Genetic Analysis of Sugar Composition and Its Relationship with Protein , Oil , and Fiber in Soybean

Soybean [Glycine max (L.) Merr.] is one of the most important crops in the world. It is a major source of vegetable oil for consumption and protein meal for animal feeds and has also been widely used in human food industries because of its nutritive and health benefits. To provide useful information for soybean quality improvement, seed individual sugars, total sugar, protein, oil, and dietary fiber were genetically analyzed in replicated trials with 323 germplasm lines grown in South Dakota and 137 cultivars and breeding lines grown in Virginia. The results indicated significant differences among the genotypes for all traits investigated. Environment effect and genotype  ́ environment interaction were also significant in most cases. Heritability estimates were high (94.45–97.79%) for all traits in the germplasm population, and higher in the population of breeding lines for most traits. High genotypic correlation existed between sucrose and total sugar, which helps improvement of digestible sugars and sweetness in soybean food. However, attention should be paid to the lines with higher sucrose but lower oligosaccharides, since stachyose was positively associated with total sugar. Genotypic correlations between seed sugars and protein were insignificant or very low in most cases, implying that alteration of seed sugars might not necessarily affect protein. In some cases, however, there might be negative correlations between seed sugars and oil or dietary fiber in soybean. This study also identified some unique germplasm lines with a desired level of a specific seed composition: one with high sucrose, five with low raffinose, 15 with high total sugar, seven with high protein, and four high in both sucrose and total sugar. G.-L. Jiang, R.A. Bowen, A. Miller, and H. Berry, Agricultural Research Station, Virginia State Univ., PO Box 9061, Petersburg, VA 23806; P. Chen, Dep. of Crop, Soil and Environmental Sciences, Univ. of Arkansas, Fayetteville, AR 72701, current address, Univ. of Missouri, Fisher Delta Research Center, Portville, MO 63873; J. Zhang, Plant Science Dep., South Dakota State Univ., Brookings, SD 57007, current address, Dep. of Agronomy, Iowa State Univ., Ames, IA; L. Florez-Palacios and A. Zeng, Dep. of Crop, Soil and Environmental Sciences, Univ. of Arkansas, Fayetteville, AR 72701; X. Wang, Plant Science Dep., South Dakota State Univ., Brookings, SD 57007, current address, College of Agriculture, Yunnan Univ., Kunming, China. Received 11 Mar. 2018. Accepted 10 July 2018. *Corresponding author (gjiang@vsu.edu, gljiang99@yahoo. com). Assigned to Associate Editor Owen Hoekenga. Abbreviations: HPAEC-PAD, high-performance anion-exchange chromatography coupled with pulsed amperometric detection; HPLC, high-performance liquid chromatography; NIFA, National Institute of Food and Agriculture; NIR, near-infrared; PI, plant introduction; RIL, recombinant inbred line. Published in Crop Sci. 58:2413–2421 (2018). doi: 10.2135/cropsci2018.03.0173 © Crop Science Society of America | 5585 Guilford Rd., Madison, WI 53711 USA This is an open access article distributed under the CC BY license (https:// creativecommons.org/licenses/by/4.0/). Published August 30, 2018

