HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in Web of Science Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Web of Science (1) Citing Articles via Google Scholar Google Scholar Articles by St. Martin, S. K. Articles by Geraldi, I. O. Search for Related Content PubMed Articles by St. Martin, S. K. Articles by Geraldi, I. O. Agricola Articles by St. Martin, S. K. Articles by Geraldi, I. O. Related Collections Crop Genetics

CROP BREEDING, GENETICS & CYTOLOGY

Comparison of Three Procedures for Early Generation Testing of Soybean

^a Dep. of Horticulture and Crop Science, Ohio Agric. Res. and Dev. Ctr., The Ohio State Univ., Columbus, OH 43210-1086
^b Dep. of Genetics, Escola Superior de Agricultura "Luiz de Queiroz," Univ. of São Paulo, Caixa Postal 83, 13.400-970, Piracicaba, São Paulo, Brazil

* Corresponding author (stmartin+{at}osu.edu)

	ABSTRACT

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION REFERENCES

 
Early generation testing as a breeding procedure for autogamous crops consists of testing heterogeneous families, followed by selection of homozygous lines from superior families. Although early generation testing has been compared with other breeding methods, there have been few comparisons of different early generation testing procedures. Our objective was to assess the effectiveness of testing F₁-, F₂-, and F₃-derived families in soybean [Glycine max (L.) Merr.]. We applied the three procedures to two soybean populations from 1986 to 1993. Selected lines (F₃-, F₄-, and F₅-derived, respectively) were obtained by applying each testing method to each population. Divided into maturity classes, these selected lines were compared with unselected lines of the same generation, population, and maturity to determine genetic gain. The three early generation treatments produced similar genetic gains in seed yield, averaging approximately 4%. Selection of F₁-derived families for yield increased plant height and lodging, but the other two selection procedures were satisfactory in this respect. To maximize genetic gain for yield while avoiding undesirable changes in lodging in an early-generation testing program, we recommend testing F₂-derived families, unless off-season nurseries permit development of F₃-derived families without further loss of time.

	INTRODUCTION

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION REFERENCES

 
EARLY GENERATION TESTING is a selection procedure based on initiating testing of genetically heterogeneous lines or families in an earlier generation than would normally be considered suitable for release. Examples include testcross evaluation of partially inbred plants in allogamous species (Jenkins, 1940) and recurrent selection procedures based on testing of lines in the S₁ or other early generation (Hull, 1945). When the concept is applied to development of homozygous cultivars in an autogamous species, selection of homozygous lines from superior heterogeneous families permits the breeder to exploit the genotypic variance provided by inbreeding and to develop cultivars of suitable uniformity. Thus, the procedure has two phases: selection among heterogeneous families and selection of homozygous potential cultivars from superior families.

Different generations have been proposed as the "early" generation in which families can be derived for testing in the first phase. Harlan et al. (1940) used yield tests of bulks of 379 barley (Hordeum vulgare L.) crosses to identify superior populations from which to select homozygous lines. In their procedure, selection among cross bulks (or, as we prefer, F₁-derived families) is the first phase. Weiss et al. (1947) used yield of F_2:3 families of soybean to predict the performance of more inbred progenies. Frey (1954) and Cooper (1990) also described procedures based on testing F₂-derived families. Procedures based on testing F₃-derived families were considered by Voigt and Weber (1960), Thorne (1974), and Schillinger (1985).

The plant breeding literature contains numerous reports of experiments comparing a single early generation testing procedure with other breeding methods, such as the pedigree, single-seed descent, and bulk methods [see Johnson and Bernard (1963) and Fehr (1987a) for reviews of the soybean literature]. Comparison of different early generation testing procedures in a common experiment has been rare. Weiss et al. (1947) tested F_1:2 through F_1:5 bulks and F_2:3 families from 17 soybean crosses. They reported that the bulks were of little value in predicting the yield or maturity of crosses, but that yield of F_2:3 families gave moderately good estimates of progeny performance. The F₁- and F₂-derived families in their experiment were not compared directly in the same tests. The objective of our experiment was to compare early generation testing procedures based on F₁-, F₂-, and F₃-derived families, as applied to two soybean populations.

