Genome-wide association and transcriptional studies reveal novel genes for unsaturated fatty acid synthesis in a panel of soybean accessions

Background The nutritional value of soybean oil is largely influenced by the proportions of unsaturated fatty acids (FAs), including oleic acid (OA, 18:1), linoleic acid (LLA, 18:2), and linolenic acid (LNA, 18:3). Genome-wide association (GWAS) studies along with gene expression studies in soybean [Glycine max (L.) Merr.] were leveraged to dissect the genetics of unsaturated FAs. Results A association panel of 194 diverse soybean accessions were phenotyped in 2013, 2014 and 2015 to identify Single Nucleotide Polymorphisms (SNPs) associated with OA, LLA, and LNA content, and determine putative candidate genes responsible for regulating unsaturated FAs composition. 149 SNPs that represented 73 genomic regions were found to be associated with the unsaturated FA contents in soybean seeds according to the results of GWAS. Twelve novel genes were predicted to be involved in unsaturated FA synthesis in soybean. The relationship between expression pattern of the candidate genes and the accumulation of unsaturated FAs revealed that multiple genes might be involved in unsaturated FAs regulation simultaneously but work in very different ways: Glyma.07G046200 and Glyma.20G245500 promote the OA accumulation in soybean seed in all the tested accessions; Glyma.13G68600 and Glyma.16G200200 promote the OA accumulation only in high OA germplasms; Glyma.07G151300 promotes OA accumulation in higher OA germplasms and suppresses that in lower OA germplasms; Glyma.16G003500 has the effect of increasing LLA accumulation in higher LA germplasms; Glyma.07G254500 suppresses the accumulation of LNA in lower OA germplasms; Glyma.14G194300 might be involved in the accumulation of LNA content in lower LNA germplasms. Conclusions The beneficial alleles and candidate genes identified might be valuable for improving marker-assisted breeding efficiency and exploring the molecular mechanisms underlying unsaturated fatty acid of soybean. Electronic supplementary material The online version of this article (10.1186/s12864-019-5449-z) contains supplementary material, which is available to authorized users.


