Evaluation of maize core collection for drought tolerance

The maize genebank (GBmaize) preserves nearly 4,000 accessions for conservation and use. The use is however restricted because the accessions do not perform as well as the elite genotypes. This problem can be reduced by pre-breeding, i.e., by extending the information on germplasm and introgressing useful alleles. Since irregular rainfall distribution and drought induce maize yield losses, drought tolerance is a main breeding target. In this study, the GBmaize accessions were evaluated for drought tolerance. Environmental factors, genotypes and the respective interactions influence the phenotypic expression. There was however no interaction genotype - irrigation level, so no accessions with different performance under the two water regimes could be identified. The performance of the following accessions was promising for a number of traits: SP154, BA166, MG099, CE002, SE025, BA154, BA194, BA085, MG076, PR053, Roxo Macapa, SE016, and AL018.


INTRODUCTION
Genebanks aim to maintain the genetic diversity and are sources of genetic variability for research. The maize genebank (GBmaize) of Brazil preserves nearly 4000 accessions from national collections, breeding programs and of exotic varieties. The maintenance of the GBmaize involves several activities, including agronomic evaluations (Teixeira et al. 2005). Core collections are representative germplasm samples preserved to maintain the genetic variability with a minimum of repetitiveness. Their main advantage is the fast evaluation and revaluation of the germplasm. The maize core collection in Brazil was established in 1997 using the maize collection maintained at Embrapa Maize and Sorghum and Embrapa Genetic Resources and Biotechnology, which at the time comprised 2280 accessions. This collection contains 300 accessions and was preliminarily classified into four strata: landraces, compounds derived from landraces, improved genotypes and introductions. The landrace stratum was divided into 27 groups, according to the geographical origin and grain type. The groups improved genotypes and introductions are further subdivided (Abadie et al. 2000.) Despite their importance, breeding programs made little use of the genebanks. Only 14% of the maize breeders regularly use genebanks and one of the reasons is the scarce amount of data on the collections. Besides, especially in the case of maize, breeders have established work collections with exceptional performance, discouraging the search for variation in the GBmaize genotypes (Nass and Paterniani 2000). This situation leads to the gap between the areas of genetic resources and breeding, which Evaluation of maize core collection for drought tolerance consequently prevents the genetic diversity preserved in BGmilho from reaching the elite collection of the breeder and the producer. Pre-breeding involves the identification of special traits in genotypes considered unimproved and the availability of such genotypes to plant breeding ( Paterniani 2000, Nass et al. 2007). Several studies have been conducted to expand the knowledge about maize germplasm (Teixeira et al. 2002, Miranda Filho andGorgulho 2003). However, the use of GBmaize is still limited, because according to Nass and Paterniani (2000), the performance of the genotypes used in breeding programs is already much better than of those of the GBmaize, which makes the elite collections far more attractive for breeding.
Among the environmental stresses that lead to yield losses, drought causes most damage in temperate regions, although the detrimental effects of water stress are more pronounced in tropical or subtropical regions than in temperate climates (Ramalho et al. 2009). Irregular rainfall distribution and drought lead to maize yield losses, indicating drought tolerance as a priority for breeding. Due to the great influence of drought on flowering, the gap between male and female flowering is a variable used in the selection of drought-tolerant genotypes (Bruce et al. 2002).
The aim of this study was to evaluate accessions of maize core collection for drought-tolerance deficit aiming at the use in breeding programs.

