21 and 6 showing the greatest AUC, and accessions 5, 12, and 4 the lowest AUC ( Supplementary Figure S5A ). In 2019, accession 9 had the greatest mean AUC and accession 5 the lowest, whereas the rest of the accessions (accs. 1, 3, 18, 22) were intermediate and statistically comparable ( Supplementary Figure S5B ).

Mean leaf chlorophyll index (CI) in T. crinita plants grown under drought stress, relative to the CI of irrigated controls, referred to as ‘relative CI’ (RCI), varied significantly among the accessions and among time-points during the drought treatment ( Figure 6 ). In 2018, little variation was observed among the accessions during the first 42 DAIDT, with most of the genotypes presenting CI values statistically comparable to their irrigated controls (data not presented), but as the drought stress progressed, the CI values of all but one of the accessions dropped, at varying rates ( Figure 6A ). A subgroup of nine accessions (accs. 5, 6, 7, 8, 18, 19, 20, 21 and 22) revealed the most rapid decay in mean RCI values, exhibiting these accessions the lowest RCI levels at the end of the drought treatment. The latter subgroup included accessions 5, 18, and 22, considered as drought-sensitive. These three materials, and accession 5 in particular, exhibited the lowest mean RCI values of all at 84 DAIDT. Conversely, accessions 9, 1, and 3 showed more gradual decays in RCI levels during the drought treatment, ending these accessions with the greatest RCI values of all at 84 DAIDT. Considering the entire germplasm collection, accession 9 was the least affected of all, exhibiting RCI values statistically comparable to the irrigated control until 63 DAIDT (data not shown), and then gradually decreased to end up with the greatest mean RCI value at 84 DAIDT, along with accessions 1 and 3. The rest of the accessions exhibited intermediate performances between these two sets of contrasting materials. Supplementary Figure S6A presents mean RCI values integrated throughout the drought treatment, expressed as AUC, for all the accessions in 2018. Such integrated analysis revealed eight significantly different groups, with accessions 5 and 9 being the plant materials with lowest and greatest mean AUC, respectively, in full agreement with their time-course performances shown in Figure 5A .

In the 2019 experiment, RCI data for the six selected accessions varied following a similar pattern as in 2018, with the drought-tolerant accessions (1, 3, and 9) presenting a delayed and more gradual decay of their mean RCI values, as compared to the sensitive accessions (5, 18 y 22), resulting the formers in significantly greater RCI values at the end of the experiment ( Figure 6B ). Consequently, the AUC for this variable was also significantly greater in all the tolerant accessions as compared to the sensitive materials, with accessions 5 and 9 representing the most contrasting extremes ( Supplementary Figure S6B ).

The photochemical efficiency (PE) of photosystem II (estimated by Fv/Fm ratio) in leaves of T. crinita plants grown under drought stress, relative to the PE of irrigated controls, referred to as ‘relative PE’ (RPE), was significantly influenced by the accession, sampling time, year of cultivation, and most of their interactions ( Table 2 ). Figure 7 depicts the time-course variation for RPE in all the accessions during the drought treatment. In 2018, little variation was observed among the accessions during the first 35 DAIDT, with most of the accessions presenting PE values statistically comparable to their irrigated controls (i.e., mean RPE values were close to 100%), but as the drought stress progressed, RPE values of all the accessions began to decrease, at varying rates, with the most abrupt decay observed between 56 and 63 DAIDT for the accessions exhibiting the fastest decline in RPE levels ( Figure 7A ). From 56 DAIDT to the end of the experiment, the drought-stress response of the accessions became evidently and significantly different, varying from genotypes that maintained high levels of RPE (>70%) until the end of the drought treatment (e.g., accs. 3, 1, and 9) to genotypes that rapidly decreased their mean RPE, reaching minimum levels at 84 DAIDT close to 0% (e.g., accs. 5, 6, 18, 19, 21, 22, and 24). Supplementary Figure S7A presents integrated mean RPE values throughout the drought treatment and expressed as AUC for all the accessions. As depicted, accession 5 had significantly lowest mean AUC, whereas a group of six accessions (accs. 3, 1, 9, 11, 14, and 13) had the greatest AUC values.

In 2019, RPE data for the six selected accessions varied following a similar pattern as in 2018, with the drought-tolerant accessions (1, 3, and 9) presenting a delayed and more gradual decay of their mean RPE values, as compared to the sensitive accessions (5, 18, and 22), which exhibited an abrupt decay in RPE between 56 and 70 DAIDT, to reach minimums of less than 2% at the end of the experiment ( Figure 7B ). Coincidently with these results, the AUC for RPE in all the tolerant accessions was significantly greater than in the sensitive accessions ( Supplementary Figure S7B ). Altogether, the RPE data from both years suggest accessions 3, 1, and 9 as the most drought-tolerant genotypes, and accession 5 as the most sensitive one.

Mean stomatal conductance (g_s) in T. crinita plants grown under drought stress, relative to the g_s of the irrigated control plants, referred to as ‘relative g_s’ (Rg_s), was significantly influenced by the accession, sampling time, year of cultivation, and most of their interactions ( Table 2 ). In 2018, the effect of the drought stress on Rg_s was evidenced very early in the experiment, showing significant decreases -to varying extents depending on the genotype- for all 21 accessions in the first week of treatment ( Figure 8A ). In this short period, mean Rg_s values dropped from ~100%, before initiation of the drought treatment, to 24% in acc. 23, and up to 68% in acc. 22. In decreasing order, accessions 23, 24, 3, 1, 7, 11, 14, 9, and 10 were the most affected at this time point, exhibiting absolute stomatal conductance values in the range of 10.4-15.2 mmol m² s^-1 (data not presented). However, after this sudden drop in mean Rg_s values, these accessions stabilized and –in some cases- even increased their Rg_s throughout the rest of the drought treatment, exhibiting at the end of the experiment significantly greater stomatal conductance than most of the other accessions. In contrast, the accessions that had a mild decrease in Rg_s at the beginning of the drought stress (7 DAIDT) (e.g., accs. 5, 18, 22, 20, and 24, among others), continued their decreasing trend throughout the rest of the experiment, reaching the lowest Rg_s values of all the accessions, with means in the range of 12.3-16.8% (6.4-9.2 mmol m² s^-1, in absolute g_s values), at the end of the drought period. Supplementary Figure S8A presents mean Rg_s values integrated throughout the drought treatment and expressed as AUC for all the accessions, showing accessions 3 and 1 with significantly greatest AUC, whereas eight accessions (accs. 4, 12, 13, 19, 17, 23, 5, and 20) comprised the lowest AUC group.

