ZmAGA1 enhances seed germination through raffinose hydrolysis and ZmSnRK1α-mediated starch mobilization in maize

Abstract

Background

Raffinose family oligosaccharides are key endogenous factors influencing seed germination, with raffinose as the sole member in maize. However, the mechanisms by which raffinose regulates germination remain largely unexplored.

Results

A metabolite-based genome-wide association study using a maize CUBIC population identifies six candidate genes involved in raffinose metabolism in maize kernel. Among these, ZmAGA1, encoding an alkaline α-galactosidase, is localized in the cytosol and exhibits high expression levels in the embryo during seed germination. CRISPR-Cas9-mediated knockout of ZmAGA1 leads to delayed germination, reduced germination rates, and impaired seedling growth. Integrative spatial metabolomics and single-nucleus transcriptomics analyses at 18, 24, 36, and 48 h after imbibition in germinating seeds reveal that ZmAGA1 not only mediates raffinose catabolism and spatiotemporal sugar dynamics, but also positively correlates with ZmαAMY2 and ZmαAMY3 expression in aleurone cells and elevates α-amylase activity. Moreover, ZmAGA1 physically interacts with the energy-sensing kinase ZmSnRK1α in the cytosol, and both genes are co-expressed across embryonic tissues. In zmaga1 protoplasts, ZmSnRK1α shifts toward nuclear enrichment, where it interacts with the transcription factor ZmMYBS1, which directly activates ZmαAMY3 transcription. Consistently, ZmαAMY3 transcript levels are reduced in zmaga1 and zmsnrk1α protoplasts. These findings support a ZmAGA1–ZmSnRK1α–ZmMYBS1 cascade linking raffinose hydrolysis to α-amylase activation during seed germination.

Conclusions

Our integrative spatial-omics analyses reveal that ZmAGA1 facilitates seed germination by modulating sugar metabolism through the hydrolysis of raffinose and via ZmSnRK1α-dependent regulation of starch mobilization, with both processes providing essential energy for seed germination and early seedling establishment in maize.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13059-026-04208-6.

Background

Seed vigor is a multifactorial trait governed by complex genetic and environmental factors, and it critically determines crop yield and quality. The final cumulative germination percentage and germination speed are key proxies for seed vigor, and among the intrinsic factors underlying these traits, raffinose family oligosaccharides (RFOs) play an essential role [1]. RFOs enhance seed viability during desiccation by stabilizing cellular membranes and protein structures [2, 3], and mitigating reactive oxygen species (ROS) accumulation during dehydration [3, 4], thus contributing to improved seed storability. Seed germination is an energy-demanding developmental process [5]. Sucrose non-fermenting-related protein kinase 1 (SnRK1) plays a central role in regulating energy metabolism, particularly sugar metabolism, and serves as a key hub for maintaining energy homeostasis and metabolic balance in plants [6, 7]. During the early germination phase prior to activation of the tricarboxylic acid cycle, RFOs are rapidly hydrolyzed into sucrose and galactose by α-galactosidase, providing an immediate energy source and influencing the final germination percentage [8].

Alkaline α-galactosidase (AGA) is a key enzyme responsible for RFO hydrolysis, and has been implicated in seed vigor regulation. In both chickpea (Cicer arietinum) and maize (Zea mays), AGA activity increases during early germination stages [9, 10]. Inhibition of AGA activity using 1-deoxygalactonojirimycin (DGJ) significantly reduced seed germination percentage to ~ 25% in pea (Pisum sativum) [11], while in soybean (Glycine max), DGJ treatment delayed the completion of germination without affecting final germination percentage [12]. Similarly, ClAGA2 knockout in watermelon (Citrullus lanatus) increased seed raffinose content, but had no impact on final germination percentage [13]. Conversely, overexpression of ZmAGA1 in Arabidopsis thaliana reduced seed raffinose levels, accelerated germination, and did not alter final germination percentage [14]. Collectively, these findings suggest that while RFOs may play a primary role, AGAs are important regulators of seed germination, although their relative contributions exhibit species-specific variation. Moreover, the precise molecular mechanisms by which RFOs and their hydrolase AGA modulate seed germination remain largely unresolved.

