Genome-Wide Association Study for Agronomic Traits in Wild Soybean (Glycine soja)

Abstract

The agronomic traits of soybean are important because they are directly or indirectly related to its yield. Cultivated soybean (Glycine max (L.) Merr) has lost genetic diversity during domestication and selective breeding. However, wild soybean (G. soja) represents a useful breeding material because it has a diverse gene pool. In this study, a total of 96,432 single-nucleotide polymorphisms (SNPs) across 203 wild soybean accessions from the 180K Axiom^® Soya SNP array were employed in the association analysis. Wild soybean accessions were divided into four clusters based on their genetic distance using ADMIXTURE, principal component analysis, and neighbor-joining clusters. The linkage disequilibrium decayed rapidly in wild soybean. A genome-wide association study was conducted for days to flowering (DtF), days to maturity (DtM), the number of pods (NoP), and the 100-seed weight (100SW), which are major agronomic traits for wild soybean accessions. A total of 22 significant SNPs were found to be associated with DtF, DtM, and the 100SW. Based on the detected SNP markers, Glyma.12g210400, a gene related to DtF, Glyma.17g115300, a gene related to DtM, and Glyma.14g140200, a gene related to the 100SW, were selected as candidate genes. The SNP markers related to agronomic traits identified in this study are expected to help improve the quality of soybean cultivars through selective breeding.

1. Introduction

Soybean (Glycine max L.) is one of the world’s most valuable legumes due to its use as a source of food, oil, livestock feed, and industrial materials. Because soybean is a self-pollinated crop, its homozygosity increases with successive generations, resulting in lower genomic variation [1]. Its genetic diversity has also been reduced due to genetic bottlenecks during domestication and selective breeding [2,3]. Therefore, a genetically diverse germplasm is required to improve the agronomic and seed quality characteristics of soybean. Wild soybean (Glycine soja Siebold & Zucc.) is the wild ancestor of cultivated soybean (G. max) and represents a valuable gene pool for improving the agronomic traits of soybean [4], particularly in terms of retrieving lost alleles. In addition, wild soybean may have unique alleles because they are found across a variety of climates and geographic topographies [5].

The individual growth stages of soybean can affect its seed composition, harvest time, and yield. In particular, flowering, which refers to plant transition from vegetative stage to reproductive stage [6], is related to photoperiod response and vernalization and is the most important agronomic characteristic of soybean in that it determines regional adaptation and yield [7,8,9]. In addition, the number of pods and the seed weight have a positive correlation with soybean yields [10]. Soybean is a short-day plant, and its growth and flowering are regulated by the photoperiod [11]. The flowering and maturity of soybean are controlled by multiple genes, with 11 major genes (E1 through E11) having been identified across various soybean chromosomes to date [12,13,14,15,16,17,18,19]. The E1 gene, which encodes a legume-specific transcription factor, is located on chromosome 6 and has been reported to be involved in the regulation of flowering and maturity in soybean [12]. In addition, the E1, E2, E3, E4, E5, and E7 genes are sensitive to long days and/or light quality [13,20,21]. The mechanism of action or function of these genes has been confirmed to be homologous with Arabidopsis thaliana, which is widely used as a model for plant research [22,23,24]. In addition, the seed size, seed weight, and number of pods are controlled by the GmCYP78A10 gene in soybean, with a 7.2% difference in seed weight and 5.8% difference in the number of pods between the wild soybean allele GmCYP78A10A and the cultivated soybean allele GmCYP78A10B [25].

Bi-parent QTL analysis generally uses populations derived from both parental lines and is limited to a specific phenotype. On the other hand, a genome-wide association study (GWAS) typically uses natural populations and allows for a comparison of various genotypes and phenotypes [26]. GWASs have been successfully employed to identify QTLs and genes in various crops, such as rice, wheat, maize, and soybean [27,28,29,30,31,32]. Factors such as selection, migration, local adaptation, geographic isolation, and genetic shifts affect the population structure, leading to differences in allele frequencies, and these differences in the population structure can lead to spurious associations between traits and markers in GWASs [33]. In addition, limited linkage disequilibrium (LD) between pairs of adjacent markers can originate from a set of unrelated genotypes, which requires a fairly large number of markers to resolve [32].

