Molecular atlas of soybean flowers identifies candidate abscission and reproductive pathways associated with flower abortion

Abstract

Background

Soybean exhibits great variance of flower abortion, with 20–80% of flowers aborting before pod set, posing a major constraint to yield improvement. Although physiological factors contributing to flower drop are well recognized, the underlying genetic and molecular mechanisms governing abscission and floral retention remain poorly understood. To elucidate the molecular determinants of flower abortion, we performed a comparative transcriptomic analysis of high and low abortion soybean genotypes across four reproductive stages—bud, closed flower, open flower, dry flower, and the leaf as vegetative tissue for reference.

Results

Differential expression and functional enrichment analyses revealed distinct transcriptional landscapes between high and low abortion plants, particularly in genes associated with hormonal signaling, abscission regulation, cell wall remodelling, and reproductive failure. The high abortion plants showed an elevated expression of ethylene responsive and abscission-associated genes, including Inflorescence Deficient in Abscission (IDA), Inflorescence Deficient in Abscission-Like (IDL), Homeobox Protein Knotted-1-Like 6 (KNAT6) and Arabidopsis thaliana Homeobox Gene 1 (ATH1/BEL-1), along with increased activity of pectin-degrading proteins. In contrast, low abortion plants exhibited upregulation of Mitogen-Activated Protein Kinase Phosphatase 1 (MKP1) and KNAT1, consistent with repression of MAPK activity and maintenance of boundary integrity.

Conclusion

Our findings identify distinct transcriptional programs associated with contrasting flower abortion phenotype and highlight coordinated changes in hormone signaling, cell wall remodeling, and boundary specification pathways.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12864-026-13230-8.

Background

Soybean (Glycine max L. Merr.) is a globally important crop, providing essential protein and oil for human consumption, animal feed, and industrial purposes [1]. Despite steady genetic and agronomic improvements of soybean, advances based on genetic studies is insufficient to meet the demands of growing populations and erratic environmental conditions [2]. Among the primary constraints to improved soybean productivity, flower abortion (premature flower loss before pod set) remains one of the most critical and persistent challenges, due to its substantial impact on final yield and the complexity of factors influencing floral retention. Although soybean is not inherently reproductively inefficient, it exhibits considerable variation in the proportion of flowers that develop into pods. While each plant is capable of producing hundreds of floral buds, it typically aborts a large fraction of these during the reproductive phase. In most environments and genotypes, flower abortion rates range broadly from 20 to 80% [3–6]. This extensive genotypic and environmental variation reveals untapped potential for improving flower retention and thus overall productivity. Harnessing the genetic diversity underlying flower abortion rates offers a strategic avenue to increase soybean yield by translating more of the plant’s reproductive potential into realized grain production. However, it is important to note that there are very few studies on this topic which date back to more than four or five decades. Comparatively, little research has been conducted since then, highlighting a significant knowledge gap that needs to be addressed to fully exploit this potential.

Abortion may be due to developmental failure of a flower, ovule or embryo, whereas abscission is the physical shedding of the flower at an abscission zone (AZ). While abortion can precede and promote abscission, it might not be an invariable consequence as abscission can occur even in pollinated and fertilized flowers [7]. Physiological studies have established that soybean exhibits consistent abortion of a large proportion of flowers and young pods even under favorable growing conditions in soybean [8]. Environmental stresses during flowering, such as high temperatures, water deficit, shading, and photoperiod sensitivity strongly exacerbate abortion and abscission [9–14]. Historical reports have demonstrated that the frequency of abortion is highest just after fertilization, during early embryo or proembryo (mass of cells that have not yet differentiated) development [3, 15]. Moreover, it has been reported that first formed flowers (flowers initiated at the lower nodes during the initial stages of flowering) experience around 40% abortion, whereas late flowers (arising at upper nodes during later stages) exhibit up to 80% abortion, with immature pods following the same trends [16]. Overall, these studies suggest that abortion is not a random process but is influenced by developmental stage, inflorescence (raceme) position, and canopy structure in addition to environmental factors.

Flower abortion acts as an internal checkpoint in soybeans, allowing the plant to conserve energy by selectively aborting flowers before investing the substantial resources required for full pod and seed development. This mechanism helps balance reproductive success with overall plant fitness, especially under resource-limited or stressful conditions. Currently, two hypotheses have been proposed to explain this phenomenon. The first is the assimilate limitation hypothesis, which presumes that abortion occurs when reproductive demand exceeds assimilate supply [5]. The idea was supported by the observation that the pod setting increases under elevated photosynthesis. Conversely, defoliation increases abortion, while depodding decreases it, consistent with the source-sink competition. However, detailed analyses of carbohydrate and nitrogen levels in soybean flowers and immature pods challenged the notion that assimilate concentrations alone determine abortion. Heitholt and coworkers [16] reported that aborted and retained flowers often had comparable soluble sugar and starch concentrations, suggesting that carbon availability is not the sole driver of abortion. The second hypothesis emphasized on intrinsic signaling mechanisms, including hormonal control and inter-organ competition [17]. To support this hypothesis, studies have demonstrated that established pods can promote abortion of adjacent flowers through hormonal cues [18]. This perspective suggests that abortion is an active regulatory process rather than a passive consequence of resource scarcity.

In soybean, quantitative trait loci (QTL) mapping has demonstrated that flower abortion in soybean is a complex trait controlled by multiple loci [19, 20]. Tischner and coworkers [21] identified QTLs regulating the flower abortion, which co-localized with loci for flowering time, reproductive stage, and water-use efficiency, indicating pleiotropic regulation. However, despite these early efforts, mapping studies have not yet pinpointed the underlying major QTL or causal genes, leaving the molecular basis of flower abortion unresolved. Additionally, even with the documented genetic variation, the underlying regulators remain largely unknown in soybean.