Recent advances in spatial multi-omics offer potential for dissecting the spatial and temporal complexity of plant growth and development at tissue to single-cell resolution [15]. In maize, spatial transcriptomics, single-cell RNA sequencing (scRNA-seq) and single-nucleus RNA sequencing (snRNA-seq) have been successfully applied to diverse developmental and stress-related processes, including seed germination [16], root development [17, 18], endosperm development [19], shoot apical growth [20], heat stress response [21], and plant-pathogen interactions [22]. Spatial metabolomics has also been employed to profile metabolite distributions during maize seed germination [23]. These spatial-omics tools provide unprecedented resolution and context for unraveling metabolic and transcriptional regulation during germination and facilitate a deeper understanding of raffinose metabolism and its enzymatic regulation.

In this study, we conducted a metabolite-based genome-wide association study (mGWAS) in the maize CUBIC (Complete-diallel design plus Unbalanced Breeding-like Inter-Cross) population to identify genes encoding proteins involved in raffinose metabolism. We showed that ZmAGA1, as a raffinose hydrolase, plays a key role in seed germination, and elucidated its role using spatial metabolomics and snRNA-seq analyses of germinating seeds at 18, 24, 36 and 48 HAI in both wild-type (WT) and zmaga1 mutant. Additionally, we performed protein–protein interaction assays to uncover the mechanistic basis of ZmAGA1’s regulatory network. Finally, we propose a model in which ZmAGA1 interacts with ZmSnRK1α to coordinate starch degradation and soluble sugar metabolism, thereby regulating seed germination in maize. This study offers valuable insights into the metabolic and regulatory mechanisms governing seed germination through raffinose hydrolysis.

Results

Candidate genes encoding proteins involved in the raffinose metabolic pathway in maize kernels

Despite the recognized importance of the raffinose pathway in seed physiology, the natural genetic variations governing its regulation in maize kernels remain largely unresolved. In this study, we investigated the relative content of key raffinose pathway metabolites—myo-inositol, sucrose, galactinol, and raffinose—in maize kernels using the CUBIC population [24]. All four metabolites exhibited extensive variation within the population, with fold changes ranging from 4.6 for sucrose to 1,388.2 for galactinol (Fig. 1a), providing a valuable basis for the genetic dissection of this pathway.

Fig. 1.

Candidate genes regulating the raffinose metabolic pathway in maize kernels were identified by mGWAS. a Phenotype distribution of four metabolites in the raffinose pathway among CUBIC population. Dark green dashed line indicates the mean value of each metabolite. RPA, or Relative Peak Area, refers to the relative abundance of metabolites in a given sample. b Manhattan plot based on mGWAS result of four metabolites in the raffinose pathway. SNPs within 10 kb regions of those candidate genes are highlight in colors of different traits. Note: ZmMIPS2 (myo-inositol-1-phosphate synthase 2), ZmGOLS2 (galactinol synthase 2), ZmRAFS (raffinose synthase), ZmAGA1 (alkaline α-galactosidase 1), ZmGALT (galactose-1-phosphate uridyltransferase), ZmSWEET3a (sugar will eventually be exported transporter 3a). c Raffinose metabolic pathway reconstruction in maize kernels based on the candidate genes involved in raffinose metabolic pathway identified by mGWAS. d Regional plot for the GWAS result of raffinose. The pink dot indicates SNP in ZmAGA1. e Comparison of raffinose content between the two alleles for peak SNP in ZmAGA1 locus

We performed a genome-wide association study on kernel metabolite levels and identified 39 single-variant-associated QTL (sQTLs) associated with raffinose pathway regulation (Fig. 1b; Additional file 2: Table S1). Notably, genes whose protein products are known to be related to raffinose metabolism such as ZmMIPS2 (myo-inositol-1-phosphate synthase 2), ZmGOLS2 (galactinol synthase 2), ZmRAFS (raffinose synthase), and ZmAGA1 (alkaline α-galactosidase 1) -previously characterized in maize or Arabidopsis [14, 25, 26] -were recovered (Fig. 1c). In addition, we nominated novel candidate genes, including ZmGALT (galactose-1-phosphate uridyltransferase) and ZmSWEET3a (sugars will eventually be exported transporter 3a) (Fig. 1a, c), suggesting that raffinose metabolism in maize is modulated at multiple levels, encompassing biosynthesis, catabolism and transport (Fig. 1a, c).

