Introduction

Soybean [Glycine max (L.) Merr.] is one of the most important protein and oilseed crops that is widely grown worldwide (Fang et al. 2017). Increasing attention has been paid to the improvement of soybean seed quality traits in addition to the agronomic traits, such as yield, as seed quality traits directly affect the nutritional quality of soybean, especially for the food-grade cultivars (Xie et al. 2012; Gong et al. 2018; Lee et al. 2019). However, most major seed quality traits and essential agronomic traits in soybean are controlled by quantitative trait loci (QTL) with complicated interactions between alleles (Panthee et al. 2006b; Medic et al. 2014; Vaughn et al. 2014; Wang et al. 2014; Zhang et al. 2018; Lee et al. 2019). Some of the traits are linked with each other in a favorable way, but in many cases, it is very common to see undesirable associations between some traits, including the consistent negative correlation between protein and oil concentrations (Lee et al. 2019). Therefore, establishing a thorough understanding of the correlations between different traits may allow breeders to avoid or manage undesirable associations and utilize the desirable ones in their breeding populations. Moreover, combining the correlations with genotypic studies may help further in selecting favorable genes and breaking the undesirable linkages with the assistance of marker-assisted selection (Karikari et al. 2019; Yoosefzadeh-Najafabadi et al. 2022).

Among diverse seed quality traits, amino acids, fatty acids, and sugar-related traits have gained more attention from breeding programs. Amino acids and fatty acids determine the quality of soybean protein and oil, respectively (Lee et al. 2019). The sugar-related traits are critical for the soy-food flavor in addition to the health benefits (Wang et al. 2014; Dhungana et al. 2017; Lee et al. 2019). In contrast to well-documented correlations between agronomic traits, the complex correlations between some important seed quality traits on the one hand and agronomic traits on the other are not as well studied. The potential mechanisms involved in the correlations, when unknown, may lead to barriers for cultivar improvement (Bachlava et al. 2008; Kumar et al. 2010; Zhang et al. 2018; La et al. 2019; Lee et al. 2019). Four essential amino acids, including lysine (Lys), cysteine (Cys), methionine (Met), and threonine (Thr), five major fatty acids (linoleic, oleic, linolenic, palmitic, and stearic acids), and the three sugar-related traits (sucrose, stachyose, and raffinose), together with the major agronomic traits: yield, days to maturity (DTM), lodging, and height were selected to be the focus for this correlation study as they highly affect the quality of commercial soybean cultivars (Palomeque et al. 2010; Lee et al. 2019). The strong negative association between protein and oil is well established in the literature (Hernández-Sebastià et al. 2005; Zhang et al. 2018; La et al. 2019; Lee et al. 2019). Nevertheless, there have been limited studies testing the correlations between seed quality traits per se, and various conclusions were presented based on different analysis and assessment methods (Li et al. 2018). In general, protein concentration has been reported as being positively correlated with amino acid concentrations when calculated based on the total seed dry weight (Zhang et al. 2018). However, negative correlations were observed between protein and Cys, as well as Met with Cys, Thr, and Lys in several studies (Panthee et al. 2006b; Medic et al. 2014; Vaughn et al. 2014). In terms of fatty acid traits, correlations, as reported, have been relatively consistent, which included the strong positive correlation between palmitic and stearic acids, in addition to the negative associations between oleic acid and linoleic and linoleic acid concentrations (Bachlava et al. 2008; Abdelghany et al. 2020). The strength of correlations for the sugar-related traits varied in different studies (Kumar et al. 2010; Zeng et al. 2014; Bueno et al. 2018; Jiang et al. 2018; La et al. 2019). Sucrose concentration was reported as negatively correlated to the protein concentration by Lee et al. (2019) and Dhungana et al. (2017). No significant correlation was reported for sucrose with stachyose and raffinose (Wang et al. 2014).

