Crop Science 40:1559-1564 (2000)
© 2000 Crop Science Society of America
CROP BREEDING, GENETICS & CYTOLOGY
Genetic Gain in Early Stages of a Soybean Breeding Program
S.K.St. Martina and
Xie Futib
a Dep. of Horticulture and Crop Science, Ohio Agric. Res. and Dev. Ctr., The Ohio State Univ., Columbus, OH 43210-1086 USA
b Dep. of Agronomy, Shenyang Agric. Univ., Shenyang 110161, People's Republic of China
stmartin+{at}osu.edu
 |
ABSTRACT
|
|---|
Multistage testing of genotypes is an essential feature in plant breeding programs. Little research is available to help breeders test efficiently. Our objective was to determine the genetic gain obtained in selected stages of a soybean [Glycine max (L.) Merr.] breeding program and to determine ways to increase the overall gain, if possible. The mean selection differential and genetic gain and their regression coefficient were determined for tests at the F3, F4, and F6 generations of the Ohio State University-Ohio Agricultural Research and Development Center breeding program during 1985 to 1997. Genetic gain for yield averaged -1.4% in the F3 stage, 3.7% in the F4, and 9.1% in the F6 stage, where percentages are based on the mean of the common check genotypes. Overall changes in maturity were small at each stage. Improvements in lodging resistance occurred in the F4 and F6. Negative yield gains in the F3 were attributed to use of unreplicated plots and to the necessary selection pressure for early maturity. Examination of the selection differentials for individual lines indicated that selection could be intensified in the F4 and F6 generations with little risk of discarding potentially superior cultivars. The conclusion that near equality of the selection intensity across stages would be beneficial for this breeding program confirms the theoretical recommendations reported in earlier research.
|
INTRODUCTION
|
|---|
TESTING OF GENOTYPES in an adequate sample of production environments is probably the most expensive feature in most plant breeding programs. Relatively little published research is available, however, to guide breeders toward optimum allocation of resources in multistage testing programs. The term multistage refers to the cyclical (usually annual) process of testing selections, discarding inferior genotypes, and retesting the superior selections. Normally, for annual field crops, a stage corresponds to a generation.
Finney (1966) used computer simulation to examine the effect on total genetic gain of such variables as the number of stages and the selection intensity at each stage. On the basis of simulations covering a broad range of selection strategies, Finney (1966) recommended that similar selection intensities (i.e., the percentage of genotypes retained) should be used in each stage. Curnow (1961) reached a similar conclusion. Simplifying assumptions, such as the absence of genotype x stage interaction, in these earlier publications suggest that it would be helpful to confirm this recommendation with empirical data.
St Martin and McBlain (1991) described a procedure in which historical test data are used to estimate genetic gain in each stage. Applying their method of retrospective analysis to 29 yr of regional test data for soybean, they were able to make recommendations about the tests.
Our objective was to determine the genetic gain obtained in selected stages of a soybean breeding program and to determine ways to increase the overall gain, if possible.
|
Materials and methods
|
|---|
Description of Breeding Program
We used data from the soybean breeding program conducted at the Ohio Agricultural Research and Development Center from 1985 to 1997 to perform the analysis of gain by stages. The primary goal of the program was development of high-yielding lines for use as commodity-type and food-grade cultivars. Parents for these populations were primarily cultivars and elite lines chosen for yield, disease resistance, seed protein content, and seed size. In addition, some tests were devoted to the evaluation of lines carrying exotic germplasm with the goal of identifying lines that might contribute previously unexploited genes to future populations.
We used the F3, F4, and F6 stages (Table 1)
in the analysis. In the F3 stage, the progeny of individual F2 plants were tested in single-row plots approximately 1.5 m long (St Martin et al., 1990). In most cases, these tests of F2:3 lines were unreplicated, but some tests had two replications, one at each of two locations. Most tests contained 83 F2:3 lines and 7 check genotypes, for a total of 90 randomized entries. We avoided wide ranges of maturity within a test by classifying F2 plants as early, medium, or late maturity and assigning to each test progeny from plants of a single maturity class. The experimental lines in each test generally derived from numerous crosses. There were 4 to 26 different tests of F2:3 lines annually. Maturity was recorded for each plot as the date when 95% of the pods had reached their mature color. All plots were harvested with a plot combine, and yield was recorded as the weight of air-dry seed per plot at approximately 90 g kg-1 moisture content.
View this table:
[in this window]
[in a new window]
|
Table 1 Overview of the soybean breeding program at Ohio State University-Ohio Agricultural Research and Development Center; genetic gain estimates were obtained for stages in italic type
|
|
We made selections from the F3 stage by considering the performance of both crosses and individual F2:3 lines for maturity and yield. Selection was subjective. Crosses with high mean yields and early maturity were identified first, then superior lines were selected, primarily from superior crosses. From 1993 to 1997, best linear unbiased prediction was used to help identify superior crosses (Panter and Allen, 1995).