To further refine our analysis, we interrogated maize transcriptome data from ZEAMAP [27] (http://www.zeamap.com/) and found that among the six candidate genes, only ZmAGA1 exhibited high expression levels in germinating seeds (Additional file 1: Fig. S1) with its transcript accumulating under heat (42 ℃) and dehydration stress conditions, and specifically localized to the scutellum and embryo axis [9, 10]. Given its functional annotation as a catabolic enzyme in the raffinose pathway and its location on chromosome 7, we selected ZmAGA1 for detailed analysis. The mGWAS results indicated that ZmAGA1 accounts for approximately 6% of the phenotypic variation in raffinose content in maize kernels (Fig. 1d, e).

ZmAGA1 knockout leads to raffinose accumulation and a reduction in final cumulative germination percentage in maize

To functionally validate ZmAGA1’s role in raffinose catabolism, we generated two CRISPR-Cas9-edited homozygous knockout lines, designated ZmAGA1-KO1 and ZmAGA1-KO3, harboring single-base insertions (T and G, respectively) in exon 5 (Fig. 2a). These frameshift mutations led to premature stop codons, and immunoblotting confirmed the absence of detectable ZmAGA1 protein in both lines (Fig. 2b). Quantification of raffinose levels revealed a 3.7–3.9-fold increase in kernels of the two KO lines relative to the wild type (WT), while raffinose content in leaf remained unaffected (Fig. 2c). This indicates that ZmAGA1 plays a kernel-specific role in raffinose degradation.

Fig. 2.

ZmAGA1 knockout increased raffinose accumulation and reduced final cumulative seed germination percentage in maize. a The exon–intron structure of ZmAGA1 and two CRISPR-Cas9-generated ZmAGA1 knockout (KO) lines, namely KO1 and KO3. The PAM sequence, which serves as the recognition site for CRISPR/Cas9-mediated ZmAGA1 knock-out, is CGG and is located in exon 5. A single-base insertion—a T in KO1 and a G in KO3—results in frameshift mutations in both lines relative to wild-type (WT). The peptide, namely ENTKFQTKPDGDG (indicated with a blue shadow), was synthesized for the generation of an antibody against ZmAGA1. b Protein expression levels of ZmAGA1 in WT and its two KO lines as detected by western blot. c Raffinose content in mature leaves and kernels of WT and two ZmAGA1-KO lines. Error bars represent the standard deviation (± SD) across three biological replicates, each analyzed in technical triplicate. p-values were calculated using one-way ANOVA test. ^**p < 0.01. NS indicates no significant difference. d The representative ear morphology of maize plants in WT and KO lines. Scale bar = 5 cm. e Germination percentages of WT and KO lines. Error bars at each time point from 24 to 120 HAI represent the ± SD calculated from three independent biological replicates, with 50 seeds per replicate. f Comparison of root and shoot lengths of 10-day-old seedlings between WT and KO lines. Error bars represent ± SD from three biological replicates, with 50 seeds per replicate initially; seedlings from each replicate were used for morphometric evaluation. p-values were calculated using one-way ANOVA test. ^**p < 0.01. g Estimation of seed viability in WT and KO lines at 48 HAI. Error bars represent the mean ± SD derived from three biological replicates, with 50 seeds per replicate. p-values were calculated using one-way ANOVA test. ^**p < 0.01. NS indicates no significant difference. h Tissue-specific expression of ZmAGA1 in germinating seeds (36 HAI) and young seedlings (72 HAI) analyzed by GUS staining. The experiments were repeated twice, with at least 5 seeds per replicate. Scale bar = 4 mm. i Subcellular localization of ZmAGA1 determined by transient transformation of transgenic tobacco leaves with NLS-mCherry as a fluorescent marker, with results confirmed across three independent biological replicates. Scale bar = 20 μm

Phenotypic evaluation showed no significant differences in major agronomic traits such as plant height, ear height, leaf size, or hundred-grain weight between KO lines and WT (Fig. 2d; Additional file 1: Fig. S2a). However, seed germination speed and final germination percentage were both significantly reduced in KO lines, with final germination percentages declining by 10–12% compared with WT (Fig. 2e). Exogenous application of 5 mM raffinose enhanced the germination percentage and speed in wild-type seeds, whereas it suppressed germination in KO lines (Additional file 1: Fig. S2b). These contrasting phenotypes indicate a genotype-specific shift in raffinose function: in wild-type seeds, raffinose acts as a positive metabolic signal promoting germination, whereas in the zmaga1 mutant background, impaired raffinose catabolism leads to its accumulation, which consequently inhibits germination. Additionally, early seedling growth was markedly impaired in the KO lines (Fig. 2f; Additional file 1: Fig. S2d).