Lack of detailed understanding regarding correlations among seed quality traits and their correlations with agronomic traits poses challenges for breeding programs to effectively improve desirable seed traits without affecting others (Zhang et al. 2018). Zhang et al. (2018) indicated weak and nonsignificant correlations between protein or oil and DTM. A negative association between yield and protein was also reported, although it was not as strong as the correlation between protein and oil (Vaughn et al. 2014). In a study conducted by Bachlava et al. (2009), yield was determined to be negatively correlated with oleate concentration but positively correlated with linoleate and linolenate concentrations. There is much less information available in the literature on the correlations between seed quality traits and agronomic traits. Therefore, the objective of this study was to examine and identify correlations between major seed quality traits as well as agronomic traits across the two mega-environments (MEs) in Canada and Ukraine. By understanding the complex trait correlations, the genomic information can be combined with the phenotypic correlations to directly select preferable traits or indirectly select associated traits with the aid of molecular marker technologies to facilitate soybean cultivar development.

Materials and methods

Plant materials and experimental design

The soybean diversity panel consisting of 200 accessions within maturity group 0 (MG 0) was used in this study. The majority of the panel was made up of 127 University of Guelph accessions, which were not sister lines, accounting for 63% of the total. The rest of the panel included 35 accessions from the Agriculture and Agri-Food Canada, 19 accessions from Le Centre de recherche sur les grains inc. in Québec, Canada, 10 Northern US cultivars, and nine diverse ancestral soybean cultivars. Sixteen accessions were removed from the original panel consisting of 200 accessions due to the poor emergence in the field or seed mixtures in the source in 2018. These modifications resulted in some changes to the total accessions used in further analyses bringing it to the final number of 184. All the accessions were planted in three locations in Ontario and two locations in Ukraine, respectively. The three locations in Ontario were: Elora Research Station (ERS), Woodstock Research Station (WRS), and St. Paul's (STP), ON, for both 2018 and 2019. The two Ukrainian locations were Lyubar and Kodyma for 2018, and Lyubar and Kotovsk for 2019. Two replications were planted per location using the randomized complete block design (RCBD). In the three Ontario locations, the plots were 5 m x 1.65 m (8.25 m²) with four rows spaced at 35 cm between rows, and 500 seeds were planted per plot. In Ukraine, the plot sizes were 5.425 and 15.736 m² for 2018 and 2019, respectively, to accommodate the local equipment settings. In both countries, the whole plot was harvested to ensure enough seeds for data collection in the seed lab upon harvest. The planting and harvest dates are summarized in Table 1.

Table 1.

Planting and harvest dates for Elora, Woodstock, and St. Paul's in Ontario, and Kodyma, Lyubar, and Kotovsk in Ukraine in 2018 and 2019.

Results

Correlations between seed quality traits and agronomic traits under Canadian mega-environment and Ukrainian mega-environment

In general, similar correlations were consistently observed in the separated years and combined-year data as presented in Tables 2–4. Nonsignificant correlations were left out from the tables to avoid overloading the tables. In all three datasets, strong positive correlations were consistently observed between protein and the four amino acids, and also within the four amino acids, with r values ranging from 0.82 to 0.99 (Tables 2–4). Stearic acid was positively associated with palmitic acid, showing an r value greater than 0.50 (Tables 2–4). Moderate positive correlations also existed between sucrose and linolenic acid, besides sucrose and raffinose (Tables 2–4). Some other positive correlations were presented as well but were not significant in all years and the combined ANOVA, which included stearic acid and oleic acid (Tables 3 and 4), along with stachyose and raffinose (Table 3). In contrast, oil concentration was negatively correlated to protein and the four amino acids (Tables 2–4). Linoleic acid was negatively correlated with oleic acid and stearic acid (Tables 2–4). Consistent moderate-to-strong negative correlations were found between sucrose and the four amino acids as well as the protein concentration (Tables 2–4). In 2019, the stachyose and oil concentrations were indicated to negatively affect each other (Table 3). In terms of the agronomic traits, DTM consistently showed moderate positive associations with yield and plant height (Tables 2–4). Lodging and height were positively correlated (Tables 2–4). In addition, yield and height (Tables 2 and 4), as well as DTM and lodging (Tables 2 and 4), were positively correlated with each other, respectively, but not in 2019. There was no significant strong correlation found between the agronomic traits and the seed quality traits in this genome-wide association study (GWAS) panel across Ontario locations (Tables 2–4).

