Introduction

Soybean [Glycine max (L.) Merr] is one of the most important oilseed crops produced worldwide (FAOSTAT 2018). Over the past three decades, the total global production of soybean has tripled to > 300 million metric tons in 2016/17 (FAOSTAT 2018). Genetic improvements in the yield potential of modern soybean varieties and the implementation of improved management practices have enabled enormous increases in productivity from 700 kg ha−1 in 1924 to more than 3000 kg ha−1 in 2017 (USDA 2018). Recent studies in the USA have estimated genetic gains in soybean yield range from 12 to 27 kg ha−1 year−1 over the last century, close to 1% year−1 (Rowntree et al. 2013; Rincker et al. 2014; Rogers et al. 2015), while average yield increases worldwide were 1.3% per year (Ray et al. 2013). A pedigree analysis indicated that most modern US public cultivars could be traced back to 35 ancestral lines (Gizlice et al. 1994). Therefore, the lack of genetic diversity in modern soybean breeding programs is of major concern and focus for ensuring continued genetic gains in soybean grain yield well into the future.

Soybean was domesticated from its wild progenitor (Glycine soja Sieb. and Zucc.) 3000–5000 years ago in China (Carter et al. 2004; Doebley et al. 2006). During the domestication process, more than half of the genetic diversity found in G. soja was lost, decreasing pairwise diversity (π) from 2.94 × 10−3 in Glycine soja to 1.05 × 10−3 in modern cultivars (Hyten et al. 2006; Zhou et al. 2015). Currently, it is estimated that ~ 8500 unique accessions of G. soja and 45,000 accessions of Asian landraces of G. max are maintained in germplasm banks around the world (Hyten et al. 2006; Wen et al. 2009). The USDA Germplasm Resources Information Network (GRIN) collection contains ~ 1100 G. soja accessions (Song et al. 2015). However, only a small fraction of the publicly available G. soja germplasm has been used in commercial breeding aimed at improving elite soybean variety development.

Wild soybean exhibits diverse undesirable agronomic characteristics such as excessive lodging, late flowering, small black seeds, hard seed coat reducing uniform germination and emergence, a lack of uniformity for plant maturity and excessive pod shattering (Liu et al. 2007). Despite the many unfavorable traits found in G. soja germplasm, useful genetic diversity have been reported within G. soja germplasm including those associated with resistance to soybean aphid (Aphis glycines) (Zhang et al. 2017c) and soybean cyst nematode (Heterodera glycines) (Wang et al. 2001; Winter et al. 2007; Zhang et al. 2017b) and with tolerance to salt and drought (Lee et al. 2009; Chen et al. 2013; Ning et al. 2017; Nisa et al. 2017), and with improved seed composition such as increased protein content and improved fatty acid and amino acid profiles (Ha et al. 2014; Yan et al. 2014; Leamy et al. 2017; La et al. 2019).

Although much work has been done recently to understand the implications of domestication as well as the genetic architecture of wild soybean (Liu et al. 2007; Zhou et al. 2015), little research has been done to investigate the potential of wild soybean to be used for improving grain yield. Ertl and Fehr (1985) evaluated two populations derived from G. soja × G. max backcrosses across two years and four environments and observed no significant grain yield increases when comparing recombinant inbred lines (RILs) to the G. max recurrent parent. Akpertey et al. (2014) reported G. soja-derived lines that were not significantly different in yield form the G. max parent. In another study using a similar approach, Concibido et al. (2003) mapped QTL associated with grain yield from G. soja on chromosome 14. The yield QTL significantly increased yield by 9.4% when averaged across all test environments. Li et al. (2008) utilized two backcrossed populations from G. max × G. soja and identified a QTL associated with grain yield on chromosome 5 (Satt511). Grain yield of the lines carrying the G. soja allele were averaged across all environments, and those lines demonstrated a 6.3% yield increase over lines carrying the G. max recurrent parent allele. Thus, a small amount of evidence exists that G. soja germplasm may harbor useful genes for improving the agronomic performance of modern day soybean cultivars, but more research is clearly needed.

Quantitative trait loci (QTL) mapping and genome-wide association studies (GWAS) have been used extensively to dissect the genetic architecture of agronomic traits (Holland 2007). One drawback of QTL mapping is that the estimated QTL interval is usually relatively large, due to the relatively large and extensive linkage disequilibrium (LD) present in these soybean populations, and the traditionally low marker density throughout the genome (Xu et al. 2017). QTLs are often population- and environment-specific, which can generate non-reproducible results in non-related environments, populations and individuals (Kang et al. 2009; Bernardo 2010). As an alternative, GWAS relies on historical LD to increase the analysis power (Zhu et al. 2008) and to detect marker associations. GWAS studies have identified nucleotide variants associated with complex traits in soybeans including seed oil and protein content (Hwang et al. 2014; Cao et al. 2017), flowering time and maturity (Zhang et al. 2015; Mao et al. 2017) and soybean cyst nematode resistance (Heterodera glycines) (Zhang et al. 2016). However, this method has a relatively low power to detect associations with rare alleles and can be confounded by population structure (Price et al. 2010; Auer and Lettre 2015). Accordingly, maize geneticists developed an approach to overcome some of the pitfalls of both conventional GWAS and linkage mapping techniques (Yu et al. 2008). The design consisted of multi-cross recombinant inbred lines (RILs) where each family has a unique parent crossed with a common “hub” parent, (or “founder”). The NAM design combines aspects of two genetic approaches: linkage mapping through genetic recombination and association mapping through historical linkage disequilibrium presented in the populations, improving the resolution and the power for detection of rare alleles in the analysis (Yu et al. 2008).