With the development of high-throughput genotyping technology, the use of single-nucleotide polymorphism (SNP) genotyping has become easier. In Korea, a 180K Axiom^® Soya SNP array has been developed from more than 4 million high-quality SNPs identified from the resequencing of 16 Korean and 31 Chinese genetic resources [34]. This array has been used in the structural analysis of genetic diversity [35] and in association analysis for various traits, such as the flowering time, ultraviolet B resistance, and the soyasaponin biosynthesis of soybean [36,37,38]. The purpose of this study was to investigate the genetic diversity of soybean using high-density SNP chip data from 203 wild soybean accessions and to detect candidate genes for major agricultural traits that can affect soybean yield using GWAS analysis.

2. Materials and Methods

2.1. Plant Materials and Experiment Field Management

A total of 203 wild soybean accessions were obtained from the Rural Development Administration (RDA) in South Korea. These accessions were collected from South Korea (167), China (22), Japan (11), and Russia (1), with two from an unknown location (Table S1). They were cultivated in an experimental field at Chonnam National University (Gwangju, 36°17′ N, 126°39′ E, Republic of Korea) in 2015 and 2016. Seeds were sown in late June and harvested from September to November based on the flowering time and maturity. Each accession was planted in a single hill plot at a spacing of 1 × 1 m with two replications. Compound fertilizer composed of nitrogen, phosphorus pentoxide, and potassium oxide at a ratio of 8:8:9 was applied at 40 kg per 1000 square meter before sowing, and pesticides and fungicides were alternately applied every two weeks until one month before harvest.

2.2. Phenotypic Evaluation and Data Analysis

The agronomic traits investigated were days to flowering (DtF), days to maturity (DtM), number of pods (NoP), and 100-seed weight (100SW). The DtF was recorded as the date when the first flower for each individual plant bloomed. The DtM was recorded as the number of days between flowering and the date when approximately 90% of the pods exhibited the color associated with maturity. The NoP was measured as the number of pods per raceme. For 100SW, mature seeds were dried in a dry oven at 30 °C for 72 h, and then 100 of them were randomly selected and weighed. Descriptive statistics and correlation analysis were performed with SPSS 25.0 software, Microsoft Excel 2016, and Python 3.9 using the phenotypic data for each trait. The broad-sense heritability (h²) was calculated using QTLmaxV2 software [39] using the following equation:

$$h^2=\frac{{\mathsf{\sigma }}_{\mathrm{u}}^2}{{\mathsf{\sigma }}_{\mathrm{u}}^2+{\mathsf{\sigma }}_E^2}$$

where ${\mathsf{\sigma }}_{\mathrm{u}}^2$ is the variance component for random genetic effects and ${\mathsf{\sigma }}_E^2$ is the variance component for environmental effects.

2.3. DNA Extraction and SNP Genotyping

Fresh leaf tissue was collected for DNA extraction at the beginning of growth and ground in a mortar using liquid nitrogen. Genomic DNA was isolated from 20 mg of lyophilized leaf tissue using a DNeasy Plant Mini Kit (Qiagen, Valencia, CA, USA) according to the manufacturer’s protocol. The quality and quantity of the extracted total DNA were verified using a Nano-MD UV-Vis spectrophotometer (Scinco, Seoul, Korea). The extracted DNA was stored in a freezer at −80 °C until further use. A total of 203 wild soybean accessions were genotyped using the 180K Axiom^® Soya SNP array [34]. SNPs with genotypes observed in less than 95% of the sample were removed from further analysis (CallRate < 0.95). Similarly, one individual had missing genotypes for more than 5% of the SNPs, so it was excluded from the sample. SNPs with a minor allele frequency (MAF) under 5% were also eliminated to remove low-quality SNPs, while duplicated parts of the raw data were eliminated using R software [40]. Missing genotypes were imputed using BEAGLE software 3.0 [41]. As a result, a total of 96,432 SNPs were used for analysis.

