Introduction

The genetic base of modern soybean [Glycine max (L.) Merr.] cultivars in North America is narrow due at least in part to the small number of ancestors that formed the base of this germplasm and subsequent breeding and selection during cultivar development. Over 80 % of the genes present in modern North American soybean cultivars could be traced to 17 plant introduction (PI) ancestors and their first progeny (Gizlice et al. 1994). Since the establishment of the North American germplasm base, most soybean yield improvements have been made using crosses among elite germplasm rather than crosses with exotic germplasm or wild relatives (Carter et al. 2004); however, soybean breeders have introduced new germplasm into the North America soybean gene pool to improve resistance to diseases and pests and to attempt to increase yield (Carter et al. 2004). Exotic germplasm has proven to be an important source of genes especially for disease and pest resistance (Carter et al. 2004); however, it has been difficult to improve yield using PIs. This difficulty stems from the lower average yield performance of PIs compared to elite breeding lines which makes it difficult for breeders to obtain similar or higher yields in selected progeny from crosses with PIs when compared with the progeny obtained using only adapted parents (Smalley et al. 2004). In addition, there is a lack of reliable methods to predict whether a PI actually carries yield increasing alleles. Despite these difficulties, there continues to be a need to identify and use genetic variability in soybean germplasm that can improve soybean yields (Diers and Kim 2008) and experimental lines derived from exotic germplasm that yield significantly more than the best public cultivars indicate that useful yield genes do exist in exotic germplasm (Nelson and Johnson 2012).

Genetic mapping with molecular markers and marker-assisted selection (MAS) are widely used in soybean breeding programs. For both soybean breeding and research, single nucleotide polymorphism (SNP) markers are becoming the marker of choice because of their high frequency, widespread distribution throughout the genome as well as their suitability for high-throughput automated genotyping (Choi et al. 2007; Hyten et al. 2010). Multiple Illumina GoldenGate assays with 384–1,536 SNP markers have been developed (Hyten et al. 2008, 2010) and used in developing genetic maps (Hyten et al. 2010), mapping genes conferring resistance to Asian soybean rust (Hyten et al. 2009; Chakraborty et al. 2009), soybean cyst nematode (Heterodera glycines Ichinohe) (Vuong et al. 2010; Kim et al. 2011), and soybean aphid (Aphis glycines Matsumura) (Jun et al. 2012), and mapping loci involved in isoflavone concentration (Gutierrez-Gonzalez et al. 2011).

Understanding the genetic architecture of complex traits is a major challenge in the post-genomic era, especially for quantitative trait loci (QTL) by QTL interactions (epistasis), QTL by environment interactions, epistasis by environment interactions and more complex higher order interactions (Yang et al. 2008). The genotypic effect of one locus on a phenotype might depend on the genotype at several or many other loci, and QTL with minor or no individual effect can also involve epistasis, a finding that is well documented for a number of physiological traits in Drosophila melanogaster (Montooth et al. 2003). Strong interactions between QTL have been detected in maize (Lukens and Doebley 1999) and soybean (Lark et al. 1995), which have implications in cultivar development programs. If alleles involved in positive epistatic interactions are not transferred together to the cultivar that is being developed, yield improvement will be unsuccessful because high yield is conditional on the presence of epistatic effects (Lark et al. 1995).

QTL alleles from exotic soybean germplasm that significantly increase seed yield have been reported previously. Kabelka et al. (2004) identified nine positive yield QTL alleles that trace to the exotic soybean germplasm accessions FC 04007B and PI 68508. Wang et al. (2004) reported four positive yield QTL alleles from G. soja PI 468916; however, the QTL were only identified when the significance threshold was reduced and the data were analyzed with simple linear regression. Li et al. (2008) reported one positive yield QTL allele from G. soja and the QTL mapped to the same region on chromosome 5 where Kabelka et al. (2004) also reported a yield QTL. Guzman et al. (2007) identified eight positive yield QTL alleles from PIs but all of them mapped to the same regions where yield QTL were reported previously. Although these results suggest that it may be difficult to identify new positive yield QTL from exotic germplasm, there is a need to identify these positive alleles to help increase the rate of yield improvement of future cultivars. The objective of this study was to identify QTL and epistatic interactions associated with important agronomic traits in soybean using two backcross populations that each has a different PI as the donor parent.

Materials and methods

Plant material

Two populations of lines developed through backcrossing were used in the study. The first population (E population) included 116 BC2F3-derived lines developed using Elgin (PI 548557) as the recurrent parent and PI 436684 as the donor parent. Elgin was developed by the Iowa Agriculture and Home Economics Experiment Station and was released in 1984 because of its superior yield compared to public cultivars of similar maturity (Fehr and Bahrenfus 1984). Elgin is a maturity group (MG) II cultivar and it has resistance to bacterial pustule (Xanthomonas axonopodis pv. glycines). PI 436684 (MG III) is the Chinese cultivar Tie feng No. 8, which was released in 1970 by the Liaoning Academy of Agricultural Sciences, Shenyang, Liaoning, China (Cui et al. 1999) and introduced in the US in 1979 (http://www.ars-grin.gov/npgs/acc/acc_queries.html/; accessed 30 April 2012). It was selected as a parent based on its yield potential in germplasm evaluations conducted at Urbana, IL, USA, in 1983 and 1984 (Nelson et al. 1988). The cross of PI 436684 by Elgin was made in 1985. Progeny from this cross were advanced through early generation testing for yield that included testing F2-derived lines in the F3 and F4 generations and F5-derived lines. LG90-2847 was selected from the original cross and used as a parent in 1992 to develop a BC1 population. Early generation testing for yield was again employed and LG98-1351 was selected as a F4-derived line and crossed to Elgin to develop the BC2 population in 2001. This population was advanced by single seed descent and BC2 F3-derived lines were harvested in Chile in the spring of 2005.

