Transcriptomic Analysis on Developing Seed Uncovers Candidate Genes Associated with Seed Storage Protein in Soybean

Abstract

Soybean [Glycine max (L.) Merr.] is a globally significant crop that provides essential meal protein and vegetable oil for human consumption. The protein content in soybean seeds is a critical factor that affects nutrition regarding human dietary needs as well as livestock feed. Therefore, identifying the key genes that affect the soybean seed protein content is one of the major goals in soybean research. To identify candidate genes and related pathways involved in soybean seed storage protein during seed development, an RNA-seq analysis was conducted in two soybean varieties that differ in protein content. A series of pathways related to seed protein metabolism, including “Photosynthesis”, “TCA cycle”, and “Starch and sucrose metabolism” pathways, were identified through Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis. Seven candidate genes exhibiting two different gene regulation patterns were identified, six of which are directly related to the seed storage protein pathway, and one of which is related to the carbon binding pathway. An integrated analysis of transcriptomic and candidate gene expression trend suggested that 40 days after flowering (DAF) might be a crucial period for seed protein accumulation in soybean. Through a Weighted Gene Co-expression Network Analysis (WGCNA), two modules and two novel hub genes were found, which may be highly correlated with seed protein development. These findings might be valuable for a complete understanding of the genetic basis of seed protein content and lay a theoretical foundation for future gene functional identification and breeding efforts in soybean.

1. Introduction

Soybean [Glycine max (L.) Merri.] is one of the most valuable crops in the world for its high protein and oil contents in seeds. Protein and oil are two major economical components of soybean seed, with approximate proportions of 40% and 20%, respectively [1]. With well-balanced amino acids and a high concentration, soybean protein not only serves as a major part of poultry and livestock meals but also provides essential nutrition for humans through soy foods like tofu and soymilk [2]. Nowadays, the changes in people’s dietary habits require higher protein intake, which largely increases the human demand for soy protein. Soybean is an important oil crop as a major edible oil source and a raw material for industry [3]. With long-term research, it has been found that there is a negative correlation between protein and oil in soybean seeds [4,5,6,7]. Thus, breeding soybeans with higher contents of both protein and oil through genetic approach is an important goal for soybean breeders. However, our current understanding of the genetic mechanisms of protein underlying soybean seed development remains very limited, which impedes the progress of soybean breeding and genetic engineering to meet the strong demand.

Soybean seed storage protein is mainly composed of conglycinin (7S) and glycinin (11S), which account for about 70% of the total soy seed protein and jointly affect the content and quality of soybean seed protein. Usually, the 7S and 11S components of soybean storage protein are frequently referred to as β-conglycinin and glycinin, respectively [8,9]. Therefore, changes in the composition or the synthesis, transport, and deposition of β-conglycinin and glycinin will lead to a change in seed storage protein. With the availability of high-quality reference genomes (William 82, Zhonghuang 13, and the wild soybean W05), functional gene identification and studies of its regulation mechanism have become research hotspots in soybean. Although the SoyBase database contains over 300 protein-related QTLs (https://www.soybase.org/), only a very small number of transcription factors such as FUSCA3 (FUS3), ABSCISIC ACID INSENSITIVE3 (ABI3), LEAFY COTYLEDON1 (LEC1), and LEC2, and genes such as GmSWEET39, CCT, and GmOLEO1 [10,11,12,13] were cloned, and a few genes were functionally identified to be associated with seed protein, such as POWR1, GmST05, GmSWEET39, GmSWEET10a, and GmSWEET10b [10,14,15,16,17]. In a recent study, three genes, Glyma.03g219900, Glyma.14g119000, and Glyma.17g074400, associated with the Kompetitive Allele-Specificv Polymerase Chain Reaction (KASP) markers were identified as potential key candidate genes for seed protein content [18]. Thus, discovering more candidate genes using different methods is still needed for the complete understanding of protein accumulation in soybean.