Table 2.

Pearson's correlation coefficient values among 18 seed quality and agronomic traits combining the data from three locations (Elora, Woodstock, and St. Paul's) in Canada in 2018.

Table 3.

Pearson's correlation coefficient values among 18 seed quality and agronomic traits combining the data from three locations (Elora, Woodstock, and St. Paul's) in Canada in 2019.

Table 4.

Pearson's correlation coefficient values among 18 seed quality and agronomic traits combining the 2018 and 2019 data from three locations (Elora, Woodstock, and St. Paul's) in Canada.

The combined year and location GT biplot (Fig. 1) for Ontario visually presented the correlations between all the seed compositions and agronomic traits. The biplot explained 50.6% of the variation, with 31.8% explained by PC1, and 18.8% by PC2 (Fig. 1). The significant and strong correlations observed in the biplot were consistent with the correlation coefficients calculated earlier (Fig. 1 and Tables 2–4).

Fig. 1.

Genotype-by-trait biplot for a combined year and location data from Elora, Woodstock, and St. Paul's in 2018 and 2019 for 18 seed quality and agronomic traits.

For Ukraine, the data from Kodyma in 2018 and Lyubar in 2019 were dropped from the combined and correlation analyses as the data were incomplete and invalid according to the ANOVA results. The Pearson's correlation coefficients of the traits for the separate years and the combined-year data are presented in Table 5. The negative correlation between protein and oil (r = −0.67 for 2018, 2019, and combined-year data) was consistently shown for the GWAS panel as expected (Table 5). Combining the two-year data, protein and yield showed significant and negative correlation (r = −0.15), while oil and yield showed a significant positive correlation (r = 0.19) (Table 5). Moreover, oil was positively associated with the yield in 2019 in Kotovsk (r = 0.20) (Table 5). However, these three significant correlations were weak in value (Table 5). The combined GT biplot (Fig. 2) accounted for 89.1% of the variation, with PC1 = 58.3% and PC2 = 30.8%. The negative association between protein and oil was confirmed in the biplot as well (Fig. 2 and Table 5).

Fig. 2.

Genotype-by-trait biplot for a combined year and location data from Lyubar and Kotovsk in Ukraine in 2018 and 2019 for protein, oil, and yield.

Table 5.

Pearson's correlation coefficient values between protein, oil, and yield for 2018, 2019, and combined year and location data in Ukraine.

Correlations between seed quality traits and agronomic traits across Canadian and Ukrainian mega-environments

The correlations across the two countries were studied among limited number of traits due to the lack of available data from Ukraine as only protein and oil data were collected by cooperators in Ukraine. The Pearson's correlation coefficients are listed in Table 6. Combining the data across the two MEs, protein and oil consistently interacted with each other in a moderately negative way, with r value ranging from −0.62 to −0.58 (Table 6). A significant negative correlation between protein and yield was observed in the 2-year combined analysis, and a positive correlation between oil and yield was found in 2019 (Table 6). However, the correlations between yield and protein or oil were weak and inconsistent (Table 6). The trait correlations derived from combined locations and years (Fig. 3) were consistent with Pearson's correlation results (Table 6). A total of 85.7% variation was explained in the biplot, with 52.5% by PC1 and 33.2% by PC2 (Fig. 3).

Fig. 3.

Genotype-by-trait biplot for a combined year and location data from Elora, Woodstock, and St. Paul's in Canada, and Lyubar and Kotovsk in Ukraine in 2018 and 2019 for protein, oil, and yield.

Table 6.

Pearson's correlation coefficient values between protein, oil, and yield for 2018, 2019, and combined year and location data across Canada and Ukraine.

