Analysis of Direct and Indirect Selection and Indices in Bread Wheat (Triticum aestivum L.) Segregating Progeny

1Department of Agronomy, Faculty of Natural, Life and Earth Sciences and the Universe, University of Mohamed El Bachir El Ibrahimi, 34034 Bordj Bou Arréridj, Algeria 2National Agronomic Research Institute of Algeria (INRAA), Unit of Sétif, 19000 Sétif, Algeria 3Department of Ecology and Plant Biology, Valorization of Natural Biological Resources Laboratory, Faculty of Natural and Life Sciences, University of Ferhat Abbas Sétif-1, 19000 Sétif, Algeria


Introduction
In Algeria, most of wheat producing areas are located in the High Plateaus which are characterized by cold winters, insufficient and erratic rainfall, frequent spring frosts, and late-season sirocco occurrence [1,2].In addition to these climatic stresses, there are some other technical constraints that essentially arise from the use of unproductive varieties and often bad agronomic practices.Selection for a better adaptation to environmental stresses is, therefore, more promising outcome in the field of wheat breeding.Breeders are continually seeking to improve the selection methods in order to develop superior wheat varieties with high grain yield, good end-use quality, and tolerance to biotic and abiotic stresses.
Direct selection based on grain yield is mainly practiced in wheat breeding programs without considering the adaptive traits that are crucial production regulators under variable environments [3][4][5].In these environments, the presence of genotype × environment interactions reduces the efficiency of using grain yield as the sole selection criterion and, thus, complicates the efforts of selection [6,7].In addition to the environmental effects, other factors such as polygenic nature, low heritability of grain yield, linkage, and nonadditive gene action may make the selection less efficient mainly in early segregating generations.
In order to overcome these difficulties, breeders are focusing on other traits that can be used in parallel or independently of yield in a multitraits approach.Indirect selection uses some yield components that are more heritable than yield itself and more stable in relation to genetic and environmental factors affecting them.When these components are measured without error and expressed in appropriate units, their product is yield.This has created new opportunities for plant breeders to use certain morphological, physiological, and biochemical traits during selection for grain yield.In the literature, several authors have reported the use of many of 2 International Journal of Agronomy these traits to improve grain yield in diverse environments [8][9][10].
Selection-based index is another approach, certainly complex, but can avoid the limits of the single-trait selection, particularly the undesirable between-trait relations that present an additional nuisance in breeders' work [11].Selection-based index approach targets the simultaneous improvement of several traits at the same time, including the grain yield [12,13].The indices allow the use of a single value in the selection process, since the analysis is carried out by means of linear combinations of phenotypic data of different traits of agronomic interest with the genetic properties of a population [14].The objective is to guarantee the improvement of the population genotypic values and consequently the efficiency of the selection process, maximizing the expected genetic gain.In this purpose, many selection indices have been used as an effective selection criterion in plant breeding programs on different crops [15].However, the conditions determining the usefulness of an appropriate selection index may vary with individual plant breeder.The objective of this research paper was to evaluate the efficiency and applicability of different selection criteria based on the estimation of genetic gain in F 4 segregating populations of bread wheat evaluated under semiarid conditions.

Plant Material and Experimental
Design.This experiment had 609 genotypes, comprising 20 F 4 -derived families and their 9 parents.The history of these families is that, after the initial crosses in 2010/11 [18], 30 F 2 lines were selected in each family by using pedigree method and were evaluated during the 2012/13 cropping season [19].A total of 600 lines were planted and harvested in bulk during two consecutive growing seasons 2013/14 and 2014/15.The F 4 lines along with their parents were planted in Sétif Research Unit (36 ∘ 15  N; 5 ∘ 87  E; 1081 masl) of the Algerian National Institute of Agronomic Research (INRAA) in a modified randomized complete block design with three replications.The experimental unit consisted of a single row plot of 2 m length with rows spaced 0.2 m apart.The plots were fertilized with 100 kg ha −1 of 46% superphosphate, before sowing, and 75 kg ha −1 of 35% urea, at tillering stage.Weed control was performed chemically using 12 g ha −1 of the Granstar [tribenuron methyl] herbicide.According to Chennafi et al. [20], the climate of the region is a semiarid type, continental with cold winter and hot and dry summer.Total annual rainfall is around 350 mm.The soil used was classified as a silty clay with high CaCO 3 content [20].