and sugars such as sucrose, raffinose, and stachyose (Choct et al., 2010).The proportion of seed composition determines the uses of soybean.For instance, cultivars high in oil are preferred by vegetable oil and soy-diesel industries, whereas soy food products usually need lower oil but higher protein and sugar contents.
Soybean seed protein and oil contents have been extensively investigated, particularly in plant breeding and genetics from quantitative genetics to molecular mapping and candidate gene identification (Wang et al., 2014;Hwang et al., 2014;Zhang et al., 2018).Relatively, studies on sugar and fiber content in soybean are less reported.Sucrose, a disaccharide and the most important component of total sugar in soybean, is a free or digestible sugar and is very important for food soybean (Kumar et al., 2010;Song et al., 2013).Similarly, monosaccharides glucose and fructose are also easily digested, and thus they should be worthy of exploring, in particular for food use of soybeans like edamame (Song et al., 2013).However, two oligosaccharides, raffinose and stachyose, cannot be digested in monogastric animals and cause flatulence (Choct et al., 2010).A decreased concentration of both raffinose and stachyose is preferred in soy food industries such as soymilk and tofu production (Kumar et al., 2010;Saldivar et al., 2011).
Using high-performance anion-exchange chromatography coupled with pulsed amperometric detection (HPAEC-PAD), Bainy et al. (2008) reported varietal differences in carbohydrates in defatted soybean flour and soy protein isolate byproducts among 12 soybean lines.Hou et al. (2009) analyzed five individual sugars and total sugar in 241 germplasm accessions of three maturity groups using high-performance liquid chromatography (HPLC).They identified some plant introductions (PIs) considerably low or high in individual sugars.Cicek et al. (2006) reported a high heritability for sucrose content, but relatively low heritabilities for stachyose and raffinose contents in a recombinant inbred line (RIL) population derived from an interspecific cross.Quantitative trait loci associated with sucrose and oligosaccharide contents were also explored (Maughan et al., 2000;Kim et al., 2005).
Strong correlations between sugars were previously reported (Cicek et al., 2006;Hou et al., 2009).In the study by Hou et al. (2009), the absolute values of simple correlation coefficients among individual sugar and total sugar contents varied from 0.59 to 0.999, except between total sugar and glucose (r = −0.27)or fructose (r = −0.24).Hymowitz et al. (1972) evaluated 60 soybean lines from maturity groups 00 through IV for protein, oil, total sugar, and individual sugar content.Their results suggested that total sugar content and oil content were positively associated, and each was negatively correlated with protein content.Sucrose and raffinose content were positively correlated with oil content, whereas stachyose content was positively associated with protein (Hymowitz et al., 1972).Wilcox and Shibles (2001) reported that concentrations of carbohydrates were not associated with seed yield, but increased protein was coupled with decreases in oil, total carbohydrates, and sucrose.In an evaluation of 23 conventional and food-grade cultivars, Geater and Fehr (2000) suggested that total sugar was highly correlated with the sum of protein and oil.In a study with 30 vegetable soybean genotypes, negative correlations were found between protein and total sugar (r = −0.52)or sucrose (r = −0.43),whereas no significant correlations were detected between oil and total sugar or sucrose (Li et al., 2012).Yu et al. (2016) evaluated 35 soybean germplasm lines (mostly from China) for five individual and total sugars, as well as protein.They found that protein content was positively correlated with total sugar and sucrose contents but negatively correlated with fructose and glucose contents.
There have historically been fewer carbohydratefocused studies than studies focused on protein and/ or oil content, and the studies that considered all proximate nutrients including sugars, protein, oil, and fiber were even more limited.In most of the previous studies on seed sugar content in soybean, the number of genotypes used was relatively small compared with the studies on protein and oil research.Inconsistencies between the studies existed to some extent.In addition, there is lack of understanding of genetic variability of fiber content in soybean, although it is an important trait in vegetable soybeans that are consumed directly for fresh market and other food-grade soybeans that are used by food product manufacturers (Redondo-Cuenca et al., 2007).Genotypic correlation between sugars and other traits has rarely been discussed.Therefore, there is a need to further investigate the genetic feature of sugars and their relationships with other seed composition in soybean.To provide useful information for quality improvement and related research, using two populations, one consisting of 323 soybean germplasm accessions grown in South Dakota and one of 137 cultivars and breeding lines grown in Virginia, we characterized seed individual sugar, total sugar, protein, oil, and dietary fiber content in the present study.We also analyzed the genotypic correlations between the traits.

Genotypes and Field Trials Experiment 1
In total, 323 soybean germplasm accessions or PIs were obtained from the US Soybean Germplasm Collection without specific criteria of selection but mainly limited to early maturity groups, since soybean is highly photoperiod sensitive and a given trial cannot accommodate a wide range of maturities.Approximately 91% of the PIs are maturity group 0 and 9% belong to maturity group 00; 91% originated from China, and the remaining from other countries or unknown origins (Supplemental Table S1).
automated sampler with a 25-mL injection loop, and a Chromeleon Chromatography Management System (Dionex), was used to identify and quantify sugars.Sugar separation was done with an analytical CarboPac PA10 pellicular anion-exchange resin column (25 ´ 4 mm) preceded by a CarboPac PA10 guard column (50 ´ 4 mm) and an AminoaTrap column ( 30´ 3 mm, Dionex).Before HPLC injection, a 24-mL aliquot of sugar sample was diluted with 576 mL of distilled water.Sugars were eluted with 90 mM NaOH at a flow rate of 1.0 mL min −1 under isocratic conditions.Glucose, fructose, sucrose, raffinose, and stachyose in the extracts were determined and quantified based on the retention time and the regression curve established for each of five standard sugars (Hou et al., 2009).