In 1990, visual selection was used to identify superior F2:3 lines, and undesirable lines were discarded prior to harvest. This made it impossible to obtain an unbiased estimate of the mean of the unselected population, and therefore data from the F3 stage in 1990 were excluded from the analysis.
In the F4 stage, we tested the selected F2-derived lines in replicated, multiple-row plots. Until 1992, plots consisted of two rows, spaced 76 cm apart. Beginning in 1992, row spacing was reduced to 38 cm and there were three rows per plot. Plots were planted to a length of 4.5 to 5 m, depending upon the year and location, and end-trimmed after physiological maturity to a length of 3 m. Each selection from the F3 tests was assigned to one of four F4 tests, depending on maturity (early, i.e., maturity group II or early maturity group III, or late, i.e., late maturity group III or early maturity group IV) and breeding objective. Early and late F4 tests of lines intended for use as commodity cultivars were conducted at three locations; early and late tests of grain-type or exotic lines were conducted at two locations. There were 25 to 100 entries per test, including 3 to 5 check genotypes. Maturity and yield were measured on each plot as in the F3 tests, and lodging score was also recorded on a scale of 1 (erect) to 5 (prostrate). Selection of superior lines from the F4 test was based on yield, maturity, lodging score, and, in some cases, grain protein content, seed size, and disease resistance.
Selected lines from the F4 tests were continued for the following year in the same test with newly selected F2:3 lines. Thus, F2:4 and F2:5 selections were in a common test. Handling of selections in the F5 generation was complicated, and therefore we did not attempt in this paper to assess the genetic gain made in the F5. While continuing to test selected F2:5 lines, we also grew F4:5 progeny from these lines in a separate nursery, using unreplicated short-row plots. Performance in the F2:5 tests was the primary selection criterion, with a small amount of attention to the yield, maturity, and disease resistance of the individual F4:5 lines. We saved only F4:5 lines for continued testing, discarding all F2-derived lines after the F5 generation.
In assessing the F6 generation for this paper, we used only data from commodity-type lines, because, unlike food-grade and germplasm lines, they were handled in the same way throughout the period of 1985 to 1997. We tested F4:6 lines in two- or three-row plots, identical to those used in the F4 stage, at three locations, with two replications per location. We selected lines primarily on the basis of yield, maturity, and lodging, informally considering F2 family means along with line performance per se. In the F7 to F9, selections were tested statewide at four to six locations in bordered row plots with two or (usually) three replications per location. Row length and spacing varied in the statewide tests.
Overall selection intensity at the F3 stage for the years covered by this study was 10.8% (1326 lines selected from 12317 lines tested). Selection intensity was 21.0% at the F4 stage and 18.3% at the F6 stage.
Description of Methods of Analysis
We estimated the mean and standard error of the selection differential and genetic gain for each stage using the procedure of St Martin and McBlain (1991). Briefly, selection differential (I) was obtained as
I = 
s -
, where 
s is the mean of selected experimental lines and is the mean of all experimental lines in the test. Genetic gain (G) for a test was estimated by means of the information from both the test itself and its successor (a successor of a test is a test, generally in the following year, in which selections from the test are entered for further testing):
G = 
- 
, where s' is the mean in the successor test of strains that were retained from the original test, c' = the mean of the check genotypes in the successor test, and c mean of the check genotypes in the original test. The values c' and c were calculated only from check genotypes that were common between the test and its successor. The regression coefficient H, which represents the proportion of phenotypic superiority that is realized as genetic gain, was obtained by regressing G on I, subject to the constraint that the regression line must pass through the origin. In these expressions, yield was expressed as a percentage of the mean of the common check genotypes, maturity as number of days after 31 August when 95% of pods had reached their mature color, and lodging on a scale of 1 (erect) to 5 (prostrate).
For a given test, we obtained as many estimates of selection differential (I) and genetic gain (G) as there were successors to the test. For example, we obtained three estimates for an F3 test that provided selections that were assigned to three different F4 tests (e.g., F4 early commodity, F4 late commodity, and F4 early food-grade). Mean values of I and G, and the estimate of H (regression of G on I) were based on 294 estimates from the F3 stage, 38 from the F4 and 30 from the F6. In addition to the overall mean value for each stage, we also calculated mean values for each type of successor test (i.e., early or late, commodity-type or food-grade–exotic).