To determine whether the reduced germination resulted from reduced seed vigor, we next performed TTC staining. Although total seed viability was similar between WT (97.5%) and KO lines (~ 95.8%), a greater proportion of strongly viable seeds was observed in WT (79.2%) than in KO lines (70.8–71.7%) (Fig. 2g; Additional file 1: Fig. S2c), suggesting that prolonged germination in KO lines may result from diminished seed vigor (Additional file 1: Fig. S2).

ZmAGA1 is localized to the cytoplasm, and ZmAGA1 is highly expressed in the embryo during germination

To further elucidate the biological role of ZmAGA1, we examined its expression pattern and subcellular localization. ZmAGA1 belongs to the glycoside hydrolase 36 (GH36) family of alkaline α-galactosidases and is one of five homologs (ZmAGA1–ZmAGA5) in maize (Additional file 1: Fig. S3a). Analysis using the ZEAMAP transcriptomic data showed that ZmAGA1 exhibits peak expression during seed germination (Additional file 1: Fig. S3b).

To validate the spatial expression pattern of ZmAGA1, we generated transgenic maize lines harboring the pZmAGA1::GUS reporter construct. Histochemical staining revealed GUS activity in germinating seeds at 36 h after imbibition (HAI) and established seedlings at 72 HAI, as well as in young roots, leaves, internodes, mature leaves, and tassels (Fig. 2h; Additional file 1: Fig. S3c), consistent with the transcriptomic data and supporting a role for ZmAGA1 in seedling establishment.

To examine the subcellular localization of ZmAGA1, constructs pZmUbi::ZmAGA1–eGFP and pZmUbi::eGFP (control vector) were transiently expressed in tobacco (Nicotiana benthamiana) leaves via Agrobacterium-mediated transformation, as described previously [28]. Under laser confocal microscopy, the green fluorescent signal from the ZmAGA1–eGFP fusion protein was specifically localized to the cytoplasm of tobacco leaf cells, in contrast to the control eGFP signal, which was observed in both the nucleus and cytoplasm (Fig. 2i).

ZmAGA1 regulates the spatiotemporal dynamics of soluble sugars during germination

To elucidate the role of ZmAGA1 in seed germination, we conducted metabolic profiling, mass spectrometry imaging (MSI), and snRNA-seq on maize seed samples collected at 18, 24, 36, and 48 HAI. An overview of the experimental workflow is shown in Fig. 3a. GC–MS-based metabolite profiling revealed that raffinose was predominantly accumulated in the embryo and represented the most abundant soluble sugar at early germination stages. Its content was significantly higher in KO lines compared with WT, and it declined rapidly in both embryo and endosperm tissues of both genotypes as germination progressed to completion at approximately 36 HAI and seedling establishment commenced (Fig. 3b, c).

Fig. 3.

ZmAGA1 regulated the metabolism and the spatiotemporal distribution of soluble sugar during seed germination and seedling establishment in maize. a Schematic workflow of metabolomic, MSI imaging, and snRNA-seq data acquisition from germinating maize seeds. The maize embryo and endosperm were separately isolated at 18, 24, 36, and 48 HAI for LC–MS-based metabolomics analysis. Additionally, whole seeds at the aforementioned four time points were collected for MSI and snRNA-seq analyses. b-c Quantitative analysis of soluble sugar content and its dynamics during maize seed germination in WT and KO lines: b and c raffinose, sucrose, maltose, fructose, and glucose levels in the embryo and endosperm at 18, 24, 36, and 48 HAI, respectively. EM, embryo; EN, endosperm. Error bars indicate ± SD from three biological repeats, each assayed in technical triplicate. p values were determined based on one-way ANOVA test. ^**p < 0.01. d The dynamic spatial and temporal distribution of monosaccharides, disaccharides and trisaccharides by AFADESI-MSI during maize seed germination (18 and 24 HAI) and seedling establishment (36 and 48 HAI) in WT and KO3 lines. For each condition, at least three cryosections—each prepared from an independently imbibed seed—were selected for subsequent AFADESI-MSI analysis. e Procedure for integrating LCM with nanoLC–MS/MS to identify disaccharides and trisaccharides metabolites observed in specific tissues by AFADESI-MSI. Three tissue sections—each derived from (a) separately imbibed seed—were subjected to LCM followed by nanoLC–MS/MS analysis. Raffinose, sucrose, maltose, fructose, and glucose were successfully identified using nanoLC–MS/MS