Background
Soybean (Glycine max L. Merr.), a major oil crop worldwide, contributes 31% of the global supply of edible vegetable oil [1,2]. Soybean oil consists of saturated fatty acids (FAs), including palmitic acid (PA, 16:0) and stearic acid (SA, 18:0) and unsaturated FAs such as oleic acid (OA, 18:1), linoleic acid (LLA, 18:2), and linolenic acid (LNA, 18:3). The amounts and relative proportions of the five FAs, especially unsaturated FAs directly determine the flavor, stability, and nutritional value of soybean oil [3]. Consumption of soybean oils with higher OA levels is desirable because this monounsaturated FA could improve storage time, which increases the price of the soybean oil and in the cost of generating unhealthy trans fat for humans [4]. LLA and LNA can't be synthesized by the human body, and in turn have to be obtained from the daily diet [5]. Although the nutritional value of unsaturated FAs benefits humans, the high levels of LLA and LNA in soybean oil results in low oxidative stability and rapid rancidity, as well as reduces storage time of soybean oil [6]. Hence, the increase in OA and decrease in LLA and LNA contents could improve the quality of soybean oil for human consumption and thus has become an important goal of soybean breeders.
The unsaturated FAs in soybean are typical quantitative traits controlled by major and minor genes/quantitative trait loci (QTLs) and are easily affected by environment or by genotype X environment interaction factors [7]. Identifying unsaturated FA QTLs based on biparental mapping populations through linkage mapping accelerates the development of soybean lines with reasonable unsaturated FA content and relative proportions to meet the widespread demand for soybean oil. Currently, nearly 200 QTLs associated with unsaturated FA concentrations have been reported (http://www.soybase.org), which distributed across 20 chromosomes (Chr.) or linkage groups (LG) of soybean. However, the utilization in breeding programs of these QTLs for developing soybean lines with reasonable unsaturated FA content is limited due to poor resolution and accuracy of these QTLs, which is caused by lower recombination events based on biparental mapping populations or genomic regions with higher linkage disequilibrium (LD) [8]. Another reason for the reported lower utilization efficiency of these QTLs is lower consistency across different genetic backgrounds. Unlike linkage analysis, genomewide association studies (GWAS) have been utilized as an alternative method to identify unsaturated FA QTLs [8]. GWAS based on the natural population have more extensive recombination events, thereby resulting in shorter LD segments and therefore increased resolution and accuracy of marker-phenotype associations. The developments in genome sequencing and single nucleotide polymorphism (SNP) genotyping technology have promoted the applicability of GWAS in soybean, which has been used to determine genomic regions that are associated with resistance to biological stress, including soybean cyst nematode [9][10][11] and white mold [12], abiotic stress [13,14], yield-related traits, including seed weight [15], flower time [16], and seed composition such as seed oil content [17] and seed protein content [18]. For unsaturated FAs, presently only two GWAS based on the high-throughput technologies have been reported. A targeted GWAS was used to analyze the genetic architecture of unsaturated FAs using only 1536 SNPs and 421 diverse accessions, and the results showed that 8, 12, and 5 Quantitative Trait Nucleotide (QTNs) were associated with LLA, LNA, and OA content [19]. Compared to previously identified known genes/QTLs, only a few novel significant association signals were detected due to the utilization of only thousands of markers [19]. Leamy et al. [20] conducted a GWAS of unsaturated FAs using nearly 30,000 SNPs and found 9, 5, and 5 QTNs were associated with LLA, LNA, and OA content. The results of Leamy et al. [20] did not fully agree with those of other studies because different cultivated soybeans were used as parents in each linkage analysis. However, by late 2017, no studies have been conducted to identify QTNs and candidate genes underlying unsaturated FA content of cultivated soybean based on enormous SNP markers phenotype evaluation in multi-environments.
The present study performed a GWAS of unsaturated FAs based on 36,981 SNPs and 194 germplasms from Northeastern China. The aim of the present study was to identify QTNs associated with unsaturated FA levels, to screen candidate genes located within SNP regions, and to verify the role of candidate genes related to soybean FA metabolism.

Results
Phenotyping and statistical analysis for seed unsaturated FA content There was a wide range of phenotypic variations in the levels of the three unsaturated FAs, particularly those of OA in 194 accessions ( Fig. 1 and Additional file 1: Table S1). Genetic parameters, including mean, standard deviation, coefficient of variation, skewness, and kurtosis for OA, LLA, and LNA content in the tested GWAS panel were calculated (Additional file 1: Table S1). The differences in the levels of the three unsaturated FAs among the three tested environments were not statistically significant, as indicated by the coefficient of variation (Additional file 1: Table S1), suggesting that the contents of the three unsaturated FAs coincided with the three study environments. Among the three tested unsaturated FAs, LNA was the least variable, whereas OA and LLA had more variation (Additional file 1: Table S1). The wide range of variations in content of the three unsaturated FAs in the 194 accessions was found and continuous the distribution of the three unsaturated FAs in this GWAS panel was generally normal.

Distribution of SNP markers and LD
The genotyped samples consisted of a set of 194 soybean germplasms representing a diverse range of unsaturated FA contents (Additional file 2: Supplementary Data). A total of 36,981 SNPs with minor allele frequencies (MAF) > 0.04 and missing data of < 10% were used for the estimation of LD of the soybean accessions. These SNP markers were evenly distributed among 20 soybean chromosomes, resulting in an average SNP density of 1 SNP every 25.67 kb in the unique genomic region. The mean LD was 227 kb (Additional file 3: Figure S1).