Measurements.
The following traits were measured as per plot basis.Chlorophyll content (CHL, Spad) was determined at heading stage using the SPAD-502 chlorophyll meter (Minolta Camera Co., Osaka, Japan).Flag leaf area (FLA, cm 2 ) was obtained using the formula described by Spagnoletti Zeuli and Qualset [21]: FLA (cm 2 ) =  (cm) ×  (cm), where  is the flag leaf length and  refers to the flag leaf width.Heading date (HD, days) was recorded as the number of calendar days from January first to the date when 50% of the spikes were half way out from flag leaf.Plant height (PH, cm) was measured at maturity from the soil surface to the top of the spike, awns excluded.Above ground biomass (BIO, g m −2 ) was estimated from a harvested area of 0.5 m long × 0.20 m interrow spacing, which served also to obtain the grain yield (GY, g m −2 ), number of spikes (SN, m −2 ), and spikes weight (SW, g m −2 ).1000-kernel weight (TKW, g) was obtained from the count and weight of 250-kernel.The number of grains per spike (GN) was derived from estimated values of grain yield, number of spikes, and 1000-kernel weight.Harvest index (HI, %) was estimated by the ratio between grain yield and above ground biomass.

Data Analysis.
Data collected were subjected to analysis of variance following the procedures described by McIntosh [22]; the skeleton of the analysis of variance is shown in Table 1.The statistical model used considered the complete randomized block design as   =  +   +   +   , where   is the observation of the th genotype, evaluated in the th replicate;  is the overall mean of the experiment;   is the effect of the th genotype;   is the effect of the th block; and   is the effect of the th plot.
In case -test was significant, standard error and critical differences were calculated by using the least significant difference test at 0.05 probability level (LSD0.05)according to Steel and Torrie [23]: LSD0.05 =  0.05 √2 2  /, where  0.05 is the tabulated value of the -test at 0.05 probability level for ( − 1)( − 1) residual degrees of freedom;  2  is the residual variance; and  refers to the number of replications or blocks.
The genotypic variance ( 2  ) and residual variance ( 2  ) were calculated and served to determine the following genetic and nongenetic parameters for each trait.The coefficient of experimental variation (CV) was calculated by CV (%) = √ 2   /, where  2  is the residual variance and  is the general mean of the trait.The coefficient of genetic variation (CV  ) was calculated by the following equation: CV  (%) = √ 2  /, where  2  is the genotypic variance and  is the general mean of the trait.The variation index was determined as the ratio of CV  /CV  .Broad-sense heritability of average progenies (ℎ 2 bs ) was estimated by the expression [24]: Narrow-sense heritability of individuals within families (ℎ 2 ns ) was determined based on the parent-offspring regression.To do so, two methods were used: the first was by the linear regression of F 4 on the parental F 3 individual values [16], while the second approach was performed with standardized data of offspring (F 4 ) versus standardized of the corresponding parent (F 3 ) according to Frey and Horner [17].
The selection gains were estimated among families based on three selection criteria: direct selection, indirect selection, and selection-based index considering the selection intensity of 5% of top families.The expected gains by direct selection for each trait evaluated were estimated by the expression [25]: , where Δ  is the gain with the direct selection carried for the th trait; ℎ 2  is the heritability of the th trait;   refers to the differential selection of the th trait;   is the mean of the th trait for the selected individuals; and  0 is the mean of the th trait in the base population.The expected gain of direct selection, expressed as a percentage of the population mean, is given by Δ  % = (Δ  × 100)/ 0 .
Gains from indirect response to selection were calculated using the following expression [25]: , where GS () is the gain on the th trait, with selection based on the th trait;   is the mean of the th trait for the selected individuals based on the th trait;  0 is the mean of the th trait; ℎ 2  is the heritability of the th trait; and DS () refers to the differential selection of the th trait, in which the selected lines presented the best performance for the th trait.The expected gain of indirect selection, expressed as a percentage of the population mean, is given by GS () % = (GS () × 100)/ 0 .
For the selection-based index, the following methodologies were used for gains estimation: the classic index proposed by Smith [26] and Hazel [27], the base index of Williams [28], the free weights and parameters index of Elston [29], the index of desired gains of Pesek and Baker [30], the multiplicative index of Subandi et al. [31], the sum of ranks index of Mulamba and Mock [32], and the genotype-ideotype distance index proposed by Cruz [25].Each index displays certain particularities in its calculations and, as such, application is generally laborious due to the need to assign adequate economic weights to each trait.Based on the different analytical procedures of selection, the best populations were identified and the gains from selection were calculated.All statistical analyses were carried out using Genes software [33] and a Microsoft Excel © spreadsheet.