The phenomenon of superfluous flowering followed by colossal abortion is not unique to soybean but is widespread across angiosperms, particularly in woody perennials [22]. Olive (Olea europaea) trees, for example, may initiate up to half a million flowers per tree during an “on” year, yet only 1–2% mature into fruit [22]. Similar patterns of pistil or ovule abortion have also been documented in apricot, litchi, and peony [23–26]. These studies emphasize that abortion is not simply an agronomic anomaly, but an evolved reproductive strategy shaped by ecological and evolutionary pressures. Hypotheses predict that selective abortion to favor higher-quality seeds, bet-hedging to cope with variable environments, and surplus flower production to enhance male fitness through pollen dispersal [22].

Despite extensive agronomic, physiological, and genetic studies, the molecular basis of flower abortion in soybean remains poorly characterized. Unlike fruit crops such as apricot and litchi, where transcriptomic datasets have provided candidate regulators [23–25], soybean abortion has not been systematically dissected using transcriptome-wide approaches. Most prior soybean studies have focused on physiological correlations with carbohydrate concentrations or source–sink manipulations, leaving a critical gap in linking gene expression networks with the observed patterns of abortion. To bridge this gap, we performed a comprehensive transcriptome profiling of selected soybean genotypes contrasting in flower abortion rates across successive floral developmental stages. By integrating differential expression analysis, functional enrichment, and candidate gene identification, our aim was to unravel the molecular regulators associated with abortion in soybean. Specifically, we sought to (i) identify transcription factors, hormone signaling components, and metabolic pathways associated with flower abortion; (ii) determine stage-specific patterns of gene expression linked to abortion events; and (iii) provide new candidate genes and pathways for breeding strategies to enhance pod set and yield stability.

Material and methods

Plant material and flower sampling

Flower abortion percentage in diverse soybean genotypes was evaluated in 2023 (Texas Tech University Quaker Farm, Lubbock 33° 41′ 36.4596″ N, 101° 54′ 18.612″ W) and 2024 (New Deal Research Farm, New Deal 33° 44′ 13.76″ N, 101° 43′ 58.04″ W) [27]. Flower count data from these two seasons (2023–206 genotypes and 2024 – a representative subset of 48 genotypes) were combined to identify two contrasting soybean genotypes with consistently high (HA) and low (LA) flower abortion rates. Flower abortion was calculated as:

The two selected genotypes PI506862 (cultivar originally from Japan, low abortion, LA) and PI563678 (Breeding material originally from Liaoning, China, high abortion, HA) were used for gene expression analysis. Each genotype was planted in 4 rows of 2.5 m length with a spacing of 76.2 cm between rows and a planting density of approximately 26 seeds per meter. For transcriptomic analysis, sampling was conducted (in 2024 – year II) for four critical developmental stages: (i) bud, (ii) closed flowers, (iii) open flowers, and (iv) dry flowers (wilted petals post-anthesis), and leaves for gene expression comparison. For each stage, flowers were collected from three tagged plants per replicate and pooled to represent one biological replicate stage. A total of 4 biological replicates were sampled, resulting in 12 plants sampled per stage per genotype across replicates. The same tagged plants were resampled across all four stages. Sample collection was done over an approximately 5-day interval, at about same time of the day (9:00am) during the cooler morning hours to minimize variation associated with increasing daytime temperatures. Immediately after excision, samples were snap-frozen in liquid nitrogen at the field site, transported to the laboratory, and stored at –80 °C until RNA extraction. Floral tissues from the same contrasting genotypes were independently harvested in 2025 (year III) under comparable field conditions for RT-qPCR validation.

RNA extraction, library preparation and transcriptome sequencing

Approximately 100 mg of floral and leaf tissue from the samples was used for RNA isolation with the RNAeasy® Plant Mini Kit (Qiagen, Hilden, Germany) following the manufacturer’s protocol. The contaminating DNA was removed using RNase-Free DNase I enzyme (Qiagen, Hilden, Germany) treatment as per the manufacturer’s instructions. RNA quality was assessed using a Nanodrop (Multiskan SkyHigh ThermoScientific), and RNA integrity was verified by agarose gel electrophoresis. High-quality RNA was used to construct transcriptome libraries with the Illumina TruSeq RNA Sample Preparation Kit. Libraries were sequenced on the Illumina HiSeq 2500 platform to generate 150 bp paired-end reads.

Raw reads were subjected to quality assessment using Fastp v0.23.2 with the “-detect adapter for pe”, “-trim poly x”, and “trim poly g” parameters to remove adapters and filter removal of low—quality reads [28]. The resulting clean reads were aligned to the soybean reference genome (Glycine max Wm82.a4.v1) downloaded from https://phytozome-next.jgi.doe.gov, using kallisto v0.48 [29]. Gene level read counting was subsequently performed using feature Counts v2.0.2 [30] to generate expression profiles for the downstream analyses.

