Selection of cowpea populations tolerant to water deficit by selection

- The objective of this work was to use different selection criteria to identify cowpea genotypes tolerant to water deficit to be used in recurrent selection programs aiming to reduce the cowpea crop cycle and improve its productive components. Six cowpea genotypes were crossed in a complete diallel scheme. The experiment was conducted in an incomplete block experimental design in a triple lattice arrangement, with the 30 F 2 populations of cowpea. The genetic variability of 10 traits was verified, and the following selection criteria were used: direct and indirect selection, classical index, index based on desired gains, and index based on sum of ranks. The number of days to flowering (NDF), number of days to maturity (NDMD), pod weight (PW), grain weight per pod (GWP) and 100-grain weight (100GW) were classified as primary traits, and the others as secondary. The direct selection based on the pod weight was the most efficient criterion to reduce the crop cycle and improve productive components of the cowpea genotypes subjected to water deficit conditions. The genotypes Pingo-de-Ouro-1-2 × MNC99-510F-16-1, CNCx698-128G × MNC99-510F-16-1, MNC99-510F-16-1 × Pingo-de-Ouro-1-2, Pingo-de-Ouro-1-2 × BRS Paraguaçu, Santo-Inácio × Pingo-de-Ouro-1-2, BRS Xiquexique × Pingo-de-Ouro-1-2 and Pingo-de-Ouro-1-2 × Santo-Inácio were the most suitable to be used in interpopulation recurrent selection programs aiming to reduce the crop cycle and improve productive components of cowpea grown under water deficit conditions.


INTRODUCTION
Walp] is one of the most important and strategic food sources in tropical and subtropical regions of the world (AGBICODO et al., 2009). Brazil is the third largest world producer of this crop, which is cultivated in the North and Northeast regions of the country and is the basic food of low-income populations (OLIVEIRA et al., 2013). However, according to Almeida et al. (2014), there are frequent supply deficits in these regions due to the cowpea low national average yield (300 kg ha -1 ).
The low yield of cowpea in Brazil is due to several factors, such as the use of unimproved genotypes, the low fertility of the soils and irregular rainfall distribution (NASCIMENTO et al., 2011). Water deficit is common in these regions of Brazil and is one of the factors that reduce the cowpea yield (MENDES et al., 2007). Agbicodo et al. (2009) point out that the cowpea 'drought scape' by shortening its cycle is one of the main mechanism used by this species to avoid water deficit. Therefore, identifying genotypes with early cycles is one of the most promising strategies to minimize the water deficit in this crop.
Different agronomic traits have been used to investigate the tolerance of cowpea genotypes to water deficit (BASTOS et al., 2011;NASCIMENTO et al., 2011). The simultaneous selection of traits increases the chance of success in breeding programs. Selection indexes allow the generation of a genotypic aggregate as an additional trait resulting from the linear combination of traits chosen by the breeder as desirable to perform the simultaneous selection; thus allowing the identification of superior genotypes, regardless of correlations between the traits (CRUZ; REGAZZI; CARNEIRO, 2012).
In this context, the objective of this work was to use different selection criteria to identify cowpea genotypes tolerant to water deficit to be used in intrapopulation recurrent selection programs aiming to reduce the cowpea crop cycle and improve its productive components.

MATERIAL AND METHODS
The experiment was conducted in the experimental area of the Embrapa Meio-Norte, in Teresina PI, Brazil (05° 05'S, 42° 48'W and 74.4 m), from July 26 to October 11, 2011. Teresina has annual average relative humidity of 77.02%, annual average rainfall of 1,388.9 mm, and annual average air temperature of 26.97 ºC (22.08 ºC to 33.52 ºC) (INMET, 2011). The climate of the region is Aw-tropical with wet and dry seasonsaccording to the Köppen classification. The soil of the experimental area was classified as eutrophic Yellow Argisol (Ultisol), according to the SiBCS, with sandy loam surface texture.
Six genotypes of cowpea-three cultivars (BRS Paraguaçu, BRS Xiquexique and Santo-Inácio) and three lines (Pingo-de-Ouro-1-2, CNCx698-128G and MNC99-510F-16-1)-were selected based on previous studies on tolerance to water deficit (BASTOS et al., 2011;NASCIMENTO et al., 2011) and used as parents (Table  1). These genotypes were crossed in a complete diallel scheme, according to Method 1 of the model proposed by Griffing (1956). The seeds of the 30 F 1 hybrids were seeded to obtain the F 2 generation.
