According to the analysis of variance (Table 1), the variables plant height (PH), ear height (EH), flowering interval (ASI), male flowering (MF), female flowering (FF), number of rows per ear (NRE), number of grains per row (NKR), spike length (EL), 1000-grain weight (1000GW), grain weight per plant (GWP), relative spike position (RPE) and grain yield (GY) were significant at 1% (p ≤ 0.01), and that of ear weight per plant (EWP) was significant at 5% (p ≤ 0.05).
The significant variables contributed to differentiate the 78 corn genotypes, showing that there is genetic variability in the materials evaluated, and, according to Cruz et al. (2014), the differentiation of genotypes is essential in a breeding program, because it favors the selection of genotypes (Tavares et al. 2018).
In the 1000GW and GY variables, the means of the experimental and commercial genotypes showed statistical difference at 5% (p ≤ 0.05) by contrast analysis (Table 1), which indicates variability between the two types of cultivars for these variables.
The coefficient of variation (CV) ranged between 2 and 22.74% (Table 1), being within the ideal for a corn experiment, which shows homogeneity in data collection (Fritsche Neto et al. 2017). On the other hand, the CV for IF was considered very high (41.5%), probably due to the wide variability of maturity for the genotypes evaluated. These results corroborate with those found by Suthamathi and Nallathambi (2018), who when evaluating maize strains in a semi-arid region, showed CV values equal to 39.73%.
Distinguishing the performance of commercial and experimental genotypes is important in the context of identifying the standards that should or should not be improved in cultivars already available to the producer. This distinction, which can be translated into genetic variability among genotypes, enables the implementation of the breeding program with these materials in the target region (Sheikh et al. 2017). In addition to this, one can better observe secondary traits, i.e., those that influence PG (Sheikh et al. 2017).
In the distribution analysis, using the boxplot, it is possible to observe the trend by means of the median and quartiles of the distribution of the commercial and experimental genotypes, and to demonstrate the variability within each type (Figure 1).
In the FF, MF and ASI traits, relative to the beginning of the reproductive stage, the trend of higher averages was observed for the experimental genotypes (Figure 1e, 1f and 1g). This may be an explanation for the drop in the averages of the experimental ones in grain yield (Figure 1d), since, according to Paterniani et al. (2019) and Rocha et al. (2021), genotypes that present a high ASI and later cycle are more susceptible to abiotic stresses, especially water deficit, which can cause sterility or severely reduce the number of grains per ear.
For the GY variable, the commercial genotypes presented the highest averages, having genotypes with yields above 10,000 kg ha-1, which may be a consequence of the earliness and lower ASI, observed for most commercial genotypes (Figure 1d).
The correlation coefficient is used to identify the interaction between variables, such as the positive or negative influence that one character has on another and can assist the breeder in selecting materials with good performance in variables that are difficult to measure (Miot 2018).
According to Figueiredo Filho and Silva Junior (2009), the closer the correlation coefficient is to 1, the greater the positive correlation between the characters, similarly, the closer to -1, the greater the negative correlation. Thus, six positive and significant correlations were identified between the variables (Figure 2), they are: PH and EH (0.7), FF and MF (0.81), RPE and EH (0.74), GWP and EWP (0.77), and for the economic value variable, GY was positively correlated with GWP and EWP (0.98 and 0.82, respectively). Besides this, there were also negative and significant correlations between GY and FF (-0.60), GY and ASI (-0.41), FF and GWP (-0.50) and FF and EWP (-0.44), which allows us to observe that, the increase in ASI and FF leads to low corn grain yield (Figure 2).
The negative correlations corroborate with those obtained by Melo et al. (2018), who, when evaluating maize genotypes under water stress, found that earlier genotypes with lower ASI present higher grain yields and, consequently, greater adaptability in regions with characteristics like the semiarid.
According to the distribution of means (Figure 1), the population of early genotypes is more productive and, consequently, more adaptable to the conditions of the semi-arid region of Sergipe, a fact proven in the correlation analysis (Figure 2), with a negative correlation between FF and GY (Figure 2). Thus, it is affirmed that the FF and ASI characteristics can be used to help the indirect selection of individuals. Thus, the selection of more precocious genotypes with lower ASI is suggested to obtain more productive materials for the semiarid region of Sergipe (Carvalho et al. 2020).