The second population (W population) has 93 BC3F3-derived lines developed using Williams 82 (PI 518671) as the recurrent parent and PI 90566-1 as the donor parent. Williams 82 was developed by the USDA-ARS and the Illinois Agricultural Experiment Station through backcrossing the phytophthora rot (Phytophthora megasperma Drechs. f. sp. glycinea Kuan and Erwin) resistance gene Rps1k from Kingwa into the cultivar Williams (Bernard and Cremeens 1988). PI 90566-1 is a MG III soybean accession originating from Liaoning, China and introduced in the United States in 1930 (http://www.ars-grin.gov/npgs/acc/acc_queries.html/; accessed 30 April 2012). PI 90566-1 was tested in 1978 in a cooperative project involving both public and private soybean breeders to evaluate the yield potential of accessions from the USDA Soybean Germplasm Collection. This project was organized by Dr. Clark Jennings of Pioneer Hi-Bred International and used the existing yield data from previous general germplasm evaluations as the initial selection criterion (Bernard et al. 1998). PI 90566-1 was one of 26 MG III accessions advanced for testing at four locations in 1979 and was one of 21 accessions used as parents in 1979 when PI 90566-1 was crossed with L77-1779, which was later released as Williams 82. Progeny from this cross were advanced through early generation testing for yield potential as described for the E population. An F4-derived line, LG84-1022, was selected for first backcross which was made in 1986. The early generation testing procedure was repeated and an F5-derived MG III line, LG91-7654, was selected and crossed in 1995 to Williams 82 to develop the BC2 population. The early generation testing procedure was again repeated and an F5-derived MG III line, LG98-2080, was selected for use as a parent in the third backcross, which was made in the greenhouse during the spring of 2002.

Field trials

The E population was field tested at DeKalb and Bellflower, IL and Wooster, OH in 2005. In 2006, the field trials were conducted at Fisher and Bellflower, IL and Wooster, OH. The recurrent parent Elgin, experimental line LG98-1351, which was the donor parent used to develop the BC2 population, and the high yielding cultivar IA2065 were included as checks in the E population tests. The W population was tested at Hume and Ivesdale, IL and Wooster, OH in 2005. In 2006, the population was tested at Fisher and Hume, IL, Portageville, MO and Wooster, OH. The recurrent parent Williams 82, experimental line LG98-2080, which was the donor parent used to develop the BC3 population, and the high yielding cultivar IA3023 were included as checks at all locations and years of the W population tests. When two populations were evaluated at the same location, they were evaluated in separate tests.

All field trials of the two populations were arranged in randomized complete-block designs (RCBD) with two replications at each location. In the field tests at Wooster, OH, during 2005 and 2006, each plot consisted of eight rows. The middle six rows were spaced 19 cm apart and were harvested for seed yield. The two border rows were 0.76 m from the outside harvest rows. The plots were planted to a length of 6.4 m and were end trimmed to 4.88 m at maturity. The seeding rate was 10 seeds m−1 of row. The plots in both Missouri and Illinois were four rows wide with a 76-cm row spacing and the middle two rows were harvested to estimate seed yield. At Portageville, MO, the plots were 4.42 m long and the planting rate was 33 seeds m−1 of row. At the DeKalb, Bellflower, Fisher, Ivesdale, and Hume locations, the plots were 3.6 m long. Thirty seeds per meter were planted in the Illinois locations. Conventional tillage and herbicide practices were followed at all locations to maintain weed-free environments and recommended fertilization levels were applied. The plots were rated for maturity date, plant height, and lodging. Maturity date was recorded as the day when approximately 95 % of the pods had reached mature pod color (R8; Fehr et al. 1971). Plant height (cm) was measured at maturity as the average distance from the soil surface to the apex of the main stem. Lodging was scored at maturity on a scale of 1–5 with 1 designated as all plants standing erect and 5 as all plants prostrate. Plots were harvested to measure seed yield (kg ha−1) and yield values were adjusted to 130 g kg−1 moisture.