Sucrose also showed a trend to decrease in both tissues during seed germination; however, its level in the endosperm was markedly lower in KO lines when compared with the WT (Fig. 3b, c); By contrast, maltose displayed the opposite trend to sucrose (Fig. 3b, c). Fructose and glucose, which serve as major energy sources during seedling establishment, increased over time following the completion of germination in the embryo, but their levels were significantly lower in KO lines compared with WT from 36 to 48 HAI (Fig. 3b, c). No significant differences in fructose and glucose content were observed in the endosperm between WT and KO lines, although their dynamics differed: fructose levels declined, while glucose remained stable (Fig. 3b, c). Collectively, these results suggest that ZmAGA1 knockout leads to excessive raffinose accumulation in the embryo, thereby disrupting the metabolism and mobilization of key soluble sugars necessary for energy supply during seed germination and seedling establishment.

To visualize the spatiotemporal dynamics of these sugars, we employed air flow-assisted desorption electrospray ionization mass spectrometry imaging (AFADESI-MSI) to monitor their in situ distribution in embryo and endosperm tissues at four time points. Consistent with the metabolite profiling data (Fig. 3b, c), MSI results showed stronger signals for monosaccharides (glucose and fructose) and disaccharides (sucrose and maltose) in WT seeds, whereas trisaccharide (raffinose) signals were slightly stronger in KO lines (Fig. 3d). Notably, monosaccharides, as immediate energy sources, were mainly localized in the endosperm from 18 to 24 HAI and subsequently translocated to the embryo from 36 to 48 HAI in both genotypes (Fig. 3d). Disaccharides, which serve as indirect energy sources, exhibited an inverse spatial pattern—present in both embryo and endosperm during germination (18–24 HAI), but later restricted mainly to the embryo post-germination (36–48 HAI) (Fig. 3d). Trisaccharide signals, attributed to raffinose, were detected in the embryo and diminished precipitously to below the limit of detection following the completion of germination (Fig. 3d).

To validate that the MSI signals corresponded to specific sugars in targeted tissues, we excised high-signal regions in embryo and endosperm tissues using laser capture microdissection (LCM), and analyzed them by nanoLC–MS/MS (Fig. 3a). By comparing MS² spectra of these samples with a raffinose standard, we confirmed that the trisaccharide signals in the radicle and plumule were indeed derived from raffinose (Fig. 3e).

ZmAGA1 knockout suppresses α-amylase activity in aleurone layer cells and delays starch mobilization during germination

To determine whether sugar metabolism at the cellular level reflects organ- and tissue-level changes during seed germination and early seedling establishment in maize, we performed snRNA-seq on germinating seeds at 18 and 24 HAI and establishing seedlings at 36 and 48 HAI to profile gene expression at single-nucleus resolution (Fig. 3a). Analysis of barcode and unique molecular identifier (UMI) data identified 44,777 individual nuclei (Additional file 2: Table S2), with expression detected for 39,036 genes, of which 7,439 were differentially expressed. Dimensionality reduction via Uniform Manifold Approximation and Projection (UMAP) clustered nuclei into 10 distinct cell types: aleurone layer (AL), scutellum (SC), plumule (PL), radicle (RA), root apical meristem (RAM), shoot apical meristem (SAM), embryonic axis (EA), coleorhiza (CR), root cap (RC) and unknown (UN) (Fig. 4b, c; Additional file 2: Table S3). These cell types were annotated using known marker genes from prior spatial transcriptome studies and anatomical references (Fig. 4d; Additional file 2: Table S4), with high marker specificity and enrichment confirming accurate classification (Fig. 4d).

Fig. 4.