QTNs associated with the content of three unsaturated FAs identified by GWAS
A total of 120 SNPs that covered all 20 soybean chromosomes and represented 62 genomic regions were determined to be associated with the contents of three kinds of unsaturated FAs in soybean seeds by GWAS ( To confirm the beneficial alleles of each SNP associated with unsaturated FA content in soybean seeds, the average FA content of OA, LLA, and LNA of soybean accessions that carried each allele of peak SNPs were analyzed. Of the 73 peak SNPs that were associated with the three target traits, 73 beneficial alleles were identified. The ' Allele 1' of each peak SNP in Table 1 showed the effect of increasing the phenotypic value of unsaturated FA content. Inversely, the ' Allele 2' of each peak SNP in Table 1 was showed the effect of decreasing the phenotypic value of unsaturated FA content (Table 1). Concretely, for seed OA content, the average phenotypic value of accessions with ' Allele 1' were 1.98 to 15.71% higher than that with 'allele 2' (Table 1). For seed LLA content, the average phenotypic value of accessions with ' Allele 1' were 1.26 to 12.31% higher than that of accessions with ' Allele 2' (Table 1). For seed LNA content, the average phenotypic values of accessions with ' Allele 1' were 0.25 to 3.82% higher than that with ' Allele 2' (Table 1). To increase of unsaturated FA concentrations in soybean oil is desirable to improve human cardiovascular health. However, LNA increases the oxidation of food oils, causing an off-flavor and reducing the shelf life of the oil [3]. Therefore, the important focus of soybean breeding for FAs is to increase the OA and LLA acid contents and to reduce LNA content in seed oil. Hence, the ' Allele 1' of each peak SNP associated with OA and LLA content were beneficial alleles and the ' Allele 2' of each peak SNP associated with LNA content were beneficial alleles for marker-assistant selection (MAS) for the improvement of the seed FA compositions in soybean oil (Table 1).
Fifteen pairs of pleiotropic loci were detected both in 2014 and 2015. Fourteen pairs of pleiotropic loci (showing peak SNP here rs50012108 on Chr.01, rs35817332 on Chr.03, rs29355727 on Chr.04, rs50110801 on Chr.06, rs16802809 on Chr.13, rs36081420 on Chr. 16 and rs47531251 on Chr.20) were found controlling seed OA and LLA content and one pair of pleiotropic loci (rs35024325 on Chr.17) was found controlling OA and LNA content, simultaneously. The allelic effects of all the pleiotropic SNP were analyzed (Fig. 3). Eleven SNPs   were detected controlling OA and LLA content in soybean seed in 2015. Except for rs29355727 on Chr. 04, the direction of the contribution to the phenotype value of all the rest 10 SNPs was opposite according to the value of each allelic effect. The same tendency was also found in 2014. All the four SNPs identified in 2014 that associated with OA and LLA content of soybean seed showed the opposite direction of the contribution to the phenotype value. There into, the allelic effect values of SNP loci, rs16802809 on Chr.13, rs50012108 on Chr.01, rs3937684 on Chr.07, and rs16802831 on Chr.13 were positive for OA content and negative for LLA content in 2015. The allelic effect values of three SNP loci, rs19310064 on Chr.08, rs20308530 on Chr.08 and rs20595231 on Chr.09, were positive for OA content and negative for LLA content in 2014. Therefore, the above seven SNPs could be useful for assisted selection for soybeans with high OA content and low LLA content at the same time.