GoldenGate assay

A bulked leaf sample from at least 30 greenhouse grown plants of each line and parent from the two populations was used to extract DNA with the CTAB (hexadecylatri methylammonium bromide) method described by Saghai Maroof et al. (1984) with slight modifications in speed and time of centrifugation. DNA concentrations were quantified with a ND-1000 Spectrophotometer (NanoDrop Technologies, Wilmington, DE) and diluted to 100 ng μl−1. DNA quantity and quality were confirmed by electrophoresis in 3 % agarose gels and staining with a 1 μg/ml ethidium bromide staining solution (BMA, Rockland, ME, USA). The DNA samples from the E and W populations together with the parents were tested with SNP markers using the Golden Gate 1,536 Universal Soy Linage Panel 1.0 according to methods described in Hyten et al. (2010). The GoldenGate assay data were scored with the Illumina software BeadStudio v.3.2 and visually inspected to ensure that homozygous and heterozygous clusters were properly assigned.

Statistical analysis

Agronomic traits were analyzed by the PROC GLM functions of SAS 9.2 (SAS Institute, 2002). Lines, locations, replication within locations, and the line by location interaction were analyzed as random effects in each population. Each location by year combination was considered a separate environment in the analysis (Kim and Diers 2009). Broad-sense heritabilities of additive effects for yield and other agronomic traits were calculated based on the results from PROC GLM in SAS 9.2 according to Hill et al. (1998). Pearson correlation coefficients among all traits were calculated from the mean of lines across the environments using PROC CORR function in SAS 9.2.

A genetic linkage map was constructed for each population with JOINMAP 3.0 (Van Ooijen and Voorrips 2001) using the Kosambi mapping function. A logarithm (base 10) of the odds (LOD) score of 3.0 was used as the threshold to group markers into linkage groups. QTL analysis in the two populations was conducted using both the composite interval mapping (CIM) function in MapQTL 4.0 (Van Ooijen et al. 2002) and the mixed-model-based composite interval mapping (MCIM) function in QTLNetwork v2.1 (Yang et al. 2008). Average trait values across the environments were used for CIM while raw data including all individual observations from each location were used for MCIM. For each trait and environment, LOD thresholds in CIM and critical F values in MCIM corresponding to an experiment-wide threshold of P = 0.05 were determined by 1,000 permutations. For MCIM, QTL effects were estimated using Markov chain Monte Carlo method and the genome scan was performed using a 10 cM window size and a 1 cM walk speed. QTLNetwork calculates additive and dominance effects and epistatic interactions that include both of these effects. Because BC2F3 or BC3F3-derived lines that had undergone further inbreeding after derivation were evaluated in the field tests, there was very little heterozygosity in the plants grown in the yield trials that could have contributed to dominance estimates. Therefore, although dominance effects and dominance interactions are reported in the text, only additive effects and additive by additive (AA) interactions are reported in tables.

Single marker analysis (SMA) and regression analysis (SRA) were conducted to detect QTL using PROC GLM in SAS 9.2 when segregating SNP markers were not joined onto linkage groups by JOINMAP 3.0. Multiple regression analysis for QTL was conducted using PROC REG function with markers linked to significant QTL and epistatic interactions identified by CIM, MCIM, and SMA to determine the total phenotypic variance explained (R 2) by QTL and epistatic interactions. The proportion of the genotypic variance for yield explained by all significant QTL in the multivariate model was estimated from the ratio R 2/H 2 (Schön et al. 1994).

To test the impact of maturity on yield QTL, yield estimates of lines in both populations were adjusted using maturity as a covariate with PROC MIXED in SAS 9.2. The adjusted yield values were used to map yield QTL with CIM and SMA.

Results

Field data analysis

E population

There were significant (P < 0.0001) differences for seed yield, maturity date, plant height, and plant lodging (P = 0.004) among the three check genotypes in the E population tests across the six environments. The yield of LG98-1351, the BC1 parent of the BC2 population, was significantly greater (P = 0.05) than the recurrent parent, Elgin. The yields of the check and parental genotypes were 4,435 kg ha−1 for IA2065, 4,212 kg ha−1 for LG98-1351, and 3,620 kg ha−1 for Elgin. LG98-1351 matured significantly (P = 0.05) later than Elgin (3 days) and IA2065 (4 days).

There were significant (P < 0.0001) effects of lines, environments, and the interaction of lines by environments for seed yield, days to maturity, plant height, and lodging score in the population across the six environments. Seed yield was positively correlated with plant maturity (r = 0.55, P < 0.0001) and plant height (r = 0.69, P < 0.0001) but not significantly (P = 0.05) correlated with lodging. The average yield of the lines in the population across the six environments was 3,915 kg ha−1 (Table 1) and the average yields for environments ranged from 4,671 kg ha−1 at DeKalb in 2005 to 3,012 kg ha−1 in Wooster in 2006. The average yield of the population in 2005 was significantly greater than that in 2006 (Table 1). The broad-sense heritability for yield was 0.66 across environments in 2005, 0.54 in 2006, and 0.60 over the 2 years (Table 1). Across environments, 81 lines in the population yielded significantly (P = 0.05) more than the recurrent parent Elgin while no line yielded significantly greater than LG98-1351.