A water deficit test was conducted to evaluate t h e 30 F 2 populations of cowpea and their parents. An incomplete block experimental design in a triple lattice arrangement was used. The experimental plot consisted of six 2-m rows spaced 1.0 m apart with 0.50 m between plants. Irrigation was performed with a fixed conventional system with sprinklers in lateral lines arranged in a spacing of 12 m x 12 m, with pressure of 250 kPa, nozzle diameter of 3.6 mm x 2.6 mm, and flow rate of 1.19 m 3 h -1 .
Soil moisture was monitored using a capacitance probe (DIVINER 2000 ® ) with 12 access tubes that were installed to a depth of 0.70 m and distributed in both tests, with readings at each 0.10 m depth of the soil. The soil water retention curve of the experimental area-0.0 to 0.4 m layer-was determined using the model of Genuchten (1980). The area had field capacity (FC) of 21.2 cm 3 cm -3 and permanent wilt point (PWP) of 9.05 cm 3 cm -3 . The water deficit test (WDT) was conducted with suspension of irrigation from the pre-flowering at 35 days after sowing (DAS) to the reproductive stage at 49 DAS.
Data were collected from 15 random plants from each plot. The mean values of the following traits were evaluated in each plot: number of days to flowering (NDF)-number of days from sowing to the beginning of flowering stage; number of days to maturation (NDM)number of days between flowering and the physiological maturity stage of the pod; number of days to field maturity (NDMD)-number of days from sowing to the beginning of the drying process (appearance of the first pod with color change); number of peduncles per plant (NPeP); number of pods per plant (NPoP); number of pods per peduncle (NPPe); pod weight per plant (PWP)-using 3 pods per plant (g); pod weight (PW)-using 3 pods per plant (g); pod length (PL)-using 3 pods per plant (cm); number of grains per pod (NGP)-using 3 pods per plant; grain weight per pod (GWP)-using 3 pods per plant (g); 100-grain weight (100GW) (g); grain yield (GY) (g); and grain index (GI)-grain to pod weight ratio (%). Selection of cowpea populations tolerant to water deficit by selection index The data obtained for each trait were subjected to analysis of variance, according to the model: Y ijk = m + g i + r j + b k(j) + e ijk , wherein Y ijk is the observed value in the plot that received the treatment i in the block k within the replication j; m is the overall mean; g i is the random effect of treatment i (i = 1, 2, ..., 36); r j is the effect of the replication j (j = 1, 2 and 3); b k(j) is the effect of the block k (k = 1, 2, 3, ..., 6) within the replication j; and e ijk is the experimental error of the observation Y ij with ~ NID (0,σ 2 ). The following genetic parameters were estimated for each trait according to the procedures described by Cruz, Regazzi and Carneiro (2012): heritability in the broad sense (h²), genetic standard deviation (SDg), genetic coefficient of variation (CVg) and relative coefficient of variation (CVr).
First, a multicollinearity diagnosis was performed in the matrices of phenotypic (P) and genotypic variance and covariance, in which the number of conditions showed a weak multicollinearity (NC<100). The economic weights and desired gains were established from the experimental data, as recommended by Cruz, Regazzi and Carneiro (2012), to predict the genetic gain in the main traits. The adopted desired gain was equivalent to the genetic standard deviation (SDg), and the adopted economic weights were the genetic coefficient of variation (CVg) estimates. The selection of the best genotypes and estimations of genetic progress were performed with the following methods: direct and indirect selection, classical index (HAZEL, 1943;SMITH, 1936), index based on the desired gains (PESEK; BAKER, 1969) and index based on the sum of ranks (MULAMBA; MOCK, 1978).
The direct and indirect selection is used expecting gains in a single trait, which is focused in the selection, and traits of secondary importance may have favorable or unfavorable responses. The expected gain from the direct selection in the i th trait was estimated by: GS i = DS i h i 2 , being , wherein X si is the mean of the selected genotypes for the trait i; X 0i is the original mean of the diallel; DS i is the selection differential practiced in the diallel; and h 2 1 is the heritability of the trait i. The indirect gain in trait j by the selection of the trait i was given by: GS j(i) = DS j(i) h 2 i , wherein DS j(i) is the indirect selection differential obtained with the mean of the trait of the genotypes, whose superiority was shown based on other trait on which direct selection is practiced.