Predicting and verifying of candidate genes for unsaturated FAs
A total of 1283 genes were found in the 100-kb flanking genomic region of each peak SNP related to three unsaturated FAs according to soybean genome v2.1 (Phytozome 10.0) (Additional file 4: Table S2). Of them, 915 genes were categorized into 29 groups based on BLAST sequence homologies and gene ontology (GO) annotations using MapMan [21] (Additional file 5: Figure S2). Thirty-four genes involved in lipid metabolism. Among the 34 lipid metabolism-related genes, 12 genes that were involved in unsaturated FA synthesis of soybean were considered as candidates for expression profile analysis during the accumulation of soybean seeds (Table 1). To predict possible roles of candidate genes on unsaturated FA content regulation, dynamic accumulation of three kinds of FAs in four characteristic soybean accessions were determined during the development stages of seeds (R5-R8) (Figures 4a-b and 5a). Furthermore, a quantitative RT-PCR was performed to analyze the expression of eight candidate genes in the same four soybean accessions from R5 to R8. All the candidates could express in immature and mature soybean seeds, although the expression levels were largely different among soybean accessions and development stages (Figs. 4c-h and 5b-c).
According to the result of quantitative RT-PCR, the expression levels of Glyma.13G068600 and Glyma.16G003500 in low OA carriers ('Line3' and 'Line 4') of each sampling stage were much higher than those in high OA carriers ('Line1' and 'Line2'). The expression levels of Glyma.16G200200 in low OA carriers (Lines 3 and 4) were much higher than those in high OA carriers ('Line1' and 'Line2') during R5 and R6 stages. The expression levels of Glyma.07G151300 in high OA carriers ('Line1' and 'Line2') were much higher than those in low OA carriers ('Line3' and 'Line4') during R7 stage. The similar tendency was found in LLA for the genes mentioned above. No obvious rules were found in the dynamic expression of the other candidate genes.
For LNA, the peak expression of Glyma.14G194300 and Glyma.07G254500 occurred at the early developmental stages of soybean seeds (R5 to R6 stages), which is similar to the trend of dynamic changes in the phenotype of LNA. The expression levels of Glyma.14G194300 in high LNA carriers ('Line1' and 'Line3') were much higher than those in low LNA carriers ('Line2' and 'Line4') indicating that the gene might promote the accumulation of LNA. The obvious difference in expression of Glyma.07G254500 between high and low LNA carriers was detected only in R5 stage. The regulation pattern of candidate genes could not be sufficient to determine only by analyzing the expression patterns. Therefore, correlation analysis was conducted between on the four tested strains of dynamic phenotypic data of unsaturated FA content and the relative expression of candidate genes (Table 2).
According to correlation analysis results, for the dynamic content of OA in soybean seeds, the positive correlations were found between the relative expression amount of Glyma.07G046200 and Glyma.20G245500 and dynamic OA content during the development of seeds of four tested soybean accessions. There into, the relationship between the relative expression amount of Glyma.07G046200 and dynamic OA content of 'Line1' and 'Line3' was significantly positive correlation. Meanwhile, the relationship between the relative expression amount of Glyma.20G245500 and dynamic OA content of 'Line4' was significant positive correlation indicating that Glyma.07G046200 and Glyma.20G245500 genes should promote the OA accumulation in soybean seed. Positive correlations were found between the relative expression amount of Glyma.13G068600 and Glyma. 