The SoyNAM project (http://www.soybase.org/SoyNAM) has the objective of understanding the genetic architecture of seed yield in soybean using NAM RILs derived from crosses between IA3023 (hub parent) and 40 different G. max adapted cultivars specifically selected for improved grain yield. A significant QTL for canopy coverage identified on chromosome 19 was associated with an increase in grain yield of ~ 50 kg ha−1 (Xavier et al. 2017). In another study using the same NAM panel, Xavier et al. (2018) identified a genomic region positively associated with grain yield stability across a range of environments located on chromosome 18, and insight was gained about the QTL × environment interaction prevalent in most reported QTL for grain yield. Again using the same soybean NAM panel, Diers et al. (2018) identified 23 significant regions associated with seed yield and 29 associated with seed weight, demonstrating the complexity of grain yield-related traits. NAM populations have also been used to dissect complex traits in other important crop species such as rice (Oryza sativa) (Fragoso et al. 2017), barley (Hordeum vulgare) (Sharma et al. 2018) and wheat (Triticum aestivum) (Bajgain et al. 2016). Although the large G. max NAM panel contained exotic germplasm in the population structure, there continues to be a need also to investigate the potential of alleles from wild soybean utilizing modern population structures. The objectives of this study were to identify genomic regions associated with complex agronomic traits in wild soybean germplasm and to identify alleles from G. soja accessions that may be used to improve agronomic traits in soybean using a nested association mapping panel.

Materials and methods

Germplasm development and phenotyping

Our NAM panel contains 392 F4-derived RILs, developed from three biparental cross-combinations between the G. max hub parent Williams 82 (Bernard and Cremeens 1988) and three G. soja plant introductions PI458536, PI464890B and PI522226. The G. soja accessions were among eight diverse accessions selected for crossing that year based on origin and extreme values for plant and seed composition traits. PI458536 (MG 0) was from Heilongjiang, China, and had a protein concentration greater than 53 g kg−1 and moderate resistance to Soybean Mosaic Virus. PI464890B (MG I) originated from Jilin, China, and had moderate resistance to Soybean Mosiac Virus. PI522226 (MG 000) came from Primorye, Russia, and had extremely narrow leaflets. The four parental lines were obtained from the USDA Soybean Germplasm Collection. The crosses were made in Urbana, IL, during the summer of 2010, and the true F1 hybrid plants were grown at the same environment in 2011 and bulk harvested. Approximately, 10,000 F2 plants were grown at the Bradford Research Center in Columbia, MO, during the summer of 2012 for each of the three populations, and F2 plants were bulk harvested. In 2013, approximately, 100,000 F3 plants were grown at the Bradford Research Center in Columbia, MO, for each of the three populations, and ~ 500 single plants were harvested separately from each F3 population. The 500 F3 single plants were selected during the fall of 2013 within each population based on a visual selection of appropriate maturity date, upright plant growth and resistance to pod shattering. The F3:4 plant rows were grown at the Bay Farm Research Facility in Columbia, MO, during 2014, and a single plant was harvested from each row. Single plants within a row were selected based on critical agronomic traits needed for yield evaluation including maturity date, erect plant growth, pod shattering resistance and resistance to soybean Bacterial Pustule (Xanthomonas axonopodis pv. glycines). The F4:5 plant rows were grown at the Bay Farm Research Facility in Columbia, MO, during the summer of 2015, and entire plant rows were again selected based on agronomic characteristics that would enable high-quality yield evaluations. Each plant row was bulk harvested for development of the F4-derived RILs. Only lines with sufficient seed for multi-environment testing were carried forward, and RILs were randomly chosen to represent each of the three populations to reduce the NAM panel size to a manageable field experiment.

During the growing seasons of 2016 and 2017, RILs were planted in four-row plots with a row spacing of 76 cm, a row length of 3.6 m and a 1.2 m fallow alley. The environments included the Hundley-Whaley Research Center in Albany, MO (40° 14′ 53″ N 94° 19′ 51″ W), the Bay Farm Research Facility in Columbia, MO (38° 57′ 50″ N 92° 19′ 42″ W), the Greenley Research Center in Novelty, MO (39° 57′ 28.23″ N 92° 10′ 40.45″ W) and the Graves-Chapple Research Center in Rock Port, MO (40° 16′ 18.09″ N 95° 28′ 26.38″ W) (Supplementary Fig. 1). The plots were arranged in an augmented incomplete block experimental design with one replication in 2016 and two replications in 2017, with two check cultivars per block in both years, IA3023 for MG III and IA4005 for MG IV. Genotypes were randomly assigned to a block and replication, and blocks were randomly ordered in the field across all environments. The experiment was divided into two field blocks at each environment for MG III and MG IV RILs to accommodate combine harvesting.