As one of the common candidate gene-mining tools, transcriptomics has been used by many studies to investigate various soybean quality traits, including the seed oil content, protein content, and seed size. Using this approach, researchers have identified some candidate genes, like GmGA20OX and GmNFYA, and have elucidated the underlying molecular mechanisms governing these traits [19,20,21,22]. To identify the candidate genes and related pathways involved in soybean seed protein accumulation, two landraces with significant differences in protein content were selected. Soybean samples of the lines at four developmental periods were chosen for RNA-seq. A GO and KEGG enrichment analysis was performed to identify genes and pathways related to seed protein metabolism. The aim of this study was to identify candidate genes involved in seed storage protein accumulation. The results might enrich the understanding of the regulation of storage protein metabolism in the early soybean developmental seeds and could provide a valuable gene resource for the genetic improvement in soybean seed protein.

2. Materials and Methods

2.1. Plant Materials

Two soybean landraces from the Yunnan province of China were planted in the greenhouse at the same environmental conditions (13 h:11 h light/dark cycle, 28 °C:25 °C day/night cycle) in both 2022 and 2023 in the Agricultural Station of Yunnan University in Kunming (24.82° N, 102.84° E), Yunnan Province, China. The lines showed no significant differences in the main agronomic traits, i.e., plant height, growth period, seed coat, and seed size, but there was a significant difference in protein content. The dry seed protein content was measured by the Kjeldahl method. RNA-seq samples were collected from different plants from two groups of soybean plants named HP (high-protein landrace) and LP (low-protein landrace) in 2022. Three biological replications of each seed sample were collected 20, 30, 40, and 50 days after flowering (DAF), respectively. In total, 24 seed samples were collected, including HP_20_(1_3), HP_30_(1_3), HP_40_(1_3), HP_50_(1_3), LP_20_(1_3), LP_30_(1_3), LP_40_(1_3), and LP_50_(1_3). The RNA samples were collected and immediately frozen in liquid nitrogen and stored at −80 °C for further analysis.

2.2. Transcriptome Sequencing and Data Analysis

Total RNAs from the 24 seed samples were extracted using the TIANGEN RNA easy fast plant tissue kit (TIANGEN Biotech Company, Beijing, China) according to product specification. The samples were tested for quality, including purity, concentration, and integrity, prior to being sent out for RNA-seq. Sequencing was performed using the Illumina high-throughput sequencing platform, which is based on sequencing using synthesis technology, by Sangon Biotech Company, Shanghai, China. The clean reads were aligned to the Glycine_max_v4.0 soybean reference genome using HISAT2 [23], and StringTie was used to assemble and quantify [24]. Transcripts per million (TPM) was employed to quantify the transcription or gene expression level [25]. Reproducibility among samples was assessed by conducting a Pearson correlation coefficient analysis utilizing the R package Corrplot V4.3.3.

2.3. Analysis of Differentially Expressed Genes (DEGs)

Differential expression analysis was carried out using the DESeq2 V1.48.1 package in R. DEGs were filtered with |log2 fold change| ≥ 1 and p-value < 0.05. For each comparison group (HP_20 vs. LP_20, HP_30 vs. LP_30, HP_40 vs. LP_40, and HP_50 vs. LP_50), MapMan V3.6.0 was used to analyze the enrichment of metabolic pathways among these DEGs. Functional annotation of all expressed genes was conducted using the GO and KEGG database, and significant GO terms and KEGG pathways were identified.

2.4. Quantitative Real-Time Polymerase Chain Reaction (qRT-PCR) Verification

To evaluate the relative expression of ten DEGs across various periods and to confirm the RNA-seq findings, qRT-PCR was employed. Supplementary Table S2 included the gene-specific primers used for qRT-PCR. And in this study, the gene TUBB3 (β-tubulin, NCBI Gene ID: 547844) from soybean served as the internal control, and each reaction was performed with three technical replicates. The relative expression fold changes of the genes in HP compared to LP were determined using the 2^−ΔΔCt method [26]. In RNA-seq analysis, the fold differences of genes between HP and LP were counted via the TPM method. The log2 fold change values from both qRT-PCR and RNA-seq for DEGs were employed for graphical representation.