Discussion

The negative correlation between protein and oil was consistently observed in both countries representing MEs. This correlation could be explained from both physiological and genetic perspectives (Hernández-Sebastià et al. 2005; Zhang et al. 2018; La et al. 2019; Lee et al. 2019). The competition for the carbon skeletons between protein and oil synthesis pathways could be the potential contributor for the inverse correlation (Hernández-Sebastià et al. 2005). In terms of genetic control, multiple studies confirmed the pleiotropy that exists in QTL that govern protein and oil concentrations (Zhang et al. 2018; La et al. 2019; Lee et al. 2019). There were 23 significant single nucleotide polymorphisms (SNPs) identified in common for both protein and oil, which showed negative effects located on chromosomes (Chr) 15 and 20 (Lee et al. 2019). This was also indicated in a study conducted by Zhang et al. (2018), in which a QTL tagged by the marker ss715622170 on Chr 15 exhibited significant associations with both protein and oil in addition to other 18 seed quality traits (Zhang et al. 2018).

Correlations between all targeted traits were determined from the data obtained in Ontario locations. The strong positive correlations between protein and the four amino acids, as well as the four amino acids with each other, were in accordance with the literature when calculated based on the dry seed weight basis (Assefa et al. 2018; Zhang et al. 2018; La et al. 2019; Ma et al. 2019). There are strong natural correlations between protein and amino acid concentrations, since amino acids are the major components of protein (Zhang et al. 2018). These correlations could result from diverse reasons such as the shared biosynthetic pathways, co-located QTL, pleiotropic QTL, and gene linkages (Shaul and Galili 1992; Hacham et al. 2007; Kastoori Ramamurthy et al. 2014; Ma et al. 2019). In plants, Lys, Met, and Thr are synthesized through the aspartate (Asp) family biosynthesis pathway by different branches (Hacham et al. 2007; Warrington et al. 2015). In soybean, Cys is the intermediate product of sulfur assimilation of protein synthesis, which also relates to the Met synthesis involving Asp and various enzymes (Wang et al. 2015; Ma et al. 2019). Therefore, these traits may share some common genetic-regulating factors and enzymes belonging to similar biosynthetic pathways (Ma et al. 2019). Wang et al. (2015) found four QTL associated with Cys and Met were also co-localized with the previously identified QTL for protein on Chr 3, 4, 17, and 20. There was a major genomic region on Chr 7 associated with Met, Cys, and protein identified by Ma et al. (2019). Similarly, several other QTL controlling both Cys and Met concentrations were identified on Chr 10 and 20 (Kastoori Ramamurthy et al. 2014; Warrington et al. 2015).

Several studies have identified QTL and biosynthetic pathways that independently associated with some of the traits described above without affecting the others. One QTL tagged by ss715602750 on Chr 8 was found to be related to the synthesis of Asp-related amino acids but unrelated to the protein concentration (Zhang et al. 2018). Similarly, a very close region on Chr 8 associated with marker ss715602763 was identified as responsible for Cys, Lys, and Thr but not the total protein concentration (Vaughn et al. 2014). Comparing these results, the AK-HSDH gene with described functions in this region was suggested to be the candidate gene regulating the Asp-related amino acid without affecting the total protein concentration (Zhang et al. 2018). This finding was meaningful since some of the correlations between protein and amino acids become undesirable when amino acid concentrations are expressed as the proportion of total crude protein (Zhang et al. 2018; La et al. 2019). Another study conducted on tobacco plants indicated a potential solution to enhance both Lys and Met without reducing the Thr concentration in legume and cereal crops (Hacham et al. 2007). In brief, the increased Lys can alter the expression level of S-adenosylmethionine synthase, resulting in upregulation of Met concentration without reducing Thr in the transgenic tobacco plants expressing both dihydrodipicolinate synthase and Arabidopsis cystathionine γ-synthase (Hacham et al. 2007). These findings provided background information and potential solutions to break the undesirable correlations and utilize the beneficial ones.