Experiment 2
About 50 g of seeds per plot was randomly taken with a small cup and analyzed for protein, oil, dietary fiber, sucrose, raffinose, and stachyose content using NIR spectroscopy in a DA 7250 NIR analyzer (Perten Instruments) in the Soybean Breeding and Genetics Laboratory at Virginia State University.Glucose and fructose were not evaluated because no appropriate calibration was available in the system.

Statistical Analysis
Data of the seed composition were presented as milligrams per gram on a dry-weight basis.Analysis of variance was performed using PROC GLM in SAS version 9.4 (SAS Institute, 2013), and frequency distribution was computed in Microsoft Excel 2013.Heritability was estimated on a genotype mean basis (Fehr, 1987) as h 2 = s g 2 /(s g 2 + s e 2 /l) for Exp. 1, where s g 2 is genotypic variance, s e 2 is environmental variance, and l is the number of environments or locations.Phenotypic and genotypic correlations between traits were computed as described by Holland (2006).Genotypic and phenotypic correlation coefficients were tested for significance using a t test as suggested by Robertson (1959) and Sharma (1988).For Exp. 2, heritability was computed as h 2 = s g 2 /[s g 2 + s ge 2 /l + s e 2 /(rl)], where s ge 2 is genotype´ environment interaction variance and r is the number of replications (Fehr, 1987).In addition, heritability for protein, oil, and fiber content in Exp. 1 was also estimated in the same way with individual plot data.

Genetic Variation and Heritability in Soybean Germplasm
Analysis of variance showed that there were highly significant differences (P < 0. 01) among the germplasm accessions for all traits investigated.Overall means and variation of the traits evaluated across three environments are shown in Table 1.Comparatively, glucose and fructose exhibited larger relative variations, whereas the smallest relative variations were observed in protein and fiber.Of total sugar that averaged 103.61 mg g −1 , sucrose (50.65 mg g −1 ) was the largest component (48.90%), followed by stachyose (39.26%) and raffinose (9.38%).Glucose and fructose together contributed a very small part (2.46%) to total sugar content.The result was consistent with the These PIs are all cultivated soybeans (G.max) and represent landraces and obsolete and improved cultivars.All the PIs were planted in a randomized complete block design with three replications at three locations: Aurora in 2011, Brookings in 2012, and Watertown in 2012, all in South Dakota.Plots were composed of a single 3-m-long row with 0.76-m row spacing, and 80 seeds were planted per plot.The field management was similar to general soybean production.After full maturity (R8), all plots were individually bulk harvested.Then the seeds were dried in an air-drying chamber for a week.

Experiment 2
One hundred and thirty-seven cultivars and breeding lines of maturity groups IV, V, and VI were grown at the Virginia State University Randolph Research Farm in Ettrick, VA, in 2014 and 2015, respectively.Of them, 130 were breeding lines developed by Virginia State University soybean and edamame program and derived from 12 different crosses each with 10 to 12 F 3:4 or F 4:5 lines evaluated (Supplemental Table S2), and the remaining were released cultivars.The breeding lines were selected on the basis of agronomic performance in plant rows and preliminary yield trials, with connections to some extent among those lines derived from similar crosses.The released cultivars (Asmara, Mooncake, N6202-8, Osage, Owens, Randolph, and UA 4805) were used as the checks of edamame or general-purpose soybean.The breeding lines are mostly unrelated to the seven released cultivars and the early-maturing germplasm lines used in Exp. 1. Four-row plots with a 4.8-m length and 0.76-m row spacing were planted in a three-replicate randomized complete block design, and ?100 plants were planted per row.After full maturity, one row per replication was intended to be bulk harvested in 2014, but only one replication was harvested due to severe deer and weed damages.In 2015, two central rows were combine harvested for each replication.