ZmAGA1 involved in sugar metabolism and associated with α-amylase activity in maize. a Vertical cryosection of a maize kernel from WT plants at 36 HAI and spatial distribution of cell clusters in maize seeds. Scale bar = 1 mm. PE, pericarp; EN, endosperm; AL, aleurone layer; SC, scutellum; CO, coleoptile; PL, plumule; RA, radicle; RAM, root apical meristem; SAM, shoot apical meristem; EA, embryonic axis; CR, coleorhiza. The meanings of subsequent abbreviations in panels (b – e) are consistent with those in panel (a). b Uniform manifold approximation and projection (UMAP)-based dimensionality reduction and visualization of 44,777 single-nucleus transcriptomes from germinating maize seeds of WT plants, clustered into 10 transcriptionally distinct cell types, i.e., SC, EA, AL, RA, PL, CR, RAM, SAM, RC (root cap) and UN (unknown). Each dot represents a single cell, and colors indicate different cell-types. c Bar plot illustrating the proportion of cell counts for each cell type identified by snRNA-seq at 18, 24, 36, and 48 HAI. The same cell type across different time points is indicated by the same color. d Dot plot illustrating the expression patterns of 39 representative marker genes specific to 10 cell-types. The size of the circle indicates the relative percentage of cells expressing the marker genes, while the color represents the relative expression level. e Dot plot illustrating the expression levels of genes encoding proteins involved in sugar metabolism across AL, SC, PL, EA and RA cell types in germinating maize seeds (18–48 HAI, from left to right). The size of the circle indicates the relative percentage of cells expressing the gene, while the color represents the relative expression level. f Estimation of α-amylase activity in WT and KO lines during maize seed germination from 18 to 48 HAI. Error bars represent ± SD from three biological replicates, each assayed in technical triplicate. p-values were calculated using one-way ANOVA test. ^**p < 0.01. NS indicates no significant difference. g Quantitative RT-PCR analysis of the expression levels of ZmαAMY2 and ZmαAMY3 in WT and KO lines during maize seed germination. Error bars represent ± SD from three biological replicates, each assayed in technical triplicate. p-values were calculated using one-way ANOVA test. ^**p < 0.01. ZmGAPDH (Zm00001eb173410) was used as an internal control for normalization. h Estimation of starch contents in the endosperm of WT and KO lines during maize seed germination from 18 to 48 HAI. Error bars represent ± SD from three biological replicates, each assayed in technical triplicate. p-values were calculated using one-way ANOVA test. ^**p < 0.01. NS indicates no significant difference

Among these, AL cells, positioned between the endosperm and embryo, are known for synthesizing hydrolytic enzymes and mediating nutrient transport [29]. We compared the expression levels of ZmAGA1 and representative sugar metabolism genes across AL, SC, PL, EA, and RA cell types. We examined the expression of ZmAGA1 and several sugar metabolism-related genes, including those encoding α-amylase 2 and 3 (ZmαAMY2 and ZmαAMY3), β-glucosidase 8 (ZmGLU8), trehalose-6-phosphate synthase 6 (ZmTPS6), the sugar transporter ZmSWEET4, and the MYB transcription factors ZmMYBS1, ZmMYBS2 and ZmMYBS3. ZmAGA1 was broadly expressed across these tissues and exhibited a dynamic pattern, with initial induction between 18 and 24 HAI or between 24 and 36 HAI, followed by a subsequent decline (Fig. 4e). Its expression was notably higher in SC, PL, and RA cells (Fig. 4e), consistent with the GUS staining pattern observed in the pZmAGA1::GUS reporter assay (Fig. 2h; Additional file 1: Fig. S3c).

ZmαAMY2 and ZmαAMY3 were expressed in all tested cell types, with ZmαAMY3 showing the highest expression in AL cells (Fig. 4e). Experimental validation revealed that the expression levels of ZmαAMY2/3 and α-amylase activity increased throughout germination and early seedling establishment in both WT and KO lines, but they were consistently and significantly lower in the KO background (Fig. 4f, g). These findings suggest that the suppression of α-amylase gene expression and enzymatic activity in AL cells, along with subsequent delay in starch mobilization in the ZmAGA1-KO lines (Fig. 4h), contributes to the impaired germination phenotype observed in ZmAGA1-KO seeds compared with WT seeds.

A ZmAGA1–ZmSnRK1α–ZmMYBS1 cascade links raffinose hydrolysis to α-amylase activation during germination

To investigate whether ZmAGA1 interacts with ZmSnRK1α to modulate sugar metabolism during maize seed germination and early seedling establishment, we predicted the three-dimensional structures of both proteins and their potential interaction using AlphaFold3 [30]. As shown in Fig. 5a, the predicted interaction model yielded a high-confidence pTM + ipTM score of 0.96 (Additional file 1: Fig. S4a), indicating the formation of a stable and reliable complex between ZmAGA1 and ZmSnRK1α. Structural analysis revealed seven pairs of hydrogen-bonded interacting residues between the two proteins (Fig. 5b; Additional file 1: Fig. S4b, c). Notably, two of these residues (Lys¹⁸⁹ and Glu²⁵³) are located within the S_TKc domain of SnRK1α, while Tyr⁴²⁴, Lys⁴²⁷, Tyr⁴⁷⁴, and Lys⁴⁷⁵ reside in the KA1 domain (Fig. 5a, b; Additional file 1: Fig. S4c).