Identification of differentially expressed genes (DEGs) and functional enrichment analysis

Differential expression analysis was conducted using the DESeq2 package (v1.42.1) [31] in R v4.0.0, with read counts from kallisto abundance files as input. For each genotype, expression levels in different floral stages were compared to the leaf tissue. Genes with a log2 fold change ≥ 2 and adjusted p-value ≤ 0.05 were considered significantly upregulated, while those with log2 fold change ≤ –2 and adjusted p-value ≤ 0.05 were considered significantly downregulated (Supplementary Fig. 1). The stringent threshold was chosen to emphasize robust transcriptomic changes. Subtle but biologically important regulators may have been excluded. Gene Ontology (GO) and pathway enrichment analyses were performed using ShinyGO (http://bioinformatics.sdstate.edu/go/) to identify significantly enriched biological processes and pathways. To further explore expression dynamics, K-means clustering was performed in R based on normalized expression values, and the resulting clusters were functionally annotated using GO and KEGG categories derived from ShinyGO.

Plant transcription factor (TF) prediction

To investigate the transcriptional regulation underlying flower abortion in soybean, we performed a comprehensive TF prediction analysis. DEGs identified in our dataset were first filtered to extract TFs, which were then categorized into respective TF families using the Plant Transcription Factor Database v5 (PlantTFDB; http://planttfdb.gao-lab.org/). Expression profiles of these TFs were subjected to K-means clustering (Euclidean distance) in R to classify them based on co-expression patterns.

Validation of DEGs by reverse transcription-quantitative polymerase chain reaction (RT-qPCR)

DNase I treated total RNA (1 µg) was used to synthesize first-strand cDNA using Verso cDNA Synthesis Kit (Thermo Fisher Scientific, USA) following the manufacturer’s instructions. Quantitative real-time PCR was performed on a CFX96 Real-Time System (Bio-Rad, Hercules, CA, USA) using Kapa SYBR® FAST (Roche, Darmstadt, Germany) in a 10 μl reaction volume. The qRT-PCR conditions included an initial denaturation at 95 °C for 3 min, followed by 40 cycles of 95 °C for 15 s and 60 °C for 30 s. A melt curve analysis was performed to confirm the specificity of amplification. Relative transcript abundance was calculated with the 2^−∆∆Ct method [32] by normalizing with soybean reference gene Actin (Glyma.18G290800). To independently validate RNA-seq derived expression patterns, RT-qPCR was performed using RNA isolated from floral tissues collected from year III. Three biological replicates with two technical replicates each were analyzed. Statistical significance was determined using Welch’s t-test, with p < 0.05 considered significant. Primer sequences and target gene details are provided in supplementary Table 1.

Results

Comparative phenotypes and global expression profiles of contrasting soybean genotypes

Two contrasting soybean genotypes, PI506862 (low abortion) and PI563678 (high abortion), were selected from a two-year field evaluation for phenotypic (Fig. 1) and transcriptomic (Fig. 2) analyses to elucidate mechanisms of flower abortion. Quantification of flower abortion across two consecutive growing seasons (2023 and 2024) confirmed differences in flower abortion rates, with the LA genotype maintaining consistently significant lower abortion rates (20–35%) compared to the high abortion rates (60–70%) observed in the HA genotype (Fig. 1D). Importantly, the abortion did not vary significantly across years, indicating that the observed differences are strongly genotype-dependent rather than environmentally driven.

Fig. 1.

Phenotypic variation between contrasting soybean genotypes differing in flower abortion rates. A Mature seeds from low-abortion (PI506862, left) and high-abortion (PI563678, right) genotypes. B Field images showing flowers in PI506862 (left) and PI563678 (right). C Developmental stages of soybean flowers from bud to dry flowers in PI506862 (top row) and PI563678 (bottom row). D Percentage of flower abortion in the two genotypes across two growing seasons (2023 and 2024). Bars represent mean ± SE (n = 5). Different letters above bars indicate statistically significant differences at p < 0.05 based on one-way ANOVA followed by Tukey’s HSD test

Fig. 2.

Principal component analysis (PCA) and differentially expressed genes (DEGs) across high- and low-abortion soybean genotypes. A PCA of HA genotype samples only, illustrating distinct clustering of all replicates in bud, closed flower, open flower, dry flower, and leaf tissues. B PCA of LA genotype samples only, showing distinct separation among floral stages and leaf tissue. C Stacked bar plot showing the number of DEGs across tissues and stages in HA and LA genotypes. Bars represent the total number of DEGs (grey), with up-regulated genes in red and down-regulated genes in blue

To establish a transcriptomic framework for understanding these differences, soybean flowers at distinct developmental stages (Fig. 1C) were carefully staged and harvested along with leaf tissue for RNA-seq analysis. After the filtering of low-quality reads and contaminating adapters, an average of 96.67% reads were retained with approximately 90.97% mapping efficiently (Supplementary Table 2) onto the soybean reference genome dataset (Wm82.a4. v1). Principal component analysis (PCA) of the transcriptomic data revealed clear tissue specific clustering patterns (Supplementary Fig. 2) where the leaves of both genotypes formed a distinct cluster separate from the floral samples. Additionally, for both the HA and LA genotypes, the closed and open flowers grouped, together indicating an overlapping expression landscape (Fig. 2A-B). Across all samples, PC1 captured the major variance attributable to tissue type and floral developmental stage, whereas PC2 highlighted differences between genotypes. Interestingly, the PCA comparison of corresponding flower developmental stages between LA and HA revealed distinct genotype-specific clustering patterns (Supplementary Fig. 3). At each stage, samples from the two genotypes formed clearly separated groups, indicating that differences in gene expression were maintained consistently throughout floral development. Differential expression analysis further supported these trends, with thousands of genes identified as differentially expressed across developmental stages in both genotypes (Fig. 2C). Both the total number as well as upregulated DEGs were consistently higher in the HA genotype compared to LA, underscoring the heightened molecular perturbations associated with flower development.