The correlations between fatty acids were consistent with previous reports (Maestri et al. 1998; Bachlava et al. 2008; Li et al. 2015; La et al. 2019; Abdelghany et al. 2020). The positive correlations between stearic and palmitic acids, in addition to the negative correlations between linoleic acid with oleic and stearic acids, were well established in different studies (Maestri et al. 1998; Cardinal and Burton 2007; Bachlava et al. 2008; Li et al. 2015; Abdelghany et al. 2020). Cardinal and Burton (2007) suggested the GmFATB1a gene might also possess activities toward stearoyl acyl carrier protein substrates besides encoding the palmitate thioesterase, which was a potential explanation for the positive correlation between palmitic and stearic acids. Moreover, Zhang et al. (2018) identified two QTL responsible for both palmitic and stearic acids on Chr 5 and 14, which were associated with markers ss715592503 and ss715618427, respectively (Zhang et al. 2018). The deletion of the candidate gene SACPD-C tagged by ss715618427 could lead to the elevation of stearic acid concentration in soybean (Gillman et al. 2014). The negative associations between linoleic acid on one hand and stearic acid, oleic acid on the other could be attributed to the fatty acid desaturation pathway (Bachlava et al. 2008; Li et al. 2015). During this conversion, stearic acid is the precursor of oleic acid, and then linoleic acid is further synthesized from oleic acid by the microsomal ω-6 desaturase enzymes coded by the Fad2-1 and Fad2-2 genes (Rubel et al. 1972; Bachlava et al. 2008, 2009; Li et al. 2015).

In addition to the biochemical and genetic factors described above, various QTL with pleiotropic effects were described by previous studies regarding the tight associations between the fatty acids (Hyten et al. 2004; Panthee et al. 2006a; Bachlava et al. 2008; Li et al. 2015; Zhou et al. 2019). Panthee et al. (2006a) identified two markers (Satt185 and Satt263) on Chr 15 consistently associated with both oleic and linoleic acids, which was likely related to the modifier QTL responsible for certain enzymes in the fatty acid desaturation pathway. Four SNPs identified by Li et al. (2015) presented significant associations with both oleic and linoleic acids but with inverse effects (Li et al. 2015). There was an interval on Chr 19 harboring significant QTL for palmitic acid, which was also significantly associated with linoleic, oleic, and linolenic acids (Hyten et al. 2004). A total of seven influencing SNPs for oleic and linoleic acids were located on Chr 1, 7, 8, 9, and 13 (Zhou et al. 2019). The inconsistency observed in different year–location combination for the correlation between stearic acid and oleic acid could be the result of the strong interaction of these traits with the environmental factors such as heat and drought conditions (Hou et al. 2006). These correlations between fatty acids should be considered and properly utilized depending on the specific breeding purposes, such as for human consumption and oil processing, since the emphasis on specific fatty acid may change based on the product in mind (Li et al. 2015; Zhou et al. 2019).

The strong negative correlation between sucrose and protein was described in detail by different research groups (Dhungana et al. 2017; Patil et al. 2018; La et al. 2019). As the corresponding traits for protein and oil, it was not surprising to expect correlations existing between sucrose and amino acids as well as linolenic acid (La et al. 2019). The production of sucrose, protein, amino acids, and oil resulted from integrated pathways including nitrogen assimilation, carbohydrate production, and carbon assimilation, which may provide a potential explanation regarding the correlations (Paul and Foyer 2001; Li et al. 2012). The syntheses of amino acids and protein require carbon skeletons supported by sucrose, which leads to the negative correlations between them (Paul and Foyer 2001). In addition, common genetic control may exist between protein and sucrose, as the identified protein QTL Seed protein 7-g10 was located close to a significant sucrose SNP at the 34 025 091 bp location on Chr 12 (Zhang et al. 2017). Very limited number of studies reported correlations between sucrose and linolenic acid as well as the underlying mechanism, and the positive correlation identified in this study was not consistent with previous studies (La et al. 2019; Zhao et al. 2019). Moreover, the previously reported results were not consistent with each other, and most of the correlations were nonsignificant (Zhang et al. 2017; Zhao et al. 2019). Consequently, further studies are required to confirm these identified correlations, which might be highly affected by environmental factors (Dhungana et al. 2017). The significant positive correlation between sucrose and raffinose at different levels was previously reported (Wang et al. 2014; Bueno et al. 2018; Jiang et al. 2018; La et al. 2019). This correlation was considered to be the consequence of the raffinose oligosaccharide metabolic pathway, in which the raffinose synthesis depends on the sucrose substrate (Hannah et al. 2006; Bueno et al. 2018). From the genetic perspective, the marker Satt359 on Chr 11 was indicated to be common for both sucrose and raffinose, which further supported the correlation between them (Wang et al. 2014). However, the sucrose concentration seemed to exhibit strong interactions with environments, but the raffinose concentration tended to be more genotype dependent, which possibly led to fluctuations in correlations between these sugar-related compositions (Kumar et al. 2010).

