The plant material consisted of 45 F₁ hybrids developed using a diallel mating scheme between 10 S₇ popcorn lines. The lines were selected from a set of 20 lines that had previously been evaluated in the field under WS. Based on the performance of the 10 lines, four were classified as agronomically efficient in terms of their water use (P2, P3, P6, and P7), with four classified as inefficient (L61, L63, L65, and L75) and two classified as intermediate (L71 and L76) (Kamphorst et al., 2018).

The genealogy of popcorn inbred lines (S7) was derived from germplasm adapted to tropical (L61, L63, L65, and L71 from the BRS-Angela population) and temperate/tropical conditions (P7, from the commercial hybrid Zélia; P2 and P3, from the compound CMS-42; P6, from the commercial hybrid IAC-112; and L54, L55, L75, and L76, from the Barão de Viçosa population) (Vittorazzi et al., 2018).

To generate the hybrids, the lines were crossed in pairs, in 6.00 m rows spaced 1.00 m apart, with the plants spaced 0.40 m apart. The ears in each line were covered with plastic bags prior to stigma exertion. The mature tassels were covered with a brown paper bag on the day preceding the crosses to avoid contamination of the crosses with undesired pollen. Approximately 65 d after planting, the plants were hand pollinated. The hybrid seeds were harvested in April 2019, at the Experimental Station of Colégio Estadual Agrícola Antônio Sarlo, in Campos dos Goytacazes, Rio de Janeiro State, Brazil (lat. 21° 42”48” S; long. 41° 20” 38” W, 14 m asl).

To evaluate the early hybrid development under each water regime, two experiments were conducted in a complete randomized block design with four replications of 20 seeds per replication. The experiments were conducted at the Crop Science Laboratory of the State University of Northern Rio de Janeiro Darcy Ribeiro (LFIT/UENF). The seeds from the 45 hybrids were germinated under two water regimes (WC). These were: i) adequate substrate moisture (well-watered: WW), where the substrate (germitest paper) was wettened to a moisture content of 70% of the water retention capacity; and ii) water stress (WS), where the substrate was wettened to a moisture content of 25% of the water retention capacity. To adjust the moisture content of each WC, the dry paper was weighed and multiplied by three, which was equivalent to the weight of the water-saturated paper. Thereafter, the moisture was adjusted to 70% and 25% of the substrate retention capacity ( Supplementery Figure 1 ).

Once the water content of the germitest paper had been adjusted, the seeds were arranged in four rows of five seeds, that is, 20 seeds per hybrid and replication. Each germitest paper was rolled up into a cylindrical tube and packed in two plastic bags with openings in opposite directions. This prevented water from escaping into the external environment. The paper rolls were placed in germinators, with alternating temperatures (20–30 °C) and an 8:16 h light–dark photoperiod (MAPA-Brasil, 2009). Based on the initial weight of each replication, which comprised the weight of water, seeds, and plastic cover bags, water was replenished daily to maintain the desired moisture level for each WC.

The traits evaluated were divided into three groups related to germination and the shoot and root systems. In each group of germination traits, germination on the seventh day (GER) and the germination speed index (GSI) were evaluated. Germination on the seventh day was evaluated as the percentage (%) of normal seedlings that emerged on the seventh day. On that day, the number of abnormal seedlings and non-germinated seeds were also counted. Seeds with no sign of germination were considered dead. Those with malformed structures and aspects of abnormalities were considered anomalous. Seedlings with strong structure formation, without any deformities and abnormalities, were recorded as normal (MAPA-Brasil, 2009). The GSI was based on the count of the number of seedlings that emerged daily until the stand stabilized. This occurred on the seventh day after sowing, with radicle emergence being considered germination. The GSI (seedlings day⁻¹) was computed using the equation proposed by Maguire (1962), by counting the number of seeds with a radicle length of 0.50 cm every two days.

The shoot traits of the seedlings evaluated were the shoot length (SL) and the shoot dry weight (SDW). The shoot length (SL) (in cm seedling⁻¹) was measured with a millimeter ruler, considering the distance from the seedling mesocotyl to the tip of the primary leaves (Da Silva et al., 2016). Shoot dry weight (SDW) was measured as the mean weight of all seedlings, in milligrams per normal seedling (mg seedling⁻¹). For these evaluations, 10 seedlings that had been oven-dried at 70°C for 48 h were used.