The classical index proposed by Smith (1936) and Hazel (1943) consists of a linear combination of several economically important traits whose weighting coefficients are estimated in order to maximize the correlation between the genotypic aggregate and the index. The genotypic aggregate is established by another linear combination, involving the genetic values that are weighted by their respective economic weights (CRUZ; REGAZZI; CARNEIRO, 2012). The selection index of Smith (1936) and Hazel (1943) is given by: , wherein n is the number of traits evaluated; b' is the vector of dimension 1×n of the weighting coefficients of the selection index to be estimated; x is the matrix of dimension n×p (genotypes) of phenotypic values of the trait. The genotypic aggregate is estimated by wherein a' is the vector of dimension 1×n of economic weights previously established; and g is the matrix of dimension n×p of unknown genetic values of the n trait. Thus, the vector b = P -1 Ga, wherein P -1 is the inverse of the matrix of dimension n×n of phenotypic variance and covariance between the traits; and G is the matrix of dimension n×n of genetic variance and covariance between the traits.
The Pesek and Baker (1969) index is based on the desired gains for the traits, defined by . Substituting Δg, the vector of the estimated gains, by Δg d , the vector of the desired gains, and eliminating the scalar i/σ 1 , which does not affect the proportionality of the coefficients b 's , b is estimated by the expression , whose coefficients give the maximization of the gains in each trait based on the specification of the desired gains. Subsequently, the genotypic aggregates were estimated by , wherein n is the number of traits evaluated; b' is the vector of dimension 1×n of the weighting coefficients of the selection index to be estimated; x is the matrix of dimension n×p (genotypes) of the phenotypic values of the trait.
The index based on sum of ranks (MULAMBA; MOCK, 1978) consists of classifying the genotypes regarding each of the traits in order of improvement. Then, the orders of each genotype are summed, resulting in the selection index , wherein r j is the rank of the genotype i regarding the j th trait; n is the number of traits in the index.
Eight genotypes were selected for each case, corresponding to an intensity of approximately 22%. Statistical analysis were performed with the Genes (CRUZ, 2013) and SAS (SAS INSTITUTE, 2002) programs.

RESULTS AND DISCUSSION
The evaluated traits of the genotypes presented significant differences (p<0.05), except NDM, NPeP, PWP and GY ( Table 2). The existence of genetic variability in a population is a determining factor for any breeding program, thus, the cowpea genotypes evaluated are promising for recurrent selection for tolerance to water deficit. Recent studies also identified genetic variability among cowpea genotypes (BASTOS et al., 2011;CORREA et al., 2012;HAMIDOU;ZOMBRE;BRACONNIER, 2007;NASCIMENTO et al., 2011;RODRIGUES et al., 2016;SANTOS et al., 2014;TEODORO et al., 2016;TORRES et al., 2015).
According to Cruz, Regazzi and Carneiro (2012), the experimental precision is high when the experimental coefficient of variation (CVe) estimates of the traits with continuous distribution are lower than 20%, as found in this study. The lattice design presented high efficiency-except for NGP and GI-according to the criteria of Ramalho, Ferreira and Oliveira (2005), Selection of cowpea populations tolerant to water deficit by selection index who define as adequate the use of this experimental design in situations where the efficiency is higher than 100.
The genetic coefficient of variation (CVg) quantifies the genetic variability available for selection (CRUZ;REGAZZI;CARNEIRO, 2012). The ratio between CVg and CVe is the relative coefficient of variation (CVr). The CVg obtained resulted in a CVr higher than 1 for NDF, NDMD, NPPe, PW, PL, GWP, 100GW and GI. This is a favorable situation for the selection of superior genotypes in the populations. The heritability (h²) estimates for these traits-except NDMD-can be considered high (>70%) because they are polygenic traits, governed by several genes of small effect on the phenotype.