2.5. Gene Expression Trends Cluster Analysis

Based on the TPM values of each group (HP and LP from 20 to 50 DAF), MapMan V3.6.0 was used to cluster the gene expression trends within each group using Euclidean distance. For genes in different clusters with distinct gene expression trends, the R package Pheatmap V1.0.13 was used for the graphical presentation of gene expression.

2.6. WGCNA

WGCNA was performed using R package WGCNA V1.72.5 [27]. A gene cluster dendrogram was constructed with a power value of 15. The gene modules were further classified and clustered by a similarity threshold of 0.8 and a minimum module size of 30. For co-expression network construction, the top 100 gene pairs with the highest weighted value in each module were chosen, and the gene ranked highest by the degree method in cytoHubba (a Cytoscape V3.9.1 plugin) was designated as the hub gene. The network of trait-specific modules was visualized using Cytoscape V3.9.1.

3. Results

3.1. Comparison of Protein Content Between HP and LP

In this study, two soybean landraces, HP and LP, were analyzed, which both have yellow seed coats and similar seed sizes (Figure 1A). There was a significant difference in the dry seed protein content between these two landraces across different years (Figure 1B). Specifically, in 2022, the LP group exhibited a protein content of 38.1%, while the HP group showed a higher protein content of 43.6%. Similarly, in 2023, the protein content of the LP group was 34.1% compared to 40.63% in the HP group.

3.2. Analysis of RNA-Seq Data

Soybean seeds 20, 30, 40, and 50 DAF were collected from HP and LP in 2022, respectively. A total of 24 libraries were generated and analyzed, yielding 105 Gb of clean data. Over 94.47% of the bases had a Q30 quality score (exceeding 99.9% accuracy), and in each sample, 93.78–95.63% of the reads were uniquely aligned with the soybean genome. To check the reliability of the biological replicates, a correlation analysis was carried out among these replicates (Supplementary Figure S1A). Two samples, LP_20_3 and LP_50_2, were eliminated from further analysis due to their low correlations with the other two replicates. Genes with normalized reads below 0.5 TPM were removed from the analysis. In total, 34,905, 32,246, 31,638, and 30,101 genes were found to be expressed at 20 DAF, 30 DAF, 40 DAF, and 50 DAF in the HP line, respectively. Similarly, 36,594, 33,791, 32,014, and 35,235 genes were identified in the corresponding samples in the LP line, respectively (Figure 1C). Around 65% of the expressed genes had TPM values between 0.5 and 10, 32% between 10 and 100, and fewer than 3% above 100. Supplementary Figure S1B shows the overlap of expressed genes in all replicated samples of HP and LP.

3.3. Identification of DEGs and Gene Enrichment Analysis

For the identification of DEGs, we carried out pairwise comparisons between HP and LP at every developmental stage. A differential expression analysis showed that 5036, 4536, 4255, and 2720 DEGs were identified at 20, 30, 40, and 50 DAF, respectively. Among the DEGs at each developmental stage, 1331, 1295, 1335, and 732 genes were up-regulated, and 3705, 3241, 2920, and 1988 genes were down-regulated, respectively (Figure 1D). These DEGs were used to analyze the overview enrichment of metabolic pathways in MapMan. The results show that although the number of DEGs decreased with seed development, the relative gene expression levels of the DEGs increased, especially at 50 DAF (Supplementary Figure S2).

In a preliminary investigation of the functions of these DEGs, a GO analysis was employed. The results show that many GO terms were significantly enriched and could be classified into three categories: biological process, cellular component, and molecular function. From 20 DAF to 50 DAF, for biological process ontology, the representative GO terms were “regulation of transcription”, “protein phosphorylation”, and “defense response”. For the cellular component ontology, the representative GO terms were “Nucleus”, “chloroplast”, and “integral component of membrane” (Figure 2). For molecular function, the representative GO terms were “protein binding”, “ATP binding”, and “metal ion binding” (Figure 2).