Enriched gene ontology processes in HA and LA flower development stages

To further explore the biological processes underlying the contrasting flower abortion phenotypes, we conducted GO enrichment analysis of DEGs in all stages for both genotypes (Supplementary Fig. 4) and identified a core set of common DEGs across all samples (Fig. 3). In both the genotypes, the DEGs associated with photosynthetic complex, protein folding, response to oxidative stress, regulation of DNA transcription, carbohydrate metabolic process, microtubule based movement, and cell wall modification were recurrently enriched in all stages of flower development, underscoring their fundamental role in flower physiology. Importantly, three sugar efflux transporter genes, Glyma.02G088400, Glyma.08G009900 and Glyma.08G183500 were differentially expressed in all stages for both genotypes (Supplementary Table 3). Cell-wall remodeling genes, particularly pectin methylesterases (PMEs), pectin methylesterase inhibitors (PMEIs), and pectin lyases, were strongly represented among DEGs and showed contrasting abundance between HA and LA flowers. Additional biological processes, although not consistently detected at every stage included, embryo development, response to auxin, protein phosphorylation, transmembrane and ion transport, highlighting stage specific exclusive regulatory pathways. Notably, the GO terms “vitamin B6 biosynthetic process” and “response to heat” were only identified in the HA genotype. By contrast, the LA genotype uniquely showed enrichment for GO term sulfate transmembrane transport (Supplementary Fig. 4D).

Fig. 3.

Comparative analysis of DEGs, clustering and functional enrichment. A UpSet plot showing the intersection of DEGs identified across different floral stages (bud, closed flower, open flower, and dry flower) of HA and LA genotypes. The bar graph indicates the number of unique and shared DEGs across stage comparisons. B K-means clustering of DEGs common across all developmental stages. Heatmap shows scaled expression (Z-scores) of clustered DEGs, with hierarchical grouping highlighting distinct expression profiles across samples. Red indicates up-regulation and blue indicates down-regulation relative to mean expression. C GO enrichment analysis of cluster-specific DEGs. The y-axis represents enriched biological processes, while bubble size corresponds to the proportion of genes annotated to each process (GeneRatio), and bubble color indicates significance level (FDR)

To gain deeper insight into the genes and pathways associated with floral abortion, we analyzed the overlap of DEGs shared across all floral stages. The upset plot analysis revealed that 3,783 DEGs were common across all developmental stages in both HA and LA genotypes, representing a core molecular signature underlying flower development (Fig. 3A, Supplementary Table 3, Supplementary Table 4). Additional intersections showed smaller subsets of DEGs shared among two or three stages highlighting a stage or genotype specific transcriptional signature. The 496 common DEGs in HA and LA buds included clusters associated with regulation of metabolic and biosynthetic processes, protein phosphorylation, pollen wall assembly, pollen exine formation, and morphogenesis-related cellular component assembly (Supplementary Fig. 5A-B). Functional enrichment of the HA and LA dry flower common genes (487) revealed auxin-responsive signaling modules together with indole/glucosinolate and S-glycoside metabolic pathways, as well as genes involved in cell-wall modification (Supplementary Fig. 5C-D).

To explore the transcriptional dynamics of these common DEGs, we applied k-means clustering which grouped the genes into distinct expression modules across floral stages in both genotypes (Fig. 3B).

Cluster 1 was categorized by genes showing higher expression during early floral stages, gradually declining towards the dry flower stage in both HA and LA genotypes, suggesting their broad role in early flower development. Synonymously, it included genes in the photosynthetic-related pathways such as the electron transport, light harvesting, and photosystem II assembly, as well as energy metabolism processes like ATP synthesis, glycolysis, and oxidation–reduction reactions (Fig. 3C). Cluster 2 contained genes with stage specific upregulation during closed and open petal stages pointing toward their involvement in petal opening and expansion. This included 25 cell wall modification genes of the pectin lyase, pectinesterase and pectinesterase inhibitors family. Cluster 3 displayed contrasting expression profiles between HA and LA genotypes; genes in this group were upregulated in HA but downregulated in LA. This striking divergence highlights genotype-specific transcriptional regulation of otherwise common DEGs and suggests that such differential regulation could underlie the contrasting flower abortion phenotypes. The cluster included genes belonging to auxin response factors (ARFs), Apetala2-like ethylene responsive TFs (AP2-ERFs), ERFs, BEL1-like homeodomain protein, WRKY family TFs, KNAT6, laccases and cytochrome P450 family proteins. Cluster 4 showed a gradual increase in expression across developmental stages, with this trend being more pronounced in HA, suggesting their potential involvement in progression of floral maturation. It included genes involved in transmembrane transport and developmental processes. Cluster 5 exhibited contrasting expression between genotypes specifically at the bud and dry flower stages, suggesting that regulatory differences at the onset and termination of floral development may be critical in shaping abortion outcomes. This group included several 30S and 50S ribosomal proteins, KNAT1, NAC domain protein, GATA and MADS box TFs. Cluster 6 comprised genes with high expression restricted to the bud stage, pointing to their exclusive role in early floral initiation and commitment. It included genes enriched in various biosynthetic processes, including methionine, spermine, and glutamine biosynthesis, alongside photosynthetic and translation related genes were also present.

Stage-specific TF families associated with flower development

To identify key regulatory components underlying the transcriptional differences between HA and LA genotypes, we examined the expression patterns of TF families across flower developmental stages (Fig. 4). A broad spectrum of TF families were represented among the individual DEG clusters, reflecting their specialization. Among these, MYB, NAC, bHLH, WRKY, ERF, and bZIP families were highly enriched, consistent with their established roles in reproductive development, stress responses, and hormone signaling. ERF and ARF family members were among the most prominent transcription factors displaying genotype-dependant expression.