Grain yield was measured by harvesting the two center rows of each four-row plot with an Almaco R1 or an Almaco SPC-40 plot combine (Nevada, Iowa). The total seed weight and seed moisture were measured in real time on the combine, and the final grain yield was calculated as kg ha−1 on a 13% moisture basis. Plant maturity was recorded as the day when 95% of the pods in the center two rows of each plot reached final color (R8 stage) (Fehr et al. 1971) and considering September 1 as day one. Lodging was rated for the center two rows of each plot at R8 stage according to the following score: (1) all plants erect; (2) 5–25% of the plants prostrate; (3) 25–50% of the plants prostrate; (4) 50–80% of the plants prostrate; (5) all plants prostrate. Plant height was recorded as the average distance (cm) between the top node and the soil surface of plants in the center two rows of each plot.

Genotyping and SNP analysis

Genomic DNA was extracted from young trifoliate leaf tissue of ten random plants for each RIL and the respective parents during the V4/V5 growth stage. A Qiagen Dneasy Plant 96 kit (QIAGEN, Valencia, CA) was used for high-quality DNA extraction, and the protocol given for the kit was followed. DNA samples were submitted to the Soybean Genomics and Improvement Laboratory, USDA-ARS, for genotyping using the Illumina Infinium BARCSoySNP6K BeadChip (Song et al. 2013). Genotypes were called using the software Genome Studio (Illumina, San Diego, California USA). SNPs with less than 90% call rate were excluded, and SNPs were imputed using the LD-KNNi method implemented in version 5 of Tassel software (Bradbury et al. 2007). A total of 5786 SNPs remained for subsequent quality control, association mapping and linkage analysis. The R package ‘ParentOffspring’ (Abdel-Haleem et al. 2013), and a principal component analysis using version 5 of the TASSEL software (Bradbury et al. 2007) were used to eliminate RILs that were found to exhibit genotypes inconsistent with expected genetic similarity (≥ 80%) to the respective parents.

Only the filtered SNP marker data were used to measure the linkage disequilibrium (LD) in the NAM panel using the sliding window method implemented in version 5 of TASSEL software. The LD SNPs were estimated using the pairwise squared correlation (r2), and then they were plotted against SNP pair distances to obtain the rate of the LD decay in the heterochromatin and euchromatin region (Song et al. 2013). We calculated the fixation index (FST) among the three populations in the NAM panel using the unbiased estimator (Weir and Cockerham 1984) implemented in the NAM package (Xavier et al. 2015).

Nested association mapping

Best linear unbiased estimators (BLUE) were calculated to account for non-genetic sources of variance associated with environmental variation. BLUE values were then used in the nested association mapping according to the model adopted by Jarquín et al. (2014): \(y_{ijkl} = \mu + g_{i} + cb_{l} + l_{j} + gl_{ij} + r_{k(j)} + b_{l(k)} + e_{ijk}\), where \(\mu\) is the trait mean, \(g_{i}\) is the genetic effect of the genotypes, \(cb_{k}\) is the interaction effect between the checks and incomplete blocks, \(l_{j}\) is the effect of environments, \(gl_{ij}\) is the interaction effect between the genotypes and environments, \(r_{k(j)}\) is the effect of the replications nested in environment, \(b_{l(k)}\) is the incomplete block effect nested within replicate, and \(e_{ijkl}\) is the residual. Broad-sense heritability (H2) on an entry-mean basis was calculated as follows: \(H^{2} = \sigma_{\text{g}}^{2} /(\sigma_{\text{g}}^{2} + \sigma_{\text{ge}}^{2} /l + \sigma_{\text{e}}^{2} /rl)\), where \(\sigma_{\text{g}}^{2}\), \(\sigma_{\text{ge}}^{2}\) and \(\sigma_{\text{e}}^{2}\) are the genetic variance, genotype × environment variance and error variance; \(l\) is the number of environments and \(r\) is the number of replications (Fehr 1991). Narrow-sense heritability was calculated using the genomic relationship matrix from GBLUP model (de los Campos et al. 2015) as follows: \(h^{2} = \sigma_{\text{a}}^{2} /\sigma_{\text{y}}^{2}\), where \(\sigma_{\text{a}}^{2}\) is the additive variance estimated from GBLUP model, and \(\sigma_{\text{y}}^{2}\) is the phenotypic variance (Tan et al. 2017).