MATERIAL AND METHODS
Accessions of the stratum of autochthonous varieties (subgroups Caatinga and Cerrado) of the core collection were tested and elite genotypes and commercial cultivars used as controls. These genotypes were divided into two groups: test 1 (T1) and test 2 (T2), according to the number of days to flowering, determined in a preliminary assessment, to facilitate the irrigation management. T1 comprised the earlier accessions and T2, the later. Below is the list of accessions in each test: Accessions in T1: SP181, SP154, BA166 and MG099 of the Cerrado group with semident grain, BA178 and BA083 in the Cerrado group with semi-flint grain; SP015 Cerrado group with flint grain BA019, PB010, PE011, BA028, MG060, CE002, SE025, BA154, AL001, BA194, and PB003 of the Caatinga group, with semident grain, BA003, BA061 and SE014 Caatinga group semi-flint grain; PE002 Caatinga group with flint grain, and Synthetic Elite Flint (SEF), Drought-tolerant Synthetics (STS) and Sertanejo as controls. Accessions in T2: MG090, MS043, SP019, MS007, SP036, BA085, MG076, PR053, Roxo de Macapá, MS030, MT009, and PR050 of the Cerrado group with semident grain; MS019 and MG010 of the Cerrado group with semi-flint grain; SP145 Cerrado group with flint grain, RN003, PE013, SE016 and AL018 Caatinga group, semident grain; BA020 Caatinga group, semi-flint grain, AL009 and PB020 Caatinga group with flint grain, and BR106, SEF and Synthetic Jaíba (SJ) as controls. It should be mentioned that the controls STS and SJ are elite maize genotypes of the Embrapa Maize and Sorghum breeding program, which at some development stage had undergone selection for drought stress.
The following environmental factors were considered: the locations (L) Janaúba-MG and Teresina-PI; years (A), planting in the dry season in 2005 and 2006, and the irrigation regimes (I) with full water supply throughout the cycle (no stress) or cutting of irrigation in the pre-flowering period (under stress). At each location and each year, experiments were implanted with two irrigation regimes. In the tests without stress, sprinkler irrigation was maintained throughout the cycle. In Janaúba, irrigation management was established as recommended by Albuquerque (2007), based on soil and climate data. In Teresina, plants were irrigated daily and the irrigation level was estimated based on crop evapotranspiration of the day before, according to the methodology proposed by Andrade Júnior et al. (1998). In the stress tests, irrigation was interrupted/halted/ at the beginning of tasseling until 20 days after pollination. After this period, irrigation was resumed by returning the soil to predetermined field capacity. The experimental design used in all evaluations was a 5 x 5 triple lattice where plots consisted of two 5 m rows, a sowing density of five plants per meter and 0.90 m spacing. Statistical analyses were performed in each location and combined analysis involving the factors studied. According to the results of analysis, means were compared using the Tukey test at 5% probability. The broad-sense heritability (h 2 ) was estimated for all traits evaluated and the phenotypic correlation among these.
The following traits were evaluated: number of days to male flowering (MF), number of days to female flowering (FF), both based on the number of days between seedling emergence and flowering of 50% of the plants in the plot; interval (in days), between anthesis and silk interval (ASI), plant height (PH) and ear height (EH), averaging the data of 10 plants per plot (in cm); prolificacy (PROL) obtained by the division between the total number of ears per plot and the plot stand, and grain yield (GY) in ton ha -1 .

FF Teixeira et al.
the analysis variance for MF, FF and ASI. This result was considered a preliminary indication of superiority of these accessions over others. One should bear in mind that none of the controls flourished under severe water stress, as imposed in 2005 in Janaúba.
Estimates of phenotypic correlations indicated a significant and high correlation between some traits. In T1, the correlations between PH and EH, PH and GY, GY and EH, and MF and FF estimates were high and positive, while GY and ASI high and negative. In T2, the correlations

RESULTS AND DISCUSSION
The overall means of the traits in each experiment (Table 1) showed that the mean variation in GY in the stressed locations was between 35.54% and 78.79% of the unstressed treatments. Thus, irrigation suspension induced GY reductions approaching 50%, as described by Bruce et al. (2002), which is the value recommended for evaluations of drought stress in maize. The mean ASI between tests with and without water stress reached 6.85 days in T1 in Janaúba, in 2005. It is worth noting that the water stress in T1 and T2 in Janaúba was very severe, leading to a considerable GY reduction, longer ASI due to the delay in FF and not reaching the flowering stage in some plots, i.e., in some plots percentage of flowering plants did not reach 50 %. Therefore, the accessions were classified into two groups: the first contains the accessions that did not reach flowering and the second, those that flowered all locations. Only the accessions: SP154, BA166, MG099, CE002, SE025, BA154, BA194, and BA061, in T1 and SE016, AL018, BA085, MG076, PR053, and Roxo de Macapá, in T2 flowered under all conditions and were considered in between PH and EH, PH and GY were high and positive. Among these results, the negative correlation between ASI and GY should be emphasized, which indicates a yield reduction with increasing ASI, in agreement with the statement of Bruce et al. (2002), indicating the use of ASI as variable underlying selection of drought-tolerant genotypes.