Fig. 4.

K-means clustering of differentially expressed transcription factors (TFs). A Differentially expressed TFs identified from the DEG dataset clustered into four groups based on their expression profiles using K-means clustering. The thick black line represents the consensus expression trend for each cluster, while colored lines indicate individual TF expression profiles. B The right panel shows the distribution of TF families within each cluster, with bar heights indicating the number of TFs belonging to each family

Cluster 1 contained TFs with low to negligible expression in the LA genotype but markedly higher expression in HA stages (except bud). These included members of the ERF, WRKY, bHLH, bZIP, and MYB-related families. Cluster 2 was enriched in bHLH, MYB and MADS TFs, which were differentially expressed across both genotypes but peaked during petal closure and open flower stages. Cluster 3 comprised TFs that were expressed almost only in the buds of both genotypes. Interestingly, cluster 4 had TFs with contrasting expression in both genotypes across all stages. Their differential activation suggests a central role in driving the divergent floral abortion phenotypes. Together, this analysis underscores that while both genotypes recruit a diverse repertoire of TFs, the preferential activation of cluster 1 TFs in HA and the contrasting regulation of cluster 4 TFs between HA and LA likely represent key regulatory checkpoints contributing to the phenotypic divergence in flower abortion.

qRT-PCR validation highlights contrasting stage-specific expression of candidate genes in HA and LA genotypes

To validate the RNA-seq results, we selected a panel of genes that were either expressed across all floral stages, present in cluster 3 (Fig. 3B, having contrasting expression pattern in LA and HA flowers) or had been previously implicated in abscission processes in Arabidopsis (Table 1). Genes related to abscission signaling (IDA/IDL), boundary specification (KNAT/ATH1), and MAPK regulation (MKP1) expression profiled in HA and LA genotypes using qRT-PCR (Fig. 5), revealing clear genotype and stage-specific transcriptional differences. Several genes such as Glyma.17G23000, Glyma.01G028300, Glyma.18G141200, Glyma.10G205400, Glyma.06G1500 and Glyma.10G226800 showed consistently significant (p < 0.05) higher expression in HA genotype across all stages, suggesting their involvement in promoting the high-abortion phenotype. In contrast, only Glyma.04G049500 displayed consistently significant (p < 0.05) higher expression in LA, particularly at close and open flower stages, indicating its role as a negative regulator for flower abortion. Interestingly, three genes namely, Glyma.10G205400, Glyma.06G151100 and Glyma.10G226800 showed a contrasting expression pattern (upregulation in HA and downregulation in LA) pointing to their direct role in abscission processes. Collectively, these results demonstrate that while both genotypes activate overlapping sets of genes, their temporal expression patterns diverge sharply, with HA showing preferential upregulation of several abscission related transcriptionally active genes, potentially driving its higher flower abortion phenotype.

Table 1.

List shortlisted candidate genes from the RNA-Seq data

Soybean Gene IDa	Descriptionb	Arabidopsis Ortholog	Function	Reference
Glyma.17G230000	BEL1-like homeodomain protein 3-like isoform X2	AT4G32980 (ATH1)	ATH1 with KNAT6 and KNAT2 control AZ structure. In the ath1 knat2	[33]
Glyma.06G051600	BEL1-like homeodomain protein 1-like isoform X3
Glyma.01G028300	homeobox protein knotted-1-like 6-like	AT1G23380 (KNAT6)	A double mutant of KNAT6 and KNAT2 disrupts AZ structure and organ shedding	[33, 34]
Glyma.02G036900
Glyma.08G284300
Glyma.18G141200
Glyma.14G091200	homeobox protein knotted-1-like 2-like	AT4G08150 (KNAT1)	KNAT1 controls inflorescence growth by restricting KNAT6 and KNAT2 expression to boundary regions and, likewise, prevents premature abscission	[35, 36]
Glyma.17G232205	homeobox protein knotted-1-like 2-like
Glyma.09G172851	homeobox protein knotted-1-like 7-like isoform X2
Glyma.04G049500	homeobox protein knotted-1-like 2-like
Glyma.05G174000	protein-tyrosine-phosphatase MKP1-like isoform X4	AT3G55270 (MKP1)	MKP1 acts to dissipate spurious phosphorylation that leads to premature shedding of floral organs	[37, 38]
Glyma.08G131200	protein-tyrosine-phosphatase MKP1-like isoform X3
Glyma.10G205400	protein IDA-like	AT1G68765 (IDA)	Overexpression of IDA, IDL2, IDL3, IDL4 and IDL5 induces early abscission. A quadruple mutant (ida idl1 idl2 idl3) retains its floral organs more strongly than the ida mutant	[39, 40]
Glyma.02G131300	protein IDA-like	AT3G25655 (IDL1)
Glyma.01G090400	protein IDA-like
Glyma.06G151100	protein IDA-LIKE 2-like	AT3G18715 (IDL4)
Glyma.10G226800	protein IDA-LIKE 2-like	AT1G76952 (IDL5)

^aChromosomal locus of the genes defined by SoyBase (https://www.soybase.org/)

^bDescription of the gene ID

Fig. 5.