	MATERIALS AND METHODS

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION REFERENCES

 
We used three early generation testing treatments (Table 1). These were based on testing of F₁-, F₂-, and F₃-derived families. The term "treatments" will be used throughout this paper to refer to these selection procedures. Selection was for yield, maturity, and lodging resistance. The products resulting from application of these treatments were groups of F₃-, F₄-, and F₅-derived lines, respectively. Estimates of genetic gain were obtained for each treatment by comparing the selections with an unselected control group of lines of the same generation.

Population HS3C1
The 14 parents of HS3C1 were crossed in 1986 at Columbus, OH, by the bulk parent method, in which crosses were made between each parental line and a bulk of the remaining 13 parents (Fehr, 1987b; p. 150–151). We recorded the one known parent of each cross. Approximately 300 F₁ seeds were obtained. A portion of the resulting F₁ plants were produced in the Iowa State University-University of Puerto Rico winter nursery facility in winter 1986-1987. A portion of the seed was used to produce F₂ populations at Columbus in 1987, and a single F₃ seed was harvested from each plant. We used this seed, along with remnant seed, to produce the F₁, F₂, and F₃ generations in 1988 at Columbus. A base population for selection was developed by selecting plants of each generation at random. A control (unselected) population was also developed by harvesting a single seed from each plant. We used the partial pedigree information to ensure, as much as possible, similar contributions of the 14 parents to both the base population and the control group.

Unreplicated tests of F_1:2, F_2:3, and F_3:4 families took place in 1989. For convenience, these were divided randomly into four sets, each set consisting of 90 entries: 6 check cultivars or elite lines (of maturity groups II, III, and IV) and 28 families of each treatment. The experimental unit was a single row, 1.5 m long with 1-m spacing between rows (St. Martin et al., 1990). Entries were randomly assigned to the plots. Maturity was recorded for each plot as the date when 95% of the pods had reached their mature color. Plots were harvested with a plot combine and yield of air-dried seed recorded. From each set x treatment combination of 28 families, 10 families were chosen on the basis of a selection index: yield - b(maturity - mean maturity), where b = the phenotypic regression coefficient of yield on maturity calculated for the 28 families. We used indices of this type throughout the experiment to obtain genetic gain for yield without undesirable changes in maturity. Control populations (unselected F₂, F₃, and F₄) were also produced in 1989 and maintained by harvesting a single seed from each plant.

Selected families (F_1:3, F_2:4, and F_3:5) were evaluated in 1990 in two-row plots (row spacing 76 cm), 3 m long, replicated twice at each of two locations (Columbus and Lakeview, OH). Seeding rate was 400 000 seeds/ha, and the same seeding rate was used for all replicated tests in both the application and evaluation of the selection treatments. Because these were unbordered plots, we intended to limit the range of maturity within a test. Accordingly, the five earliest families from each set x treatment combination were placed in one test and the five latest in another. Each test also included four cultivars or elite lines as checks, plus 20 families per treatment, for a total of 64 entries. Each test was arranged in a simple square lattice design, and entry means were adjusted for block effects. Maturity and seed yield were measured as before, and a lodging score (from 1 = erect to 5 = flat) was assigned to each plot. We used a selection index (Baker, 1974) to rank the families in each test x treatment group of 20. The index was I/s_I - (1/2)L/s_L, where I = yield - b(maturity - mean maturity), s_I = phenotypic standard deviation of I among the 20 families, L = lodging score, s_L = phenotypic standard deviation of L, and b = genotypic regression coefficient of maturity on yield, calculated for the 20 families in the test x treatment group. In this index, I represents yield adjusted for maturity. We chose the coefficient of ¹/₂ for lodging subjectively to provide half as much weight for lodging as for yield. Based on the index, the best three families and the second-best three families (i.e., families ranked 4th, 5th, and 6th) were identified.