Sampling and Seed Composition Analysis
Experiment 1 Approximately 200 g of dried seeds per plot was randomly taken and ground with Perten Laboratory Mill 3600 to prepare the flour samples for seed composition analysis.Concentrations of protein, oil and dietary fiber were determined by near-infrared (NIR) spectroscopy in a DA 7200 NIR analyzer (Perten Instruments) in the Soybean Breeding and Genetics Laboratory at South Dakota State University.To be consistent with the design for sugar analysis below, the individual-plot values of three replications estimated per genotype were averaged for protein, oil, and fiber, and then the means were used as data units for statistical analysis in a randomized complete block design with three locations used as replications.
To reduce the number of samples and save resources in sugar analysis, one mixed flour sample per genotype was prepared by mixing a similar amount (5 g) of soy flour for all three replications in each environment.Sugar sample preparation, extraction, and analysis were conducted in the Soybean Breeding and Genetics Laboratory at the University of Arkansas following Hou et al. (2009) using HPLC.Briefly, a Dionex DX500 HPAEC-PAD system, which was equipped with a GS50 pump, an ED40 pulsed amperometric detector, an AS40 previously reported results (Karr-Lilienthal et al., 2005;Hou et al., 2009;Yu et al., 2016).This might be the reason that glucose and fructose were not often included in studies on sugars in soybean (Maughan et al., 2000;Kim et al., 2005).Overall across three environments, concentrations of protein, oil, and dietary fiber averaged 418.31, 197.88, and 62.10 mg g −1 , respectively (Table 1).This is similar to previous reports (Liu, 1999;Wang et al., 2014).
The variability of seed sugar, protein, oil, and dietary fiber observed in this study falls within a normal range of variation (Liu, 1999;Karr-Lilienthal et al., 2005).Frequency distribution was continuous for all the seed compositions (Fig. 1).Except for fiber content, the frequency distributions approximated to a normal distribution, with some skewness for fructose, stachyose, and oil (Fig. 1).This is consistent with the results reported by Cicek et al. (2006) using a population of 303 RILs derived from an interspecific cross.Hou et al. (2009) identified many germplasm accessions with extremely high or low concentrations in individual sugars from a diverse population of 206 PIs originating from multiple countries and regions.In the present study, no germplasm accession with that high of glucose and/or fructose content was found.However, some unique germplasm lines with desired concentrations in specific seed composition were identified (Table 2) that could be used as parents in soybean breeding.These lines include five low-raffinose PIs (PI 603443B, PI 358323, PI 603426E, PI 603429A, and PI 612753A; raffinose content < 3 mg g −1 ), one high-sucrose PI (PI 291329; sucrose content > 64 mg g −1 ), 15 PIs with >120 mg g −1 in total sugar content, and four PIs high in both sucrose and total sugar (PI 538395, PI 458827, PI 597651, and PI 597652).These PIs have not been previously reported to possess such specific sugar features, and no genetic connection between them and previously reported lines has been evidenced by molecular marker and/or pedigree information (Hou et al., 2009;Jo et al., 2018).We also confirmed seven high-protein PIs (PI 468909, PI 612758A, PI 468910, PI 319536B, PI 612759B, PI 603712, and PI 597467) with an average of >462 mg g −1 over three environments, in spite of being coupled with a lower oil concentration (130.9-178.9mg g −1 ).Of them, PI 603712 also possesses high resistance to soybean aphids (Aphis glycines Matsumur; Bhusal et al., 2017).
As shown by ANOVA (Table 1), environmental effects were also significant in six traits but insignificant for glucose, fructose, and stachyose.In other words, locations and/or years did not significantly affect the concentrations of glucose, fructose, and stachyose.By evaluating seven genotypes grown at different geographical locations, Kumar et al. (2010) found that sucrose content was significantly higher at a cooler location, but the differences in raffinose and stachyose contents across growing locations were genotype dependent.Since no data of individual plots were available for sugars in this study, variance of genotype ´ environment interaction could not be tested for its significance but was used as the residual or error variance in the significance test for genotypic and environmental or locational effect.This might affect the results.However, for protein, oil, and dietary fiber, ANOVAs with and without individual plot data exhibited similar results to the significance tests for genotypic and environmental effects.Analysis of variance based on individual plot data also showed a highly significant genotype ´ environment interaction (P < 0. 01) for these traits (data not shown).In addition, correlation coefficients between environments or locations were high for all the traits, averaging 0.882 and ranging from 0.809 for protein between 2012 Watertown and 2011 Aurora to 0.944 for raffinose between 2012 Watertown and 2011 Aurora.Therefore, the results presented without individual plot data are reliable and should be informative.
Heritability of protein and oil content in soybean seed was estimated as high in a recent study with two RIL populations (Wang et al., 2014).In the present study, high heritability was also found for all traits, from 94.45% for protein to 97.79% for raffinose (Table 1).Interestingly, the estimates of heritability for protein, oil, and dietary fiber with individual plot data (Table 3).Among the individual sugars, low significant correlations were found between glucose and fructose (r = 0.239), sucrose and raffinose (r = 0.264), and sucrose and stachyose (r = 0.411).No correlations with |r| > 0.16 existed between other pairs of individual sugars.This is different from the results of Hou et al. (2009).They reported that there were higher simple correlations among individual sugars, either positive or negative with |r| = 0.590 to 0.989.The reason for the inconsistency might be that some germplasm lines with extremely high or low individual sugars were included in their study (Hou et al., 2009).However, significant correlations between total sugar and sucrose, raffinose, or stachyose were detected in this study, with r = 0.828, 0.541, and 0.759 (P < 0.01), respectively.
To further elucidate the relationships among traits, we performed a genotypic correlation analysis.The results indicated that both phenotypic and genotypic correlations were ignorable or very low among individual sugars in most cases, with an average of 0.141 ranging from 0.012 to 0.392 and 0.156 ranging from 0.004 to 0.422 for absolute coefficients of phenotypic and genotypic correlations, respectively (Table 4).Similar to simple correlations described above, there were weak or low not used but the means of three replications were the same as or very close to those values (94.50, 96.53, and 95.05% for protein, oil and fiber, respectively) estimated when individual plot data were used in the computation.This indicated that genotypic effects dominated in phenotypic variation, and selection would play an important role in trait improvement.In a study with 31 vegetable soybean genotypes, Mebrahtu and Mohamed (2006) reported lower estimates of heritability for individual sugars and total sugar, with h 2 = 26.6 to 56.1%.Cicek et al. (2006) reported that sucrose content exhibited a higher heritability, but stachyose and raffinose content had relatively low heritabilities in a population of 308 RILs derived from an interspecific cross.The difference might be mainly attributed to different types and numbers of soybean germplasm lines used.No heritability was previously reported for fiber content in soybean.Further investigations will be of interest.

