Crop Science 41:358-362 (2001)
© 2001 Crop Science Society of America
CROP BREEDING, GENETICS & CYTOLOGY
Family and Line Selection for Seed Yield of Soybean
Leon G. Streita,
Walter R. Fehrb and
Grace A. Welkeb
a Dep. of Research and Product Development, Pioneer Hi-Bred International, Inc., Johnston, IA 50131
b Dep. of Agronomy, Iowa State Univ., Ames, IA 50011
Corresponding author (wfehr{at}iastate.edu)
 |
ABSTRACT
|
|---|
Plant-row-yield tests (PRYT) are used by soybean [Glycine max (L.) Merr.] breeders for the initial evaluation of experimental lines. The highest yielding lines in the PRYT are advanced for additional testing in replicated tests. The objective of this study was to determine the reliability of selection for seed yield in unreplicated plots by the family and line methods of selection. Four F3-derived lines from each of 21 F2 families from four populations were grown in a PRYT during 1995 and in replicated tests at four environments in 1996. For the family method, the mean seed yield of the four F3-derived lines of each F2 family was used to identify superior families from which to select individual lines. For the line method, lines were selected without regard to the family structure. The seed yield of the selected and unselected lines on the basis of data from the PRYT was compared with their mean seed yield in the 1996 environments. The total number of lines selected by the family method was less than for the line method in all populations. The percentage of selected lines that were correctly classified was similar for both methods. There was a greater percentage of lines incorrectly rejected by the family method than by the line method. The use of replication at an individual location did not improve the selection of lines by the family or line methods of selection. For the selection of lines for seed yield in unreplicated plots, breeding methods that rely on family performance would not be more effective or efficient than methods that ignore family structure. To obtain lines for yield tests at multiple locations, selection of lines by the line method on the basis of their performance in a PRYT would be better than the use of random lines.
Abbreviations: PRYT, plant-row-yield tests
|
INTRODUCTION
|
|---|
SELECTION FOR SEED YIELD is one of the most important and difficult challenges of plant breeding. Family or line methods of selection can be used to identify superior genotypes for seed yield in a cultivar development program. For the family method, the mean performance of lines that trace to the same F2 plant is determined, and individual lines are chosen within the superior families. For the line method, family structure is ignored when selecting individual lines. Family structure is considered for the pedigree and early-generation-testing methods and is ignored for the bulk and single-seed-descent methods (Fehr, 1987). Falconer (1960) suggested that family selection is favored when the heritability of a trait is low and the number of families is large. Bravo et al. (1999) and Streit et al. (2001) reported that the family method was not more effective than the line method for selection of genotypes with elevated palmitate, or with reduced palmitate, palmitate + stearate, or linolenate in the seed oil of soybean. Other comparisons of breeding methods have been made in soybean; however, they did not involve an assessment of the value of maintaining family structure when selecting lines from unreplicated yield tests (Raeber and Weber, 1953; Voigt and Weber, 1960; Empig and Fehr, 1971; Boerma and Cooper, 1975; Luedders et al., 1973; Ivers and Fehr, 1978).
The initial yield evaluation of lines from soybean populations is commonly done in unreplicated PRYT. Hegstad et al. (1999) concluded that PRYT would be effective for identification of elite soybean lines. Byrum (1999) reported positive, but nonsignificant (P > 0.05), correlations between PRYT and replicated tests. The objective of this study was to compare the reliability of selection for seed yield in unreplicated plots by the family and line methods.