Root-related traits were evaluated on the same 10 seedlings, namely the root system length (RSL), root dry weight (RDW), and root-to-shoot ratio (RSR). The length of the root system was measured from the collar to the tip of the largest primary root (in cm plant⁻¹) (Da Silva et al., 2016). The root material from all replicates was then spread out and photographed (resolution 3.088 × 3.088 pixels) in a water-filled acrylic tray. This procedure reduced root overlap. The root photographs were analyzed with GiARoots (Galkovskyi et al., 2012). The software was calibrated to 93 pixels cm⁻¹ and used to estimate the maximum root number (MRN), root network area (RNA, in cm²), specific root length (SRL, in cm/cm³), root network length (RNL, in cm), and root network volume (RNV, in cm³).

After being photographed, the RDW was estimated by weighing the material (in mg seedling⁻¹) after being oven dried at 70°C for 48 h. Based on these values, divided by SDW, the RSR estimates were calculated using the equation: R R A = P S R P S A .

Individual analysis of variance was performed for each WC according to the statistical model Y i j = μ + g i + b j + ϵ i j , where: Y_ij = observed value of the i-th genotype in the j-th block; μ = general constant; g_i = fixed effect attributed to the i-th genotype; b_j = random effect of block j; and ϵ ij random error associated with observation Y_ij. The experimental variation coefficient was calculated as: C V e = R M S X ¯ ,   where RMS corresponds to the residual mean square and X ¯ to the mean. Then, a combined analysis of variance was performed between the WCs, according to the statistical model Y i j k = μ + G i + B / W C j k + W C j + G W C i j + ϵ i j k , where   Y i j k : Y_ijk = observation of the i-th genotype in the j-th WC and k-th block; μ = general constant; G_i = fixed effect of the i-th genotype; B / W C j k = effect of the k-th block within WC_j ; WC_j = fixed effect of the j-th WC; GWC_ij = fixed effect of the interaction between the i-th genotype and the j-th WC; and ϵ i j k = mean experimental random error associated with observation Y i j k

The proportional reduction (%) of each variable, based on the comparison between WCs, was calculated using the formula 100 − [ ( Y W S / Y W W ) * 100 ] , where: Y = overall mean of the variable, under WS and WW conditions.

For each WC, diallel analysis was performed separately, according to Griffing’s Method IV, Model 1 (Griffing, 1956), according to the expression Y i j = μ + g i + g j + s i j + ϵ i j where: Y_ij = mean value of the hybrid combination (i ≠ j) or the parent (i = j); μ = overall mean; g_i = effects of general combining ability of the i-th parent (i, j= 1, 2,…, 10); g_j = effects of general combining ability of the j-th parent (i, j= 1, 2,…, 10); s_ij = effect of specific combining ability for crosses between the i-th and j-th parent; and ϵ i j mean experimental error associated with the observation of order ij.

The quadratic components (ϕ) that express the genetic variability, in terms of general (g) and specific combining ability (s), were estimated using: ϕ_g = (GMS – RMS)/2p and ϕ_s = SMS – RMS, respectively. Here, GMS is the mean square of the general combining ability, SMS is the mean square of the specific combining ability, RMS is the residual mean square, and p is the number of parents. The effects of quadratic components were expressed as a percentage of the total sum of effects.

Variance and diallel analysis were performed using the GENES program (Cruz, 2013). The graphs were constructed using the G-Biplot Package, R Studio (R Core Team, 2018). The “which won where/what” multivariate analysis was performed based on the biplot model genotype per trait (GT), using the standardized values of all the traits evaluated (GSI, GER, SL, SDW, RSL, RDW, RSR, MRN, RNA, SRL, RNL, and RNV). To generate the GT biplot graph, the R software package GGEbiplotGUI (R Core Team, 2018) was used.

Germination traits that differed significantly under WS were GSI and GER. Under WW, the GER trait significantly differed. The experimental coefficients of variation (CV_e) ranged from 6.73% (GER under WW) to 38.65% (GER under WS). In the combined analysis, the germination traits differed statistically only for the genotype source of variation (G) ( Table 1 ). In a comparison of the WS and WW water regimes, GSI tended to decrease by 15.34% under WS, while GER increased by 0.32% under the same condition ( Figure 1 ).

Of the shoot traits, only SDW under WW differed significantly ( Table 2 ). Regarding CVs, there was a variation from 54.21% (SDW under WS) to 12.19% (SDW under WW) ( Table 1 ). The combined analysis showed significant differences for the source of variation G and water condition (WC) for the traits SL and SDW and for the genotype × water regime interaction (G*WC) for SDW ( Table 1 ). Comparing the water conditions (WCs), under WS, PSA was reduced by 61.75% and CPA by 68.06%, in relation to the WW condition ( Figure 1 ).

