Introduction

Soybean (Glycine max (L.) Merr.) is the most widely grown oilseed crop in the world, with nearly 129 million hectares (ha) planted in 2021, according to the Food and Agriculture Organization of the United Nations (FAO 2021). Soybean yields have steadily increased over the past century through selective breeding and agronomic changes (Specht et al. 2014). However, these yield increases have lagged those observed in staple cereal grains such as maize, rice, and wheat, which had major “green revolution” breakthroughs precipitated by the optimization of shoot architecture for modern agricultural systems (Liu et al. 2020). As a result, understanding the genetic bases underlying key soybean shoot architectural traits is imperative for continued productivity growth in soybean.

Soybean stem growth habit is a vital trait that lies at the intersection of shoot architecture and flowering to determine soybean yield through its effects on height, node number, stress tolerance, and adaptation to diverse environments (Ping et al. 2014). This trait is defined as the duration of vegetative growth at the shoot apical meristems (SAMs) after the induction of flowering at axillary meristems elsewhere in the plant. Determinate soybeans cease vegetative growth at the SAM almost immediately upon the plant reaching its beginning flowering (R1) stage as the SAM transitions to a reproductive inflorescence that forms a characteristic terminal cluster of flowers. Indeterminate soybeans in contrast maintain vegetative growth at the SAMs for the length of the growing season as resources allow even as reproductive growth occurs simultaneously at other meristems. Semideterminate soybeans are an intermediate class in which the vegetive SAM is maintained for a period of time after R1 before transitioning to a terminal raceme like that observed in determinate varieties (Ping et al. 2014). In general, determinate soybean varieties are grown in areas where long growing seasons allow sufficient time for distinct vegetive and reproductive stages, while indeterminate soybean varieties are preferred at higher latitudes with shorter growing seasons (Clark and Ma 2023). Semideterminate soybean varieties are also useful in short-season environments and have been adopted in some regions such as northeastern China (Liu et al. 2008). They may be able to provide the advantages of determinate varieties, such as resistance to lodging, increased pod number per node, and uniformity of pod maturation, without sacrificing the flexibility of development and increased node number that makes indeterminate varieties desirable (Ping et al. 2014). The compact architecture of semideterminate soybeans echoes the semi-dwarf phenotypes utilized to launch the “green revolution” in cereal crops and consequently could be a major target for soybean improvement.

Classical genetic analysis identified two genes, Dt1 and Dt2, which control stem growth habit in soybean (Bernard 1972). Dt1 and dt1 govern indeterminate or determinate growth, respectively, and the indeterminate allele Dt1 is dominant over the determinate allele dt1. Dt2 and dt2 modulate semideterminate and indeterminate growth, respectively, in the Dt1 backgrounds. dt1 is recessively epistatic to Dt2, so that soybeans with the genotype dt1dt1 are determinate regardless of what allele is present at the Dt2 locus. The functional Dt2 allele conferring a semideterminate growth habit is completely dominant over the dt2 allele conferring an indeterminate growth habit (Bernard 1972; Ping et al. 2014). Depending on the environment and genetic background, terminal flowering at the SAM caused by the dt1 allele may be indistinguishable from terminal flowering caused by the Dt2 allele, so that while these are distinct phenotypic classes, in practice semideterminacy refers to stem termination in soybeans which do not carry the genotype dt1/dt1 (Clark and Ma 2023). Dt1 was later identified to be Glyma.19g194300, which is the functional ortholog of Arabidopsis thaliana TERMINAL FLOWERING 1 (TFL1) encoding a phosphatidylethanolamine-binding protein (PEBP) and maintains indeterminate growth at the shoot apex (Tian et al. 2010; Yue et al. 2021). At least four recessive loss-of-function mutations in Dt1 result in determinate growth (Tian et al. 2010). Dt2 is Glyma.18g273600, which encodes a MADS-domain transcription factor that represses Dt1 by directly binding to its promoter (Ping et al. 2014; Liu et al. 2016). Dt2 also acts as a transcriptional activator of genes which induce flowering including GmSOC1, GmAP1, and GmFUL (Zhang et al. 2019).