qRT-PCR validation of selected differentially expressed genes identified from RNA-seq analysis. Relative expression levels of representative candidate genes were quantified across floral stages in soybean genotypes with high and low abortion rates. Expression values are shown as log₂ fold change (log₂FC) normalized to the internal control. Error bars represent the mean ± SE of three biological replicates. Statistical significance was determined using Welch’s t-test (p ≤ 0.05). Levels of significance are indicated as: p ≤ 0.05 (*), p ≤ 0.01 (**) and not significant (ns)

Discussion

Reproductive development is one of the most critical determinants of yield in flowering plants. Largely, in angiosperms, the number of flowers initiated during the reproductive phase greatly exceeds the number of fruits ultimately harvested, indicating that flower abortion and flower drop are pervasive biological processes. In soybean which is a major source of protein and oil worldwide, this phenomenon is particularly pronounced. Despite prolific flowering, 20–80% of flowers and young pods typically abort/abscise before reaching maturity. Therefore, understanding the underlying causes and regulatory mechanisms of flower abortion has important implications for crop improvement and global food security. In our pursuit to deep dive into the biological mechanisms spanning flower drop in soybean, we conducted a stage-wise transcriptome analysis of flowers from genotypes contrasting in their floral abortion pattern.

Cell wall remodeling genes exhibit consistent expression across floral developmental stages in both soybean genotypes

Cell-wall remodeling is a plausible driver of flower drop because abscission requires controlled loosening of middle lamellae and disruption of cell–cell adhesion [41]. In our dataset, seven pectin-centric cell wall categories encompassing 71 genes were represented across developmental stages in both genotypes. The most enriched class was the pectin PMEIs followed by pectin lyases, PMEs, VANGUARD1 homolog 2 and methylesterase PCR A (Supplementary Fig. 6). We focused on these genes because pectin chemistry is a primary lever for cell–cell adhesion in the middle lamella, the interface that is enzymatically loosened in abscission [42]. The PMEs demethyl-esterify homogalacturonan (HG), exposing Ca²⁺ crosslinked blocks which create attack sites for pectin lyases and polygalacturonases to loosen the cell wall [43]. In parallel, pectin lyase-like genes execute β-elimination cleavage of de-esterified HG and consistently appear among the cell-wall hydrolases upregulated during abscission programs [44]. In our study, a higher PME and pectin lyase expression was observed in the HA genotype which might be a key contributor to its higher propensity for flower abortion. The PMEIs proteins physically block PME active sites and tune PME activity spatiotemporally [45]. The balance between PME and PMEI sets wall mechanics and thereby organ fate. We found a total of 47 PMEIs to be differentially expressed in our study, with peak activities at the closed and open flower stages, potentially stabilizing pectin methyl-esterification before and after anthesis, reducing the propensity for premature lamella loosening. As suggested by their function, these genes showed higher expression in the LA genotype as compared to HA flowers, potentially serving as a stopgap for premature abscission. The Arabidopsis VANGUARD1 encodes a PME that enhances pollen tube growth in the style [46]. Consistently, in our study, this gene (Glyma.02G008300) showed higher expression in the LA bud, close and open flowers suggesting a higher fertilisation rate in the low abortion genotype. However, further studies are needed to validate this theory.

Hormonal and transcriptional activation is associated with increased flower abortion in HA genotype

In our clustering analysis of common genes, cluster 3 (Fig. 4B) was the module in which HA consistently showed higher expression than LA. This was our cue to critically examine the genes that constituted this cluster. The repertoire included six ethylene responsive (ERFs) TFs, Glyma.05G214400, Glyma.08G020900, Glyma.15G077000, Glyma.17G114500, Glyma.20G155200 and the Apetala2-like ERF Glyma.07G038200 which indicated a hyperactivation of an ERF centric abscission program in HA. ERFs are classically known to bind to GCC-box cis elements to regulate downstream transcriptional responses [47, 48]. In tomato, SlERF52 knock-down delays flower abscission and reduces cellulase and polygalacturonase accumulation, demonstrating that ERF activity is positively required for pedicel abscission [49]. Similarly, RhERF1/4 in rose integrate ethylene-auxin crosstalk to modulate pectin-degrading activity during petal drop, reinforcing the idea that ERFs sit at the hub of hormone-guided AZ activation [50]. Additional evidence from litchi shows that LcERF10 binds to an AZ Na⁺/H⁺ antiporter promoter, altering cellular pH and positively regulating fruitlet abscission [51]. Notably, a recent study in rice provides evidence that an AP2/ERF, OsSHAT2 regulates seed shattering an abscission analogous process. Loss of function Osshat2 mutants generated via CRISPR-Cas9 exhibited reduced seed shattering and incomplete abscission layer formation [52]. Given this functional parallel, the up-regulation of the AP2/ERF (Glyma.07G038200) in HA flowers likely represents a similar regulatory mechanism. Taken together, the enhanced up-regulation of ERFs in HA is consistent with a pro-abscission transcriptional state offering a tenable molecular explanation for stronger abscission tendencies in the HA genotype.