Nested association mapping was conducted using the R package NAM (Xavier et al. 2015) developed from the SoyNAM project. The mixed linear model designed for multiple parent intercross populations was used for the SNP and haplotype-based association (Wei and Xu 2016): \(y = \mu + X\alpha + g + \text{cov} + e\), where \(\mu\) is the intercept, \(X\) is the allele matrix from SNP/haplotype data and family information, \(\alpha\) is the SNP/haplotype effects, \(g\) is the population structure effect, cov is the covariate, and \(e\) is the residual effect. The R8 date was used as a covariate in the NAM analysis to account for maturity effects on grain yield, plant height and lodging. A false discovery rate (FDR) threshold at \(\alpha\) ≤ 0.05 level was used to declare SNP significant in the association mapping.

Linkage mapping analysis

We used the consensus genetic linkage map constructed by Song et al. (2016) based on 21,478 SNPs mapped in the Williams 82 × PI479752 (G. soja) population to define SNP positions. QTL mapping was performed using the composite interval mapping (CIM) functionality qtl package in R (Broman and Sen 2009) individually in each population of the NAM panel. To fit multiple QTL models, we used the function fitqtl which drops one QTL at a time. The logarithm of the odds (LOD) for defining QTL significance was determined by a 1000 permutation test analysis at \(\alpha\) ≤ 0.05 significant level and estimated confidence intervals of QTL via used 1.5-LOD support intervals. Additive allelic effects for NAM and CIM analysis were estimated by family relative to the common parent (Williams 82).

We searched for candidate genes, published genes, and published QTLs using the soybean reference genome available at SoyBase (www.soybase.org) in the Wm82.a2.v1 assembly. The SNPs and QTL intervals that were highly associated with traits were considered as a potential region for candidate genes. SoyBase was also used to search for QTLs previously found to affect the traits.

Results

Phenotypic variation, linkage disequilibrium and population structure

The analysis of variance (ANOVA) results for all traits evaluated in this study are shown in Supplementary Table 1. Significant genetic differences were found for all measured traits, and both environmental effects and RIL × environment interaction were significant (p < 0.01) for all the traits. The entry-mean-based estimates of heritability (h2) were 0.51–0.64 for grain yield (GY), 0.87–0.93 for plant maturity (PM), 0.82–0.93 for plant height (PH) and 0.85–0.91 for lodging score (LDG).

There was a broad, continuous and normal distribution for all phenotypic traits (Supplementary Table 1 and Supplementary Fig. 2) measured in this study, and transgressive segregation was observed in both directions when comparing the phenotypic values of the hub parent (Williams 82) and RILs, suggesting the effect of multiple genes controlling the traits and appropriate distribution of alleles among hub parent and G. soja parents to move forward with subsequent mapping efforts to identify marker associations for significant QTL.

Linkage disequilibrium (LD) decay varied between euchromatin and heterochromatin regions (Fig. 1a). The LD decayed to half its maximum r2 value (0.44) at an average 619.5 kb in euchromatin and 892.6 kb in heterochromatin. Since we estimated LD in experimental biparental populations, a greater extent of LD is expected when comparing expected LD to natural soybean population (Zhou et al. 2015) due to the limited amount of recombination events and selection in experimental population development (Hyten et al. 2007). Principal component analysis (PCA) was used to evaluate the variation in genotypic data of the NAM population (Fig. 2b). The PCA could not distinguish clearly the subpopulation structures likely due to the half-sib family relationship among the three subpopulations, yet PC1 explained 10.1% of the variation, while PC2 explained 6.3% of the genotypic variation.

Fig. 1
figure 1

Linkage disequilibrium (LD) decay of the NAM panel (a), distribution of the NAM RILs under PC1 and PC2 (b), and fixation index of the NAM panel over the entire genome (c), where the red line represents the 99th percentile (0.29)

Full size image
Fig. 2
figure 2

Manhattan plots of nested association mapping (NAM) analysis for grain yield, plant maturity, plant height and lodging plotted against positions on each of the 20 chromosomes, across four environments within 2016 (red dots) and 2017 (blue dots). The significant SNPs were distinguished by the FDR 0.05 (3.8 × 10−5) threshold line (color figure online)

Full size image

More than 70% of the SNPs had an estimated FST value smaller than 0.1 (Fig. 1c). Just 0.02% of the SNPs had an FST value larger than the 99th percentile (red line). To distinguish the SNPs under selection and potential genetic drift in the NAM population, we utilized the smoothing kernel (blue lines) procedure implemented in the NAM package (Flori et al. 2009; Xavier et al. 2015). The results revealed significant FST values on chromosomes 3, 5, 6, 9, 11 and 20. The SNPs ss715594391, ss715593858 and ss715594468 on chromosome 6 had the highest FST values 0.67, 0.66 and 0.63, respectively. ss715593858 is ~ 400 kb from the region of E1 (Glyma06g23026) (Xia et al. 2012), a well-known photoperiod response/plant maturity gene. This is likely a result of fixing the maturity date during the inbreeding process.