Phenotypic and Genotypic Correlations between Traits in Soybean Germplasm
Pearson correlation analysis showed that the simple correlations between traits investigated were not or slightly significant (|r| < 0.18) for two-thirds of 36 pairs of traits genotypic correlations between glucose and fructose, and between sucrose and raffinose or stachyose.In addition, total sugar content exhibited a moderate genotypic correlation with raffinose and a higher genotypic correlation with sucrose or stachyose, but no noticeable genotypic correlation was found between total sugar and glucose or fructose.This is consistent with the proportion of individual sugars contributing to the total sugar content.
High genotypic correlation between sucrose and total sugar helps improvement of digestible sugar and sweetness in food-grade soybean, whereas increases in total sugar might be accompanied by increased indigestible oligosaccharides (raffinose and stachyose) to some extent.Thus, more attention should be paid to the lines with higher sucrose but lower oligosaccharides, such as PI 597652 and PI 291313 (Table 2).1.79 † Bolding indicates that the value fell within the range of desired concentration for a given seed constituent (i.e., raffinose content < 3 mg g −1 , sucrose content > 64 mg g −1 , total sugar content > 120 mg g −1 , and protein content > 462 mg g −1 ).Fructose, sucrose, stachyose, and total sugar exhibited low but different phenotypic and genotypic correlations with protein (Table 4).No noticeable genotypic correlations were found between other sugars and protein.
In a trial with 20 high-oil and 20 high-protein soybean genotypes, Hartwig et al. (1997) also revealed that the correlation between protein and stachyose + raffinose was negative but nonsignificant.In our study, the genotypic correlations between the concentrations of sugars and oil, as well as dietary fiber, were ignorable or very low in most cases.This indicated that the improvement of oil did not obviously affect sugar content in soybean, and alteration of protein or dietary fiber would have no or limited impact on sugars as well.Similarly, selection for sugars might not have a greatly negative impact on protein, oil, or dietary fiber.There were higher negative phenotypic and genotypic correlations between protein and oil content (Table 4).This is consistent with previously reported results (Wang et al., 2014).Dietary fiber content exhibited higher negative phenotypic and genotypic correlations with oil concentration, but lower positive correlations with protein content.It implied that compared with protein, alteration of oil had a larger impact on dietary fiber content in soybean.In addition, the phenotypic and genotypic correlation coefficients were very similar, and they were also close to the simple correlation coefficients (Table 3), suggesting that the effects of environment on relationships among the traits might not matter much in similar situations.