Table 1 Population mean, their standard errors and broad-sense heritability estimates, proportion of phenotypic variance explained in multiple QTL models with and without epistasis effects for four agronomic traits in the E and W populations
Full size table

W population

Across the seven environments, there were significant differences (P < 0.0001) in seed yield, maturity date, and plant height, and lodging score (P = 0.03) among the three check genotypes in the W population tests. The BC2 parent of the population, LG98-2080, yielded significantly more than the recurrent parent Williams 82. The average yield across environments for IA3023 was 4,083, 3,520 kg ha−1 for LG98-2080, and 3,275 kg ha−1 for Williams 82.

Across environments, there were significant (P < 0.0001) effects of lines and environment for seed yield, days to maturity, plant height, and lodging score. The line by environment interaction was significant for seed yield, days to maturity but not for plant height or lodging score. Seed yield was positively correlated with plant maturity (r = 0.45, P < 0.0001) while negatively correlated with plant height (r = −0.37, P = 0.0002) and there was no significant correlation with lodging score. The average yield of the W population was 3,442 kg ha−1 across environments (Table 1) and the average for environments ranged from 3,986 kg ha−1 for Fisher in 2006 to 2,864 kg ha−1 for Hume in 2006. The lines at Fisher in 2006 also had the highest lodging score (2.3). The average yield of the population in 2005 was 3,520 and 3,384 kg ha−1 in 2006 (Table 1). The board-sense heritability for yield was 0.64 across the 2005 environments, 0.37 for the 2006 environments and 0.66 over the 2 years. Like the E population, there were lines in the W population that yielded greater than both parents with 21 lines yielding significantly (P = 0.05) more than Williams 82 and one line yielding significantly more than LG98-2080.

Genetic map construction

E population

Of the 1,536 SNP markers in the GoldenGate assay, 513 were polymorphic between Elgin and PI 436684, the original donor parent of the population. Of these polymorphic SNP markers, 106 (21 % of the polymorphic markers) were segregating in this BC2 population. The segregating markers mapped to 16 chromosomes while chromosomes 3, 7, 19 and 20 were fixed for the Elgin alleles. The genetic map covered a length of 469 cM out of a total map size of 2,241 cM and the relative positions of the markers were generally consistent with the G. max consensus map 4.0 (Hyten et al. 2010; http://soybase.org). Chromosome 9 had the largest number of segregating SNP markers (16) covering approximately 74 cM while chromosomes 10 and 11 had only one segregating SNP marker each.

W population

Four hundred and three out of 1,536 SNP markers in the GoldenGate assay showed polymorphisms between Williams 82 and PI 90566-1, and 83 (21 % of polymorphic markers) SNP markers segregated in this BC3 population. The genetic map covered a distance of 238 cM. Chromosome 15 had the largest number of segregating SNP markers (18) covering 40 cM, while chromosomes 1, 5, 7, 10, 11, 14 and 19 were fixed for the Williams 82 alleles. Like the E population, the positions of markers on the genetic map were generally consistent with the G. max consensus map 4.0.

QTL identified in the E population

Yield QTL

Across the six environments, three yield QTL were identified by CIM with MapQTL in the E population with an experiment-wide threshold of P = 0.05 (Table 2). For the chromosome 4 and 18 yield QTL, the alleles from PI 436684 conferred significantly greater yield than the Elgin allele while the Elgin allele conferred greater yield than the PI allele for the chromosome 14 QTL (Table 2). Across the 2 years, the QTL on chromosomes 4, 14, and 18 explained 12.2, 7.8 and 10.6 % of the phenotypic variance for yield and their additive effects were 62, 39, and 61 kg ha−1, respectively (Table 2). Out of six locations in which the population was tested, the QTL on chromosomes 4 and 14 were significant at two locations, and the chromosome 18 QTL was significant at three locations (Table 2).

Table 2 Yield QTL detected by both CIM and MCIM in the E population
Full size table

The yield QTL on chromosomes 4 and 18 identified with CIM were also detected with MCIM; however, the chromosome 14 QTL was not detected with MCIM. The MCIM analysis showed that the QTL on chromosome 4 had significant additive, dominance and additive by environment interaction effects while the QTL on chromosome 18 had additive and additive by environment interaction effects (Table 2). The magnitude of the effects of yield QTL identified by MCIM was similar to what was observed by CIM with the PI 436684 alleles for the QTL on chromosome 4 and 18 having an additive effect of 67 and 62 kg ha−1 for yield, respectively (Tables 2).

Markers in the E population that were not placed in linkage groups and therefore were not included in either the CIM or MCIM analysis were tested for associations with agronomic traits with SMA. The SNP marker BARC-044481-08709 (BARC8709) on chromosome 5 was significantly (P = 0.0022) associated with yield across the environments (Table 3). The other 12 segregating SNP markers on the chromosome 5 were grouped together in one genetic linkage map by JOINMAP 3.0 while BARC8709 was not. Based on the G. max consensus map 4.0, there was a distance of at least 8.8 cM between BARC8709 and other SNP markers that were grouped together. BARC8709 had an additive effect on yield of 44 kg ha−1 and the yield increasing allele was from PI 436684. This marker was not significantly associated with any other trait and no other significant QTL was identified with the non-linked markers by single factor analysis in the E population. When the three significant yield QTL identified by CIM and the BARC8709 were placed into a multivariate model, all QTL were significant (P < 0.001) except BARC8709 and the total variance explained was 0.23 (Table 1).