Alongside the ERFs, cluster 3 (Fig. 3B) also contained two auxin response factors (ARFs), Glyma.13G354100 and Glyma.14G217700, both showing higher expression in the HA genotype. ARFs act as transcriptional regulators of auxin-inducible genes, binding to auxin response elements (AuxREs) within target promoters, orchestrating developmental and hormonal responses [53]. Their elevated expression in HA tissue suggests a transcriptional reconfiguration of the auxin signaling cascade within the AZ. In Arabidopsis, AtARF1 and AtARF2 have been shown to regulate floral organ abscission and senescence, with double mutants exhibiting delayed organ shedding [54]. Similarly, in tomato, SlARF6 and SlARF8 were among the most highly expressed ARFs in the pedicel AZ, where their transcripts increased during ethylene-induced abscission and declined upon auxin depletion [55]. This dynamic reflects the hormonal antagonism between auxin and ethylene, wherein declining auxin levels heighten AZ sensitivity to ethylene signaling which accelerates abscission [56]. In the HA genotype, sustained up-regulation of ARFs may indicate a persistent auxin signaling feedback, maintaining transcriptional activity of auxin-responsive genes even as ethylene levels rise. Such concurrent activation of auxin and ethylene pathways has been proposed to prime AZ cells for coordinated cell-wall disassembly, integrating hormonal cues that promote abscission [55]. Therefore, the enhanced ARF expression observed in HA likely underpins a hormone integrative regulatory circuit, wherein auxin response factors interact with ethylene-responsive transcription factors to activate abscission-related genes, reinforcing the transcriptional predisposition of HA flowers toward premature detachment and abortion.

Coordinated regulation of abscission pathway signaling genes distinguish high and low abortion in soybean genotypes

The final phase of floral organ separation in Arabidopsis is governed by a conserved signaling network that integrates the IDA-HAE/HSL2-MAPK module with downstream cell wall modifying enzymes [38]. In our study, several orthologs representing this core abscission module and its regulators were differentially expressed between the HA and LA genotypes, suggesting their potential roles in modulating floral retention versus shedding (Fig. 6).

Fig. 6.

Proposed regulatory model underlying differential flower abortion in soybean genotypes. Schematic representation depicting the hormonal and genetic regulation associated with floral abscission. The left panel represents the low flower abortion (LA) genotype, while the right panel represents the high flower abortion (HA) genotype. Differential expression of hormone-related pathways, auxin (IAA), abscisic acid (ABA), jasmonic acid (JA), and ethylene (ET) and key regulators including MKP1, IDA, KNAT1, ATH1, and KNAT6 is illustrated with corresponding heatmap insets (red: high; blue: low). In the HA genotype, enhanced ethylene and abscission-related gene activity promotes floral organ separation and increased abortion, whereas the LA genotype maintains higher auxin signaling and suppressed ethylene responses, favoring lower flower abortion. Downstream activation of hydrolytic enzymes and cell wall remodeling genes mediates organ detachment at the abscission zone, contributing to the observed phenotypic differences between genotypes. Because endogenous hormone concentrations were not directly measured, this model should be interpreted as a hypothesis generating framework

The MAP kinase phosphatase 1 (MKP1) acts as a key negative regulator establishing a signaling threshold prior to activation of abscission cascade. Functionally, MKP1 dephosphorylates MPK3/6 thereby dissipating basal MAPK activity that might otherwise trigger premature cell wall hydrolysis and organ detachment [38]. In our dataset, one of the soybean orthologs (Glyma.08G131200) of AtMKP1 was upregulated and had comparatively higher transcripts in LA, which might be associated with reduced activation of abscission related responses, and enhanced floral retention (Fig. 5).

Inflorescence Deficient in Abscission (IDA) and its close homologs, the IDA-like (IDL) peptides bind to a receptor complex on the plasma membrane formed by the leucine-rich repeat receptor-like kinases (LRR-RLKs) HAESA (HAE) and HAESA-like2 (HSL2) [38]. They in turn heterodimerize with co-receptors from the somatic embryogenesis receptor-like kinase (SERK) family to form an active receptor complex. IDA binding activates the receptor complex which triggers a MAP kinase cascade, leading to the transcriptional induction of cell wall hydrolase genes [37, 39, 57–59]. In the current study, IDA and IDL genes were upregulated in the HA genotype, suggesting that these plants may exhibit stronger abscission-associated transcriptional regulation relative to LA (Fig. 5). The LA genotype, on the other hand, had lower IDA and IDL expression and higher levels of MKP1, a negative MAPK regulator, consistent with attenuated pathway activation. Even though HAE/HSL2 receptors were not differentially expressed in the floral tissues, the elevated IDA levels in HA may indicate stronger association with this signaling module. However, because transcriptomic profiling was performed on whole floral tissues rather than isolated AZ cells, these patterns should be interpreted as indicative signatures than direct evidence.

Beyond the MAPK cascade and IDA-HAE/HSL2 signaling core, abscission is tightly coordinated with the developmental patterning machinery that defines organ boundaries and tissue fate. In Arabidopsis, class I KNOTTED-like homeobox (KNOX) genes such as KNAT1/BREVIPEDICELLUS (BP) and KNAT6, together with the BELL-family transcription factor ATH1, constitute a transcriptional axis that regulates boundary specification and spatial restriction of abscission competence [34, 35, 60, 61]. In our soybean dataset, we observed a distinct expression pattern of these homeobox genes between the HA and LA genotypes. KNAT1 transcripts were elevated in the LA genotype (Fig. 6), aligning with its known role in maintaining tissue integrity and repressing cell separation. In contrast, KNAT6 and ATH1 genes were upregulated in the HA genotype, suggesting a transcriptional profile associated with altered boundary specification or abscission related developmental regulation. KNAT6 functions in meristem maintenance but is spatially restricted by KNAT1/PNY during reproductive development [35, 62]. Its expression promotes boundary expansion and cell separation, and the removal of KNAT6 rescues the knat1 mutant phenotypes (shortened internodes and downward pointing siliques) in Arabidopsis [35]. Moreover, KNAT6 acts downstream of the IDA-HAE/HSL2 cascade [34], linking peptide-mediated signaling to transcriptional activation of genes involved in wall hydrolysis and remodeling. Elevated KNAT6 expression in HA soybean flowers may thus enhance boundary competence and sensitize cells in the floral base to separation cues. Similarly, the ATH1 homeobox gene, which acts in concert with KNAT1 and KNAT6 to define floral organ boundaries [63, 64], was more strongly expressed in HA. High ATH1 expression in the HA genotype likely reinforces cell fate transitions within the abscission zone, supporting the upregulation of cell wall modifying genes observed in HA.