16G200200 and dynamic OA content of 'Line1' and 'Line2'. No obvious correlation was found between the relative expression amount of Glyma.13G068600 and Glyma.16G200200 and dynamic OA content of 'Line3' and 'Line4' , indicating that these two genes might have the effect of increasing the OA accumulation in high OA carrier germplasms and the effect was not obvious in low OA carrier germplasms. The dynamic OA content of high OA carrier germplasms ('Line1' and 'Line2') showed positive correlation with the relative expression amount of Glyma.07G151300. On the contrary, the dynamic OA content of low OA carrier germplasms ('Line3' and 'Line4') showed negative correlation with that of Glyma.07G151300. The result suggested that Glyma.07G151300 could promote the accumulation of OA in germplasms with relatively higher OA content and suppress that in germplasms with relatively lower OA content. The relative expression amount of Glyma. 13G068600 showed a weak positive correlation with the dynamic OA content of all the four tested accessions suggesting that the gene has less influence on the OA accumulation.
For the dynamic LLA content, the relative expression amount of Glyma.20G245500 was positively correlated with the dynamic LLA content of four tested accessions. Among the four positive relationships, a significantly positive correlation was found between the relative expression amount and LLA content of 'Line4'. It is speculated that Glyma.20G245500 might have a positive regulatory effect on LLA accumulation. The relative expression amount of Glyma.16G003500 showed a positive correlation relationship with LLA content of Lines 1 and 2 (high LLA carrier) but showed a weak positive or negative correlation relationship with 'Line3' and 'Line4' (low LLA carrier) indicating that Glyma.16G003500 has the effect of increasing the LLA accumulation in relatively higher LLA germplasms with and has no effect in relatively lower LLA germplasms. Positive correlations were found between the relative expression amount of Glyma.16G200200 and dynamic LLA content of 'Line1' and 'Line3'. There into, the very significant positive correlation was found between the relative expression amount of Glyma.16G200200 and dynamic LLA content of 'Line1'. Meanwhile, no relation was found between the relative expression amount of Glyma.16G200200 and dynamic LLA content of 'Line2' and 'Line4'.
For dynamic content of LNA, the dynamic LNA content of germplasms with lower LNA content ('Line1' and 'Line3') showed negative correlation with the relative expression amount of Glyma.07G254500. On the contrary, no obvious correlation was found between the relative expression amount of Glyma.07G254500 and the dynamic LNA content of high LNA carrier germplasms ('Line2' and 'Line4'). The result suggested that Glyma.07G254500 could suppress the LNA accumulation in germplasms with lower OA content. The relative expression amount of Glyma.14G194300 was very significantly correlated with the dynamic LNA content of low LNA carrier germplasms ('Line1' and 'Line3'). While in germplasms with higher LNA content ('Line2' and 'Line4'), the positive correlation was not obvious. The result (See figure on previous page.) Fig. 4 The relationship between expression level of candidate genes and oleic acid content as well as linoleic acid content detected during the seed developmental stages in characteristic soybean accessions. a-b: the dynamic oleic-and linoleic-acid content of four characteristic soybean accessions. For the ten sampling points on x axis, 1-4 corresponds with R5 stage, 5-6 corresponds with R6 stage, 7-8 corresponds with R7 stage, 9-10 corresponds with R8 stage. c-h: The relative expression patterns of candidate genes that are possibly responsible for oleic-and/or linoleicacid content Fig. 5 The relationship between expression level of candidate genes and linolenic acid content detected during the seed developmental stages in characteristic soybean accessions. a: the dynamic linolenic acid content of four characteristic soybean accessions. b-c: The relative expression patterns of candidate genes that are possibly responsible for linolenic acid content Table 2 Correlation coefficient (R) between phenotypic value and relative expression amount of candidate genes during the development of soybean seed in the representative accessions with relative higher and lower oleic-and linolenic-acid content suggested that Glyma.14G194300 might be involved in the accumulation of LNA content in germplasms with lower phenotypic value. But the regulation effect was not obvious in high LNA content germplasms.