There is considerable quantitative variation in degree of soybean stem termination within the three broad phenotypes of determinate, semideterminate, and indeterminate. We wondered whether additional genes influence the range of phenotypic values classified as semideterminate. In this study, we employed allelic analysis of soybeans in the USDA soybean Germplasm Collection and a series of genome wide association studies (GWAS) to further dissect the genetic basis underlying semideterminacy in soybeans. We focused particularly on semideterminate soybeans which did not possess the known stem termination alleles of dt1 or Dt2. We then selected four of such accessions, crossed them to indeterminate cultivars, and through linkage mapping, found that dt3, a recessive allele(s) at a novel locus on chromosome 10, is responsible for their semideterminacy. This study provides a starting point for a more comprehensive understanding of the genetic underpinnings of stem growth habit in soybeans and will aid soybean breeders in marker assisted selection (MAS) for this key trait.

Materials and methods

Plant materials

A total of 67 publicly available landraces which were classified as semideterminate according to publicly available phenotypic data from the USDA soybean germplasm collection but which don’t carry known stem termination alleles (i.e., Dt2 or dt1) are listed in Table S1. Five biparental mapping populations were developed and used in this study. The first consists of 270 F2 lines derived from a cross between the high-yielding indeterminate cultivar HS6-3976 and the semideterminate landrace PI 548406 (Richland). The second consists of 108 F2 lines derived from a cross between the indeterminate cultivar Williams 82 and the semideterminate landrace PI 297520 (Iregi Universal). The third consists of 108 F2 lines derived from a cross between the indeterminate cultivar Williams 82 and the semideterminate landrace PI 561350B (Yong ji qun zhong da dou). The fourth consists of 153 F2 lines and F3 families derived from a cross between the indeterminate cultivar Williams 82 and the semideterminate landrace PI 597403B (Krasnodar 391–89). The fifth population consists of 50 F2 lines derived from a cross between the indeterminate cultivar Williams 82 and the semideterminate landrace PI 248409 (Subotica).

Mapping population development

Four of the 67 publicly available semideterminate accessions and PI 561350B were selected for crossing once the terminal raceme was visible during R2 (full flowering stage) and crossed to Williams 82 or HS6-3976. Crossing was conducted in the field at the Purdue Agronomy Center for Research and Education (ACRE), with F1 seeds grown in the greenhouse. Markers polymorphic between the pairs of parents were used to confirm the hybrid status of the F1 seeds. F2 plants were phenotyped twice, at the R5 (beginning seed) and R7 (Full maturity) growth stages and classified as either indeterminate or semideterminate based on, respectively, the absence or presence of a terminal cluster of pods. F2:3 families were planted the following year at ACRE and phenotyped in the same manner for the Williams 82 by Krasnodar 391–89 population. The 67 publicly available accessions were phenotyped in the same manner for two consecutive years in the field at ACRE.

Genome-wide association study

Genotypic data was drawn from the Illumina Infinium SoySNP50K BeadChip database (Song et al. 2013), and phenotypic data was drawn from the USDA Germplasm Resources Initiative (GRIN) accessed through SoyBase. 1000 accessions were randomly selected using R. Phenotypes were set as determinate = 0, semideterminate = 1, and indeterminate = 2. GWAS was performed with the TASSEL 5 software (Bradbury et al. 2007) using the Mixed Linear Model (MLM). The 170 indeterminate accessions used in GWAS after filtering out dt1 were randomly selected using R. P-values were considered significant at a threshold of 0.05 divided by the population number, n, within each GWAS.