Genetic Variability and Heritability.
The analysis of variance revealed significant genotype effect for almost all the traits under study except for flag leaf area and grain yield, which were not significant at 0.05 probability level (Table 2).This provides evidence of the presence of sufficient genetic variability among parents and hybrids that can be exploited in wheat breeding program through selection.Partitioning the genotype effect indicated significant differences between all parents for HD, PH, SN, TKW, SW, GN, BIO, and HI, and significant interpopulation differences for HD, PH, and GN.The contrast "Parents versus F 4 ," was highly significant for HD, PH, and SN, while the F 4 populations exhibited significant differences for nearly all the observed quantitative phenotypic traits, except for CHL, FLA, GN, and HI, which were not significant at the 5% probability level.The results of this study corroborate those of Abd El-Shafi [34], who reported significant and highly significant differences among genotypes (families + parents) and families for all studied traits across three segregating generations F 2 , F 3 , and F 4 .This author, also, reported that greater response to selection can be expected from selection in cross having greater phenotypic and genotypic variances.
The coefficient of variation (CV) presented values between 0.7 and 26.8% for heading date and grain yield, respectively (Table 2).The CV above 20% is considered high, indicating high dispersion of the experimental data, which may have been caused by the genetic and phenotypic differences between the studied materials.High CV estimate obtained for grain yield can be explained by the fact that it is quantitative trait, governed by several genes and highly influenced by the environment.
The variances values, coefficient of variation, and genetic parameters estimates for wheat traits studied are presented in Table 3. Broad-sense heritability is the proportion of total phenotypic variation due to all genetic effects.The knowledge of the genotypic determination coefficient (ℎ 2 bs ) allows establishing an estimate of the genetic gain to be obtained and defines the best strategy to be used in the plant breeding program [35].In this study, the estimated broad-sense heritability varied from 0.00 to 91.80%.The highest values were found for heading date (91.80%) followed by plant height (80.82%), 1000-kernel weight (72.90%), biomass (65.33%), and number of grains per spike (61.64%), indicating that these traits are highly heritable among the genotypes evaluated.These results can be confirmed with the values obtained by the CV  /CV  ratio that were close to or greater than 1 for these traits, suggesting satisfactory conditions for selection [36].Moderate estimates of ℎ 2 bs occurred for the number of chlorophyll contents (40.31%), number of spikes (52.00%), spikes weight (54.85%), and harvest index (40.67%).On the other hand, the lowest values of ℎ 2 bs were found for the flag leaf area (0.00%) and grain yield (19.39%).These traits exhibited also low CV  /CV  ratio values, indicating the dominant effect of the environment on crop.
Generally, literature indicates widely varying narrowsense heritability estimates.Mesele et al. [37] reported high heritability values for days to heading, days to maturity, and 1000-kernel weight; moderate estimates for grain filling period, spike length, number of spikelets per plant, grains per spike, and harvest index; and low values for number of tillers per plant, biomass yield and grain yield.Evaluating seven F 2 populations derived through cross combinations of five parental varieties/lines of bread wheat, Saleem et al. [38] found low to high broad-sense heritability values ranging from 4.75% to 92.6% depending on the trait and the cross.The findings