The root-related traits with significant differences were RDW, SRL, and RNV under both WCs and RSR, MRN, RNA, and RNL under WW conditions only. RSL and RSR differed significantly for the three sources of variation (G, WC, and G*WC) and the traits RDW, MRN, RNA, and SRL only for the source of variation G. Meanwhile, RNL and RNV showed significance for sources of variation G and WC. Comparing WS to the WW condition, there was a decrease in CRS of 15.09%, in RNA of 16.99%, in RNL of 17.60%, and in RNV of 19.16% ( Figure 1 ). However, RNA increased by 164.54% ( Figure 1 ).

Germination traits with significant differences for GCA estimates were GSI under WS and GER under WW conditions ( Table 2 ). Significance was observed for SCA in GSI and GER under WS and GER under WW conditions ( Table 2 ). It was found that the quadratic components associated with SCA (ϕ_s) were the most important in explaining the observed variability in GSI under WS and in GER under WW conditions ( Figure 2 ). Given the high percentages of residues, a strong environmental influence was detected for GSI, with a magnitude of 77% ( Table 2 ).

For the shoot trait SDW under WW, significant differences for GCA estimates and significance for SCA were observed ( Table 2 ). The quadratic components associated with SCA (ϕ_s) explained most of the variability observed in SL under WS ( Figure 2 ). As a result of the high residual percentages, a strong environmental influence on SL and SDW was observed with estimates of 94% and 89%, respectively ( Table 2 ).

The root-related traits that differed significantly in GCA estimates were RDW, RSR, RNA, SRL, RNL, and RNV under both WCs ( Table 2 ). Under both WCs, significant SCA effects were observed for RDW and RNA, SRL, RNL, and RNV under WW conditions ( Table 2 ). The quadratic components associated with SCA (ϕ_s) explained most of the observed variability in RDW, MRN, and SRL under both WCs ( Figure 2 ). As a result of the strong estimates of experimental residues, consistent environmental influences on CRS and RNA were detected, as shown by a CV_e of 79% and 83%, respectively ( Table 2 ).

For GSI, under both WCs, the parents with positive GCA estimates were L63 (0.08 under WS and 4.21 under WW), L65 (0.42 under WS and 0.98 under WW), P3 (3.45 under WS and 2.41 under WW), and P7 (2.86 under WS and 0.55 under WW). For GER, the parents L63 (1.60 under WS and 0.81 under WW), L65 (5.07 under WS and 6.22 under WW), P2 (4.44 under WS and 2.68 under WW), P3 (5.90 under WS and 7.68 under WW), and P7 (6.94 under WS and 7.06 under WW) had positive GCA effects ( Figure 3 ).

For shoot-related traits, the parents with positive GCA estimates under both WCs were L61 (0.66 under WS and 1.29 under WW) and L63 (0.23 under WS and 0.97 under WW) for SL. For SDW, estimates were positive for L65 (0.01 under WS and 0.04 under WW), P2 (0.0003 under WS and 0.0183 under WW), and P3 (0.0073 under WS and 0.0233 under WW) ( Figure 3 ).

Regarding the root-related traits, the parents with positive GCA effects under both WCs were P2, P3, and P6 for CRS; L76, P2, P3, and P7 for RDW; and L76, P2, P3, and P7 for RSR. For MRN, positive estimates were predicted for P2, P3, and P7. For RNA, the same situation was detected for P2, P3, and P7. In SRL, positive GCA values were found for L61, L71, L75, and P6. For RNL, the parents L76, P2, P3, and P7 had positive GCA estimates and, equally, P2, P3, and P7 in relation to RNV ( Figure 3 ).

For germination traits, under both WCs, the hybrids with simultaneously highest SCA estimates and means (among the 15 highest values), were L75 × P6 and L63 × P3 for GSI and L61 × P6, L61 × P7, and L75 × P6 for GER ( Figures 4 , 5 ).

For shoot traits and considering both WCs together, the hybrids with more marked SCA estimates and means among the 15 highest values were L65 × L71 and L71 × P3 for SL and L61 × P2, L71 × P3, and L71 × P7 for SDW ( Figures 4 , 5 ).