Nested association mapping

A QTL region was determined by the marker with the greatest LOD score presented in the region associated with a trait. Additive allelic effects were estimated by family relative to the hub parent (Williams 82) where a positive effect represents an increase in the value of the trait as a result of substituting the hub parent (G. max) allele with the respective G. soja allele, and a negative effect decreases the value of the trait when the hub parent allele is substituted with the G. soja allele. Four SNPs on Chr 16 were significantly associated with grain yield (GY) in 2017 across the average of four environments (Fig. 2a), with allelic effects ranging from − 116.2 to 59.4 kg ha−1. This marker and the associated locus has been identified to harbor (~ 200 kb) the gene responsible for regulating pod shattering, Pdh1 (Funatsuki et al. 2014). Hence, the relatively large negative allelic effects on grain yield.

Forty-nine SNPs significantly associated with plant maturity (PM) were identified on chromosomes 6, 11 and 12 in 2016 and 2017 (Fig. 2b). These regions were consistently associated with the trait across environments, and SNPs associated with PM on Chr 6 and 12 were observed in all the eight environments, and in Albany 2016 (ALB2016), Columbia 2016 (CLM2016) and Novelty (NOV2016) for the region on Chr 11 (Supplementary Fig. 3b). The allelic effect estimates for PM ranged from − 2.0 to 5.7 days, depending on the SNP associated and population (Table 1 and Supplementary Table 2). A higher frequency of estimated positive PM alleles was evident from the G. soja founder parents. The three most significant SNPs associated with PM on Chr 6 were ss715594029, ss715593853 and ss715594119. These SNPs were co-located in the same region (~ 350 kb apart) as the known E1 photoperiod response/plant maturity gene (Xia et al. 2012). Also, the genomic region associated with PM on Chr 11 and 12 were reported to be correlated with flowering/maturity time (Zhang et al. 2015; Fang et al. 2017; Mao et al. 2017), and two homologs GmPRR3A (~ 30 kb) and GmPRR3B (~ 100 kb) have recently been cloned which underlie these regions (Li et al. 2019).

Table 1 Summary of significant SNPs from NAM analysis for grain yield, plant maturity, plant height and lodging and their respective allelic effect with negative and positive effects relative to the common parent (Williams 82) across four environments within 2016 and 2017
Full size table

The three regions associated with PM (ss715593853 on Chr 6, ss715608829 on Chr 11 and ss715613171 on Chr 12) explained 36%, 16% and 18% of the genetic variation, respectively (Fig. 3a–c). The RIL carrying the genotype CC (G. soja) at ss715593853 exhibited a positive increment of 4.3 days in PM compared to the TT (G. max) genotype (Fig. 3a). A similar trend was observed for ss715608829, and ss715613171, where the lines carrying the G. soja allele were 3.6 and 3.9 days higher PM than the lines carrying G. max allele, respectively (Fig. 3b, 3c). The allelic effect for the SNPs on Chr 6, 11 and 12 varies among G. soja parents.

Fig. 3
figure 3

Difference in genotypic values (a–c) between lines segregating at SNP ss715593853 (a), ss715608829 (b), and ss715613171 (c) associated with plant maturity in the NAM panel. The difference in genotypic values of plant height associated with SNP ss715616047 (d), and lodging associated with SNP ss715602804 (e), between lines in the NAM panel across eight environments. The difference in genotypic values between lines segregating at SNP ss715625973 for grain yield at Pop3-Williams 82 × PI522226 across eight environments (f). The least square difference (LSD) and differences in mean (∆m), and the probability of a greater F statistic (Pr > F) and R2 are in the figures

Full size image

Seven SNPs were found to be associated with plant height (PH) across four environments in 2016 and 2017 (Fig. 2c), one SNP on Chr 10 and six on Chr 13. The region on Chr 13 was identified in five of the eight environments across years (Supplementary Fig. 3c), with an allelic effect ranging from − 2.86 to 13.07 cm (Supplementary Table 2). The most significant SNP was ss715616047, which explained 12% of the genetic variation and exhibited an increase in PH of 19.6 cm when comparing lines carrying the AA (G. soja) genotype to the CC (G. max) genotype. The region on Chr 13 was located ~ 200 Kb of qPH13.1; a confirmed QTL associated with increased plant height in G. soja (Zhang et al. 2018).

A novel region on Chr 8 (ss715602804 and ss715602662) was significantly associated with lodging (LDG) in 2016. In 2017, four SNPs were associated with LDG, ss715579524 on Chr 1, ss715602804 on Chr 8, ss715630642 on Chr 18 and ss715633186 on Chr 19 (Table 1). The QTL region associated with ss715602804 was more stable across environments and was also identified at Novelty 2016 and Columbia 2017. The region on Chr 8 was responsible for explaining 17% of the phenotypic variation, and lines carrying the TT allele (G. soja) exhibited a significant increase in 0.46 in LDG score when compared to the lines with the CC allele (G. max) (Fig. 3e).