of Yaqoob [39] showed that heritability estimates were low for number of tillers per plant (20%), grains per spike (26.81%), days to maturity (30.13%), spike length (36.66%), and 1000-kernel weight (38.68%), moderate for plant height (45.79%), and high for heading date (84.73%)CHL: chlorophyll content (Spad), HD: heading date (days), FLA: flag leaf area (cm 2 ), PH: plant height (cm), SN: number of spikes, TKW: 1000-kernel weight (g), SW: spikes weight (g m −2 ), GN: number of grains per spike, BIO: above ground biomass (g m −2 ), GY: Grain yield (g m −2 ), HI: harvest index (%), ℎ 2 bs : broad-sense heritability, ℎ 2 ns (SK): narrow-sense heritability according to Smith and Kinman [16], and ℎ 2 ns (FH): narrow-sense heritability according to Frey and Horner [17].and grain yield (99.83%).His results also indicated that most of these traits exhibited low heritability under drought stress conditions, suggesting the presence of high genotype × environment interactions that affected the crop behavior.
Narrow-sense heritability is the proportion of the total phenotypic variation that is due to the additive effects of genes.This component of variation is important because it is the only variation that natural selection can act on.Hence, ℎ 2 ns determines the resemblance of offspring to their parents and the population's evolutionary response to selection.There can be considerable nonadditive genetic variance, but this does not contribute to the resemblance between parents and offspring, or the response to selection.Low to moderate narrow-sense heritability values were recorded in this study.Chlorophyll content, heading date, plant height, 1000-kernel weight, and number of grains per spike recorded the highest estimates.These traits were less influenced by the environmental factors and would respond positively to a selection pressure in the current breeding program.The low heritability values can be explained by the change occurring in the segregating lines behavior from the precedent to the current generation.This change may be due to nonadditive gene action and/or high environmental factors effects.
Means of the variables measured showed that the best values varied depending on the cross and the trait, and the few populations had the best performances for several traits at the same time (Table 4).The best grain yielding population (557.5 g) was Ain Abid × El-Wifak which had also the highest average for the number of grains per spike (31.47 grains), spikes weight (857.3 g), above ground biomass (1350.3g), and harvest index (41.3%).Ain Abid × Hidhab cross combination had the longest vegetative cycle with an average of 127.8 days and presented the highest mean for the flag leaf area (21.0 cm 2 ).Acsad 1069 × El-Wifak had the highest average for the chlorophyll content (54.0 Spad).Acsad 1069 × Mahon-Demias was the tallest (75.23 cm), while Acsad 1135 × Rmada  ), GN: number of grains per spike, BIO: above ground biomass (g m −2 ), GY: grain yield (g m −2 ), and HI: harvest index (%).
exhibited the highest average for the number of spikes (503.7 spikes).For 1000-kernel weight, Acsad 1135 × Mahon-Demias was the best population with an average of 39.8 g.As the contrast "Parents versus F 4 " was significant for PH, SN, TKW, SW, BIO, and GY, and, compared with the least significant difference (LSD0.05),significant differences existed between the parents and hybrids values for these measured traits.These differences were 1 to 3 times higher than the LSD0.05 and were in favor of some F 4 lines, indicating that they perform better than the parents and suggesting the possibilities of making significant changes through effective selection.These findings were in harmony with those obtained by Abd El-Shafi [34], Löffler and Busch [40], and Alexander et al. [41].