Fig. 5.

ZmAGA1 interacted with ZmSnRK1α and a working model in which the ZmAGA1–ZmSnRK1α module regulates seed germination in maize. a Predicted 3D protein structures of ZmAGA1 and ZmSnRK1α and their interactions using AlphaFold 3. b Key interaction residue sites and their corresponding interaction distances between ZmAGA1 and ZmSnRK1α. c Y2H analysis of the interaction between ZmAGA1 and ZmSnRK1α, as well as its Domains. The pair pGADT7-T and pGBKT7-p53 served as a positive control, whereas the pair pGADT7-T and pGBKT7-Lam served as a negative control. AD, activation domain; BD, binding domain; SD/-THL, synthetic dextrose medium lacking tryptophan, histidine, and leucine; SD/-THLA + X-α-Gal, synthetic dextrose medium lacking leucine, tryptophan, histidine, and adenine supplemented with X-α-Gal. Each colony was diluted in 50 µL of sterile water and subsequently serially diluted up to 10⁻². At least three colonies were tested for each combination. d LCA Analysis of the interaction between ZmAGA1 and ZmSnRK1α in N. benthamiana leaves. Results shown are representative of three independent biological replicates. e BiFC Analysis of the interaction between ZmAGA1 and ZmSnRK1α in maize leaf protoplasts. The CDS of ZmAGA1 and ZmSnRK1α were respectively cloned into the pSPYNE and pSPYCE vectors, resulting in fusion to the N-terminal and C-terminal of YFP, respectively. The constructs pCaMV 35S::bZIP63-VN and pCaMV 35S::VC-bZIP63 served as positive controls for protein interaction, whereas pCaMV 35S::NLS-mTBFP was used as a positive control for nuclear localization. Results shown are representative of three independent biological replicates. Scale bars = 50 µm. f Co-IP Analysis of the interaction between ZmAGA1 and ZmSnRK1α in maize leaf protoplasts. Immunoprecipitated samples were detected using anti-GFP and anti-FLAG antibodies, respectively. The symbols “ + ” represent presence corresponding construct; symbols “ − ” represent free construct. Three independent experiments were performed with similar results. g A working model proposing that ZmAGA1 regulates seed germination and seedling establishment in maize through raffinose hydrolysis and ZmSnRK1α-dependent starch mobilization

To experimentally validate the predicted interaction, we employed a combination of yeast two-hybrid (Y2H), luciferase complementation assay (LCA), bimolecular fluorescence complementation (BiFC) and co-immunoprecipitation (Co-IP). In Y2H assays, ZmAGA1 (used as bait) interacted with several ZmSnRK1α constructs expressed as prey, including full-length ZmSnRK1α and its UBA-KA1 and Loop-KA1 domains (Fig. 5c). Among these, the strongest interaction was observed between ZmAGA1 and the UBA-KA1 domain. In addition to ZmSnRK1α, three paralogous genes i.e., ZmSnRK1αB, ZmSnRK1αC, and ZmSnRK1αD were identified in maize, and their interactions with ZmAGA1 were assessed. ZmAGA1 was also found to interact with ZmSnRK1αC (Additional file 1: Fig. S5). Moreover, ZmSnRK1α is capable of forming both homodimers and heterodimers with ZmSnRK1αC (Additional file 1: Fig. S5c, d). The interaction between ZmAGA1 and ZmSnRK1α was further confirmed by LCA, BiFC and Co-IP assays, which additionally revealed a cytosolic localization of the protein complex (Fig. 5d, e, f).