Discussion
Different concentration and relative proportions of unsaturated FA in soybean seed may play a decisive role in the nutritional value of soybean oil. Breeding soybean lines with improved unsaturated fatty acid level was primary goal of some soybean breeding programs. In the present study, a total of 194 samples, mainly collected inside China, were evaluated unsaturated fatty acid level. Among these, three and one accessions were found with higher OA (> 30%) and lower LNA level (< 4%), respectively, which both have perfect agronomic traits. Therefore, these accessions with reasonable unsaturated FA level have great potential value for in the future MAS.
To date, about 200 unsaturated FA QTL have been reported (www.soybase.org) based on linkage analysis, most of which were identified via different cross populations constructed by some specific materials. Of these identified QTLs, many were related to the mutant of two major delta 12 FA desaturase enzymes genes (FAD2-1A and FAD2-1B), which participated in conversion of OA precursors to LLA precursors in the soybean seed lipid biosynthesis pathway [22]. In this study, 11 and 29 SNPs located 23 genomic regions of 16 chromosomes, 11 and 29 SNPs located 27 genomic regions of 19 chromosomes, and 35 and 7 SNPs located 12 genomic regions of 12 chromosomes, were identified to be associated with OA, LLA, and LNA level in 2014 and 2015, respectively. Most of these genomic regions were firstly reported to be associated with unsaturated FA concentrations. Among the 11 and 29 association signals for OA content in 2014 and 2015, a total of five SNPs were overlapped with or near the known OA QTL (Table 1). For instance, two genomic regions in rs31881460 and rs43329219 of Chr.08 were significantly associated with OA level, and the association between these two genomic regions and OA content had been verified by previous studies [23]. Similarly, three SNPs (rs3937684, rs1748955, and rs5683686 located on Chr7, Chr15, and Chr18, respectively) of this study were located near the two marker intervals reported by other studies [24]. For the identified 11 and 29 association signals for LLA in 2014 and 2015, one and two genomic regions defined by SNP marker rs31318746, and rs16802831and rs39135305, have similar genomic regions with that of the previous study [25]. Panthee et al. (2006) [26] reported one LNA QTL Linolen4-2, has similar genomic regions as that of this study (rs52833743). These consistent genomic regions across China and North American genetic background might play important roles in regulating unsaturated FA level.
FAD2-1A and FAD2-1B were expressed in developing seeds and many studies have shown that FAD2-1A and FAD2-1B could take part in regulating unsaturated FA concentration [27]. Therefore, most soybean breeder selected FAD2-1A and FAD2-1B as target genes to improve unsaturated FA level, especially OA level. Except for these two significant genes, other genes were still difficult to predict and confirm as the candidates. However, GWAS was an effective way to identify and confirm candidate genes especially within 150-200 kb LD block [19]. In this study, the mean LD was 227 kb, which ensure the candidate gene to be effectively identified. Therefore, we screened candidate genes in the 100-kb flanking regions up-and downstream of OA, LLA, and LNA peak SNPs. We first focused whether the FAD family genes existed in LD blocks when we predicted candidate genes. There were 20 FAD genes in soybean genome including FAD2, FAD4, FAD6 and FAD8 (Phytozome 10.0). However, most of the FAD genes were beyond or just barely beyond the LD interval except for two FAD8 genes (Glyma.07G151300 and Glyma.14G194300). Our study found that the biological function of different copies of FAD8 genes in soybean was differentiated. Based on our results, Glyma. 07G151300 promotes OA accumulation in higher OA germplasms and suppresses that in lower OA germplasms. Glyma.14G194300 might be involved in the accumulation of LNA content in germplasms with lower phenotypic value, but the regulation effect was not obvious in high LNA content germplasms. The FAD8 gene was identified as a light-and temperature-sensitive ω3 desaturase and the activity of which was cold inducible [28][29][30]. The functional differentiation FAD8 gene among different germplasms might be due to differential responses to light and temperature conditions. Previous studies suggest that ubiquitination may play an important role in plant tolerance against various abiotic stresses [31][32][33][34]. Also the ubiquitin-dependent protein degradation pathway is involved in photomorphogenesis, hormone regulation, floral homeosis, senescence, and pathogen defense [35,36]. Interestingly, we found a ubiquitin-conjugating enzyme gene (Glyma.07G046200) and a putative proteasome (Glyma.20G245500) were promote the OA accumulation in soybean seed meantime. It was the first time to find the function of ubiquitination on the accumulation of FA in soybean. However, the accurate function and the regulatory network of the candidates will be discussed in a future work.