Trait Genetic Characterization and Relationships in Soybean Breeding Lines
To effectively perform selection and further improvement for quality traits in breeding materials, we evaluated 130 breeding lines selected from different crosses of maturity groups IV, V, and VI and seven released cultivars for the concentrations of seed sugars, protein, oil, and dietary fiber.As discussed above, these breeding lines are not genetically related to the seven released cultivars and the PIs evaluated in Exp. 1, although some lines were derived from similar crosses.Among the breeding lines and released cultivars, all traits evaluated exhibited a highly significant difference (P < 0. 01) and a relatively large range of variation (Table 5), in spite of smaller values compared with those of soybean germplasms in Exp. 1. Similar to Exp. 1, coefficients of variation for raffinose, sucrose, and stachyose were larger than others.Year or environment effects and genotype ´ environment interaction were significant for all the traits as well.Higher estimates of heritability were found for protein, oil, sucrose, stachyose, and total sugar content than for raffinose and fiber content (Table 5).This indicated that further selection for the traits in current breeding populations would be effective, although raffinose and dietary fiber content were more likely affected by environments.Basically, the results were consistent with those in Exp. 1 in most cases, although the estimates of heritability in this experiment were smaller than the values in Exp. 1.For raffinose, low heritability estimated in Exp. 2 might be largely attributed to a small range of variation, compared with Exp. 1.
Correlation analysis indicated that no or very weak phenotypic and genotypic correlations existed between sucrose and stachyose, protein, or oil and between raffinose and protein, oil, or fiber, with an R 2 ranging from 0.002 to 0.044 for phenotypic correlation and from 0.008 to 0.067 for genotypic correlation (Table 6).Similar to Exp. 1, total sugar content was positively associated with sucrose and/or stachyose content but had no noticeable association with raffinose.Different from Exp. 1, however, there were moderately negative phenotypic and genotypic correlations between total sugar or stachyose and oil or dietary fiber, but a weak positive genotypic correlation was found between total sugar and protein.It could be assumed that this phenomenon is mainly due to the difference in materials used in the two experiments.Therefore, the potential reverse impact of selection for total sugar concentration on oil should be taken into consideration in further improvement of the breeding materials.Negative genetic association was also found between protein and oil in this experiment with breeding lines and released cultivars, which was consistent with Exp. 1 and previously reported results (Wang et al., 2014).Dietary fiber exhibited a negative correlation with protein but a positive correlation with oil, which was inconsistent with Exp. 1.

CONCLUSION
Significant differences in concentrations of seed individual sugars, total sugar, protein, oil, and dietary fiber existed among the soybean germplasm accessions or released cultivars and breeding lines.All the traits exhibited a moderate to high heritability except for raffinose in breeding materials, implying that selection for desired phenotypes would be effective.Unique germplasm lines with desired concentration of a specific seed composition, such as high sucrose, low raffinose, high total sugar, and high protein, were identified and could be used as parents in soybean breeding.Higher positive genotypic correlation existed between sucrose and total sugar, helping with improvement of digestible sugar and sweetness in soybean food, whereas increases in total sugar might be accompanied by increased indigestible oligosaccharides (especially stachyose) to some extent.Genotypic correlations between seed sugars and protein were ignorable or relatively low in most cases.Therefore, improvement of seed sugars may not necessarily affect protein.In some cases, however, there could be a negative correlation between seed sugars and oil or dietary fiber in soybean.

Fig. 1 .
Fig. 1.Frequency distribution of 323 soybean germplasm accessions in seed composition based on average over three environments.

Table 1 .
Mean, variation, and estimate of heritability of seed composition trait in 323 soybean germplasm accessions across three environments.Significant at the 0.01 probability level.† F Gen , and F Env are F values of significance tests for genotype and environment effects, respectively. **

Table 2 .
Means of seed composition traits in soybean germplasm accessions with low raffinose, high sucrose, high total sugar, or high protein content over three environments.

Table 3 .
Coefficients of simple correlations between seed composition traits based on 323 soybean germplasm accessions grown in three environments.

Table 4 .
Coefficients of phenotypic (above diagonal) and genotypic (below diagonal) correlations between seed composition traits based on 323 soybean germplasm accessions grown in three environments.,** Significant at the 0.05 and 0.01 probability levels, respectively. *