Linkage mapping

Twelve QTLs across eleven chromosomes in three biparental populations were identified using the composite interval mapping (CIM) in 2016 and 2017 (Fig. 4). Three novel QTL were associated with GY on Chr 17, 18 and 20 (Fig. 4a); four QTL with PM on Chr 6, 11, 12 and 20 (Fig. 4b); three QTL with PH on Chr 1, 2 and 13 (Fig. 4c); and one QTL associated with LDG on Chr 9 (Fig. 4d).

Fig. 4
figure 4

Graphical display of significant QTL detected in three populations from composite interval mapping (CIM), for a grain yield, b plant maturity, c plant height and (d) lodging across eight environments. Pop1-Williams 82 × PI464890B; Pop2-Williams 82 × PI458536; Pop3-Williams 82 × PI522226. The dashed dark-red horizontal line indicates the threshold logarithm of the odds (LOD) based on a 1000 permutation test

Full size image

Four QTL were detected for PM in multiple environments in 2016 and 2017 (Supplementary Fig. 4c-d). A significant QTL on Chr 6 was detected in all three populations, and the phenotypic variance ranged from 28.3 to 52.6% dependent on the specific G. soja parental line. The additive effect of qPM-6-1, qPM-6-2 and qPM-6-3 ranged from 2.34 to 5.5 days (Table 2). Three minor QTLs were detected in specific populations; qPM-11 on Chr 11 in Williams 82 × PI464890B (Pop1), qPM-12 on Chr 12 in Williams 82 × PI522226 (Pop3) and qPM-20 on Chr 20 in Pop1. The confidence interval for the QTL qPM-6-1, qPM-6-2, qPM-6-3, qPM-11 and qPM-12 overlapped with the SNPs associated with PM on chromosomes 6, 11 and 12 in the NAM analysis (Table 1). Additive effect represents the impact of substituting the G. max allele with the G. soja allele for each respective SNP. The QTL qPM-20 was responsible for 12.5% of the phenotypic variation with a negative effect of − 2.87 days on the trait. qPM-20 was found close to the reported position of the E4 gene (Liu et al. 2008). Also, an epistatic interaction was significant between qPM-6-1 × qPM-20 in 2017, accounting for more than 5.6% of the phenotypic variation for PM.

Table 2 Summary of QTL results from composite interval mapping (CIM) method in three populations (Pop1- Williams 82 × PI464890B; Pop2- Williams 82 × PI458536; Pop3- Williams 82 × PI522226) for four agronomic traits across four environments within 2016 and 2017
Full size table

Three QTL were identified for PH in multiple environments on Chr 1, 2 and 13 (Fig. 4c). qPH-1 on Chr 1 was observed in the population Williams 82 × PI458536 (Pop2), qPH-2 on chromosome 2 was identified in the population Williams 82 × PI464890B (Pop2), qPH-1 accounted for 18.7% of the variation, while qPH-2 accounted for 15.2% to 20.1% of the phenotypic variation (Table 2). The G. soja alleles for both qPH-1 and qPH-2 showed a negative effect on PH, from − 8.9 to − 14.2 cm. qPH-13 was identified in the population Williams 82 × PI522226 (Pop3), and the phenotypic variance explained was 23.3% with the G. soja alleles having a positive effect of 11.87 cm. The qPH-13 interval overlapped with the SNPs previously associated with PH on chromosome 13 in the NAM analysis (Table 1) that region had G. soja alleles associated with greater plant height (Fig. 3d).

One QTL on chromosome 9 was associated with LDG in the population Williams 82 × PI458536 (Pop2) (Fig. 4d and Table 2). The phenotypic variation explained for qLDG-9 was 14.3%, and the additive effect for the G. soja alleles was 0.24 on the 1–5 scale. The qLDG-9 interval region overlaps a QTL (Lodging 5–9) reported by Lee et al. (1996).

Three QTL associated with GY were identified on chromosomes 17, 18 and 20 explained 24.4%, 21.2% and 17.5% of the total phenotypic variance, with additive effects for the G. soja alleles ranging from − 297.2 to 166.1 kg ha−1 (Table 2). Only for qGY-17 did G. soja alleles have a positive impact on GY, and this was found across two environments in 2016: Columbia and Albany. Lines carrying the G. soja allele (BB) presented an increment of 197.1 kg ha−1 or 6% in grain yield when comparing to the lines carrying G. max allele (AA) across all environments (Fig. 3f).

Discussion

In this study, we evaluated a nested association mapping panel by using both linkage mapping in three biparental populations, as well as association mapping with the entire panel of RILs. The NAM analysis identified more markers associated with traits than the CIM approach. The NAM design has been successfully used for gaining an understanding of the genetic architecture for several traits in soybean, as well as in other crops (Yu et al. 2008; Fragoso et al. 2017; Li et al. 2017; Diers et al. 2018; Xavier et al. 2018). One advantage of the NAM design is that it allows for comparing allelic effects for the same marker from multiple parents, in relation to a common parent, enabling the distinction of allelic effects in our study for each G. soja parent relative to the Williams 82 (hub parent) (Xavier et al. 2015). The allelic effect varies between G. soja parents for markers associated with traits in the NAM panel, similar to the results observed in other NAM studies (Diers et al. 2018; Chen et al. 2019). In general, the parent-specific additive effects are smaller for the NAM analysis than the effects observed using the CIM method. It is important to mention that the CIM method comprises a relatively small number of RILs compared to the full NAM panel, which likely increases the risk of overestimating the individual QTL effect (Maurer et al. 2017; Vatter et al. 2017). From that perspective, it is important to evaluate the wild soybean allelic effects across the NAM panel and within each family for a better understanding of the genetic background and environment effect, and a truer estimate of allelic effects relative to the hub parent.