Taken together, these patterns indicate that in soybean, the differential regulation of KNAT1, KNAT6 and ATH1 establishes opposing transcriptional states that influence floral retention. The LA genotype maintains a KNAT1 like regime that preserves structural integrity and suppresses separation, while the HA genotype transitions toward a KNAT6/ATH1 dominated state that primes abscission-zone differentiation and accelerates the separation process. This interplay mirrors the antagonistic relationships characterized in tomato and Arabidopsis [34, 35, 65] and underscores a conserved developmental logic linking meristem-derived patterning genes to abscission signaling.

Reproduction associated and ovule development genes

Beyond abscission-related signaling, reproductive failure in soybeans as for most plants may also stem from intrinsic defects in stamen-pollen, pistil-ovule development and early gametophytic-embryo processes [66]. In our study, 8 ovule abortion (OVA) genes; Glyma.06G101300, Glyma.04G099400, Glyma.16G179800, Glyma.03G10900, Glyma.06G088000, Glyma.17G026600, Glyma.20G099600 and Glyma.05G059500 were found to be differentially expressed in both genotypes (Supplementary Table 5). The OVA class was originally identified in Arabidopsis through loss-of-function mutants of aminoacyl-tRNA synthetases (AARSs) that exhibited high frequencies of ovule abortion due to impaired mitochondrial translation and gametophyte viability [67]. These genes encode enzymes central to protein synthesis across cytosolic, mitochondrial, and plastid compartments, and their disruption leads to arrested embryo sacs or failed fertilization [67, 68]. In our dataset, these genes were mostly downregulated in both genotypes making it a probable cause of abortion. Given that OVA phenotypes in Arabidopsis arise from ovule defects and brief post pollination windows, bulk flower RNA-seq likely diluted the signals. We therefore infer that resolving the contribution of reproductive tissue specific genes will require sampling isolated stamens, pistils/ovules and time-stamped pollinated pistils.

Conclusion

Our findings identify candidate hormonal, developmental, and abscission associated regulatory networks that may contribute to contrasting flower-abortion phenotypes in soybean. Hormonal cues in terms of auxin and ethylene response factors appear to interface with the conserved IDA–MAPK signaling core, whose activation is dynamically buffered by the phosphatase MKP1. Downstream of this signaling module, the TALE homeodomain network comprising KNAT1/BP, KNAT6 and ATH1 translates positional and developmental patterning cues into either abscission competence or suppression. We propose that the LA genotype maintains higher KNAT1 together with enhanced MKP1 activity, reinforcing boundary integrity and restraining premature wall-degrading programs. In contrast, elevated KNAT6 and ATH1 expression in the HA genotype likely promotes abscission-zone differentiation and primes cells for IDA-mediated MAPK signaling, accelerating the induction of hydrolytic enzymes and floral detachment. The convergence of hormonal sensitivity (auxin depletion and ethylene rise), peptide–receptor signaling, phosphatase gating, and homeodomain transcriptional control constitutes a robust but tunable regulatory circuit distinguishing high and low flower-abortion phenotypes in soybean. Further, whole flower RNA-seq captures integrated floral transcriptional responses rather than tissue-specific AZ signals. Detailed studies involving AZ sampling, laser microdissection or single-cell transcriptomics and anatomical comparison of AZ cellular morphology between HA and LA genotypes will be essential. Future functional studies, including CRISPR-mediated knockouts, cell-type–specific transcriptomics, and targeted manipulation of hormone signaling pathways and abscission zone specific genes, will be critical for validating these mechanisms.

Abbreviations

AARSs

Aminoacyl-tRNA synthetases

ARFs

Auxin response factors

ATP

Adenosine triphosphate

AZ

Abscission zone

DEGs

Differentially expressed genes

ERFs

Ethylene response factors

GO

Gene ontology

HA

High abortion

KEGG

Kyoto encyclopedia of genes and genomes

LA

Low abortion

PCA

Principal component analysis

QTL

Quantitative trait loci

TF

Transcription factor

Author’s contributions

**Anita Kumari:** Writing – original draft, Visualization, Data curation, Methodology, Investigation, Formal analysis. **Leonidas D’Agostino:** Data analysis, Methodology, Writing – review and editing. **Senthilkumar Palanisamy:** Methodology. **Juliana Maria Espíndola Lima:** Methodology. **Kuldeep Kumar:** Formal analysis **,** Writing – review and editing. **Impa Somayanda**: Supervision, Writing – review and editing. **Gunvant B. Patil:** Conceptualization, Funding acquisition, Resources, Supervision, Writing – review and editing. **S.V. Krishna Jagadish:** Conceptualization, Funding acquisition, Resources, Supervision, Writing – review and editing.

Funding

The authors thank the financial support from the Multiregional Soybean Checkoff Program and the United Soybean Board.

Data availibility

The raw RNA-Seq data that support the findings of this study have been deposited in NCBI SRA database with BioProject accession number PRJNA1426070. **Reviewer link**—https://dataview.ncbi.nlm.nih.gov/object/PRJNA1426070?reviewer=n8q2sg3ilb029t406q6navgkhk.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.