Afterward, the DEGs were matched to the reference canonical pathways in the KEGG database. Based on the KEGG annotations, about 20 pathways were significantly enriched. At 20 DAF, “Plant hormone signal transduction”, “Starch and sucrose metabolism”, “Glycerolipid metabolism”, and “Linoleic acid metabolism” were significantly enriched. “Photosynthesis”, “Photosynthesis—antenna proteins”, “TCA cycle”, and “Starch and sucrose metabolism” were significantly enriched at 30 DAF. “Photosynthesis”, “Starch and sucrose metabolism”, “Fatty acid degradation”, and “Fatty acid biosynthesis” were significantly enriched at 40 DAF. At 50 DAF, “Photosynthesis—antenna proteins”, “Photosynthesis”, “DNA replication,” and “Carbon fixation in photosynthetic organisms” were significantly enriched. These enrichments are depicted in Figure 2.

3.4. Gene Cluster Analysis

Genes exhibiting normalized read counts exceeding 1.0 TPM in both the HP and LP groups were selected for cluster analysis employing Euclidean distance (Figure 3A,B). A total of 10 clusters were identified (from C1–C10). The genes in cluster C3 showed a significant relationship with the seed storage protein activity pathway. In contrast, the other clusters did not display strong specificity or representativeness. Within cluster C3, 12 and 9 genes were clustered in the HP group and LP group, respectively (Supplementary Table S1). Out of the 12 genes in the HP group, 7 candidate genes were DEGs at 20 DAF with a p-value < 0.05, showing no significant difference at 30–50 DAF. And among these seven genes, five were found to be overlapping with the LP group, and two were uniquely clustered in the HP group. Impressively, all genes within this subset exhibited enhanced expression levels relative to the LP group, suggesting their potential regulatory roles in the regulation of seed storage protein activities (

Table 1. Seven candidate genes with functional annotation, GO term enrichment with annotation description, and log2 (fold change) with p-value. Candidate genes were annotated in Glyma.a4.v1 in SoyBase (https://www.soybase.org) and Phytozome (http://www.phytozome.net); log2FC = log2 (fold change).

Gene_ID	Function Annotation	GO Term	Annotation Description	Log2FC_20	p-Value
Glyma.02g012600	Legume lectin domain	GO:0030246	carbohydrate binding	4.5041315	4.57 × 10−2
Glyma.10g246300	Beta-conglycinin alpha prime subunit 1 protein	GO:0045735	nutrient reservoir activity	8.6359036	3.81 × 10−2
Glyma.13g123500	Glycine protein	GO:0045735	nutrient reservoir activity	10.936369	3.30 × 10−2
Glyma.13g194400	Albumin I protein	GO:0045735	nutrient reservoir activity	6.9725067	1.26 × 10−3
Glyma.20g148200	Beta-conglycinin, beta chain-like protein	GO:0045735	nutrient reservoir activity	4.784346	3.21 × 10−15
Glyma.20g148300	RmlC-like cupins superfamily protein	GO:0045735	nutrient reservoir activity	5.4897572	2.56 × 10−2
Glyma.20g148400	Beta-conglycinin alpha prime subunit 2 protein	GO:0045735	nutrient reservoir activity	5.4796174	2.60 × 10−2