To investigate the consequences of changing the selection intensity at any stage, we calculated a selection differential for maturity and yield in the F3 and F4 for each individual line that produced one or more F4-derived selections that reached the F7 stage. We also calculated such selection differentials for the F6 test of these successful selections. The degree of success of each F4-derived line was noted by the number of years (1, 2, 3, or more) that it was retained in statewide tests. Lines retained more than 3 yr were released as cultivars.
|
Results and discussion
|
|---|
Despite an average selection differential of 21.8% for yield, mean genetic gain in the F3 stage was negative, -1.4% (Table 2) . In addition, selections became slightly later (gain 0.49 d) despite selection pressure for earliness
. The regression coefficient (H) of G on I was nearly zero for yield (0.02) but relatively large for maturity (0.73). The early successor tests, both commodity-type and food-grade–exotic, showed strong negative gains for yield (Table 2), as the earliest selections were placed in these successor tests. These early selections were generally lower yielding than the late selections, as shown by the smaller selection differentials for the two early successor tests. Gain for yield was positive (6.5%) in the late commodity test and near zero in the late food-grade–exotic test (Table 2).
View this table:
[in this window]
[in a new window]
|
Table 2 Selection differentials (1), genetic gain (G), and regression (H) of G on I for tests conducted in the F3 stage of a soybean breeding program
|
|
Yield gain averaged 3.7 ± 2.4% in the F4 stage (Table 3) , while change in maturity was only 0.30 ± 0.40 d. There was a small improvement in lodging score (-0.08 ± 0.06 units). Selection differentials for maturity and yield were in the same direction as in the F3 but about half the magnitude. Values of H were moderately high for maturity and lodging and, at 0.40, moderate for yield. Among individual successor tests, the early commodity test showed little improvement in any trait, and had near-zero H values for yield and maturity. The largest yield gains were obtained in the two food-grade–exotic tests, which also had the highest selection differentials.
View this table:
[in this window]
[in a new window]
|
Table 3 Selection differentials (I), genetic gain (G), and regression (H) of G on I for tests conducted in the F4 stage of a soybean breeding program
|
|
Commodity and food-grade–exotic tests had possibly different response trends in the F3 and F4. The commodity tests had greater yield gain in the F3 than the food-grade–exotic tests, but the reverse trend occurred in the F4. It is possible that selection for traits important to food-grade cultivars, such as seed size and protein content, reduced yield gain in the F3. If so, greater genetic variability for yield may have been passed along to the food-grade–exotic F4 tests, enhancing the opportunity for yield gain in the F4.
Smaller selection differentials for the F4 stage, compared to the F3, were due to smaller phenotypic variances, because of the increased replication. As in the F3, the correlation between maturity and yield was evident. In the F4, all tests had a negative selection differential for maturity, indicating continued selection pressure for earliness.
In the F6 stage, relatively large gains were evident for yield in both early and late tests (Table 4)
, with an average gain of 9.1 ± 1.0%. There was also significant improvement in lodging score, particularly in the late test, while overall changes in maturity were small. In the late test, selection pressure was for earliness (mean
I=-0.87 days
), seemingly at the expense of yield (mean
I=-0.1%
), but significant gains for yield occurred nonetheless. Values of H were close to 1.0 for both maturity and lodging; the value for yield was moderate (0.37 ± 0.16). The relatively small standard errors for yield gains in the F6 indicate a low variance of response rather than a large number of tests sampled.
View this table:
[in this window]
[in a new window]
|
Table 4 Selection differentials (I), genetic gain (G), and regression (H) of G on I for tests conducted in the F6 stage of a soybean breeding program
|
|
Necessarily, selection in breeding programs is partly subjective. Thus, breeding programs probably vary in the degree to which they approach optimum allocation of testing resources. Any breeder, therefore, can potentially benefit from a retrospective analysis like ours, but the conclusions and benefits may be variable from one breeder to another.
The near-zero or negative gains for yield in the F3 stage are somewhat disturbing, but perhaps these are approximately what should be expected of selection based on one or two replications. The positive correlation between maturity and yield (Burton, 1987) is also an important factor in these results. Strong selection pressure for earliness must be combined with selection for yield to retain the potential to develop high-yielding, early cultivars. Where lines were to be placed in early successor tests, the selection pressure for earliness was so strong that yield was reduced; for later successor tests, selection for early maturity was relaxed. The low H values reflect the result of using a selection criterion that combined two traits of high (maturity) and low (yield) heritability. St Martin et al. (1990) presented estimates of variances and covariances for yield and maturity from tests of random F2-derived lines in short-row plots. In their study, the genotypic correlation for maturity (measured in short-row plots) and yield was 0.69. If their values are used to predict genetic gain for a selection index designed to hold maturity constant while increasing yield, the result is a negative yield response of approximately 1%, similar to the direction and magnitude of yield response we observed (Table 2). Soybean breeders attempting to develop cultivars in a range of maturity groups are familiar with the need to select for early maturity to retain potential early cultivars. The breeding program studied here has developed successful cultivars in maturity groups II, III, and IV. Relaxation of selection pressure for earliness, although it would likely enhance gain in yield, seems inadvisable.