Table 3 Yield QTL detected by only SMA in the E or W population
Full size table

Maturity, plant height, and plant lodging QTL

Five QTL controlling maturity, six QTL for plant height, and two for plant lodging were mapped in the E population with either CIM or MCIM methods. QTL for maturity, plant height and lodging were mapped within 9 cM of the yield QTL on chromosome 4 (Table 4). The allele for later maturity increased plant height and greater lodging was from the PI parent, which was the source of the yield increasing allele. QTL controlling maturity and plant height also were mapped to the same positions as the yield QTL on both chromosomes 14 and 18 (Tables 2, 4). Similar to what was observed for the QTL on chromosome 4, the source of the allele that increased yield conferred later maturity and greater plant height for both chromosomes. Additional QTL for both maturity and plant height were mapped on chromosomes 9 and 17 (Table 4). The allele for later maturity and taller plants was from Elgin for the chromosome 9 QTL, while the allele for later maturity and taller plants was from PI 436684 for the chromosome 17 QTL (Table 4). An additional QTL for plant height was mapped on chromosome 2 and a QTL for lodging was mapped on chromosome 1.

Table 4 Quantitative trait loci significantly associated with agronomic traits in the E population
Full size table

QTL identified in the W population

Yield QTL

Across environments, only one QTL, located on chromosome 3, was significant for yield with CIM (Table 5). For this yield QTL across environments, the SNP marker BARC-060031-16308 had the greatest LOD score (7.3) and the allele from PI 90566-1 had an additive effect of 80 kg ha−1 greater yield than the allele from Williams 82 (Table 5). This QTL was significant at six of the seven locations in which the population was evaluated based on the CIM (Table 5).

Table 5 Yield QTL detected by both CIM and MCIM in the W population
Full size table

A similar trend was observed for the chromosome 3 yield QTL with the MCIM analysis. The MCIM analysis revealed a 77 kg ha−1 additive effect (Table 5) and it was significant in four of the seven test locations. The QTL also showed significant dominance and additive by environment interaction effects.

The only non-linked marker that was significant in the SMA was BARC-059943-16234 (BARC16234) on chromosome 16 and this marker was significantly (P = 0.0264) associated with yield (Table 3). The other 14 markers on this chromosome formed a cluster that was at least 20.2 cM from BARC16234 based on the G. max consensus map 4.0. This QTL had an additive effect of 40 kg ha−1 and the positive effect was from Williams 82. When the significant QTL identified by CIM and the second QTL identified by SMA were placed into a multivariate model, both QTL were significant and together their R 2 value was 0.34 (Table 1).

Maturity, plant height, and plant lodging QTL

Three QTL for maturity, two QTL for plant height and two QTL for plant lodging were identified with either CIM or MCIM analysis in the W population (Table 6). QTL for both maturity and plant height were mapped to the same region on chromosome 3 as the significant yield QTL. The allele from PI 90566-1 had an additive effect of 1.5 days later maturity and 2.8 cm shorter height than the allele from Williams 82. QTL controlling maturity, plant height, and lodging also were mapped to chromosome 18 with the Williams 82 allele having an additive effect of 1.1 day later maturity, 1.4 cm greater plant height, and 0.7 less lodging than the allele from PI 90566-1 (Table 6). QTL on chromosome 15 were detected for both maturity and lodging with the Williams 82 allele having an additive effect of 0.2 days later maturity, and 0.1 less lodging than the allele from PI 90566-1 (Table 6).

Table 6 Quantitative trait loci significantly associated with agronomic traits in the W population
Full size table

Epistasis identified by MCIM in QTLNetwork

E population

Across the six environments, significant epistasis (experiment wide probability of P < 0.05) for maturity, plant height and lodging was detected with MCIM, while no significant epistasis was found for seed yield (Table 7). An AA epistatic interaction for maturity was identified between loci on chromosomes 2 and 14. These two loci were not detected individually with the MCIM analysis; however, the locus on chromosome 14 was detected individually by CIM (Table 4). Epistasis for plant height and lodging was identified between loci that had no individual effect (Tables 4, 7). The epistasis for plant height was detected between loci on chromosomes 1 and 5 (Table 7). The epistasis for lodging was detected between loci on chromosomes 15 and 17 and had only AD with environmental interaction (Table 7). The proportion of the phenotypic variation for maturity explained by the three major QTL identified by both CIM and MCIM was 0.46, while total amount of the variation explained by the effects of the major QTL together with epistasis was 0.63 (Table 1). For plant height and lodging, epistasis could explain 10 and 13 % of phenotypic variation beyond what was explained by main effect QTL, respectively (Table 1).