Each family tested in 1990 was also planted in a separate nursery that year for harvest of individual F₃, F₄, or F₅ plants. Control populations for each of the same generations were also harvested in the same way. In 1991 we used short-row plots to increase seed of the selected and control F₃-, F₄-, and F₅-derived lines.

Final selections for each treatment were made based on a pair of tests (of early- and late-maturing lines) conducted at a single location (Plain City, OH) in 1993. (A 1992 test of the same materials was abandoned because of poor stands.) The early test consisted of selections from the early test of 1990, and the late test was for selections from the late test of 1990. Entries for each test consisted of 9 cultivars and elite lines as checks, 10 F₃-derived lines from each of the best three F₁-derived families, 5 F₃-derived lines from each of the second best three F₁-derived families, 10 F₄-derived lines from each of the best three F₂-derived families, 5 F₄-derived lines from each of the second best three F₂-derived families, 10 F₅-derived lines from each of the best three F₃-derived families, and 5 F₅-derived lines from each of the second best three F₃-derived families, for a total of 144 entries in each test. Each test was arranged as a simple square lattice design. Each plot consisted of three rows, spaced 38 cm apart, 3 m long. Spacing between rows of adjacent plots was 51 cm. Entry means, adjusted for blocks, were used to select the best 12 lines from each treatment in each test, again based on the selection index, yield - b(maturity - mean maturity), with the regression coefficient, b, determined from an analysis of all entries except the checks. Regardless of its index value, however, no line that was more than 3 d later than the latest check cultivar (Corsica) was retained in the selected group. The control lines were again increased for use in the evaluation experiments.

Population ED11
Application of early generation treatments to ED11 was handled similarly to the procedures described for HS3C1, except that ED11 was initially offset by 1 yr, beginning with crosses made in 1987. We attempted to make all 55 possible crosses of the 11 parents (considering reciprocals equivalent), but obtained sufficient seed of only 49 combinations. There were five sets of families in the initial unreplicated, short-row test (1990), each set consisting of seven checks, 49 F_1:2, 49 F_2:3, and 49 F_3:4 families. Each set was assigned one entry from each of the 49 useable crosses. The best 12 families were selected from each set based on the same selection index used for HS3C1 in 1989. The resulting 60 selected families per treatment were sorted into three classes by maturity: early, medium, and late, and these maturity classes were retained for the families and their derived lines for the remainder of the experiment. Three tests of F_1:3, F_2:4, and F_3:5 selections, one for each maturity class, were conducted in 1991 at Lakeview and Columbus. There were 64 entries in each test, whose entry structure, design, and plot size were the same as the 1990 tests of HS3C1. The best three and second best three entries were identified based on the selection index I/s_I - ¹/₄L/sL, where the coefficient ¹/₄ rather than ¹/₂ reflects our judgment that ED11 was more lodging resistant than HS3C1 and that therefore less selection pressure for lodging was needed. In 1992, F_3:4, F_4:5, and F_5:6 lines from each selected family were increased, along with random control sets of lines of the same generations, derived by single seed descent. These lines were tested at Plain City in 1993, the 144-entry tests having the same structure and design as the 1993 tests of HS3C1 lines, except that three tests (for early-, medium, and late- maturing lines) were used for ED11, rather than two. The best 12 lines from each generation were selected in each test, by means of a selection index calculated in the same way as the HS3C1 tests.

Evaluation of Early Generation Testing Treatments
Design
Beginning in 1994, we conducted experiments to compare the selections derived from each early generation testing treatment with the unselected control lines of the same generation. There were five experiments altogether, corresponding to early and late lines from HS3C1 and early, medium, and late lines from ED11. Each experiment consisted of 72 entries: six cultivars or elite lines as checks, 12 selected lines from each treatment (F₃-, F₄-, and F₅-derived) and 10 random unselected lines from each of the same three generations. Maturity of random lines was recorded during seed increase and they were divided into early and late groups (HS3C1) or early, medium, and late groups (ED11) corresponding to the selections. Ten lines were then chosen at random within each group to provide the unselected entries.