2.4. Population Structure and Genetic Diversity

The population structure was determined using ADMIXTURE v1.30 [42]. In this analysis, the optimal K value was assigned to different admixture models (K = 2 to 10) based on the cross-validation error. The box plots were generated using the “sort by Q” command based on the best K with the lowest cross-validation error. For the proportional member probability (Q), the cut-off assignment for each cluster was >0.55. The phylogenetic trees were drawn based on the maximum-likelihood approach using 1000 bootstrap replicates and inter-site ratios for the automatic initial trees (neighbor-joining) in MEGA11 [43]. A three-dimensional principal component analysis (PCA) plot was used to identify the distribution of the genetic variation between geographic locations using PLINK v1.07 [44]. The neighbor-joining tree and PCA used the same color based on the classification from the ADMIXTURE analysis.

2.5. Linkage Disequilibrium Estimation and Candidate Gene Identification

LD estimation was conducted using TASSEL with an LD sliding window of 500 within individual chromosomes [45]. The identification of candidate genes was carried out in the region considering the LD for the SNP detected as a result of the GWAS and in the downstream and upstream regions. Glyma.Wm82.a2.v1, a soybean reference genome from www.soybase.org (accessed on 1 November 2022), was used as the gene model for candidate gene identification.

2.6. Genome-Wide Association Study Analysis

The GWAS was conducted with a linear mixed model using the restricted maximum-likelihood (REML) algorithm and considering the population structure from ADMIXTURE analysis and an identity-by-state (IBS) similarity matrix. The IBS matrix was obtained using the proportions of IBS alleles and used as a kinship matrix in PLINK v1.07. Genomic control was also conducted to correct for genomic inflation, which means that the p value did not follow a chi-square distribution with 1 degree of freedom and that the statistical significance was exaggerated in QTLmaxV2. In order to increase the resolution between environments, association analysis was conducted using a merged phenotype [39]. The −log₁₀(p) threshold of the Manhattan plot was calculated using the Bonferroni method (p = α/n). With 96,432 SNPs employed in this study, at α = 1 and α = 0.05, the Bonferroni-corrected thresholds for the p values were 1.04 × 10⁻⁵ (α = 1) and 5.19 × 10⁻⁷ (α = 0.05), with corresponding −log₁₀(p) values of 4.98 for the suggestive threshold and 6.29 for the significance threshold [46].

3. Results

3.1. Phenotypic Variation and Correlation Analysis

To evaluate the major agronomic traits of soybean, a total of 203 wild soybean accessions were cultivated in 2015 and 2016. The descriptive statistics and heritability (h²) for each trait are summarized in

Table 1. Descriptive statistics for the agronomic traits in 2015 and 2016.

Trait	Year	Min	Max	SD	Mean	CV (%)	Skew	Kur	h2
DtF	2015	29	80	7.69	64.61	11.90	−2.33	6.39	0.89
	2016	26	81	8.71	61.11	14.26	−1.34	3.11
	Mean	28	81	7.90	63.20	12.50	−1.91	5.03
DtM	2015	44	82	6.22	51.98	11.98	2.28	6.04	0.55
	2016	42	90	9.02	60.29	14.96	0.71	0.33
	Mean	43	83	6.67	56.15	11.87	1.33	2.23
NoP	2015	2.0	11.4	1.47	5.36	27.51	0.91	1.58	0.39
	2016	1.8	10.2	1.60	4.63	34.55	0.76	0.61
	Mean	2.3	10.2	1.19	4.99	23.88	0.69	1.53
100SW	2015	1.30	10.57	1.20	3.15	38.00	2.80	11.01	0.75
	2016	1.15	7.69	1.01	2.88	35.00	2.01	5.85
	Mean	1.26	7.83	1.02	3.00	33.99	2.33	7.17

DtF, days to flowering; DtM, days to maturity; NoP, number of pods; 100SW, 100-seed weight; Min, minimum; Max, maximum; SD, standard deviation; CV, coefficient of variation; Skew, skewness; Kur, Kurtosis; h², broad sense heritability.

.

In 2015 and 2016, the average DtF was 28 to 81 d, with an average of 63 d, and DtM was 43 to 83 d, with an average of 56 d. The NoP was 2.3 to 10.2, with an average of 4.99, and the 100SW was 1.26 to 7.83 g, with an average of 3 g. Of the four traits, h² was the highest for DtF (0.89) and the lowest for the NoP (0.39), while the coefficient of variation was the highest for the 100SW (33.99) and lowest for DtM (11.87). As shown in Figure 1, these traits were quantitatively inherited with an approximately normal distribution in the two environments.