|
MATERIALS AND METHODS
|
|---|
Four populations were developed for this study. The parent lines YA7343Z006 and AX8154A370 with the fap1 fap1 fap3 fap3 genotype for reduced palmitate content and the parent cultivars 9282 and 9322 with the fan1(A5) fan1(A5) fan2 fan2 genotype for reduced linolenate were selected for high yield from 1993 yield tests. Each of the parents was crossed to multiple F2:3 lines with reduced palmitate and linolenate that were selected as individual F2 plants from segregating populations. The purpose of the crosses was to develop high-yielding cultivars with reduced palmitate and linolenate. The F2:3 lines were selected from crosses of lines with reduced saturates to lines with reduced linolenate in order to combine the two traits. The parentage of the F2:3 lines is too complex to report here because of the multiple generations of crossing and selection used to transfer the reduced saturates and reduced linolenate traits into agronomically desirable backgrounds. The crosses to form the four populations were made in March 1994 at the Iowa State University-University of Puerto Rico soybean breeding nursery at Isabela, Puerto Rico. The soil type was Coto clay (Very-fine, koalinitic, isohyperthermic, Typic Haplorthox). The crosses of F2:3 lines with 9282 were collectively designated AX11056, with 9322 were AX11063, with YA7343Z006 were AX11080, and with AX8154A370 were AX11104. The F1 seeds were planted in Puerto Rico in May 1994, and each F1 plant was harvested and threshed individually. Within each cross, F2 seeds from confirmed hybrid F1 plants were bulked. In October 1994, F2 seeds selected for reduced palmitate, stearate, and linolenate were planted in Puerto Rico; the F2 plants were harvested individually; seed from each was analyzed for fatty ester composition; and the 50 plants with the least palmitate, stearate, and linolenate were selected. The F3 progeny of the selected F2 plants were planted as individual rows in Puerto Rico in January 1995. Within each F2 family, four random F3 plants from each family were harvested individually.
The four F3:4 lines of the 50 F2 families of each population were grown in a PRYT at Johnston, IA, during the summer of 1995. The experiment for each population also included 12 check lines and cultivars. The 212 entries in each experiment were subdivided into 53 blocks. The four F3-derived lines from the same F2 family were in the same block, and the checks were randomly assigned to three blocks. Blocks and entries within blocks were randomized in each replication. The four experiments were planted as a randomized complete-block design. The soil type was a Waukegan loam (Fine-loamy, over sand or sandy skeletal, mixed, superactive, mesic Typic Hapudoll). A plot was a single row 108-cm long, with a 77-cm spacing between rows and a 92-cm alley between the ends of plots. The seeding rate was 33 seeds m-1 of row. At maturity, each F3:4 line was harvested individually with a self-propelled combine, the weight and moisture of the grain were measured, and the seed yield was calculated in kg ha-1 on a 13%-moisture basis.
The number of lines in the PRYT exceeded resources available for replicated tests in 1996. Therefore, 21 of the 50 F2 families from each population were selected at random without regard to the results of the PRYT. Each population was grown as a separate experiment that consisted of four random F3:5 lines from the 21 F2 families and 12 check lines and cultivars. The 96 entries in each experiment were subdivided into 24 blocks. The four F3-derived lines from the same F2 family were in the same block, and the checks were randomly assigned to three blocks. Blocks and entries within blocks were randomized in each replication. The four experiments were planted as a randomized complete-block design with two replications at Ames, Atlantic, and Washington, IA, and Bethany, MO, in 1996. The soil types were a Nicollet loam (Fine-loamy, mixed, mesic Aquic Hapludoll) at Ames, a Marshall silt loam (Fine-silty, mixed, superactive, mesic Typic Hapludoll) at Atlantic, a Mahaska silty clay loam (Fine, smetitic Aquertic Argiudoll) at Washington, and a Haig silt loam (Fine, smetitic, mesic Vertic Argiaquoll) at Bethany. A plot consisted of paired rows 3.7 m long, with a 77-cm row spacing and a 92-cm alley between the ends of plots. The seeding rate was 31 seeds m-1 of row. Each plot was harvested with a self-propelled combine, the weight and moisture of the grain were measured, and the seed yield was calculated in kg ha-1 on a 13%-moisture basis.
For comparison of the family and line methods, selection was practiced for 
87% seed yield of the mean of the 10 check lines and cultivars that were common to all the experiments. The criterion for seed yield was chosen so that 
50% of the lines would be selected when averaged across populations and selection environments. For the family method, the mean seed yield of the four F3-derived lines within an F2 family was determined. Within families that had a seed yield of 87% of the checks, lines with 87% seed yield were selected. For the line method, individual lines with 87% seed yield were selected without regard to the family performance.
Selection for seed yield by the family and line methods was conducted on the basis of the yield from the PRYT in 1995 and on the basis of the yield of the individual replications at the four locations in 1996. To compare the usefulness of replication at a single location, selection by the family and line methods also was based on the mean of the two replications at the 1996 locations. The PRYT performance was compared with the mean seed yield of the lines at the four locations in 1996. Selection practiced at one location in 1996 was compared with the mean seed yield of the lines at the other three locations in 1996.