Variability among the traits included in the index is essential to employ selection indices in plant breeding (CRUZ; REGAZZI; CARNEIRO, 2012). Therefore, NDM, NPeP, PWP and GY were not included in the indices used in this work; NDF, NDMD, PW, GWP and 100GW were considered as primary and the others as secondary parameters for the prediction of genetic gain. This choice was made considering that an ideal cowpea genotype has a reduced cycle, which makes it possible to be cultivated in times with lower water availability, and have high grain weight, since this trait affects its commercialization.   = not significant by the F test, * = significant at 5% probability by the F test, and ** = significant at 1% probability by the F test. SV = sources of variation; DF = degrees of freedom; CVe = experimental coefficient of variation; h² = heritability; SDg = genetic standard deviation; CVg = genetic coefficient of variation; CVr = relative coefficient of variation The gains predicted by the direct selection were superior to the indirect gains for all the traits, and close in situations in which the traits are highly correlated, for example, for PW, GWP and 100GW (Table 3). Falconer (1987) reports that indirect selection can promote greater gains than direct selection, when the auxiliary trait presents greater heritability than the main one, and when the genetic correlation between them is positive and of high magnitude. The highest individual gains were observed for 100GW (14.30), NPPe (13.17) NPoP (9.83), and PW (8.38%).
Among the traits used for direct and indirect selection, PW was the most promising for use in recurrent selection programs with these genotypes; it promoted reduction in the NDF and NDMD and increased in the PW, GWP and 100GW. The genotypes selected according to this criterion were Pingo-de-Ouro-1-2 × MNC99-510F-16-1, CNCx698-128G × MNC99-510F-16-1, MNC99-510F-16-1 × Pingo-de-Ouro-1-2, Pingode-Ouro-1-2, BRS Paraguaçu, Santo-Inácio × Pingo-de-Ouro-1-2, BRS Xiquexique × Pingo-de-Ouro-1-2, Pingode-Ouro-1-2 × Santo-Inácio. Thus, these genotypes can be used in intrapopulational recurrent selection programs aiming to improve resistance to water deficit. These results are desirable in breeding programs of cowpea because the reduction of the crop cycle is one of the main physiological mechanisms of tolerance to water deficit (AGBICODO et al., 2009). In addition, the indirect response in the other traits suggests the possibility of selecting genotypes with high grain yield.
The estimates of gains predicted by the Smith (1936) and Hazel (1943), Pesek and Baker (1969), and Mulamba and Mock (1978) indexes are described in Table  4. In general, the estimates using the Smith (1936) and Hazel (1943) and Mulamba and Mock (1978) indexes are similar in magnitude and direction. This generated the coincident selection of the genotypes 2, 13, 15, 16, 18, 28, 33 and 34. The Pesek and Baker (1969) index showed gains with similar magnitudes to the other indices used for PW, GWP and 100GW, however, it showed different magnitudes and direction from the predicted gains for the other traits.
According to Cruz, Regazzi and Carneiro (2012) the use of selection indexes is advantageous compared to the direct selection, since it allows more distributed gains in all evaluated traits without providing significant loss in the main traits. However, in the present work, none Table 3 -Estimates of the means (Ẋ 0 ), heritability (h²) and selection gains (%) obtained by direct and indirect selection in the traits number of days to flowering (NDF), number of days to maturity (NDMD), number of pods per plant (NPoP), number of pods per peduncle (NPPe), pod weight (PW), pod length (PL), number grains per pod (NGP), grain weight per pod (GWP), 100-grain weight (100GW) and grain index ( Table 4 -Estimations of selection gains (%) obtained by the Smith (1936) and Hazel (1943), Pesek and Baker (1969) and Mulamba and Mock (1978) , 16, 15, 13, 28, 33, 34 and 18 6, 11, 27, 1, 9, 25, 30 and 7 18, 2, 33, 16, 28, 34, 32 and 13 Selection of cowpea populations tolerant to water deficit by selection index of the selection indices used was efficient in selecting genotypes that simultaneously provide reduction in cycle related traits and increase in the traits PW, GWP and 100GW. Similar results were observed by Santos and Araújo (2001) using different selection indices in cowpea genotypes; they found that the direct selection in productive traits provided greater gains on the others, when compared to the Smith (1936) and Hazel (1943), Pesek and Baker (1969) and Mulamba and Mock (1978) indexes.