When evaluating the root traits together, concomitantly under both WCs, the hybrids with the highest SCA and mean values among the 15 highest values were L63 × P3 and L65 × P6 for CRS; L61 × P2, L71 × P7, and L76 × P3 for RDW; L61 × P7, L63 × P2, and L75 × P3 for RSR; L71 × P3, L71 × P7, L75 × P7, P2 × P6, and P6 × P7 for MRN; L61 × P2, L71 × P3, L76 × P3, and P2 × P6 for RNA; L61 × L75, L63 × L65, L63 × L75, L65 × L76, and L71 × P6 for SRL; L61 × P2, L63 × P7, L71 × P3, and P2 × P6 for RNL; and L61 × P2, L71 × P3, L71 × P7, and P2 × P6 for RNV ( Figures 4 , 5 ).

The main principal components had values exceeding 70%, regardless of the WC ( Figure 6 ). Under WS ( Figure 6A ), the sum of the values of the first two principal components was 77.42% and 76.19% under WW ( Figure 6B ).

Under WS, five groups were formed ( Figure 6A ). The first only included RSR, for which the estimate of hybrid P3 × P7 (44) stood out. The second group, with high estimates for hybrid L71 × P7 (30), consisted of trait SRL. In the third group, with trait SL, the estimate of hybrid L61 × L75 (4) was the most appropriate. In the fourth group, with SDW, the estimates of hybrids L71 × P3 (28) and L76 × P6 (38) were the highest. In the fifth group, which comprised GER, GSI, RDW, RNA, RNV, MRN, RNL, and RSL, the hybrid estimates were highest for L61 × P2 (6) and P2 × P6 (41).

Under WW conditions, five groups formed ( Figure 6B ). The first comprised RSL, RSR, RDW, RNA, RNV, and RNL, with appropriate estimates for the hybrids L76 × P3 (37) and P2 × P3 (40). In the second group, with trait SRL, the hybrids L61 × L65 (4), L76 × L75 (12), and L71 × P6 (29) were noteworthy. In the third group, for SL, the estimates of hybrid L65 × P3 (22) stood out. In the fourth group, which included SDW, GER, and GSI, hybrid L71 × P3 (28) performed most successfully. In the fifth group, with MRN, no high estimates were predicted for any hybrid.

For the germination traits GSI and GER, the applied water restriction was insufficient to differentiate WS from the WW condition, although GSI tended to decrease. However, GER was not influenced. In a similar study on maize seedlings, De Abreu et al. (2019) observed no significant differences between water stress and well-irrigated conditions for the same traits. The trend towards a decrease in GSI was because of a greater mobilization of the metabolism to mitigate the adverse effects of drought. This led to a greater energy expenditure and favored adaptation to WS during seed germination if the process occurred under WS (Vaz-de-Melo et al., 2012). In relation to GER, estimates are expected to decrease, because, given factors induced by water limitation, fewer reserves are mobilized, with less synthesis and enzymatic activity, with consequent changes in cell turgor (Bewley and Black, 1994).

For the seedling shoot traits, SDW and SL, the water restriction applied was sufficient to differentiate WS from WW conditions by causing an impact of a more than 60% reduction in these traits. The response in the shoot traits of the seedlings was one of the main causes of low productivity, for directly causing photosynthetic rate decline (Lisar et al., 2012). The observed decrease in SDW and SL was consistent with the earlier description as a response to water stress that led to a decrease in vegetative growth and leaf area, as well as leaf curling and reduced turgor (Pezzopane et al., 2015).

The root traits RSL, RNA, RNL, and RNV were affected by the experimentally imposed water stress, as shown by the significant difference in the source of variation for WC. The root system is related to the plant perception of water deficit in the soil, causing physiological changes in different metabolic pathways (Rahnama et al., 2011). According to Sharp et al. (2004), the lower values of root-related traits were due to root thinning. Thinning is a response that has already been observed in plants exposed to WS and is a morphological modification considered an adaptive response as a result of restriction of the lateral expansion rate of cortex cells (Sharp et al., 2004). However, the increase in RSR indicated a response of the root system to artificial water stress. The plant tends to decrease investments in leaf growth and intensify root development (Ribeiro, 2021). According to Lisar et al. (2012), under WS, the RSR increases because of root elongation, which deepens the root system to take up water and, consequently, to maintain the osmotic pressure. Root elongation was also observed by De Abreu (2016) in an evaluation of the same trait under water stress.