). Notably, a total of nine genes were identified as overlapping between the two groups (Supplementary Table S1), with seven of these genes displaying enrichment in GO terms associated with storage protein functions. Among these seven overlapping genes, six were enriched in GO:0045735, a functional annotation corresponding to nutrient reservoir activity, which are involved in storage protein metabolism. The gene functional annotations were derived from William82.a4.v1, which is available in Soybase (https://www.soybase.org/, accessed on 26 December 2024) and Phytozome (https://phytozome-next.jgi.doe.gov/, accessed on 26 December 2024). Specifically, Glyma.10g246300 and Glyma.20g148400 were functionally annotated as Beta-conglycinin alpha prime subunit proteins. Glyma.13g123500 was annotated as a glycine protein, and Glyma.20g148300 was identified as a member of the cupin superfamily. Glyma.13g194400 and Glyma.20g148200, which exclusively clustered within the HP group, were annotated as Albumin I chain b and Beta-conglycinin and beta chain-like proteins, respectively. The remaining gene, Glyma.02g012600, was identified as containing a legume lectin domain (Lectin_legB), which was enriched in GO:0030246 related to carbohydrate binding (Table 1).

The transcriptomic profiles of these seven genes are illustrated in the heatmap (Figure 3C). Interestingly, the two genes unique to the HP group exhibited a distinct expression pattern in comparison to those in the LP group. Two genes, Glyma.13g194400 and Glyma.20g148200, displayed up–down expression patterns in the HP group, contrasting with the up-regulation observed in the LP group (Figure 3C). The remaining five genes exhibited a uniform up–down expression pattern.

3.5. WGCNA and Visualization

A total of 23,594 genes with an average TPM > 1 across all samples were used to perform WGCNA. A gene cluster dendrogram was generated and partitioned into distinct modules by examining the correlation coefficients among these genes. Each module represents a cluster of highly interconnected genes with robust correlation coefficients inside the cluster. Ultimately, nine unique distinct modules (labeled with different colors) were identified (Figure 4A). Of particular interest, two of these modules demonstrated significant associations with LP_20 and HP_50 during soybean seed development (Figure 4B). The gene expression patterns within these modules showed notable differences between the HP and LP groups (Supplementary Figure S3). The blue module, containing 321 identified genes, showed a significant positive correlation with LP_20. Within this module, Glyma.03g086200 was identified as a candidate hub gene (Figure 4C), annotated as peptidyl-prolyl cis-trans isomerase FKBP62. In contrast, the green module, consisting of 86 genes, revealed a negative relationship with HP_50. Similarly, Glyma.02g272400 emerged as a candidate hub gene, annotated as abscisic aldehyde oxidase (AAO3) (Figure 4D).

3.6. Validation of DEGs by qRT_PCR

To confirm the accuracy and reproducibility of the transcriptome analysis results, 10 DEGs with p < 0.05, i.e., Glyma.02g061800, Glyma.03g248400, Glyma.08g127600, Glyma.10g088300, Glyma.11g153587, Glyma.11g177419, Glyma.13g048500, Glyma.16g123400, Glyma.18g214500, and Glyma.20g213600, were randomly selected for validation via qRT-PCR. The expression trend of these genes showed consistency with that of the RNA-seq sequencing results, indicating that the RNA-seq sequencing results should be reliable (Figure 5).

4. Discussion

Soybean is a primary source of high-quality vegetable protein. Identifying candidate genes and understanding the molecular mechanisms that regulate protein profiles is crucial for the improvement in seed nutritional value in soybean. In this study, two soybean landraces, i.e., HP (high protein content) and LP (low protein content), were used to perform transcriptomics of soybean seeds from 20 DAF to 50 DAF, which includes the critical time points of seed development from early to full development. These four periods coincided with soybean development stages R5 (beginning seed) to R6 (full-sized seed). During these periods, the seed protein content showed a decline and then stabilized. The highest protein content was observed in the period of 20–30 DAF, and stabilization occurred at 40 DAF [28,29,30]. Thousands of DEGs between HP and LP group were identified, including the published gene GmSWEET10b identified at 20 DAF. And a sharp reduction in the number of up-regulated DEGs was observed at 50 DAF (Figure 1C). The finding was consistent with previous results showing that the period from 10 to 40 DAF was critical for the rapid accumulation of lipids and storage proteins and that actual protein levels increase rapidly during this time [31].