Table 7 Epistasis detected by MCIM in QTLNetwork in the E and W populations
Full size table

W population

In the W population, two significant AA epistatic (P < 0.05) interactions for plant maturity were identified and neither was found to interact with the environment (Table 7). There was no significant epistasis for plant height, lodging and seed yield. The first epistatic interaction was between regions on chromosomes 13 and 18. A maturity QTL was mapped to the same region on chromosome 18 through both CIM and MCIM and no individual effect maturity QTL was mapped to the chromosome 13 region. The second epistatic interaction was detected between loci on chromosomes 16 and 20 and these had no significant individual effects for maturity based on CIM and MCIM analysis (Table 7). The total amount of phenotypic variation for maturity explained by the two major QTL identified through both the CIM and the MCIM analysis and two epistatic interactions was 0.72, whereas the amount of variation explained by only the two major QTL on chromosomes 3 and 18 was 0.56 (Table 1).

Discussion

The backcross populations used in this study were developed through selections made over 25 years. The development of these populations was begun before technology was available for large-scale QTL mapping and they were not initially intended for that use. This is not a recommended strategy for yield QTL mapping but the end result of this backcrossing was the development of lines that outperformed the recurrent parent. An advantage of using these backcross populations is that positive alleles from the exotic parents are segregating in a very limited proportion of the recurrent parent genome. Another advantage is that the donor alleles are segregating in a more elite genetic background compared to using a population developed from a standard two-way cross. Even when an exotic parent has alleles that could potentially improve agronomic traits, exotic accessions often have poor overall agronomic performance that makes the resulting populations difficult to evaluate for agronomically important traits. Negative features of QTL mapping using a backcross population are the inability to assay the entire genomes of either parent for QTL because much of the genome of the backcross population is fixed for alleles from the recurrent parent.

Our study is the first report of mapping QTL controlling yield using the combination of Illumina GoldenGate assays and backcross populations in soybean. Although GoldenGate assays were previously used for genetic mapping in soybean, it has been widely used to map simply inherited traits such as disease or pest resistance controlled by a single gene or a few genes including Asian soybean rust (Hyten et al. 2009; Chakraborty et al. 2009), soybean cyst nematode (Heterodera glycines Inchinoe) (Vuong et al. 2010; Kim et al. 2011), and soybean aphid (Aphis glycines Matsumura) (Jun et al. 2012). Our study shows that the GoldenGate assays are a powerful tool to quickly map major QTL since the assay is capable of testing 192 DNA samples with 1,536 SNPs in 3 days (Hyten et al. 2008).

Some QTL were identified by both CIM and MCIM, others were identified by either CIM or MCIM, and still others by only SMA (Tables 2, 3, 4, 5 and 6). It is not surprising that different mapping algorithms may lead to different results, even when the same phenotypic and genotypic data were used (Kassem et al. 2006). Because the QTL identification is based on a statistical approach, it is also possible to identify false positive and false negative QTL (Mackay and Powell 2007; McElroy et al. 2006). However, reliability of identified QTL can be enhanced using more than one analysis method (Ravi et al. 2011). This is the reason that both MapQTL and QTLNetwork were employed to identify QTL in the present study. The yield QTL identified by both programs were mapped to almost the same genomic regions which further strengthens our confidence in the reliability of these QTL. QTL detected by only one QTL mapping method may be false positives and there is a need for validation by other approaches. MCIM in QTLNetwork, which uses most sources of variation, should be more effective in detecting both QTL with major and minor effects than the analysis done with CIM in MapQTL, which used the average trait values across environments or across replications in individual environments (Gutierrez-Gonzalez et al. 2010). In our study, the yield QTL identified by CIM and MCIM on chromosome 3 in the W population had the most consistent effects across environments (Table 5).

There is no consistent pattern in the relationship between yield and the other important agronomic traits in soybean but it has been shown that generally higher yield is associated with later maturity and taller plant height (Ablett et al. 1989; Cober and Morrison 2010; Mansur et al. 1996). For example, the regions on chromosomes 4 and 18 from PI 436684 in the E population where QTL alleles for increased yield mapped were also significant for additive effects of 0.6 and 1.1 days delay in maturity and 2.2 and 1.7 cm increase in plant height, respectively (Tables 2, 4). When yields adjusted using maturity as a covariate were analyzed with CIM and SMA, the QTL on chromosome 4, 14 and 18 were still significant for yield. The additive effects for the chromosome 4 and 14 QTL changed little from after adjustment while the effect of the chromosome 18 QTL allele from PI 436684 increased from 61 kg ha−1 before adjustment to 146 kg ha−1 after adjustment. For the W population, the yield QTL on chromosome 3 was no longer significant after analysis by CIM and SMA with the yields adjusted for maturity. These results suggest that the yield QTL on chromosome 3 in the W population could be a maturity gene that increases yield through delaying maturity. However, this is not always the case for maturity QTL as significant maturity QTL of similar magnitude on chromosomes 9 and 17 in the E population and chromosome 18 in the W population were not significantly associated with increased yield.

Identified yield QTL in the present study could explain only a portion of the total variation despite near complete SNP marker coverage of the areas of the genome segregating in the two populations. Across the environments, 23 % of phenotypic variation and 38 % of the genotypic variance for yield were explained by the three yield QTL on chromosomes 4, 14, and 18 in the E population. In the W population, 34 % of phenotypic variation and 52 % of the genotypic variance for yield were explained by the yield QTL on chromosomes 3 and 16. These results suggest that a larger number of QTL with effects too small to detect are involved in controlling the quantitative genetic variation for the traits measured in addition to other factors such as environmental interaction and epistasis. Alternatively, it might be possible that some of the remaining non-explained effects were in regions with inadequate marker coverage.