Under water stress, the mode of action of agronomically important and other secondary traits is fundamental for the elaboration of more efficient breeding strategies (Derera et al., 2008). Knowledge about the genetic mode of action is key to successful selection (Hallauer et al., 2010). Germination-, shoot-, and root-related traits of maize seedlings normally have high coefficients of variation, as those observed by De Abreu (2016). The high experimental coefficients influenced the importance of quadratic components, in many cases driving up the estimated residual percentage to higher values than _s and _g. In general, this finding was repeated for traits evaluated in this study, aside from GER and GSI under WS; and for GER, SDW, and RDW under WW. When comparing only the percentage importance of the quadratic components associated with general (_g) and specific (_s) combining ability, the effects of _s explained most of the genetic variability. In these cases, the use of hybrids is recommended, as the beneficial effects of allelic complementation are exploited.

Regarding the estimates of the mean square effects, there was no change between the WCs in the most important quadratic component for the traits evaluated in this study. This shows that the mode of trait expression is the same under both WCs. According to Lima et al. (2019), under WS and WW conditions, the genetic effects of grain yield inheritance and yield components in popcorn did not change between the tested water regimes. These results were corroborated by Kamphorst et al. (2021a). These studies suggest that the same breeding methods could be applied under both WCs, as, for example, are being proposed in this study. The same arguments were supported by Kamphorst et al. (2022) in an evaluation of growth, photosynthesis, transpiration, and root architecture traits of adult popcorn plants under different water regimes. These results are parallel to those reported here, since the genetic effects of dominance were also the most important in the trait expression of popcorn seedling growth.

The parents studied in this research had been selected for higher grain yield under drought during physiological maturation. The parents P2 and P3, previously classified as tolerant, stood out with the most successful performances for traits of germination and seedling development, based on the GCA means. Until then, it was unknown how these studied lines would perform in the early growth phase. The observed performance for grain yield, evaluated at the end of the cycle, was consistent with that observed during seed germination and seedling development. In general, the best lines preformed similarly under both tested water regimes ( Figure 3 ).

Based on the general combining ability under both water regimes, for the germination traits, lines L63, L65, P3, and P7 can be recommended as parents of superior hybrids to mitigate the harmful effects of WS in popcorn. For shoot traits, there were no lines that stood out under both WCs together. For the root-related traits, considered together, lines P2, P3, and P7 are recommended as parents of superior hybrids, with a view to reducing the harmful effects of drought on early popcorn development. Overall, for all groups of traits, lines P2 and P3 are recommended as parents of superior hybrids for water stress tolerance.

From a broader perspective, the hybrids performed similarly under both the water regimes tested ( Figures 4 , 5 ). From evaluation of the combined shoot and root traits, hybrids L71 × P3 and L61 × P2 are recommended. These hybrids responded most successfully to water stress, in terms of a stronger performance, both of roots and shoots, and best adaptation to WS was observed in the root-related traits.

The hybrids L61 × P2 and L71 × P3 were derived from lines P2 and P3, classified as tolerant, while line L61 was classified as susceptible and L71 as intermediate. This shows that the intolerant and intermediate parents, although undesirable per se for stronger performance in a water-stressed environment, represent hybrids that perform well for adaptation to dry environments, given the favorable allelic complementation observed. According to Cruz et al. (2012), the hybrid combination with the highest SCA estimate can be concluded to be the most suitable if it can be derived from a cross in which at least one of the parents has a high GCA. Therefore, our results confirm the usefulness of diallel hybrids for the development of more sustainable popcorn crops in water-stressed environments given that both parents of the most outstanding F₁ lines in terms of SCA had high GCA estimates.

In the early stage of popcorn development, the root–shoot ratio, as well as specific root length, shoot length, and shoot dry weight of the seedlings were important traits to differentiate the WS tolerant genotypes. Hybrid P3 × P7 (44) was noted for the high mean RSR; hybrid L71 × P7 (30) with a high mean SRL; hybrid L61 × L75 (4) with high mean SL; and the hybrids L71 × P3 (28) and L76 × P6 (38) with high SDW means. The other traits—GER, GSI, SDW, RSL, RDW, MRN, RNA, RNL, and RNV—seem to be closely related and were not adequate for genotypic discrimination. Meanwhile, under the WW condition, the traits RSL, RSR, SRL, SL, GER, and GSI were important to discriminate the studied genotypes. The hybrids L76 × P3 (37) and P2 × P3 (40) were with high RSL and RSR means; hybrid L71 × P6 (29) was with a high SRL mean; and hybrid L71 × P3 (28) was with high means of SDW, REE and GSI. The traits RSL, RSR, SRL, SL, GER and GSI seem to be closely related and they satisfactorily differentiated genotypes for WS adaptation. Therefore, they can be recommended for future studies to be tested from the perspective of being effectively used in early selection, as applied to popcorn in this study.