The Gene Ontology term analysis uncovered that DEGs between HP and LP from 20 to 50 DAF were significantly enriched in specific GO terms (like protein binding, chloroplast, and protein phosphorylation), which may play a critical role during seed development (Figure 2). For the KEGG analysis, the DEGs were enriched in the starch and sucrose metabolism pathway at seed development stages 20–40 DAF (Figure 2). Recent studies have reported that starch and sucrose metabolism play an essential role in regulating the oil and protein contents in seeds. For example, GmSWEET39, a sugar transporter gene, can regulate oil and protein accumulation by affecting sugar delivery from the maternal seed coat to the filial embryo [10]. GmSWEET10a and GmSWEET10b have been demonstrated to facilitate the transport of sucrose and hexose in soybean, which subsequently exert a decisive influence on the oil and protein contents as well as the seed size, thereby playing a pivotal role in determining these key agronomic traits of soybean [17]. The GmSWEET10b gene was detected as a DEG at 20 DAF, and this finding supports the notion that starch and sucrose metabolism plays a key role in seed protein accumulation. At the same seed development stage at 20–40 DAF, the KEGG results show that many DEGs were also enriched in pathways related to linoleic acid metabolism and fatty acid degradation/biosynthesis. This result indicates that 20–40 DAF was an important period for seed oil content accumulation, and this is consistent with the result in our previous study [32]. At the development stage of 30–50 DAF, the DEGs were significantly enriched in the photosynthesis pathway (Figure 2). Generally, light is found to participate in storage compound synthesis by providing significant ATP through photosynthesis in developing soybean embryos [33]. The role of light is also indicated by the canopy position of seeds, with seeds at the top of the canopy accumulating much more protein compared to bottom-canopy seeds [34].

Seven candidate genes were identified through cluster analysis in both the HP and LP groups, of which one gene was related to the carbohydrate binding pathway, and six genes were directly related to the seed storage protein pathway (Table 1). One gene, Glyma.02g012600, was functionally annotated as a legume lectin related to carbohydrate binding. And this gene was the only lectin family member gene expressed during seed development [35]. Legume lectins represent the largest family of carbohydrate-binding proteins, which are widely distributed in the seeds of most legume plants and make up approximately 2% of the total protein in soybean seeds [36]. Furthermore, numerous reports have indicated that lectins were involved in many biological processes related to stress signal transduction and defense and were capable of fulfilling specific functions inside the plant cell or in the interaction with other organisms [37,38,39]. This indicated that the Glyma.02g012600 gene might influence seed protein composition and total content by modulating legume lectin, thereby affecting seed development. Three genes, Glyma.10g246300, Glyma.20g148200, and Glyma.20g148400, were annotated as types of Beta-conglycinin, which were frequently referred to as 7S components of soybean storage protein [40]. These three genes may regulate the synthesis of Beta-conglycinin, thereby causing changes in the 7S components of soybean storage protein and regulating seed storage protein accumulation. These three genes could directly influence soybean storage protein accumulation through gene regulation. Glyma.13g123500, also known as the published gene Gy5, was related to glycine, which is frequently referred to as 11S components of soybean storage protein [41]. And this gene may regulate the synthesis of glycine, thereby causing changes in the 11S components of soybean storage protein and regulating seed storage protein accumulation. Glyma.13g194400 was annotated as albumin I, which is usually regarded as a metabolic and enzymatic protein with cytosolic function and consists of molecules that have functional roles in seed germination [42]. In a previous study, Glyma.13g123500 and Glyma.13g194400 were proven to show relatively high expression levels at the seed development stage compared with other tissues, and they were regarded as promising candidates associated with the soybean protein content [43]. This indicates that Glyma.13g194400 may influence the soybean protein content by promoting seed protein synthesis through metabolic participation or gene regulation. Glyma.20g148300 was related to the seed storage protein pathway and annotated as a RmlC-like cupin superfamily protein. The cupin superfamily of proteins is known to play crucial roles in plant development and defense and also comprises the major globulin storage proteins from peanuts, soybean, and tree nuts [44,45,46]. Therefore, the Glyma.20g148300 gene may increase the seed protein content by participating in storage protein synthesis and accumulation during seed development, and potentially via other metabolic pathways or by enhancing stress resistance. All six genes were enriched in the GO term GO:0045735, which was directly related to seed storage protein accumulation.