Conclusions
A total of 149 SNPs that represented 73 genomic regions were found to be associated with the unsaturated FA contents in soybean seeds in 2013, 2014 and 2015 according to the results of GWAS. Twelve novel genes were predicted to be involved in unsaturated FA synthesis in soybean. Of them, eight genes might be involved in unsaturated FAs regulation simultaneously but work in very different ways revealed by the relationship between expression pattern of the candidate genes and the accumulation of unsaturated FAs. The beneficial alleles and candidate genes identified might be valuable for improving marker-assisted breeding efficiency and exploring the molecular mechanisms underlying unsaturated fatty acid of soybean.

Methods
Plant materials, field design, and phenotypic evaluation A total of 194 Chinese representative accessions, comprising landraces and elite cultivars, were collected from more than 15,000 samples at the Chinese National Soybean GeneBank (CNSGB), were planted at Harbin in 2013, 2014, and 2015. Field trials were conducted using single-row plots (3-m long and 0.65-m between rows) and a randomized complete block design with three replicates per tested environment.
For association mapping, 10 randomly selected plants from each line in each plot were collected to evaluate the levels of unsaturated FAs after full maturity. To study the expression patterns of candidate genes, four characteristic accessions with relatively higher and lower LNA, OA, and LLA contents within the association panel, including 'Line1' (2.10, 19.63, and 55.34%), 'Line2' (9.79, 19.71, and 56.09%), 'Line3' (2.63, 24.38, and 63.65%), and 'Line4' (8.52, 22.51, and 61.06%) were grown in pots in 2016. Pods were picked from the fifth to the seventh nodes of the main stem every seven days from the R5 to the R8 stages of plant, with three replicates for each sample. Seeds in pods were used for the isolation of total RNA and determination of unsaturated FA contents. The unsaturated FA contents in tested samples were determined using gas chromatography (GC-14C, Shimadzu Company, Japan) according to our previous method [37].

SNP genotyping data collection
Genomic DNA was extracted from fresh leaves of each experimental plant using the CTAB method [38], and then were sequenced by specific locus amplified fragment sequencing (SLAF-seq) methodology [39]. According to the preliminary evaluation of soybean reference genome, two restriction enzymes, Mse I (EC 3.1. 21.4) and HaeIII (EC: 3.1.21.4) (Thermo Fisher Scientific, Inc., Waltham, MA, USA), were utilized to generate 50,000 sequencing tags (about 300-500 bp in length) from all tested samples. The sequencing libraries of each accession were constructed based on the sequencing tags, which were distributed in unique genomic regions of 20 soybean chromosomes. For each accession library, the 45-bp sequence reads at both ends of the sequencing tags were obtained through a barcode method using an Illumina Genome Analyzer II system (Illumina Inc., San Diego, CA, USA). The Short Oligonucleotide Alignment Program 2 (SOAP2) software was used to align raw paired-end reads onto the soybean reference genome [40]. The SLAF groups were defined using the raw reads in the same genomic position. More than 58,000 high-quality SLAF tags were obtained from each tested sample. In SNP calling, a minor allele frequency (MAF) threshold was set at 0.04. If depth of a minor allele was larger than 1/3 of sample total depth, the genotype was regarded as heterozygous. Finally, the genotype matrix was imputed using the Nearest Neighbor (KNN) algorithm with parameters of w = 70, p = 7, k = 7, and f = 0.8 [41]. Using three random test data, we estimated an accuracy rate of 96.4% with these same parameters.

Population structure evaluation and LD analysis
The population structure of the tested population was determined using the principle component analysis (PCA) approach as incorporated in the software package GAPIT [42]. The calculation of LD between pairs of SNPs was performed using a combination of SNPs with MAFs > 0.04 and missing data < 10% and r 2 (squared allele frequency correlations) using TASSEL version 3.0 [43]. In contrast to GWAS, missing SNP genotypes were not imputed with the major allele before LD analysis. Parameters in the program included MAF (≥ 0.04) and the integrity of each SNP (≥ 80%).

Association mapping
The association signals were estimated based on 33,194 SNPs from 193 tested accessions. The Bonferroni method at α ≤ 0.05 (≤ 2.58 × 10 − 4 ) was utilized to calculate the p value, which was set as the threshold to declare whether significant association signals existed [44].

Definition and verification of candidate genes
Candidate genes located within the 100-kb LD segment near a peak SNP were identified. Biosynthesis of unsaturated FAs of soybean was used for FA-related gene screening (https://www.kegg.jp/). The expression level of each FA-related gene was determined by real-time reverse transcription polymerase chain reaction (RT-PCR) analyses. Real-time RT-PCR amplifications were performed using the real-time RT-PCR kit according to the manufacturer's instructions (Takara, Japan) on CFX96 Touch™ Real-Time PCR Detection System (Bio-Rad, USA). About 1 μg of total RNA was used for each reverse transcription. Each amplification reaction was performed with 1 μL of the resulting cDNA first-strand cDNA synthesis solution, 0.2 μM of each primer, and