DNA isolation and bulked segregant analysis

In general, 10–12 of the most typical semideterminate and indeterminate F2 individuals were selected from each F2 population. Plants were harvested in individual bags and 25 seeds for each of the selected plants and the parent lines were germinated in a Petri dish for DNA isolation. Hypocotyls were cut from 15 to 20 germinating seeds for each plant. Hypocotyls for the selected plants of each phenotype within each cross were bulked and treated as a single sample. DNA was isolated using a standard CTAB method modified from Mace et al. (2003). Bulked and parent samples were genotyped using the BARCSoySNP6K (Song et al. 2020) which contains about 6000 single nucleotide polymorphisms (SNPs). Regions were identified in which the bulked sample was homozygous for one of the two parent genotypes according to the principle of BSA.

Linkage analysis

The population derived from Williams 82 and the semideterminate landrace Krasnodar 391–89 was used for linkage analysis. Insertion deletion (InDel) markers polymorphic between the parents were used for genotyping. Linkage analysis calculations were performed according to the Kosambi function using QTL IciMapping. The InDel primers used in this study are listed in Table S9.

Results

GWAS and allelic analysis identified semideterminate Dt2soybean varieties possessing neither dt1 nor Dt2

Of the 17,376 accessions within the USDA Soybean Germplasm Collection with available stem growth habit phenotypic data, 8629 are classified as determinate, 7439 indeterminate, and 1308 semideterminate (Fig. S1). To explore whether any loci other than the previously described Dt1 and Dt2 are responsible for the semideterminate phenotype, we selected 990 random accessions (Table S1) from the USDA collection and conducted a genome wide association study (GWAS) for growth habit using the Illumina Infinium SoySNP50K BeadChip database (Song et al. 2013). This revealed a single major peak on chromosome 19 corresponding to the Dt1 locus (Fig. 1a) with the strongest associated SNP marker being ss715635425 (Gm19: 45,204,441), about 19 kb downstream of the Dt1 gene based on the Williams 82 reference genome second annotation (Fig. 1a).

Fig. 1
figure 1

Genome wide association studies for stem growth habit. a analysis of 990 random accessions reveals a single major locus on chromosome 19 corresponding to Dt1. Significant SNPs at the Dt1/dt1 locus are colored red. b analysis of 170 semideterminate lines remaining after filtering out dt1 loss of function alleles paired with 170 random indeterminate lines reveals the Dt2 locus on chromosome 18. The peak markers at this region were used to filter out semideterminate accessions carrying the functional Dt2 allele conferring semideterminate stem growth and identify those accessions potentially carrying novel stem termination alleles. Significant SNPs at the Dt2/dt2 locus are colored red

Full size image

We next turned to all accessions classified as semideterminate. To exclude those accessions where the semideterminate stem growth habit classification resulted from the loss-of-function alleles of dt1, we removed all lines which had the dt1 alleles associated with determinate growth according to the GWAS at the SNP markers ss715635415 (Gm19:45,141,000) and ss715635425 (Gm19: 45,204,441). A total of 170 semideterminate accessions remained after this filtering process. We then randomly selected 170 accessions classified as indeterminate, paired these with the 170 remaining semideterminate accessions (Table S2), and again conducted a GWAS for stem growth habit. This second GWAS revealed a single major peak on chromosome 18 at the Dt2 locus, with the most significant SNP being marker ss715632223 (Gm18: 55,622,046) about 16 kb upstream of the annotated transcription start site for the Dt2 gene (Fig. 1b).

We then filtered out those semideterminate accessions which carried the dominant semideterminate Dt2 allele in the same manner as was done for the Dt1 locus, keeping only those which did not have the Dt2 allele at SNP 55622046 associated with semideterminacy. A total of 74 accessions classified as semideterminate remained after filtering out of dt1 and Dt2 (Table S3). A third GWAS consisting of these 74 semideterminate accessions and 74 randomly selected indeterminate accessions failed to identify any significant loci. Nonetheless, we proceeded to evaluate the 74 lines. Sanger sequencing of the causal SNP sites for dt1 and the SNPs associated with the Dt2 haplotype reported in our earlier study (Ping et al. 2014) was performed for each of the 74 lines. This showed that four of the 74 accessions likely carried the semideterminant Dt2 allele and three of the accessions carried the determinant dt1 allele, leaving 67 accessions classified as semideterminate but not carrying stem termination alleles at either of the two known loci.