The correlation coefficients between the agronomic traits in each environment are presented in Figure 2. In 2015 and 2016, DtF and DtM exhibited a negative correlation (r = −0.729 ** and −0.440 **, respectively). DtF and the 100SW also showed a negative correlation (r = −0.390 ** and −0.345 **, respectively). On the other hand, DtM and the 100SW had a positive correlation (r = 0.532 ** and 0.423 **, respectively). DtF and the NoP, DtM and the NoP, and the 100SW and the NoP had a significant correlation in 2015 but not in 2016.

3.2. Population Structure and Linkage Disequilibrium

To understand the population structure of the 203 wild soybean accessions, various statistical analyses were conducted. First, cross-validation was used to cluster the populations, with K = 4 selected for this classification because it had the lowest error (Figure 3a). Analysis with the parametric clustering method using ADMIXTURE also found that the model with four ancestral populations (Q1–Q4) best described the genetic structure. According to the membership coefficient (qi > 0.55), of the 203 accessions, 178 were assigned to Q1–Q, and 25 accessions were assigned to an admixed group (Figure 3b). In particular, 13 were assigned to Q1, all of which were collected in China, 16 were assigned to Q2 (15 from Korea and 1 unknown), 53 to Q3, (42 from Korea and 11 from Japan), and 96 to Q4 (90 from Korea, 4 from China, 1 from Russia, and 1 unknown). The admixed group contained 25 accessions, consisting of 20 from Korea and 5 from China. In the PCA, the 203 wild soybean accessions were divided into four main components (Figure 3c). The three-dimensional PCA plot explained 59% of the total genetic variance (28% for PC1, 21% for PC2, and 10% for PC3). Neighbor-joining cluster analysis using MEGA11 also confirmed that the 203 accessions were clearly separated into four clusters (Figure 3d). As a result, the ADMIXTURE-based PCA analysis and neighbor-joining cluster analysis strongly indicated that the 203 wild soybean accessions consisted of four well-differentiated genetic sub-populations.

By plotting r² against the distance (kb) between pairs of SNPs, it was found that the LD decay sharply decreased. Estimating the LD decay for the entire genome resulted in a decrease in the maximum r² drop to half at 6 kb (Figure 4), and the average LD decay for each of the 20 chromosomes was 4 to 10 kb (data not shown). Consequently, the size of the LD block was considered to be quite small.

3.3. Genome-Wide Association Study for Agronomic Traits

In this study, association analysis was performed with a linear mixed model for the four major agronomic traits. According to the results of the Manhattan and QQ plots presented in Figure 5, 18 SNPs across 9 chromosomes for DtF, 32 SNPs across 7 chromosomes for DtM, and 41 SNPs across 16 chromosomes for the 100SW were detected at the significance threshold level (−log₁₀(p) = 6.29). However, no suggestive or significant SNPs were detected for the NoP.

Of the detected SNPs, the most significant SNP (i.e., with the lowest p value) in a particular associated genetic region was selected as a causal SNP for the trait. For DtF, five SNP markers were identified on chromosomes 6 (AX-90393598), 11 (AX-90507114), 12 (AX-90495922), 16 (AX-90386690), and 17 (AX-90395214) (

Table 2. Significant SNP markers related to DtF in the merged phenotype.

SNP	Chr	Position	−log10(p)	Reference +	Minor	Major	MAF
AX-90393598	6	9,723,612	6.05	G	T	G	0.09
AX-90507114	11	34,284,790	6.06	G	A	G	0.07
AX-90495922	12	36,991,550	8.49	C	C	T	0.08
AX-90386690	16	1,150,092	6.50	G	G	A	0.16
AX-90395214	17	9,164,086	7.43	T	T	C	0.08

SNP, single-nucleotide polymorphism; Chr, chromosome; MAF, minor allele frequency. ⁺ gene model: Glyma.Wm82.a2.v1.

).

For DtM, six SNP markers were identified on chromosomes 12 (AX-90440044 and AX-90495922), 14 (AX-90366397), 15 (AX-90472604), 16 (AX-90351904), and 17 (AX-90338412) (

Table 3. Significant SNP markers related to DtM in the merged phenotype.