QTL associated with plant maturity

Four QTL associated with PM in the NAM panel were identified. The region associated with PM on Chr 6 (E1 gene-ss715593853/qPM-6-1, 6-2, 6-3) (Xia et al. 2012) was identified in all individual populations and in the NAM panel across all environments, and showed the largest allelic effects (Fig. 3b, Supplementary Fig. 4c-d, and Table 1). When comparing the allelic effects of ss715593853, the differences between G. soja allele (E1) and G. max (e1-as) were 4.3 days. The minor allele frequency (MAF) for the significant SNPs was relatively high, ranging from 0.15 to 0.50. The NAM design was robust enough to detect associations and dissect the genetic architecture of complex traits as compared to standard GWAS (Buckler et al. 2009; Diers et al. 2018). Also, the E1 gene was found to be under strong selection in our NAM panel, revealed by the highest FST values by SNPs surrounding the gene locus (Fig. 2c). E1 is part of the phytochrome A signaling pathway and down-regulated GmFT2a and GmFT5a, orthologs of Arabidopsis FLOWERING LOCUS T responsible for early flowering (Kong et al. 2010). Xia et al. (2012), through positional cloning located the position of E1 (Glyma.06g207800), and characterize E1 as functional, and two non-functional alleles e1-fs and e1-nl, and one not fully functional e1-as. These results impact breeding strategies for maturity and soybean adaptation in different regions in the US. Langewisch et al. (2017) evaluated a series of combinations of E1 and e1-as from maturity group (MG) 000 to X and showed a higher frequency of e1-as allele in MG IV and below and a higher allele frequency of E1 in MG V and above. E1 promotes late flowering and may be used in the southern US and e1-as promotes earlier flowering and may be used in the northern US in MG 0 to IV (Wolfgang and An 2017; Langewisch et al. 2017).

The PM QTL on Chr 12 (ss715613171) was identified in seven of eight environments, and in both mapping methods. The allelic effects for the association analysis were higher in the Pop3, ranging from 0.62 to 1.30 days. In the CIM analysis, qPM-12 was specific to Pop3 and explained ~ 22% of the phenotypic variation presented in the population for the trait. SNPs associated with the first flower were reported in the same region by Mao et al. (2017) and Fang et al. (2017). The PM QTL on Chr 11 (ss715608829/qPM-11) was detected in the association analysis and for Pop1 in the CIM analysis. The largest allelic effects were observed on Pop1 in 2016 (1.14–1.35 days). Li et al. (2017) reported a QTL for days to flowering (qDTF-11-4) using a NAM panel in the same region with an additive effect of 1.87 days. Two homologs GmPRR3A (Glyma11g15580) and GmPRR3B (Glyma12g07861) controlling growth period of soybean have recently been cloned which underlie regions associated with PM on chromosomes 11 and 12 (Li et al. 2019).

A QTL on chromosome 20 (qPM-20) was significantly associated with PM in the CIM analysis, explained 12.5% of the PM variation, and the allele form the G. soja parent had a negative effect of − 2.9 days. QTLs are many times population specific which makes the results less reproducible in diverse genetic background (Bernardo 2010). The qPM-20 is ~ 4 Mb apart from the maturity gene E4 that encodes the phytochrome protein GmphyA2 (Glyma.20g090000) (Liu et al. 2008), and recessive alleles of E4 provide earlier flowering (Xu et al. 2013), which is the phenotype observed in RILs which inherited the G. soja alleles of this QTL.

QTL associated with plant height and plant lodging

Wild soybeans are well-known for expressing excessive branching, extreme lodging and a prostrate growth habit. Until now, more than 239 and 87 QTLs have been reported to be associated with plant height (PH) and lodging (LDG) in soybean, respectively (http://www.soybase.org, “SoyBase browser,” verified 01/20/2019). We identified a significant QTL for PH on Chr 13 in both mapping methods (ss715616047/qPH-13). The lines carrying the G. soja allele showed a significant increase in PH compared to the lines carrying the G. max allele (Fig. 3d). Our findings were supported by Zhang et al. (2018), as they fine mapped a QTL from G. soja in the same region of ss715616047/qPH-13, which exhibited a greater PH for the lines that carried the wild allele. RT-PCR showed that Glyma.13g249400 demonstrated higher expression in genotypes that expressed higher PH. Also, sequence data identified a difference in six amino acids from Glyma.13g249400 between parents (Zhang et al. 2018).