The five experiments were grown in close proximity in each of four environments: Plain City (1994 and 1997) and Lakeview (1994 and 1995). The soil at Plain City was a Wetzel silt loam (Typic Endoaqualf) in 1994 and a Kokomo silt loam (fine, mixed, mesic Typic Argiaquoll) in 1997. The soil at Lakeview was a Milford silty clay loam (fine, mixed, mesic Typic Haplaquoll). Tests at Plain City in 1995 and 1996 were abandoned (except as seed increase) because of poor stands. The design of each test was an 8 x 9 simple rectangular lattice. Plots consisted of six rows, spaced 38 cm apart, but only the four inner rows were harvested. Final plot length was 3 m. Yield, maturity, and lodging were measured, and mature plant height was measured in all environments except for Lakeview, 1994.

Statistical Analysis
Each of the five experiments was analyzed separately. First, a lattice analysis of variance was conducted for each of the four environments. Then, a combined, mixed-model analysis was conducted for each set, excluding the check cultivars, with the 264 (= 4 environments x 66 entries) entry–environment means from the lattice analyses. The model included treatments (F₁, F₂, F₃), the effect of selection (i.e., selected lines vs. unselected control), and their interaction as fixed factors. Random factors were environments, the effect of selection x environments, treatments x environments, effect of selection x treatments x environments, and lines within the six selection group x treatment combinations. Adoption of lines as a random factor is a recognition that the lines in the experiment, whether selected or unselected, should be regarded as samples from conceptual populations. Designating lines as random provides a conservative error term, because the source of variation due to lines thereby contributes to the error in comparing selection treatments.

Error mean squares appeared to be heterogeneous for some traits across environments. Also, it is reasonable to assume, a priori, that variances among lines may differ across treatments and between the selected and unselected groups. Therefore, the heterogeneity of variance components was incorporated into the model, as described by McLean (1989). The solution vector for fixed (b) and random (u) effects is given as

where X and Z are incidence matrices for the fixed and random factors, respectively, y is the 264 x 1 vector of entry–environment means, G is a diagonal matrix of variance components for random effects, and R is a 264 x 264 diagonal matrix of errors for individual entry–environment means.

Elements of G were obtained by using a restricted maximum likelihood procedure included in the GLMM software package (Blouin et al., 1989) to estimate variance components for each random factor except for lines within selection group x treatment combinations. Heterogeneity of the variance among lines was ignored in this procedure. Then, the variance components for lines within selection group x treatment combinations [_L(SxT)²] and for lines x environments [_LxE(SxT)²] within selection group x treatment combinations were obtained using the same restricted maximum likelihood procedure, applied to the data set for each selection treatment-generation combination, consisting of the 40 (= 10 unselected lines x 4 environments) or 48 (= 12 selected lines x 4 environments) entry-environment means. The estimates of _L(SxT)² were used in G. The estimates of _LxE(SxT)² were used in R, but it was necessary to adjust these for the heterogeneity of error across environments. We did this by assuming that these estimates included ¹/₂ of the interplot error variance in addition to the true line x environment interaction component, the coefficient of ¹/₂ occurring because there were two replications per environment. We therefore added to each estimate of _LxE(SxT)² one-half the difference between the effective interplot error for the particular environment and the pooled effective interplot error.

Standard errors for the overall effect of selection and for the effect of selection in each generation were obtained as described by McLean (1989). Means of the effect of selection across all five experiments were obtained by weighting each experiment's results by the inverse of the square of its standard error. Genetic gain was considered significant (P = 0.05) if the absolute value of the difference between selected and unselected groups exceeded twice the standard error.