The presence of the effects of location, year and water regime, as well as most of the respective interactions for most traits showed that the phenotypic expression for different traits was influenced by these environmental factors. The effect of genotypes affected most traits, except Evaluation of maize core collection for drought tolerance PROL in T1, and MF, FF and ASI in T2, indicating that in the latter, the difference observed in ASI between locations with and without water stress was not significant (Tableo1). The decomposition of the effects of genotypes within and between the ecogeographic groups showed differences between and within the groups Caatinga, Cerrado, and controls for PH, EH, GY, MF and FF in T1, and only within the Cerrado group for ASI. In T2 however, the effects were observed within and between groups for PH, EH, PROL, and GY. The decomposition of the genotype effect within and between grain type groups in the T1 showed differences for most of the decompositions except: within the flint group for PH, EH, and GY, between the groups for EH and FF and for ASI for any decomposition factor. In T2, the effect of the decomposition of the genotype factor according to the grain type was significant for all traits, except for PH within the flint and PROL within the semiflint group.
The core collection consists of accessions representing the variability of the entire collection with a limited number of entries (Abadie et al. 2000). It was therefore expected that the divergence between these accessions were high. This expectation was confirmed in the differences found between the genotypes and their decompositions. However, the limited number of accessions used in each group must be taken into account, as discussed in the following. In general, higher variability has been reported within groups with flint than within groups with dent grain in the core and base collections (Abadie et al. 2000, Netto et al. 2004), but in this study the variability within the dent group was more pronounced. This may have been caused by the nature of the character, since in most studies aimed at quantifying the genetic divergence neutral characters are considered and in the present study, we focused on agronomically relevant traits. Another noteworthy factor is that the mean squares for the estimated effects between groups were higher, in most cases, than those obtained within groups, which was expected, mainly due to the great phenotypic divergence between the groups formed by GBmaize accessions and the control group, consisting of elite genotypes (Nass et al. 2007, Teixeira et al. 2007). This observation shows the key function of actions of pre-breeding to make GBmaize accessions with valuable variability useful for breeding.
The interactions between genotype and environmental factors were present in some situations. In T1 for GY, there were interactions between locations and genotypes and their decompositions within the Caatinga and semident groups and between grain type and year groups and genotypes. In T2, there were interactions of genotypes by years and the decompositions within the Cerrado, Caatinga, semident and flint groups and between groups of origin and grain type for PH. For EH, the same interactions were present and also the interactions locations by genotypes and of the decompositions within Cerrado and flint groups and between grain type groups. For PROL in T2 interactions between locations and genotypes and the decompositions within Cerrado and semident groups were observed, the interaction between years and genotypes and their decompositions within Cerrado, semident and semiflint groups, and the triple interaction location x year x genotype and their decompositions within the groups Cerrado, Caatinga and semident and between grain type groups; for GY interactions location x genotype were observed and their decompositions within groups Caatinga, control, semident, flint and between grain type groups; genotype x year and the decompositions within the groups Cerrado, control, semident, flint and between groups of geographical origin and grain type; and the triple interaction genotype x location x year and the decompositions within the groups Cerrado, semident, flint and between groups of geographical origin.
No interaction of genotype -irrigation regime was observed, as also reported by Silva et al. (2008), for any characters in the two tests at both locations in two years. The presence of strong interactions between genotypes and other environmental factors should be highlighted. Much of the changes in phenotypic expression are possibly differentiated responses to the environmental effects of locations and years and not to the water regime, principally when taking the influence of the interactions genotypeyear and location-genotype into account, as reported in several studies (Welcker et al. 2005, Terasawa Júnior et al. 2008, as well as the great range of climatic conditions in Brazil (Paterniani 1990). The installation costs of experimental evaluations with and without drought stress are high, since this assessment requires rigorous monitoring of irrigation and the decision on the optimum time for irrigation suspension. Moreover, the selection of treatments to be included in this test type is restricted, because genotypes with different cycles can not be evaluated in parallel, since genotypes need to be in the same development phase at the time of stress onset. Regardless of the high costs, the main obstacle is that these tests need to be conducted in the field in the dry season, ie, not in the normal corn season, which can mask the performance of these varieties due to the interactions FF Teixeira et al.