In 2019, Rg_s data for the six selected accessions varied following a similar pattern as in 2018. Thus, the drought-tolerant accessions 1, 3, and 9 exhibited a sudden and more profound decrease in mean Rg_s than the drought-sensitive genotypes (5, 18, and 22) at 7 DAIDT but, as observed in the previous year, the former group stabilized and –for accessions 1 and 3- increased their mean Rg_s levels, ending up with significantly greater Rg_s at 84 DAIDT ( Figure 8B ). Conversely, the drought-sensitive accessions continued to decrease their Rg_s during the rest of the drought treatment, reaching significantly lowest Rg_s values at the end of the experiment. Interestingly, accession 9 behaved as an intermediate material, being its mean Rg_s at the end of the drought period statistically lower than the Rg_s of accessions 1 and 3, and greater than that of the sensitive genotypes. Coincidently with these results, the AUC for Rgs in 2019 was significantly greatest in accessions 1 and 3, and lowest in accession 5, with the rest of the accessions being intermediate relative to the formers ( Supplementary Figure S8B ).

The number of panicles per plant (NPP) in T. crinita accessions grown under drought stress, relative to the NPP in their respective controls, referred to as ‘relative NPP’ (RNPP), was significantly influenced by the accession, the year of cultivation, the sampling time, and their interactions ( Table 2 ). In 2018, an early response was observed for some of the accessions, either increasing (accs. 4, 7, 10, and 23) or decreasing their RNPP (e.g., accs. 5, 18, 19, 9, and 6) in the first 28-35 DAIDT ( Figure 9A ). All but one of these accessions stabilized their RNPP values as the drought conditions progressed, to end up with mean RNPP values in the range of 58-98% at 84 DAIDT. Accessions 6 and 10 were the only ones that maintained high RNPP values until the end of the trial, showing no statistical differences with their basal values at 0 DAIDT (i.e., at 84 DAIDT, their RNPP values were ~100%). Accession 5 was the most affected of all, exhibiting an abrupt decay in RNPP early during the drought treatment, which then slowly decreased until the end of the experiment to a final value of 19% RNPP. In 2019, RNPP data generally coincided with results from the previous year, with accession 5 revealing a very similar variation pattern as in 2018, atypical from the rest of the accessions, consistently presenting the lowest mean RNPP values throughout the entire drought treatment ( Figure 8B ). The rest of the accessions followed a similar variation pattern as in the previous year. Although time-course monitoring of this variable did not reveal an early separation of tolerant versus sensitive accessions, as observed for other variables, at the end of the drought treatment all the tolerant accessions had significantly greater RNPP values than the sensitive ones, with accessions 3 and 9 being the greater relative number of panicles per plant. Additional comparative analysis of RNPP, expressed as AUC, for both years of data, is presented in Supplementary Figure S9 .

Pairwise correlation coefficient values among all the variables at the end of the drought treatment, for both years, are presented in Table 4 . In 2018, relative total (RTDM) and foliage biomass (RFDM), two of the most relevant variables reflecting forage yield under drought stress, were strongly correlated with each other (r=0.98, p<0.0001), and both traits were significantly and positively correlated with all the other variables (r=0.34-0.74, p<0.01). This suggest that, under prolonged drought conditions, the accessions with greater forage biomass (relative to irrigated controls) tend to have greater relative levels of total biomass production per plant, root biomass, foliage/root ratio, leaf area, chlorophyll content, stomatal conductance, photosynthetic performance, and inflorescences per plant. In 2019, all the variables were significantly and more-strongly correlated (r=0.55-1.00) than in the previous year, most likely reflecting a sampling bias due to the selection of the six most contrasting accessions used in the analysis. Nonetheless, the majority of the correlations found coincided -in significance and sign- between years, despite the observed differences in the strength of the associations.

Principal component analyses (PCA) with nine variables were conducted, independently, with data from years 2018 and 2019 ( Figure 10 ). In the first year, two principal components (PC) explained, jointly, 74.9% of the total variation, with PC1 accounting for 53.4% of the variation. The variables that contributed most to PC1 were, in decreasing order, RTDM, RFDM, RRDM, RFRR, RLA, RCI, RPE, RGs, and RNPP. A group of nine accessions, located in the right half of the bi-plot, were the most representative of these variables, with accessions 3, 1 and 9 showing the strongest association, whereas most of the remaining accessions were, conversely, located in the left half of the bi-plot, with accession 5 being the one with strongest negative association with the variables. In the second year, considering only the selected six most contrasting accessions, the two main components accounted for 92.2% of the total variation, with PC1 explaining 86.6% and PC2 5.6%. In general, the same variables that were associated with PC1 in 2018, where also associated with this component in 2019. The drought tolerant and sensitive accessions were located on the left and right side of the bi-plot, respectively, with accessions 3 and 5 representing the most contrasting plant materials, as found in the previous year.