SNP	Chr	Position	−log10(p)	Reference +	Minor	Major	MAF
AX-90440044	12	7,550,988	7.29	T	T	C	0.08
AX-90495922	12	36,991,550	9.82	C	C	T	0.08
AX-90366397	14	5,987,400	8.2	A	A	C	0.06
AX-90472604	15	10,746,176	8.03	G	G	T	0.07
AX-90351904	16	2,017,593	6.37	T	T	C	0.09
AX-90338412	17	9,141,000	7.43	A	A	T	0.07

SNP, single-nucleotide polymorphism; Chr, chromosome; MAF, minor allele frequency. ⁺ gene model: Glyma.Wm82.a2.v1.

).

In addition, 12 SNP markers associated with the 100SW were identified on chromosomes 1 (AX-90454597), 8 (AX-90377215), 10 (AX-90374196), 12 (AX-90408186 and AX-90370905), 13 (AX-90318417), 14 (AX-90483232 and AX-90383559), 15 (AX-90472604), 16 (AX-90416982), and 17 (AX-90375042) (

Table 4. Significant SNP markers related to 100SW in the merged phenotype.

SNP	Chr	Position	−log10(p)	Reference +	Minor	Major	MAF
AX-90454597	1	46,804,555	11.05	C	C	T	0.07
AX-90377215	8	213,783	7.87	T	T	C	0.09
AX-90374196	10	51,266,958	11.10	C	C	A	0.09
AX-90408186	12	6,713,247	13.20	A	A	C	0.06
AX-90370905	12	34,609,739	14.22	T	T	C	0.09
AX-90318417	13	14,296,524	9.26	C	C	T	0.07
AX-90483232	14	3,979,096	10.57	C	C	G	0.06
AX-90383559	14	47,044,439	11.99	A	A	T	0.07
AX-90472604	15	10,746,176	13.28	G	G	T	0.07
AX-90416982	16	1,413,306	12.80	G	G	A	0.05
AX-90375042	17	7,394,535	10.18	G	G	A	0.09

SNP, single-nucleotide polymorphism; Chr, chromosome; MAF, minor allele frequency. ⁺ gene model: Glyma.Wm82.a2.v1.

).

Interestingly, the AX-90495922 SNP marker on chromosome 12 was consistently associated with both DtF and DtM, while AX-90472604 on chromosome 15 was significantly associated with DtM and the 100SW. For each trait, the allelic effect of the SNP marker with the lowest p value was estimated. The DtF-associated SNP marker AX-90495922 on chromosome 12 had the alleles T/C, and the average DtF for the individuals with TT alleles was 64 d, 15 d longer than the average DtF for the individuals with CC alleles (49 d). The DtM-associated SNP marker AX-90366397 on chromosome 14 had the alleles C/A, and the average DtM for the individuals with CC alleles was 55 d, which was 16 d shorter than the average DtM for the individuals with AA alleles (71 d). The 100SW-associated SNP marker AX-90383559 on chromosome 14 had the alleles T/A, and the average 100SW for the individuals with TT alleles was 2.83 g, which was 3.28 g lighter than the average 100SW for individuals with AA alleles (6.11 g) (Figure 6).

3.4. Candidate Genes for Trait-Associated SNP Markers

A search for candidate genes for the SNPs identified from the GWAS results was conducted based on whether or not it was included in the gene range, because the LD for the entire genome was quite small, with the search occurring within about 50 kb. Of the total 22 SNP markers associated with DtF, DtM, and the 100SW, 20 were located within a genic region. Of these, four DtF-related SNP markers were located in the Glyma.06g119400, Glyma.11g251500, Glyma.12g210400, and Glyma.17g116200 genes (

Table 5. Candidate genes associated with DtF in wild soybean.

SNP	Chr	Position	Gene +	Location (bp)	Gene Description
AX-90393598	6	9,723,612	Glyma.06g119400	9719172..9726230	S-adenosyl-L-methionine-dependent methyltransferases superfamily protein
AX-90507114	11	34,284,790	Glyma.11g251500	34283168..34289333	squamosa promoter binding protein-like 2
AX-90495922	12	36,991,550	Glyma.12g210400	36943077..36946491	14-3-3 protein
AX-90395214	17	9,164,086	Glyma.17g116200	9181066..9183991	CCCH-type zinc finger family protein

SNP, single-nucleotide polymorphism; Chr, chromosome; ⁺ gene model: Glyma.Wm82.a2.v1.