When planted in the field, 37 of the 67 “semideterminate” accessions displayed typical semideterminate phenotypes (Fig. 2; Table S3). The remaining 30 accessions are not necessarily classified incorrectly, as the semideterminate phenotype of soybean landraces adapted to lower latitudes may be masked by their late flowering time in the midwestern USA where this study was conducted. The 37 lines displayed variation in degree of stem termination but were visually indistinguishable from semideterminate lines known to carry the Dt2 allele.

Fig. 2
figure 2

Exemplification of phenotypic difference in stem growth habit. a Williams 82, a typical indeterminate cultivar b Richland, one of one of the 37 typical semideterminate lines not carrying dt1 or Dt2

Full size image

Inheritance pattern of semideterminacy from the semideterminate accessions

Three of the 37 semideterminate accessions (Krasnodar 391–89, Subotica, and Iregi Universal) were crossed to the indeterminate cultivar Williams 82 and one, Richland, was crossed to the indeterminate cultivar HS6-3976. The F2 progeny plants were screened in the field and classified as semideterminate or indeterminate (Table 1). In the (Williams 82 × Subotica) population, 88 individuals were classified as indeterminate and 20 as semideterminate (χ2 for 3:1 = 2.420, p = 0.1198). In the (Williams 82 × Krasnodar 391–89) population, 110 individuals were classified as indeterminate and 43 semideterminate (χ2 for 3:1 = 0.786, p-value = 0.3752). In the population derived from crossing Williams 82 with Iregi Universal, 40 individuals were classified as indeterminate and 10 were classified as semideterminate (χ2 for 3:1 = 0.66667, p = 0.4142). Finally, in the population derived from crossing HS6-3976 with Richland, 199 individuals were classified as indeterminate and 71 semideterminate (χ2 = 0.242, p-value = 0.6228). These observations suggest that the semidetermancy carried by each of these four semideterminate accessions is likely to be controlled by a single gene, and that, distinct from Dt2-mediated semideterminacy, the semideterminacy in these four accessions was caused by naturally occurring recessive mutations.

Table 1 Phenotypic segregation for stem growth habit in four F2 populations
Full size table

Discovery of a new locus, Dt3, at which recessive mutation (s) resulted in semideterminacy

Bulked segregant analysis (BSA) of the four biparental populations revealed overlapping regions on the long arm of chromosome 10 from the semideterminate parents are associated with semideterminacy (Fig. 3a). The regions revealed by the (HS6-3976 × Richland), (Williams 82 × Subotica), (Williams 82 × Iregi Universal), and (Williams 82 × Krasnodar 391–89) populations were bounded by SNP markers at 40.58 and 50.04 Mb (Fig. 2 b, Table S7), 41.65 and 47.7 Mb (Fig. 2 c. Table S6), 43.02 and 45.55 Mb (Fig. 2 d, Table S6), and 44.67 and 45.83 Mb (Fig. 2e, Table S6), respectively, according to the Williams 82 reference genome. Together, these observations suggest that the semideterminacy in the four parental lines is controlled by recessive mutations that likely occurred at the same locus, designated Dt3.

Fig. 3
figure 3

Bulked segregant analysis (BSA) of five biparental mapping populations. a The location of the dt3 region on the long arm of chromosome 10 highlighted by the orange bar. b The location of the dt3 region in the HS6-3976 × Richland population defined by BSA. c The location of the dt3 region in the Williams 82 × Subotica population defined by BSA. d The location of the dt3 region in the Williams 82 × Iregi Universal population defined by BSA. e The location of the dt3 region in the Williams 82 × Krasnodar 391–89 population defined by BSA. f The location of the dt3 region in the Williams 82 × PI 561350B population defined by BSA

Full size image

An Inheritance pattern suggests a semideterminate accession carries both Dt2 and dt3