In the two late successor tests, the greater selection differential (30.0%) for yield in the commodity-type test may account for the larger yield gain achieved in that test than in the food-grade–exotic test, where the selection differential was only 17.5%.
Of the three stages we studied, most of the genetic gain for yield derived from the F6 stage. Thus, even in a program based on early generation testing, selection at the stage of homozygous lines produced the majority of the gain for yield. In the absence of comparative data, it is impossible to compare these results with those from programs based on alternative methods, such as the single-seed descent or the pedigree methods.
Each of the F4 and F6 stages produced gains of approximately -0.1 point in lodging score, indicating improved lodging resistance. The magnitude of genetic changes in maturity was less than 1 d in each stage.
Efforts to increase the overall genetic gain in a program should entail more than one stage. Gains in different stages are mutually dependent because selections from one stage pass to the next. Thus, it is not always clear how changing the gain at a single stage affects the overall gain. Curnow (1961) and Finney (1966) suggested that optimal overall gain was attained by establishing approximately equal selection intensities at each stage. Our program featured more intense selection at the F3 stage (10.8%) than in the F4 (21.0%) or F6 (18.3%).
We investigated the consequences of intensifying selection in the F4 and F6 to match that of the F3 by examining selection differentials of individual lines. The selection differential of an individual line is simply the difference between the mean performance of the line and the mean of all experimental lines in the test. The results (Fig. 1, 2, and 3)
suggest that more intense selection at the F4 and F6 stages (indicated by the subjectively drawn lines at the bottom of Fig. 2 and 3) would not have led to the discarding of truly superior lines. In contrast, the F3 selection differentials (Fig. 1) reveal little scope for additional discards. Increased selection intensity in the F4 and F6 would increase the efficiency of the breeding program by eliminating inferior material at the F4 or F6 stages rather than the F7. It would also bring selection intensity closer to equality among the stages, as recommended by Curnow (1961) and Finney (1966). Thus, our empirical results seem to validate the recommendations arising from the earlier theoretical work.
View larger version (13K):
[in this window]
[in a new window]
|
Fig. 1 Selection differential in the F3 stage for maturity and yield of F2-derived soybean lines from which homozygous selections survived to the F7 stage. Shape of point indicates number of years of testing in statewide tests (cultivar was released after 3 yr in test)
|
|
View larger version (14K):
[in this window]
[in a new window]
|
Fig. 2 Selection differential in the F4 stage for maturity and yield of F2-derived soybean lines from which homozygous selections survived to the F7 stage. Shape of point indicates number of years of testing in statewide tests (cultivar was released after 3 yr in test). Points below the solid line could have been discarded with no loss of superior material
|
|
View larger version (13K):
[in this window]
[in a new window]
|
Fig. 3 Selection differential in the F6 stage for maturity and yield of F4-derived soybean lines from which homozygous selections survived to the F7 stage. Shape of point indicates number of years of testing in statewide tests (cultivar was released after 3 yr in test). Points below the solid line could have been discarded with no loss of superior material
|
|
|
NOTES
|
|---|
Salaries and research support provided by state and federal funds appropriated to the Ohio Agric. Res. and Dev. Ctr., The Ohio State Univ. This report is Journal Article no. 99-21.
Received for publication October 25, 1999.
|
REFERENCES
|
|---|
- Burton, J.W. 1987. Quantitative genetics: Results relevant to soybean breeding. In J.R. Wilcox (ed.) Soybeans: Improvement, production, and uses (2nd ed.). Agron. Monogr. 16. ASA, CSSA, and SSSA, Madison, WI.
- Curnow R.N. Optimal programmes for varietal selection. J. Roy. Stat. Soc. Ser. B 1961;23:282-318.
- Finney D.J. An experimental study of certain screening processes. J. Roy. Stat. Soc. Ser. B 1966;28:88-109.
- Panter D.M., Allen F.L. Using best linear unbiased predictions to enhance breeding for yield in soybean: II. Selection of superior crosses from a limited number of yield trials. Crop Sci. 1995;35:405-410.[Abstract/Free Full Text]
- St Martin S.K., Dye B.W., McBlain B.A. Use of hill and short row plots for selection of soybean genotypes. Crop Sci. 1990;30:74-79.[Abstract/Free Full Text]
- St Martin S.K., McBlain B.A. Procedure to estimate genetic gain by stages in multistage testing programs. Crop Sci. 1991;31:1367-1369.[Abstract/Free Full Text]
This article has been cited by other articles:
|
|
|
|
 
S. K. St. Martin and I. O. Geraldi
Comparison of Three Procedures for Early Generation Testing of Soybean
Crop Sci.,
May 1, 2002;
42(3):
705 - 709.
[Abstract]
[Full Text]
[PDF]
|
|
|
TOP
ABSTRACT
INTRODUCTION