), while DtM-related SNP markers were located in Glyma.12g091600, Glyma.12g210400, Glyma.14g071300, Glyma.15g133700, Glyma.16g021200, and Glyma.17g115300 (

Table 6. Candidate genes associated with DtM in wild soybean.

SNP	Chr	Position	Gene +	Location (bp)	Gene Description
AX-90440044	12	7,550,988	Glyma.12g091600	7494222..7499978	OTUBAIN-LIKE DEUBIQUITINASE 1
AX-90495922	12	36,991,550	Glyma.12g210400	36943077..36946491	14-3-3 protein
AX-90366397	14	5,987,400	Glyma.14g071300	5986366..5990379	RING/U-box superfamily protein
AX-90472604	15	10,746,176	Glyma.15g133700	10744186..10748830	Glycosyl hydrolases family 31 protein
AX-90351904	16	2,017,593	Glyma.16g021200	1996084..1998423	Cytochrome P450, Family 78
AX-90338412	17	9,141,000	Glyma.17g115300	9122203..9127019	NRT1

SNP, single-nucleotide polymorphism; Chr, chromosome; ⁺ gene model: Glyma.Wm82.a2.v1.

), and 100SW-related SNP markers were located in Glyma.01g140200, Glyma.08g002900, Glyma.10g295400, Glyma.12g184700, Glyma.13g047500, Glyma.14g050900, Glyma.14g205200, Glyma.15g133700, Glyma.16g016000, and Glyma.17g094400 (

Table 7. Candidate genes associated with 100SW in wild soybean.

SNP	Chr	Position	Gene +	Location (bp)	Gene Description
AX-90454597	1	46,804,555	Glyma.01g140200	46825221..46841112	arginase
AX-90377215	8	213,783	Glyma.08g002900	211294..221624	cyclin-dependent kinase E;1
AX-90374196	10	51,266,958	Glyma.10g295400	51258630..51267380	2-isopropylmalate synthase 1
AX-90370905	12	34,609,739	Glyma.12g184700	34608320..34611560	Homeodomain-like superfamily protein
AX-90318417	13	14,296,524	Glyma.13g047500	14291927..14296667	NagB/RpiA/CoA transferase-like superfamily protein
AX-90483232	14	3,979,096	Glyma.14g050900	3977944..3980627	GDSL-like Lipase/Acylhydrolase superfamily protein
AX-90383559	14	47,044,439	Glyma.14g205200	47041931..47046048	cinnamate-4-hydroxylase
AX-90472604	15	10,746,176	Glyma.15g133700	10744186..10748830	Glycosyl hydrolases family 31 protein
AX-90416982	16	1,413,306	Glyma.16g016000	1400064..1404488	Purple acid phosphatase 29
AX-90375042	17	7,394,535	Glyma.17g094400	7393359..7395553	Homeodomain-like superfamily protein

SNP, single-nucleotide polymorphism; Chr, chromosome; ⁺ gene model: Glyma.Wm82.a2.v1.

).

The candidate genes with the strongest SNP variant for each trait were Glyma.12g210400 for DtF, Glyma.14g071300 for DtM, and Glyma.14g205200 for the 100SW, corresponding to the 14-3-3 protein, RING/U-box superfamily protein, and cinnamate-4-hydroxylase, respectively.

4. Discussion

4.1. Genetic Diversity and Origin of Wild Soybean

Wild soybean, which is the ancestor of cultivated soybean, may have a diverse gene pool that has since been lost during domestication and cultivation. The 203 wild soybean accessions used in this study show 99% coverage, with 1135 wild soybean plants as the original population, using the 180K SNP array from the RDA. This collection has high genetic diversity without overlaps. In particular, Korean wild soybean accessions are valuable as a research material because they are reported to contain many rare alleles [47].