	RESULTS AND DISCUSSION

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION REFERENCES

 
Effects of the early generation treatments are summarized in Table 2. The tendency of all methods to select for earliness in the late portion of HS3C1 was counteracted by the tendency to select for lateness in the early portion of ED11. Thus, on average, the treatments did not affect maturity, suggesting that all methods gave satisfactory control of maturity. The large standard errors for response of maturity in the medium portion of ED11 were a consequence of large interactions between the effect of selection and the environment.

The treatment means, averaged across both populations, showed similar gains in yield, although only the F₃-derived treatment had a gain that exceeded twice its standard error (Table 2). Large standard errors of the gains for yield made it difficult to recommend any one selection treatment in preference to the others as a procedure for improvement of yield. The large standard errors reflected the contributions of several sources of error in comparing selected and control groups. In our study we included genetic variation among lines within selected and control groups as a random factor. This inclusion, which we believe is necessary in order to make proper inferences about differences between groups, contributed to the large standard errors. More definite conclusions on the relative merits of different early-generation testing procedures require additional experiments, perhaps employing simulation.

The gains in yield for each selection treatment represented approximately 4% of the grand mean of the control lines. St. Martin and Xie (2000) reported much larger total gains for an actual soybean breeding program based on testing F₂-derived families. The program they described benefitted from additional replication and more intense selection than we employed, and this probably accounts for their larger gains. The moderate (25–35%) selection intensities we employed in this experiment allowed us to sample adequately the within-group genetic variability by testing a total of 60 selected and 50 control lines for each treatment.

In soybean breeding, bulk harvest of segregating families, which is a feature of early generation testing, often leads to an increase in the proportion of taller, later plants (Raeber and Weber, 1953; Mumaw and Weber, 1957). Our results showed no overall tendency of selected lines to be later, but the F₁-derived treatment, which involves testing the most heterogeneous families, produced taller, more lodging-susceptible selections than the other two treatments. Soybean breeders interested in using early generation testing may avoid this problem by deferring testing to F₂- or F₃-derived families, a recommendation consistent with the conclusions of Weiss et al. (1947).

Maximizing genetic gain per year is an important consideration in choosing a breeding method (Fehr, 1987b). The number of generations required for our F_n-derived treatment increases with n. Since the three treatments gave similar yield gains, there may be an advantage to beginning testing as early as possible, unless use of off-season nurseries allows production of the additional generations with no loss of time. Use of F₂-derived families, which avoids the problem of increased lodging, may be the most reasonable compromise.

	ACKNOWLEDGMENTS

 
The authors acknowledge the excellent technical assistance of Alfred DuBois, who contributed his expertise to this experiment for 11 of its 12 yr.

	NOTES

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION REFERENCES

 
Salaries and research support provided by state and federal funds appropriated to the Ohio Agric. Res. and Dev. Ctr., The Ohio State Univ. Salaries and scholarship for I.O.G. supported by USP, BID, and FAPESP. This report is Journal Article no. HCS 00-17.

Received for publication December 4, 2000.

	REFERENCES

TOP NOTES ABSTRACT INTRODUCTION MATERIALS AND METHODS RESULTS AND DISCUSSION REFERENCES

 