Acceptance and rejection errors were calculated for both methods of selection. Acceptance error occurred when lines were chosen with 87% seed yield based on the PRYT in 1995 or based on one or two replications of testing at one location in 1996, but the lines did not have 87% seed yield based on the mean of the other environments. Rejection error occurred when lines were not chosen because they had <87% seed yield based on PRYT in 1995 or based on one or two replications of testing at one location in 1996, but the lines had 87% seed yield based on the mean of the other environments.
The rankings of lines in the PRYT and in one or two replications at one location were compared with their rankings for seed yield in the other environments. Selection intensities required to retain the highest yielding line or one or more of the 10 highest yielding lines based on the mean performance in 1996 were determined.
The data from each experiment were analyzed as a randomized complete-block design for individual locations and across locations. All variables in the analysis of variance were considered random effects. The analyses of variance were performed using the general linear model procedure of the SAS software package (release 6.12) (SAS Institute, 1992). Variance components, heritability estimates, and their standard errors for each population were calculated from the combined analyses of variance across locations (Hallauer and Miranda, 1981). Phenotypic correlation coefficients were calculated between the PRYT and the mean of the 1996 tests at individual locations and the mean of all locations by PROC CORR of the SAS software package (SAS Institute, 1992). Chi-square analysis was used to determine if selection from the PRYT and from one or two replications at one location was better than random selection of lines.
|
RESULTS AND DISCUSSION
|
|---|
There were significant (P < 0.05) differences among families and lines for seed yield in the four populations. The range in seed yield among families averaged across locations was 2336 to 3284 kg ha-1 for AX11056, 1663 to 3554 kg ha-1 for AX11063, 2039 to 3367 kg ha-1 for AX11080, and 2554 to 3206 kg ha-1 for AX11104. The range among lines was 2061 to 3458 kg ha-1 for AX11056, 1071 to 3753 kg ha-1 for AX11063, 1650 to 3858 kg ha-1 for AX11080, and 1863 to 3509 kg ha-1 for AX11104.
The broad-sense heritability estimates for seed yield on a plot basis averaged 0.52 across populations (Table 1). The lower heritability estimate for AX11056 was associated with less variability for seed yield among genotypes than in the other three populations. The heritability estimates were comparable to those reported in other studies. Garland and Fehr (1981) obtained an average heritability estimate of 0.34 on a plot basis for lines from five intermated populations. Hegstad et al. (1999) reported a heritability estimate of 0.48 for seed yield on a plot basis averaged across five populations.
View this table:
[in this window]
[in a new window]
|
Table 1. Variance component and heritability estimates and their standard errors for seed yield of 84 F3-derived soybean lines from four populations evaluated in four environments in 1996
|
|
Phenotypic correlation coefficients for seed yield between PRYT and the mean of the 1996 tests were positive and significant (P < 0.05) for all populations. The coefficient was 0.30 for AX11056, 0.43 for AX11063, 0.38 for AX11080, and 0.53 for AX11104. Byrum (1999) reported nonsignificant (P > 0.05) phenotypic correlation coefficients of 0.10, 0.25, and 0.37 between PRYT and replicated plots for three populations. Hegstad et al. (1999) reported coefficients ranging from -0.04 to 0.41 with a mean of 0.17 between PRYT and replicated tests for five populations. The correlations for only two of their populations were significant at the 0.05 probability level.
The number of lines selected by the family method was less than for the line method in all populations because a line could not be chosen if it was in a family with <87% seed yield of the checks (Tables 2 and 3). For selection in the PRYT, the average percentage of lines selected for seed yield was 37% by the family and 49% by the line method. In the 1996 tests, the mean percentage of lines selected for seed yield based on individual replications was 49% by the family and 58% by the line method, and based on the mean of replications was 51% by the family and 63% by the line method.
View this table:
[in this window]
[in a new window]
|
Table 2. Errors associated with selection by the family and line methods for 87% seed yield of check genotypes among four F3-derived soybean lines from each of 21 familes in four soybean populations from plant-row-yield tests in 1995
|
|
View this table:
[in this window]
[in a new window]
|
Table 3. Errors associated with selection by the family and line methods for 87% seed yield of check genotypes among four F3-derived lines from each of 21 familes for each of four soybean populations averaged across four selection environments in 1996
|
|
Even though fewer lines were chosen by the family method, the acceptance error was similar to the line method in the PRYT and in the 1996 tests. For selection in the PRYT, the mean percentage of lines selected without error was 80% by the family and 79% by the line method. In the 1996 tests, the mean percentage of lines selected without error was 81% by the family and 84% by the line method.