Previous work has provided evidence showing that the effects of epistasis may vary from a large to small impact on quantitative traits. Epistasis was found to be an important factor underlying the genetic basis of complex traits such as soybean seed isoflavone content (Gutierrez-Gonzalez et al. 2010), soybean seed yield (Lark et al. 1995), maize grain yield (Ma et al. 2007), and grain protein content in wheat (Kulwal et al. 2005). In contrast, linolenic acid content in soybean seed (Han et al. 2011), seed grain yield in wheat (Reif et al. 2011), and flowering time in maize (Buckler et al. 2009) were found to be controlled primarily by additive (main) effects rather than epistasis. In the present study, epistasis for plant maturity, height and lodging was detected, but their effects were minor compared to the effects of individual QTL (Tables 1, 7). A potential reason for the relatively small role of epistasis in our study is our use of backcross populations. Some epistatic interactions that could have been observed in two-way crosses may have been missed in our backcross populations because one or both interacting regions were not segregating in the backcross populations (Li et al. 1997).

E population

In the present study, two positive yield QTL alleles from an exotic source were identified on chromosomes 4 and 18 by both CIM and MCIM, a third QTL allele with a positive effect from the adapted parent Elgin was identified on chromosome 14 by only CIM, and a fourth QTL with a positive allele from the exotic source was mapped on chromosome 5 with SMA (Tables 2, 3). In Table 8, only previously reported yield QTL mapped on the same chromosomes with the positive yield QTL identified by both CIM and MCIM in the present study were listed because they are likely more reliable than other QTL detected by only one analysis method. The chromosome 4 yield QTL was mapped to approximately 14 cM on this chromosome based on the G. max consensus map 4.0 (Table 8; Hyten et al. 2010; http://soybase.org). Several yield QTL were previously reported on chromosome 4 (Smalley et al. 2004; Guzman et al. 2007; Yuan et al. 2002; Kassem et al. 2006; Sebastian et al. 2010; Table 8); however, only the QTL by Smalley et al. (2004) was mapped within 20 cM of the yield QTL mapped in our study (Table 8). This previously mapped QTL was located between SOYGPATR at 3 cM and Satt565 at 5.7 cM on the G. max consensus map 4.0 (Table 8), respectively. These two QTL are sufficiently close that they may be the same QTL.

Table 8 Genetic and physical positions of SSR and SNP markers flanking yield QTL on chromosomes 3, 4, and 18 based on the G. max consensus map 4.0 and the G. max genome (assembly version 1.01)
Full size table

The yield QTL on chromosome 18 was mapped to an interval between 97.3 and 103.1 cM on the G. max consensus map 4.0. Kabelka et al. (2004, 2006) both mapped a yield QTL within 20 cM of this QTL using two very different populations (Table 8). The allele for greater yield was from the cultivar BSR 101 in Kabelka et al. (2004) and from the Glycine soja Sieb. and Zucc. PI 468916 in Kabelka et al. (2006). The yield increasing allele from PI 468916 maps to the same position as an allele that provides resistance to soybean cyst nematode (SCN, Heterodera glycines Ichinohe) so it is likely that this yield QTL is a secondary effect of the SCN resistance. The QTL in both former studies were detected by SMA and the QTL in Kabelka et al. (2004) was also associated with plant height and the QTL in Kabelka et al. (2006) was associated with both plant height and lodging score. The yield QTL mapped on chromosome 18 in our study was associated with additive effects of 1.1 day delay in maturity and 1.7 cm increase in plant height (Table 4). This QTL is located in a 9.6-cM gene-rich interval outside the pericentromeric region on chromosome 18 based on the G. max genome (assembly version 1.01) (Schmutz et al. 2010; http://soybase.org). This interval has a relatively high rate of recombination as the 9.6 cM corresponds to an 845 kb region or 88 kb cM−1. This recombination rate is greater than the average genetic-to-physical ratio of approximately 197 kb cM−1 for soybean euchromatic chromosome arms (Table 8; http://soybase.org).

The yield QTL on chromosome 14 was identified only by CIM and was not detected by MCIM in any single environment or across environments (Table 2). The genetic position of the marker association with the yield QTL was 85.5 cM on chromosome 14 based on the G. max consensus map 4.0. The closest known yield QTL on chromosome 14 was detected by Satt066 (Smalley et al. 2004) and its genetic position was approximately 68 cM (http://soybase.org). The yield QTL on chromosome 5 that was mapped with BARC8709 by SMA were linked close to a yield QTL previously mapped by Kabelka et al. (2004) with Satt382. The genetic locations of Satt382 and BARC8709 on the G. max consensus map 4.0 are 26 and 22.5 cM and their additive effects were 50 and 44 kg ha−1, respectively. These results suggest that it is possible that the same QTL was mapped in both studies.