Two SNPs were associated with LDG on Chr 8. The SNP ss715602804 showed a significant increase of 0.46 units in the LDG score in the lines that carried the G. soja allele (Fig. 2e). Another QTL was identified in our linkage analysis on Chr 9 (Fig. 4d and Table 2), qLDG-9 expressed a positive allelic effect of 0.24 units in the LDG score. Introgression of wild soybean alleles in currently elite germplasm may result in linkage drag of unfavorable traits, such as excessive plant growth and lodging (Zhang et al. 2017a). However, this issue could be fixed with backcrossing with more adaptable germplasm and more cycles of selection (Ertl and Fehr 1985; Concibido et al. 2003).

QTL associated with grain yield

The QTL associated with GY on Chr 16 harbors the shattering gene Pdh1 (Funatsuki et al. 2014), with the G. soja allele associated with adverse effects on the trait. Pdh1 increases pod dehiscence by promoting the torsion of the pod decreasing yield (Funatsuki et al. 2014). Breeding for shattering resistance (pdh1) allowed soybean production to expand to more arid conditions (Bandillo et al. 2017), and the selection for pdh1 soybean varieties was crucial for the expansion and the success of the industrialized soybean system (Funatsuki et al. 2014).

A novel QTL associated with grain yield (qGY-17) showed a significant 6% increase in GY for the G. soja allele as compared to the G. max allele when averaged across all RIL’s and environments, although the QTL was only significant in Pop3 (Williams 82 × PI522226) and across two environments. Similar results were found by Concibido et al. (2003) using backcross populations from G. soja × G. max, which identified positive effect of G. soja allele in two of six populations evaluated. Also, a QTL (qYIE-D2-1) in the same region was detected by Wang et al. (2014), and SNP markers were identified for pod number and seed weight (Hu et al. 2014; Yan et al. 2017). Further evaluation using different genetic background and environments will be necessary to understand better the impact of the qGY-17 on grain yield. Assessment of large yield testing in G. soja × G. max populations is challenging due to the presence of numerous deleterious agronomic traits (shattering, lodging, vine-grown) which result in difficulties in mechanic harvesting of plots.

Moreover, an impressive number of G. soja × G. max RILs yielded significantly more than the G. max parent (Williams 82) (Fig. 5 and Supplementary Table 3). Among those SA15-109, SA15-66476, SA15-63422, SA15-66599, SA15-63423, SA15-66563 and SA15-63650 yielded between 4 and 17% more than the Williams 82 (depending on the environment). The high yielding line SA15-109, from Williams 82 × PI458536, yielded 400 kg ha−1 more and matured one day later than Williams 82 across four environments in 2017. In Rock Port 2017, this line yielded 725 kg ha−1 more and matured on the same day as Williams 82. Another high yielding line SA15-66476, from the population Williams 82 × PI522226, yielded 420 kg ha−1 more and matured two days earlier than Williams 82 in 2016 and 2017 in the combined analysis (Fig. 5). Ertl and Fehr (1985) and Wang et al. (2004) were unsuccessful in increasing the yield potential in interspecific crosses using G. soja. On the other hand, Concibido et al. (2003) and Li et al. (2008) identified positive grain yield QTLs from G. soja crosses. A major outcome of this study was the identified positive transgressive segregates for grain yield using G. soja in the pedigree, and the best line significantly outperformed the Glycine max parent (Williams 82) by 400 kg ha−1 across all environments. Similar results were expected to be achieved using G. max × G. max crosses; however, using single crosses of G. max × G. soja is quite a surprise.

Fig. 5
figure 5

Selected G. soja-derived RILs exhibiting significantly higher grain yield than the NAM hub parent Williams 82 for test MG III and MG IV combined over four environments in 2016 (a), 2017 (b) and both years (2016 + 2017) (c)

Full size image

Summary

A total of 17 QTLs with alleles from G. soja that are associated with increased and decreased grain yield, extended plant maturity, increased plant height, and increased lodging were identified by the nested association mapping and composite interval mapping. A novel QTL for grain yield on chromosome 17 from G. soja was detected in the CIM analysis, and lines carrying the wild soybean allele presented an increase of 6% in grain yield. There have been very few reports in the long history of soybean genetics of QTL associated with increased grain yield derived from interspecific crosses with G. soja, and our results suggest further investigation is warranted to understand the potential utility of this QTL on breeding. Regions associated with plant maturity were identified on Chr 11, and 12 and the photoperiod response/plant maturity gene E1 was confirmed on Chr 6. A significant QTL associated with PH was identified on Chr 13, exhibiting increased plant height. The use of wild soybean germplasm for commercial breeding purposes is still quite rare for breeders, mainly due to difficulties in working with undomesticated plant germplasm. Our results indicate breeding with G. soja germplasm can successfully introduce new alleles and add valuable new beneficial alleles with the potential to increase valuable new genetic diversity to the current elite soybean gene pool.