estimated. The correspondence between the selection performed in environments with and without drought stress is considered controversial by Monneveux et al. (2006). Evaluations including tests with and without water stress are very important for identifying genotypes more and less affected by reduced water availability, in the case of genotype-irrigation regime interaction. Genotypes with coincident performance in the two conditions can also be selected in these tests, both in the presence or absence of interaction. The identification of genotypes with yields less affected by environmental variations is a major breeding target, whereas genotypes strongly influenced by environmental variations are useful in studies on physiological mechanisms related to stress tolerance.
The h 2 estimates obtained for the traits for which significantly different results were obtained in the tests were high, ranging from 53.46% for ASI to 94.10% for EH in T1 and 79.76% for PROL to 93.61% for EH in T2, indicating the possibility of successful phenotypic selection for these traits in breeding programs.
Tables 2, 3 and 4 show the means and test means for the traits evaluated in T1 and T2. For the traits with genotype-year or genotype-location interaction, means were tested based on the averages of each location and year for GY, although in T1, only the assessments of 2006 were used to discriminate accessions for GY, since in the evaluations of 2005 the treatments were similar for GY. In T1 (Table 2), the group means showed that the accessions with semi-flint and flint grain had similar performance to the controls for PH, and that the accessions of the groups Cerrado, Caatinga, semident and semiflint and GY had similar accessions as the group of controls for GY in Janaúba in 2006. In T2, despite the significant differences among genotypes for PROL, the mean tests grouped most accessions on the same level for this trait, which prevented the identification of outstanding genotypes (Table 3). It is also worth mentioning that in the GY evaluations in Teresina, the genotypes did not differ in 2005 and that in 2006, none of the GBmaize accessions reached the same GY level as the control Sintético Jaíba (Table 4). The comparison of the groups formed by the GBmaize accessions and the control group showed that in the mean, PH of the flint group was similar to that of the control group. For GY in Janaúba in 2005, the means of the Caatinga and semident group were similar to the controls.
Some considerations on the estimates of means are appropriate; firstly, the majority of accessions in the T1 and T2 have at least one trait with a similar mean to that of the controls (improved genotypes). Along with the presence of favorable characteristics, there are other unfavorable traits, which disqualifies these accessions for the direct use in breeding. No GBmaize accessions were identified with better performance than the controls for GY or other traits of agronomic performance. But no control flourished under all conditions, while some accessions did not only flourish at all locations, but also had high GY and/or other favorable characters. Thus, the potential for use in pre-breeding programs for the introduction of useful variability sources into elite germplasm was greatest in the accessions SP154, BA166, MG099, CE002, SE025, BA154, and BA194 of T1 and BA085, MG076, PR053, Roxo Macapá, SE016, and AL018 of T2, according to the principles proposed by Nass et al. (2007).
It should also be highlighted that the means of the accessions of the Caatinga and semident groups were similar to the controls for some of the traits. These findings must be interpreted with caution, since the superiority of accessions from the Caatinga may possibly have been caused by adaptation to the regions in which the assessments were conducted, which reinforces the planning of trials on a regional basis. The superiority of the semident group accessions on the other hand is possibly due to the fact that the use of this grain type increased parallel to maize improvement in Brazil (Sawazaki andPaterniani 2004, Teixeira et al. 2007). These accessions may have been modified by a greater breeding effort and consequently have a higher GY.
Evaluation of maize core collection for drought tolerance Table 2. Means and test of means of the combined evaluation of the treatments in test 1 FF Teixeira et al. Table 3. Means and test of estimated means in Test 2 for each year for plant height (PH, in cm), ear height (EH, in cm), and for each location and year for prolificacy (PROL) Evaluation of maize core collection for drought tolerance Table 4. Grain yield means of the accessions evaluated in the 2 nd test for each location and year * for each combination location/year, means followed by at least one same letter did not differ from each other by the Tukey test at 5 % probability.