Eight major maturity genes (E1–E8) have been reported and six of them were mapped and placed on chromosomes 6 (E1) (Song et al. 2004), 10 (E2) (http://soybase.org), 19 (E3) (Molnar et al. 2003), 20 (E4) (Abe et al. 2003; Molnar et al. 2003), 6 (E7) (Molnar et al. 2003), and 4 (E8) (Cober and Morrison 2010). Three maturity QTL identified by both CIM and MCIM were mapped on chromosome 4, 17, and 18 in this study (Table 4). Although the QTL on chromosome 4 is on the same chromosome as E8, it was positioned at least 40 cM from E8 based on the G. max consensus map 4.0, so it is unlikely that it is E8. Six other maturity QTL were mapped on chromosome 4 by Keim et al. (1990), Orf et al. (1999), and Lee et al. (1996); however, they were all mapped at least 30 cM from the QTL we identified based on the G. max consensus map 4.0. Therefore, QTL on chromosome 4 in our study is likely a new maturity QTL.

A maturity QTL was previously mapped on chromosome 17 by Satt186 (Guzman et al. 2007) and its genetic location was 92.2 cM on the G. max consensus map 4. BARC-021991-04246, the marker closest to the chromosome 17 QTL identified in the present study, was mapped to 72.2 cM on G. max consensus map 4.0. The distance between these QTL is sufficiently great to suggest that the chromosome 17 maturity QTL mapped in our study might be new.

Three maturity QTL were previously mapped to an interval between 43.3 and 89.4 cM on chromosome 18 based on the G. max consensus map 4.0 (Guzman et al. 2007; Kabelka et al. 2004, 2006). In the present study, the genetic locations of the SNP markers flanking the QTL associated with both yield and maturity in the G. max consensus map 4.0 were 97.3 and 103.1 cM (Table 8) which indicate that the QTL from PI 436684 may be the same QTL as previously identified by Guzman et al. (2007) and Kabelka et al. (2004).

W population

Compared to the E population, fewer QTL were identified in the W population. This was expected because the W population is in the BC3 generation and therefore should have only one half as much of the genome segregating as the E population, which is in the BC2 generation. The backcross generations of these populations are consistent with the observed size of the map in each population with the W population having a map of 238 cM and the E population having a map size of 469 cM. In addition, the number of lines in the W population was smaller than the E population, which would make it less likely to identify small effect QTL in the W population.

The only yield QTL mapped by CIM and MCIM in the W population is on chromosome 3 and several yield QTL have been previously reported on this chromosome (Smalley et al. 2004; Kassem et al. 2006; Kabelka et al. 2004; Specht et al. 2001; Wang et al. 2004). The LOD peak for the yield QTL on chromosome 3 mapped in the present study is at 92.1 cM (Table 8; http://soybase.org). The closet previously mapped yield QTL was linked to Satt257 (Wang et al. 2004; Guzman et al. 2007) and its genetic location was 74.7 cM in the G. max consensus map 4.0 (Table 8). Because the QTL we mapped is 17.4 cM from these previously mapped QTL, it is likely that this is a newly identified QTL. This QTL is relatively stable under the different environments tested and is associated with additive effects of 1.5 days delay in maturity and 2.8 cm decrease in plant height (Tables 5, 6). The positive QTL allele from PI 90566-1 had a 77–80 kg ha−1 additive effect, additive by environment interaction, and dominance effect (Table 5).

A yield QTL was identified by BARC16234 on chromosome 16 in the W population by SMA (Table 3). Guzman et al. (2007) previously identified yield QTL from PI 407720 with the SSR marker Satt547, and its genetic location on the G. max consensus map 4.0 is 74.9 cM, which is 8.1 cM from BARC16234. Although the additive effects of Satt547 (90 kg ha−1) is greater than for BARC16234 (41 kg ha−1), it is possible that the same QTL was mapped in both studies.

The maturity QTL identified on chromosome 18 was mapped near a maturity QTL identified in a previous study. Guzman et al. (2007) found that Satt191 locus from Lawrence was associated with an additive effect of 2 days delay in plant maturity by SMA and the genetic location of the marker was 89.4 cM in the G. max consensus map 4.0. The maturity QTL we identified on chromosome 18 mapped to approximately 94 cM with both CIM and MCIM and the QTL had an additive effect of 1.1 days delay in the maturity (Table 6). The genetic positions and additive effects of these QTL suggest that we may have mapped the same maturity QTL that was previously mapped.

Results from this study indicate that exotic soybean germplasm can be a good resource for improving yield in North American soybean cultivars. The QTL from PI 90566-1 was significant in both CIM and MCIM across environments as well as in each year and in most of the individual environments. No previous studies have reported a yield or maturity QTL near the QTL region on the chromosome. Confirmation of the yield QTL in the E and W population is currently being done. The SNP markers flanking the QTL will be especially useful in MAS and pyramiding of positive QTL in soybean breeding programs, because of the availability of efficient SNP marker detection assays such as TaqMan and melting curve assay. In addition, the identification of the physical location of the QTL on the soybean genome will greatly facilitate discovery of candidate genes, map-based cloning and functional characterization of the QTL.