We also crossed PI 561350B, one of the semideterminate lines identified by our screen but which was found to carry the functional allele of Dt2, to Williams 82. We found that 91 F2 individuals derived from the two parent lines were classified as semideterminate and 17 as indeterminate. Instead of a 3:1 (semideterminate to indeterminate) ratio (χ2 = 4.938 p-value = 0.0263), we observed a 13:3 (semideterminate to indeterminate) ratio (χ2 = 0.642, p-value = 0.4230). This ratio can be explained by the hypothesis that semideterminacy in PI 561350B is determined by two loci, Dt2 and dt3. As expected, BSA of the (Williams 82 × PI 561350B) F2 population revealed that the semideterminacy in PI 561350B is associated with 37.69–45.32 Mb (i.e., the dt3 region) on chromosome 10 (Fig. 2f, Table S6), and with the 54.53–57.90 Mb (i.e., the Dt2 region) on chromosome 18. These observations support our hypothesis that PI 561350B carries both the Dt2 and dt3 alleles for semideterminacy.

Linkage map of the dt3 locus

To further map the location of the dt3 locus, the 153 individual plants of the (Williams 82 × Krasnodar 391–89) F2 population were genotyped (Table S8) using four InDel markers polymorphic between the two parent lines within the region defined by BSA (Fig. 3). F2:3 families of 10–15 plants derived from the F2 individuals were used to obtain more accurate phenotypes than could be obtained from individual F2 plants. Linkage analysis integrating the genotypic and phenotypic data placed Dt3 in the ~ 195 kb region between two Indel markers at 45.111 and 45.306 Mb of chromosome 10 according to the Williams 82 reference genome (Fig. 4). Dt3 is situated 3.13 centiMorgans (cM) from InDel 45.111 and 2.39 cM from InDel 45.306 (Fig. 4).

Fig. 4
figure 4

Genetic and physical maps of the Dt3 region. a Genetic map of the Dt3 region constructed based on linkage analysis. InDel markers are listed on the left side and the genetic distance (centimorgan, cM) between adjacent markers are shown on the right side of the map. b Physical positions of the molecular markers on chromosome 10 according to the soybean reference genome (Wm82.a2)

Full size image

Candidate genes for Dt3 in the mapped region of Williams 82

There are 21 annotated genes within the defined ~ 195 kb Dt3 region of Williams 82, some of which are known to be involved in flowering or related pathways. These include Glyma.10G219500, whose best matched homolog in A. thaliana encodes CLEAVAGE STIMULATION FACTOR 64 (CSTF64) and facilitates the silencing of the floral suppressor FLOWERING LOCUS C (FLC) (Liu et al. 2010). Another gene in this region, Glyma.10G221200, whose best-matched homolog in A. thaliana encodes a Choline/ethanolamine kinase, participates in the biosynthesis of phosphatidylethanolamine and phosphatidylcholine. Modifications to the biosynthesis of phosphatidylcholine have been implicated in flowering time changes, in part because the floral integrator FT can bind to them (Nakamura et al. 2014; Susila et al. 2021). Additionally, Glyma.10G221500, which sits at the edge of the mapped region, is the maturity locus E2 and encodes a GIGANTEA-like protein (Watanabe et al. 2011) although variation in this gene has not previously been associated with changes to stem growth habit.

Discussion

We identified a novel gene, dt3, specifying semideterminate growth habit in four investigated soybean varieties which do not carry Dt2 or dt1. The inheritance pattern in this study suggests that Dt3 is necessary but not sufficient to maintain indeterminate stem growth in soybean. If this is indeed the case, we can assume that the dt1 locus is recessively epistatic to the Dt3 locus, so that the genotypes dt1dt1;dt2dt2;dt3dt3 and dt1dt1;dt2dt2;Dt3Dt3 would each be determinate. However, as we are not aware of any soybean lines which possess these two genotypes, this will need to be experimentally validated through controlled crosses. The BSA result of the cross between Williams 82 and PI 561350B indicates that the semideterminate alleles of dt3 and Dt2 have synergistic effects, resulting in a more strongly determinate stem growth habit than either of these mutations alone. The combined effects of the dt3 and Dt2 alleles however still produce a terminal stem growth habit which is less extreme than the determinate growth conferred by the dt1 loss-of-function mutations and should still be classified as semideterminate (Table 2).