Regarding, the selection of genitors, in the case of morphological evaluations in the field, is based on their respective phenotypic averages. In the case of maize, selection is mainly done by grain yield (kg ha-1), since this is the economic trait of interest in the crop (Sheikh et al. 2017). Guimarães et al. (2019), when working with selection of maize genotypes, stated that these should present the highest yields and economic gains for the region where the breeding program is being conducted.
The selection using mixed models aims to differentiate, from a production variable, the genotypes by the averages adjusted by BLUP, when considering random effect, or BLUE, when considering fixed effect (Resende and Alves, 2021). Santos et al. (2021) also used the mixed model methodology (REML/BLUP) to select genotypes from the corn grain yield variable.
In Figure 3, the "good" genotypes are those with high grain yield, based on the BLUE's, estimated via mixed models. That said, the commercial BM 930 PRO2 (44) and AS 1868 Pro3 (74) were selected as superior among the others, followed by SHS 5570 (41) and LG 3040 VIP3 (63). Six experimental genotypes were rated as "good", among them DGX 20 S03 (26) that was close to the commercial ones selected in the rank (Figure 3).
In the formation of the crossing blocks, the distribution and genetic divergence could be observed (Figure 4). The first (BLOCK I) and the fifth block (BLOCK V) accounted for the largest number of grouped genotypes (15 and 13 genotypes, respectively) (Figure 4), while the second block (BLOCK II) consisted of only one genotype (Figure 4).
The experimental genotypes were present in two blocks, BLOCK I and V. In BLOCK I were grouped the experimental XB 73530, DGX 20 S02 and DGX 20 S03, and in BLOCK II the DGX 20 S01, EX 3591 L VIP3 and LG 6036 PRO3 (Figure 4).
All the blocks formed grouped genotypes with statistically equal means for GY, FF, NRE and SD, as there was no statistical difference among the selected genotypes for these variables. BLOCK V grouped the genotypes with higher averages for PH, EH, NEP, EWP and GWP, and, also, was the only group to stand out in the lowest ASI. BLOCK I gathered the genotypes with higher averages in EL, BLOCK III with higher ones in EG and 1000GW and BLOCK IV only in 1000GW.
Considering the increased frequency of favorable alleles in the crossing group of experimental genotypes, we indicate as genitors those allocated in blocks III and V, because the crossing of these blocks will allow greater chances of achieving earlier and more productive genotypes in segregating generations (Rotili et al. 2012).
Aiming at the possibility of gains with the selection by the experimental genotypes, the genetic parameters were estimated (Table 2). It was observed that the selective accuracy was between 0.34 and 0.93. According to Resende and Duarte (2007), accuracy below 0.50 is considered low, between 0.5 and 0.7 moderate, high between 0.7 and 0.9, and very high above 0.9. Thus, it can be stated that for the variable GY (0.78) and variables for indirect selection FF and ASI (0.81 and 0.87, respectively) showed high selective accuracy (Table 2), that is, their genetic values are close to the real one.
The phenotypic variance seen in all variables was higher than the genotypic variance, which shows that the environmental variance interfered significantly in the estimates of this genetic parameter.
The CVg varied between 1.84 and 20.87 among all variables (Table 2). For FF, ASI, and GY, the CVg was 1.84, 20.87 and 13.04%, respectively, and the CVe was 1.83, 28.00 and 14.83%, respectively, showing that the environment had low influence on the estimation of genetic values in the population (Table 2). Close values of CVg and CVe obtained were presented by Tesfaye et al. (2021) when they evaluated the genetic parameters in maize productivity under semiarid conditions.
According to Resende (1995), heritability ≤ 0.5 is considered low, between 0.15 and 0.5 moderate, and above 0.5 high. With the heritability for FF, ASI and GY of 0.5, 0.35 and 0.43, respectively, it is seen that selection of the genotypes within the experimental population can be effective in achieving genetic gains for grain yield.
The heritability at mean level, ranged from 0.12 to 0.86, with 0.67, 0.53 and 0.61 observed in FF, ASI, and GY variable, respectively (Table 2). These high values of heritability point to an indication of reproducibility and phenotypic stability of the characters by the evaluated population (Freitas et al. 2006), which indicates the feasibility of inserting the six experimental genotypes selected in the population in a maize breeding program (Cruz et al. 2020).
The genetic gain obtained in GY by the population of the six experimental genotypes selected earlier was 551.15 kg ha-1 in the average of the improved population (Table 2). With the positive gain of these genotypes, it can be stated that these six experimental genotypes are potential genitors for a plant breeding program in the hinterland of the Sergipe region.