The frequency of rejection error was higher for the family than the line method. The mean percentage of lines incorrectly rejected in the PRYT was 57% by the family and 41% by the line method. The average percentage of lines incorrectly rejected based on individual replications in the 1996 tests was 39% by the family and 26% by the line method.
The results indicated no advantage in maintaining family structure for selection of seed yield in either the PRYT or the individual replications in 1996. Bravo et al. (1999) and Streit et al. (2001) reported that the family method also was not more effective than the line method for selection of elevated palmitate or reduced palmitate, palmitate + stearate, or linolenate in soybean.
A breeder would like to advance the least percentage of lines from the initial evaluation to subsequent yield tests in multiple environments. The percentage of lines that had to be selected in the PRYT to retain the highest yielding line in the replicated tests ranged from 2 to 35% for the four populations, with an average of 14% (Table 4). Byrum (1999) reported that the selection intensities required in the PRYT to retain the highest yielding line in three populations were 16, 38, and 61%. In our study, an average selection intensity of only 3% was required in the PRYT to retain one of the 10 highest yielding lines in the replicated tests, but a mean selection intensity of 61% was required to retain all of the top 10 lines. Selection intensities required to retain the highest yielding lines in the PRYT were similar to those for the use of one or two replications at individual locations in 1996 (Table 5).
View this table:
[in this window]
[in a new window]
|
Table 4. Selection intensity for seed yield required in the 1995 plant-row-yield test to recover the 10 out of 84 F3-derived soybean lines with the highest mean yields averaged across four 1996 environments
|
|
View this table:
[in this window]
[in a new window]
|
Table 5. Selection intensity for seed yield required in the 1995 plant-row-yield test and in one or two replications of the 1996 yield tests to recover the highest yielding 10 out of 84 F3-derived soybean lines averaged across four populations
|
|
Similar selection intensities were required for selecting one or more of the top 10 lines from individual replications at a location compared with the mean of the two replications (Table 5). On the average, replication did not reduce the selection intensity needed to recover the best line. At Ames, use of the replication mean improved selection intensity, at Bethany, it was intermediate to individual replications, and at Washington and Atlantic, it was worse than individual replications. The use of replication also did not improve the selection of lines by the family or line methods of selection (Table 3). Instead of replicating lines at a location, it would be more beneficial to grow more lines in a single replication or to grow one replication at two or more environments as insurance against the loss of a location and to evaluate genotype x environment interaction.
Selection for seed yield by the line method on the basis of the PRYT and on the basis of each of the two replications or the replication mean at each of the 1996 locations was compared with random selection (Table 6). To determine the expected number of random lines with 87% seed yield, the number of lines with 87% seed yield based on their mean across the four locations in 1996 was divided by 84, which was the number of lines tested from the population. That fraction was multiplied by the total number of lines selected for additional testing based on the PRYT and based on each of the two replications or the replication mean at each of the four 1996 locations. For example, there were 60 out of 84 lines (0.71) from AX11056 that had 87% seed yield averaged across the four locations in 1996. There were 37 lines selected from the PRYT, of which 33 had 87% mean seed yield in 1996. If 37 random lines from the population had been tested, 26 of them (37 x 0.71) would have been expected to yield 87%. Selection based on PRYT was better than random selection in the four populations, although the difference was not significant (P > 0.05) on the basis of the Chi-square test (Table 6). An average of 34 of the 41 lines that were selected from the PRYT had 87% mean seed yield based on the 1996 tests. If the 41 lines had been selected at random from the populations, an average of 28 of the lines would be expected to have had 87% mean seed yield in the 1996 tests. For selection conducted at individual locations in 1996, the use of data from individual replications or a replication mean was superior (P = 0.01) to random selection. An average of eight more lines with 87% mean seed yield were chosen based on data from individual replications or replication means than for random lines from the populations. The smaller plot size used for the PRYT may be partly responsible for the difference in the effectiveness of selection in the PRYT compared with the 1996 tests.