Genetic Gain from Direct and Indirect Selection and
Selection-Based Index.The gains of direct and indirect selection are shown in Table 5.The results showed a big variation in the percentages of gains among the measured traits.The total sum of gains per selection varied from −36.11% for heading date to 34.75% for flag leaf area.Generally, the direct selection based on chlorophyll content, heading date, spikes weight, grain yield, and harvest index resulted in negative total gains.On the other hand, flag leaf area, plant height, number of spikes, 1000-kernel weight, number of grains per spike, and biomass recorded positive total gains.
The gains obtained by direct selection were higher than those of indirect selection.But, sometimes the indirect selection may be more efficient, especially if the secondary trait is highly correlated with yield and is easily measurable [42].The number of grains per spike (15.86%) followed by above ground biomass (12.30%), plant height (12.10%), and number of spikes (10.84%) exhibited the greatest gains from direct selection.Inversely, heading date (−1.13%), spikes weight (−4.99%) and harvest index (−3.81%)showed negative gain values.Negative gain for heading date is desired in the case of this study as the Algerian wheat breeding program designed for semiarid regions; precocity is a crucial criterion adopted for selection.It is related to the ability of plants to shorten their cycle, so as to decrease their exposure to the late-season sirocco weathering.
The direct selection for chlorophyll content resulted in negative indirect gains for nearly all the other traits, except for the flag leaf area, number of grains per spike, and harvest index that showed positive responses.For heading date, the indirect selection gains were only positive for chlorophyll content, number of spikes, and harvest index.The correlative effects of the selection based on the flag leaf area were desirable for grain yield, although the indirect gain for the remaining traits was practically negative.
The direct selection for plant height resulted in positive indirect gains for heading date, number of spikes, 1000-kernel weight, and spikes weight.In addition, positive responses of the selection based on yield and yield components were observed for grain yield itself, heading date, 1000-kernel weight, and spikes weight.However, negative indirect gains were exhibited for other traits including chlorophyll content, flag leaf area, and number of grains per spike, indicating that indirect selection for one variable for gain in another is unfeasible, because there will be a loss in the indirectly selected variable.In cases of negative gains, the model is considered inappropriate for selection in this plant material.
The highest correlated responses for grain yield were generated though indirect selection on the base of the flag leaf area (8.04%, i.e., 33.87 g m −2 ), followed by the spike fertility (3.65%, i.e., 15.39 g m −2 ), 1000-kernel weight (3.63%, i.e., 15.31 g m −2 ), and above ground biomass (2.74, i.e., 11.54 g m −2 ).These results indicated that the indirect selection was to be more effective in improving the primary trait than the indirect selection based on other traits and/or on the direct selection based on the grain yield itself.
Several authors have estimated the genetic gains of traits involved in yield determination.The results are often inconsistent and scarce.Our results were consistent with those of DePauw and Shebeski [43] and Inagaki et al. [44], who mentioned that direct selection on the basis of yield is ineffective in early generations.Benmahammed et al. [45] reported the same findings in barley crop.Their results showed that biomass-based direct selection appeared more discriminating than yield-based selection in their plant material.However, the results of this study do not corroborate findings reported by Lalić et al. [11], Mitchell et al. [46], Lungu et al. [47], and El-Morshidy et al. [48] and who observed the effectiveness of the direct selection on grain yield.The difference in the results may be attributed to differences of breeding material and to genotype × environments.
The gains of selection-based index are shown in Table 6.The total sum of gains from the selection-based index ranged from −9.37% to 42.86%.Five out of seven used indices showed positive total gains.The base index of Williams recorded the highest total gain (42.86%) followed by the classic index of Smith and Hazel, the index desired gains of Pesek and Baker (27.46%), and the genotype-ideotype distance index of Cruz (22.07%).The sum of ranks index of Mulamba and Mock had a very low gain of 3.74%.Subandi's and Elston's indices exhibited negative total gains.
The base index of Williams yielded positive responses for all measured traits, except for chlorophyll content, flag leaf area, and harvest index, which showed negative gains.This index also achieved the highest grain yield response (6.27%, i.e., 26.41 g m −2 ) compared to the other indices employed in this study.The classic index of Smith and Hazel ranked second followed by the index of desired gains of Pesek and Baker, the genotype-ideotype distance index proposed by Cruz, and the sum of ranks index of Mulamba and Mock with selection responses of 36.93%,27.46%, 22.07%, and 3.74%, respectively.The single-trait responses varied from one index to another with, a more or less, balanced distribution of positive and negative estimates among the traits.This shows that the monotrait selection was inadequate because it led to a higher final product when considering the grain yield and generated unfavorable responses in other traits.These results indicated that methods that combine favorable expected gains should be used in the evaluation of these progenies.The gains expected through indices for grain yield per se were larger than those obtained by direct and indirect monotrait selection, except for the flag leaf area (Tables 5 and 6).Mahdy [49] mentioned that selection-based index was predicted to be ), GN: number of grains per spike, BIO: above ground biomass (g m −2 ), GY: grain yield (g m −2 ), HI: harvest index (%),  0 : trait mean value of the base population, and   : trait mean value of the selected populations; values between parentheses correspond to the genetic gain in terms of percent mean for yield and yield-related traits.
International Journal of Agronomy ), GN: number of grains per spike, BIO: above ground biomass (g m −2 ), GY: grain yield (g m −2 ), and HI: harvest index (%); and values between parentheses correspond to the genetic gain in terms of percent mean for yield and yield-related traits.superior for yield improvement as compared to the monotrait selection method.
Ain Abid × Mahon-Demias (P17), Ain Abid × Rmada (P18), and Ain Abid × El-Wifak (P20) had positive coordinates with this axis.They were characterized by a long vegetative cycle and high biomass, spikes weight, and grain yield values.The last two populations Ain Abid × Rmada (P18) and Ain Abid × El-Wifak (P20) were also distinguished by high values of chlorophyll content, flag leaf area, spikes fertility, and harvest index, relatively to PCA2 but were associated with low number of ears, relatively to PCA3 (Figures 1 and  2).On the other hand, Ain Abid × Mahon-Demias (P17) was distinguished, relatively to PCA2, by high estimates of the 1000-kernel weight and plant height, and associated with high number of spikes, relatively to PCA3 (Figures 1 and 2).From these results, it could be concluded that effective selection of superior individuals within this population certainly contributes to the improvement of yield and yield components, in a semi-late genetic background.