Table 2 Soybean stem growth habit genotypes and phenotypes (Genetic basis underlying soybean stem growth habit)
Full size table

Semideterminate growth habits have been described in several plant species, including tomato (Solanum lycopersicum L.) and two leguminous crops closely related to soybean, pigeon pea (Cajanus cajan Millsp.) and chickpea (Cicer arietinum L.) (Ambika et al. 2021; Elkind et al. 1991; Kapoor and Gupta 1991). In each of these species, two genes have been reported which modulate stem growth habit. As with soybean Dt1, the first of these genes possesses a dominant functional allele that maintains indeterminate stem growth, known as CcDt1, CaDt1, and SELF PRUNING (SP) in pigeon pea, chickpea, and tomato, respectively (Hegde 2011; Pnueli et al. 1998; Saxena et al. 2017). However, in contrast to soybean Dt3 and dt3, which modulate indeterminate and semideterminate stem growth, respectively, in Dt1/Dt1 backgrounds, the second stem-termination genes described in tomato, chickpea, and pigeon pea specify semideterminate or determinate growth in the recessive backgrounds of their respective Dt1 orthologs. In chickpea and pigeon pea, the semideterminate allele of the CcDt2/CaDt2 locus is dominant over the determinate allele, while in tomato, the determinate allele of Sdt is dominant over the semideterminate allele sdt (Elkind et al. 1991; Harshavardhana et al. 2019; Kapoor and Gupta 1991). The inheritance pattern of semideterminacy conferred by the recessive dt3 allele is thus distinct not only from that modulated by the soybean Dt2 locus but also from epistatic relationships between stem growth habit genes in other plant species.

Soybean stem growth habit is revealed to be an increasingly complicated trait, with mutations in at least three genes (dt1, Dt2, and dt3) resulting in stem termination. Our initial screening to remove the dt1 allele suggests that a substantial proportion of those accessions classified as semideterminate actually carry loss-of-function alleles of dt1. These phenotypic classifications may result from the dt1-t alleles which confer determinate growth without reducing height (Thompson et al. 1997), or from environmentally influenced phenotypic variation. Additionally, the semideterminate classification of these lines may result from quantitative genetic variation among yet-to-be-characterized genes underlying stem growth habit. When we removed dt1, Dt2 could be identified by GWAS. However, neither the Dt3/dt3 locus nor any other novel locus was identified by GWAS with or without the filtering out of the semideterminate Dt2 allele, suggesting that dt3 is a rare allele.

There has long been an interest in modifying soybean stem growth habit to increase yield, although past studies have shown mixed results based upon environment, genetic background, and management conditions (Beaver et al. 1985; Chang et al. 1982; Cober and Tanner 1995; Foley et al. 1986; Schug et al. 2022). Zhang et al. (2019) reported chromatin immunoprecipitation sequencing (ChIP-Seq) which showed that in addition to conferring semideterminate growth habits, Dt2 also regulates genetic pathways associated with a range of traits, particularly as a negative regulator of tolerance to drought stress. The recessive dt3 allele may therefore prove to be superior as a source of semideterminate growth habits or as a target for modification of this trait, as it may have less negative pleiotropic effects than the Dt2 allele while conferring the same level of stem termination. As a recessive mutation, dt3 is amenable to CRISPR-Cas9-facilitated gene editing knockouts that could easily be used to switch various soybean cultivars from indeterminate to semideterminate. Further work is needed to pinpoint the candidate gene(s) for the dt3 locus, understand its molecular role in modulating semideterminacy, and evaluate its effect on soybean yields toward the goal of producing optimized soybean shoot architecture.