View this table:
[in this window]
[in a new window]
|
Table 6. Number of lines selected by the line method based on the 1995 plant-row-yield test and the mean number selected at each of four 1996 locations based on each of the two replications or the mean of the replications compared with random selection in four soybean populations
|
|
The breeder would have to decide if selection based on a PRYT is sufficiently superior to random selection to justify the cost of conducting the test. If a PRYT is not used, a progeny row would have to be grown to obtain enough seed of a line for replicated testing. The additional costs of the PRYT compared with a seed increase include the time required to set up the yield test, weigh the seed that is harvested, and make the selections. More lines would have to be grown for the PRYT than for the seed increase because only part of the lines would be selected. This would result in additional costs for growing, harvesting, threshing, and packaging seed of individual plants, and for the greater land and labor required to grow a larger number of plots for the PRYT than for the seed increase. These additional costs would be particularly important to consider if the PRYT was grown in another region during the winter where land use, shipping, and other costs would be a consideration.
|
NOTES
|
|---|
Journal Paper No. J-18837 of the Iowa Agric. and Home Econ. Exp. Stn., Ames; Project No. 3107, and supported by the Hatch Act, State of Iowa, and Pioneer Hi-Bred International, Inc.
Received for publication May 22, 2000.
|
REFERENCES
|
|---|
- Boerma, H.R., and R.L. Cooper. 1975. Comparison of three selection procedures for yield in soybeans. Crop Sci. 15:225–229.[Abstract/Free Full Text]
- Bravo, J.J., W.R. Fehr, G.A. Welke, E.G. Hammond, and S.R. Cianzio. 1999. Family and line selection for elevated palmitate in soybean. Crop Sci. 39:679–682.[Abstract/Free Full Text]
- Byrum, J.R. 1999. Selection for soybean seed yield with molecular markers. Ph.D. diss. (Diss. Abstr. 99-40185), Iowa State Univ., Ames, IA.
- Empig, L.T., and W.R. Fehr. 1971. Evaluation of methods for generation advance in bulk hybrid soybean populations. Crop Sci. 11:51–54.
- Falconer, D.S. 1960. Introduction to quantitative genetics. Ronald Press Co., New York.
- Fehr, W.R. 1987. Principles of cultivar development: Theory and technique. Macmillian Publ., New York.
- Garland, M.J., and W.R. Fehr. 1981. Selection for agronomic traits in hill and row plots of soybean. Crop Sci. 21:591–595.[Abstract/Free Full Text]
- Hallauer, A.R., and F. Miranda. 1981. Quantitative genetics in maize breeding. Iowa State Univ. Press, Ames, IA.
- Hegstad, J.M., G. Bollero, and C.D. Nickell. 1999. Potential of using plant row yield trials to predict soybean yield. Crop Sci. 39: 1671–1675.[Abstract/Free Full Text]
- Ivers, D.R., and W.R. Fehr. 1978. Evaluation of the pure-line family method for cultivar development. Crop Sci. 18:541–544.
- Luedders, V.D., L.A. Duclos, and A.L. Matson. 1973. Bulk, pedigree, and early generation testing breeding methods compared in soybeans. Crop Sci. 13:363–364.
- Raeber, J.G., and C.R. Weber. 1953. Effectiveness of selection for yield in soybean crosses by bulk and pedigree systems of breeding. Agron. J. 45:362–366.[Free Full Text]
- SAS Institute. 1992. SAS guide for personal computers. 6th ed. SAS Inst., Cary, NC.
- Streit, L.G., W.R. Fehr, G.A. Welke, E.G. Hammond, and S.R. Cianzio. 2001. Family and line selection for reduced palmitate, saturates, and linolenate of soybean. Crop Sci. 41:63–67.[Abstract/Free Full Text]
- Voigt, R.L., and C.R. Weber. 1960. Effectiveness of selection methods for yield in soybean crosses. Agron. J. 52:527–530.[Abstract/Free Full Text]
This article has been cited by other articles:
|
|
|
|
 
T. C. Helms, J. H. Orf, and J. T. Terpstra
Resource Allocation to Select for Yield in Soybean
Crop Sci.,
September 1, 2002;
42(5):
1493 - 1497.
[Abstract]
[Full Text]
[PDF]
|
|
|
TOP
ABSTRACT
INTRODUCTION