For association analysis, it is important to determine the genetic basis of complex traits using population structure analysis. With the ADMIXTURE and PCA analysis strongly supported by neighbor-joining clustering, the 203 wild soybean accessions were clearly clustered according to their origin into four groups: (1) China, (2) Korea, (3) Korea and Japan, and (4) Korea and China. A more detailed classification of the origin was not possible. The wild soybean in Korea was divided into three clusters, which may be related to imports and exports from neighboring countries. Given that Korea and Japan were grouped together in Q3, there is a possibility that wild soybean was transferred from Korea to Japan. Similarly, Korea and China were grouped together in Q4, with the accessions collected from Heilongjiang and Shandong in China, which are relatively close to Korea. Based on the fact that the Korean accessions were divided into three clusters mixed with neighboring countries, it can be inferred that wild soybeans around the Korean Peninsula contain high genetic diversity.

Cultivated soybean has been bred to increase its vegetative growth period and maturation period and increase the seed weight to ensure high yields. In the present study, it was confirmed that the major alleles for DtF-, DtM-, and 100SW-associated SNPs detected in the wild soybean accessions delayed the flowering period, shortened the maturation period, and reduced the seed weight (Figure 6). When alleles of each trait were compared with the reference genome for cultivated soybean (William 82), DtF was identical, while DtM and the 100SW differed (Table 2, Table 3 and Table 4). This suggests that, in the process of breeding wild soybean to produce cultivated soybean, selection for flowering, maturity, and seed weight occurred.

4.2. Correlation of Major Agronomic Traits

The discovery of correlations between agricultural traits is useful for breeding programs. In the present study, traits that exhibited a correlation during both growing years were DtF and DtM, DtF and the 100SW, and DtM and the 100SW. On the other hand, there was a correlation between the NoP and other traits in 2015 but not in 2016 (Figure 2). Therefore, the NoP affected the inter-year correlation because it was greatly influenced by the environment. Liu et al. (2011) reported that the soybean yield had a negative correlation with the flowering time, which is consistent with the results of the present study [7]. In addition, the strong negative correlation between DtF and DtM and the positive correlation between DtM and the 100SW was in agreement with the negative correlation between DtF and the 100SW. On the other hand, it has also been reported that both DtM and the 100SW and DtF and DtM have a positive correlation [48]. These differences from the present results may be due to differences in climatic conditions, such as temperature and day length, because the experimental cultivation in the previous study took place in Indonesia.

The correlations between these agricultural traits are important because the characteristics of soybean growth stages can influence changes in seed composition, the harvest period, and the yield. It has been widely reported that, in soybean, the yield increases as both the maturation period and seed weight increase [49,50]. When comparing the genotype of the SNP markers detected for each trait and the genotype of the cultivated soybean reference genome, the cultivated soybean was found to have a genotype that increased DtM and the 100SW. Therefore, this suggests that cultivated soybean has been bred from wild soybean to increase its yield.

4.3. Candidate Genes for Major Agronomic Traits

It has been reported that the reduction in genetic diversity via selection during the domestication and cultivation of crops increases the LD [51]. Depending on the size of the LD, it may be easy to search for a candidate gene for a trait but, if the size of the LD is too large, an excessive number of candidate genes will be detected. In the present study, the LD was estimated using wild soybean, the ancestor of cultivated soybean, and the average LD was found to be 6 kb, which is considerably smaller than that of cultivated soybean, which is about 150 kb [1]. These results are in agreement with a previous study reporting that the LD range for wild soybean is ~27 kb, smaller than the LD of soybean landraces and improved cultivars 83 kb and 133 kb, respectively [52]. These results are also consistent with the LD of ~11 kb reported by Kim et al. (2021) [53], which was obtained from resources similar to those employed in the present study. Therefore, more precise candidate gene detection is possible.

No significant markers were found for the NoP in the GWAS results. As shown in Table 1, the heritability of the NoP was 0.39, which was quite low compared to the other traits, suggesting that it is greatly influenced by the environment. Cultivated soybean has been bred to adapt to a universal environment, while the wild soybean accessions used in the present study were collected from various environments. The gene Glyma.06g119400 containing the DtF-related SNP marker AX-90393598 was annotated as S-adenosyl-L-methionine-dependent methyltransferase superfamily protein (SAM-Mtase). It has been reported that mutations in SAM-Mtase, an orthologue of trimethylguanosine synthase1, are involved in the regulation of the vernalization response and flowering time in the legume narrow-leafed lupin (Lupinus angustifolius L.) [54]. In addition, in the study of the lilium “Siberia”, there is a report that SAM-Mtase is related to benzenoid biosynthesis during flower development [55]. In soybean, it has been reported that SAM-Mtase is involved in expressed gene interaction at temperatures of 35 °C and under dark stress during the seedling stage [56]. Soybean is a short-day plant, and its growth and flowering are regulated by the photoperiod [11]. In addition, soybean generally blooms in summer when the temperature is high. Thus, SAM-Mtase from the Glyma.06g119400 gene can be considered to affect flowering.