Selection of Superior
Genotypes for Grain Yield.Genotypes were first ranked according to grain yield, then the 5% highest yielding lines were selected, and their mean yield estimated.The 30 lines thus selected are derived from 15 out of 20 F 4 populations studied.Half part of these lines was equitably derived from Acsad 1135 × Hidhab (P11), Ain Abid × Hidhab (P19), and Ain Abid × El-Wifak (P20) crosses.The population Ain Abid × Mahon-Demias (P17), previously identified among the promising populations, contributed with four lines.Acsad 1069 × Rmada (P14) and Ain Abid × Rmada (P18) participated with two lines each.Seven other populations contributed by one line each.Relative to the standard check Hidhab, which is the most cultivated variety in Algeria, the top 30 lines selected were characterized by significant improvements in yield components, including the number of spikes (152.33 spikes m −2 ), spikes weight (324.33 g m −2 ), spikes fertility (6.75 grains spike −1 ), above ground biomass (457.00 g m −2 ), and grain yield (316.80 g m −2 ) (Table 7).Besides, they were distinguished by significant reduction of the duration in the vegetative phase with 3.10 days (Table 7).

Conclusion
The results of this study indicated appreciable genetic variability among the evaluated populations.Selection based-index was more efficient to improve grain yield compared to direct International Journal of Agronomy Williams's index was predicted to be more effective than the other selection indices for improving multiple traits at time.It brought the highest total genetic gain and the best yield gain per se associated with positive correlated responses for most of yield components.Compared to the check cultivar Hidhab, the 30 F 4 selected lines, at 5% selection intensity, were characterized by significant increase in grain yield and yield related traits.

Figure 1 :
Figure 1: Principal Component Analysis (PCA) projections on axes 1 and 2, accounting for 63.56% of total variation, for 20 F 4 populations of bread wheat.

Figure 2 :
Figure 2: Principal Component Analysis (PCA) projections on axes 1 and 3, accounting for 53.65% of total variation, for 20 F 4 populations of bread wheat.

Table 1 :
The skeleton of the analysis of variance.

Table 3 :
Genetic and nongenetic parameters of different bread wheat traits studied.

Table 4 :
Means of the measured traits for 20 F

Table 5 :
Selection gain estimates (GS%) obtained from direct (diagonal) and indirect (of diagonal) selection in different bread wheat traits.

Table 6 :
Selection gain estimates (GS%) obtained from selection-based index in different bread wheat traits.
CHL* Significant effect at 0.05 probability.