• Baker, R.J. 1974. Selection indexes without economic weights for animal breeding. Can. J. Anim. Sci. 54:1–8.
• Blouin, D.C., A.M. Saxton, and K.L. Koonce. 1989. A completely randomized design with missing cells and repeated measures in time. p. 80–103. In Applications of mixed models in agriculture and related disciplines. Southern Coop. Ser. Bull. No. 343. Louisiana Agric. Exp. Stn. Baton Rouge, LA.
• Cooper, R.L. 1990. Modified early generation testing procedure for yield selection of soybean. Crop Sci. 30:417–419.[Abstract/Free Full Text]
• Fehr, W.R. 1987a. Breeding methods for cultivar improvement. In J.R. Wilcox (ed.) Soybeans: Improvement, production, and uses. Agron. Monogr. 16. ASA, CSSA, and SSSA, Madison, WI.
• Fehr, W.R. 1987b. Principles of cultivar development, Vol. I. Macmillan, New York.
• Frey, K.J. 1954. The use of F₂ lines in predicting the performance of F₃ selections in two barley crosses. Agron. J. 46:541–544.[Free Full Text]
• Harlan, H.V., M.L. Martini, and H. Stevens. 1940. A study of methods in barley breeding. U.S.D.A. Tech. Bull. 720.
• Hull, F.H. 1945. Recurrent selection and specific combining ability in corn. J. Am. Soc. Agron. 37:134–145.
• Jenkins, M.T. 1940. The segregation of genes affecting yield of grain in maize. Agron. J. 32:55–63.[Free Full Text]
• Johnson, H.W., and R.L. Bernard. 1963. Soybean genetics and breeding. p. 1–73. In A.G. Norman (ed.) The soybean: Genetics, breeding, physiology, nutrition, management. Academic Press, New York.
• McLean, R.A. 1989. An introduction to general linear models. p. 9–38. In Applications of mixed models in agriculture and related disciplines. Southern Coop. Ser. Bull. No. 343. Louisiana Agric. Exp. Stn. Baton Rouge, LA.
• Mumaw, C.R., and C.R. Weber. 1957. Competition and natural selection in soybean varietal composites. Agron. J. 49:154–160.[Abstract/Free Full Text]
• 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]
• St. Martin, S.K., B.W. Dye, and B.A. McBlain. 1990. Use of hill and short row plots for selection of soybean genotypes. Crop Sci. 30:74–79.[Abstract/Free Full Text]
• St. Martin, S.K., and Xie Futi. 2000. Genetic gain in early stages of a soybean breeding program. Crop Sci. 40:1559–1564.[Abstract/Free Full Text]
• Schillinger, J. 1985. Strategies for cultivar development in temperate climates. p. 305–310. In R. Shibles (ed.) World Soybean Conference III: Proceedings. Westview Press, Boulder, CO.
• Thorne, J.C. 1974. Early generation testing and selection in soybeans: association of yields in F₃- and F₅-derived lines. Crop Sci. 14:898–900.[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]
• Walker, A.K., and A.F. Schmitthenner. 1984. Registration of PMGT(S1)C3 soybean germplasm. Crop Sci. 24:213–214.[Free Full Text]
• Weiss, M.G., C.R. Weber, and R.R. Kalton. 1947. Early generation testing in soybeans. J. Am. Soc. Agron. 39:791–811.

This article has been cited by other articles:

				R.-C. Yang When Is Early Generation Selection Effective in Self-Pollinated Crops? Crop Sci., October 22, 2009; 49(6): 2065 - 2070. [Abstract] [Full Text] [PDF]

				R.-C. Yang Effects of Linkage and Epistasis on Intergeneration Correlations in Self-Pollinated Species Crop Sci., November 24, 2008; 48(6): 2074 - 2085. [Abstract] [Full Text] [PDF]

				F. K. Padi and J. D. Ehlers Effectiveness of Early Generation Selection in Cowpea for Grain Yield and Agronomic Characteristics in Semiarid West Africa Crop Sci., March 19, 2008; 48(2): 533 - 540. [Abstract] [Full Text] [PDF]

				D. G. Jones and C. W. Smith Early Generation Testing in Upland Cotton Crop Sci., December 2, 2005; 46(1): 1 - 5. [Abstract] [Full Text] [PDF]

				R. Bernardo On the effectiveness of early generation selection in self-pollinated crops Crop Sci., July 1, 2003; 43(4): 1558 - 1560. [Abstract] [Full Text] [PDF]

This Article Abstract Full Text (PDF) Free Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in Web of Science Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Web of Science (1) Citing Articles via Google Scholar Google Scholar Articles by St. Martin, S. K. Articles by Geraldi, I. O. Search for Related Content PubMed Articles by St. Martin, S. K. Articles by Geraldi, I. O. Agricola Articles by St. Martin, S. K. Articles by Geraldi, I. O. Related Collections Crop Genetics