The positive correlations between DTM and yield, DTM and plant height, and lodging and height were documented previously (Cicek et al. 2006; Bachlava et al. 2008; Palomeque et al. 2010; Rossi et al. 2013). Egli et al. (1981) suggested that the increase of the DTM also allowed a longer seed-filling period, which dramatically contributed to the yield for grain crops, including soybean. The development of genetic technologies assisted recent studies in revealing the underlying genetic connections between these traits, which resulted in the correlations (Palomeque et al. 2010; Kim et al. 2012; Rossi et al. 2013; Fang et al. 2017). Kim et al. (2012) mapped several contributing QTL for DTM, height, and lodging located very close to the yield QTL on Chr 4, 14, and 18, which supported the opinion that increased yield was associated with later maturity and taller plants. A QTL linked with Satt162 was responsible for both plant height and lodging but was not in all the environments studied (Palomeque et al. 2010). Same authors also identified yield QTL tagged by Satt100, Satt277, and Sat_126 co-localized with several agronomic traits, including plant height and DTM (Palomeque et al. 2010; Rossi et al. 2013). Although inconsistency existed in the correlations and QTL for yield and other related agronomic traits, most of the previous studies agreed that major QTL were shared between these traits, showing pleiotropic or additive gene effects (Cober and Morrison 2010; Kim et al. 2012; Rossi et al. 2013).

Inconsistent patterns were observed in terms of correlations between some seed quality traits and agronomic traits, such as sugar-related compositions and amino acids or fatty acids with agronomic traits (Bachlava et al. 2008; Kumar et al. 2010; Zhang et al. 2018; La et al. 2019; Lee et al. 2019). For instance, the nonsignificant negative correlation between DTM and oil concentration was reported by Zhang et al. (2018), which was different from the previous result by Bachlava et al. (2008). Several significant correlations between seed quality traits and agronomic traits were observed in this study, such as that between yield with protein and oil, but none of them was strong. The potential influencing factors regarding the inconsistency of these correlations included the difference existing in various populations used, changeable environments, and strong interactions between genotypes and environments (Kumar et al. 2010; Dhungana et al. 2017; Jiang et al. 2018; La et al. 2019)

In conclusion, the accumulation of diverse seed quality traits was correlated with each other and showed consistent trends in both Canada and Ukraine, and between the two MEs through the experimental years as discussed in this study. Consistent strong and significant correlations were observed between the agronomic traits based on 2-year Ontario location data, but there was no strong significant correlation identified between the seed quality traits and agronomic traits in either MEs. Most of the findings in this study were in agreement with the literature. Identified undesirable correlations have provided difficult challenges for breeding programs aimed at improving desirable traits without sacrificing others (Zhang et al. 2018; La et al. 2019). These are potential challenges for breeding cultivars for Ukrainian market as well. Fortunately, an increasing number of trait-specific loci, which only affect one trait with almost no effect on others, were found for protein, oil, and several amino acids (Eskandari et al. 2013; Zhang et al. 2018; Lee et al. 2019). In addition to the potential modifications on genetic factors, manipulating the growing conditions by selecting proper locations may provide another possibility to enhance the desirable traits, for instance, sucrose (Kumar et al. 2010; Zeng et al. 2014). These findings may facilitate the proper utilization of different correlations between traits in future breeding programs targeting Ukraine.

Acknowledgements

The authors thanks current and former staff of Dr. Istvan Rajcan’s soybean research team at University of Guelph, the cooperation from Ukrainian research team and Université Laval in terms of field trial management, data collection and analytical supports. The authors also appreciate the funding supports from NSERC, Cangro Genetics Inc. and Huron Commodities Inc.