Gene Glyma.11g251500, which contains SNP marker AX-90507114, was annotated as squamosa promoter-binding protein-like 2 (SPL2). SPL family plays an important role in the flowering pathway, pollen development, and fertility for soybean [57,58]. The SNP marker AX-90495922 was detected for both DtF and DtM, and the gene Glyma.12g210400 was selected as a strong candidate gene. This marker had the highest −log₁₀(p) of 8.49 and 9.82 (Table 4 and Table 5) for DtF and DtM, respectively; in particular, for DtF, there was a 15 d difference between the major and minor alleles (Figure 6). Glyma.12g210900 encodes the 14-3-3 protein, and analysis of the rice transcription factor FLOWERING LOCUS D1 (OsFD1) revealed that 14-3-3 plays an essential role in flowers [59]. In addition, 14-3-3 protein has been found to interact with FLOWERING LOCUS T(FT) in Arabidopsis thaliana and with PRUNING(SP) and SINGLE FLOWER TRUSS(SFT) acting as anti-florigen and florigen in tomatoes [60,61,62]. In the present study, it was found that DtF and DtM were significantly correlated. Additional studies are, thus, needed to determine whether these candidate genes affect soybean by a similar mechanism.

In addition, the Glyma.17g116200 gene around the SNP marker AX-90395214 was annotated as a CCCH-type zinc finger family protein. CCCH-type zinc finger family proteins are known to regulate adaptation to various stresses, such as drought, salinity, and temperature [63]. Abiotic stress has a significant association with flowering time, which is a key plant survival strategy [64]. In addition, Glyma.16g021200, which was related to maturation, is a nitrate transporter and is known to affect early embryonic development in Arabidopsis [65]. In this study, genes located at E1 to E11 and j [12,13,14,15,16,17,18,19,66], known as soybean flowering-related loci, were not detected. It is presumed that this is not only because the accessions used in the study were different, but also because wild soybean was used.

The 100SW is an important criterion for determining and estimating the yield. The Glyma.01g140200 gene, located on chromosome 1, is an arginase and has been proven to be related to yield by playing an important role in nitrogen transport and storage in rice [67]. SNP AX-90383559 detected for the 100SW is located in the Glyma.14g205200 gene and was annotated as cinnamate-4-hydroxylase (C4H). It has been reported that C4H acts as a defense mechanism against various pathogens by catalyzing the synthesis of lignin during seed development [68,69]. During this stage, pathogens can significantly reduce the yield [11]. Therefore, it is suggested that the detected candidate gene Glyma.14g205200 indirectly affects the 100SW. However, GmCYP78A10 [25], known to be a gene related to seed weight in wild soybean, was not found in the present study. This could be because the gene has been identified in mixed accessions of wild soybean and cultivated soybean in previous analyses, but the present study only investigated wild soybean. The Glyma.16g016000 gene, located approximately 10 kb downstream of the marker AX-90416982, encodes purple acid phosphatase 29. It has been reported that an increase in the expression level of purple acid phosphatase, which plays an important role in phosphate acquisition in chickpeas, leads to an increase in seed weight [70]. Additional expression analysis of the candidate genes detected in this study is required, potentially contributing to soybean breeding.

5. Conclusions

In this study, clustering using the genetic diversity of wild soybean was conducted, and candidate genes related to four target agricultural traits were searched for using GWAS analysis. The 203 wild soybean accessions were clustered into four groups according to origin, which included China, Korea, and Japan. The LD of the 203 wild soybean accessions was quite low, which could serve as an advantage or a disadvantage in the detection of candidate genes. Glyma.12g210400, Glyma.17g115300, and Glyma.01g140200 were selected as potential candidate genes for DtF, DtM, and the 100SW. The genetic diversity of wild soybean, which can be hybridized with cultivated soybean, thus offers a useful source for soybean breeding programs.