These seven genes might influence the accumulation of seed storage protein through two distinct regulatory patterns. Five genes, i.e., Glyma.02g012600, Glyma.13g123500, Glyma.10g246300, Glyma.20g148300, and Glyma.20g148400, showed an up–down regulation pattern during seed development from 20 DAF to 50 DAF, with a turning point at 40 DAF in both the LP and HP groups (Figure 3C). However, Glyma.13g194400 and Glyma.20g148200 showed an up–down regulation pattern in the HP group, whereas an up-regulation pattern was seen in the LP group during seed development from 20 DAF to 50 DAF. In the previous study, the Glyma.02g012600, Glyma.13g123500, Glyma.20g148200, Glyma.20g148300, and Glyma.20g148400 genes showed the highest level of mRNAs at the mid stage (R6 stage) of seed development [47]. Combined with the distinct expression patterns of the candidate genes previously observed in the HP and LP groups, this might imply that the diverse regulatory patterns of genes can indeed influence the variation in seed protein content. As the KEGG results show that most DEGs were significantly enriched in pathways associated with the seed development stages of 20–40 DAF and 30–50 DAF, these results also supported that 40 DAF might be a crucial period in seed development.

WGCNA is a strong technique used in systems biology to find key genetic networks in a variety of crops [48,49,50]. In this study, the WGCNA identified two modules, i.e., the blue and green modules, that showed a significant correlation with the protein content in soybean (Figure 5). For the blue module, the hub gene Glyma.03g086200 was identified as the candidate hub gene and was annotated as peptidyl-prolyl cis-trans isomerase FKBP62 (Figure 4C). The homologous gene for AtFKBP62 was ROF1 in Arabidopsis. The expression of ROF1 is induced by heat stress and developmentally regulated [51]. Studies have reported that a higher protein content is positively correlated with higher temperatures [52,53,54,55], indicating that seed protein expression might be sensitive to thermal factors. Therefore, Glyma.03g086200 might potentially influence the seed protein content by responding to heat stress in soybean. In the green module, the hub gene Glyma.02g272400 was annotated as abscisic aldehyde oxidase (AAO3) (Figure 4D). A homologous gene for AAO3 is AT3G43600, which is involved in the biosynthesis of abscisic acid (ABA). The plant growth regulator ABA affects many aspects of plant growth and development, including embryo development, seed maturation and dormancy, as well as adaptation to a variety of environmental stresses [56]. Given the conserved function of ABA in plant biology, it is likely that Glyma.02g272400 in soybean might contribute to ABA biosynthesis, thereby affecting these critical physiological processes. With the reference of homologous gene functions annotated in Arabidopsis, these two hub genes might play an important role in regulating seed development, especially in protein accumulation.

There is significant limitation in this study. For example, the candidate genes and pathways related to protein contents were identified solely by RNA-seq analysis, which is inadequate to confirm their functions. An integrated analysis of proteomics, metabolomics, and functional validation should be carried out to confirm the gene functions in the future, which might be helpful for the understanding of the molecular regulatory mechanisms underpinning soybean protein accumulation.

5. Conclusions

In this study, two soybean varieties with significant differences in protein content were used to carry out transcriptomic analysis. Seven candidate genes with two different expression patterns were identified to regulate the accumulation of seed storage protein. The transcriptomic analysis and candidate gene expression pattern analysis indicated that 40 DAF might be a crucial period for the accumulation of seed protein. The WGCNA identified two hub genes that were highly correlated with seed development. These results might provide significant insights into the regulation of seed protein and offer valuable insights for the improvement of seed protein content in soybean.