To determine whether ZmSnRK1α interacts with MYBS transcription factors, known regulators of sugar signaling [31–35], to activate or repress downstream target genes, including key α-amylase genes [35], to thereby coordinately regulate maize seed germination, we performed a series of complementary experiments to characterize these molecular interactions. As shown in Additional file 1: Fig. S7, ZmSnRK1α localizes to both the cytosol and nucleus, whereas ZmMYBS1 is exclusively nuclear. Y2H and BiFC assays confirmed physical interactions between ZmSnRK1α and all three ZmMYBS1 isoforms (ZmMYBS1.1, ZmMYBS1.2, and ZmMYBS1.3), with the strongest interactions observed for ZmSnRK1α–ZmMYBS1.1 and ZmSnRK1α–ZmMYBS1.2, and comparatively weaker interaction for ZmSnRK1α–ZmMYBS1.3. Furthermore, all three ZmMYBS1 isoforms bound the ZmαAMY3 promoter in yeast one-hybrid (Y1H) assays; ZmMYBS1.1 and ZmMYBS1.3 exhibited stronger binding than ZmMYBS1.2 (Additional file 1: Fig. S7a). Based on these results, ZmMYBS1.1, the most strongly interacting and representative isoform, was selected for subsequent functional validation and renamed as ZmMYBS1. Dual-luciferase reporter and electrophoretic mobility shift assays (EMSA) demonstrated that ZmMYBS1 directly binds the ZmαAMY3 promoter and activates its transcription. Consistent with this, ZmαAMY3 transcript levels were significantly reduced in leaf protoplasts isolated from both zmaga1 and zmsnrk1α mutants (Additional file 1: Fig. S7b, c). Sequence analysis identified three conserved TATCC motifs (TA-boxes) within the 2000-bp upstream region of the ZmαAMY3 transcription start site; EMSA confirmed that ZmMYBS1 binds all three motifs, with the highest affinity for M1, followed by M2 and M3 (Additional file 1: Fig. S7d, e, f).

To further elucidate how ZmSnRK1α coordinates its interactions with the cytosol-localized protein ZmAGA1 and the nuclear-localized transcription factor ZmMYBS1 to regulate maize seed germination, we conducted integrated phosphoproteomic profiling [36] and subcellular fractionation analyses comparing the zmaga1 mutant with WT. Results showed that, although total ZmSnRK1α protein abundance and its phosphorylation level at the conserved Thr¹⁷⁰ residue remained largely unchanged during maize seed germination (Additional file 1: Fig. S5b, S9c), its subcellular distribution was markedly altered: in WT leaf protoplasts, ZmSnRK1α was enriched in the cytosol, whereas in zmaga1 mutant leaf protoplasts, it accumulated predominantly in the nucleus (Additional file 1: Fig. S8).

Based on these findings, we propose a working model in which ZmAGA1 regulates maize seed germination by modulating sugar metabolism via raffinose hydrolysis and by activating the ZmSnRK1α–ZmMYBS1–ZmαAMY3 signaling cascade to promote starch mobilization. These two processes collectively supply the essential energy required for seed germination and early seedling establishment (Fig. 5g). In this model, ZmSnRK1α acts as a sugar energy sensor that dynamically shuttles between the cytosol and nucleus. In WT, ZmSnRK1α predominantly localizes to the cytosol and physically interacts with ZmAGA1. In contrast, loss of ZmAGA1 function in the zmaga1 mutant impairs raffinose hydrolysis, disrupting sugar homeostasis and triggering nuclear translocation of ZmSnRK1α, where it subsequently interacts with ZmMYBS1. This interaction leads to transcriptional repression of ZmαAMY3, resulting in reduced α-amylase activity and compromised starch mobilization—the primary energy source for maize seed germination. Consequently, the zmaga1 mutant exhibits delayed germination, a reduced germination rate, and impaired seedling establishment.

Discussion

The role of raffinose and its hydrolyzing enzyme ZmAGA1 on seed germination

Seed vigor, a key agronomic trait that influences germination efficiency and seedling establishment, is orchestrated by complex metabolic and physiological processes during seed maturation, desiccation, and germination. Among the biochemical factors that affect seed vigor, RFOs and their hydrolytic enzymes have emerged as pivotal regulators of stress tolerance, seed longevity, and germination dynamics across plant species [13, 14, 25, 26, 37]. RFOs, including raffinose, stachyose, verbascose, and higher order oligosaccharides, are α-galactosyl derivatives of sucrose that accumulate during seed development in certain species. Their biosynthesis involves the stepwise transfer of galactosyl units in galactinol-dependent reactions catalyzed by GOLS, RAFS, and related enzymes. Several genes encoding key enzymes in RFO biosynthesis and degradation, such as GOLS, RAFS, and AGA, have been cloned and functionally characterized in Arabidopsis, pea, cucumber (Cucumis sativa), maize, and other species [2, 14, 25, 38, 39]. In this study, we identified 39 sQTLs associated with raffinose content in maize. These include four previously reported genes (ZmMIPS2, ZmGOLS2, ZmRAFS, and ZmAGA1) and two novel candidate genes, ZmGALT and ZmSWEET3a, which are involved in the RFO metabolic pathway and its upstream components. These findings underscore the metabolic diversity of RFOs and the complexity of their regulatory architecture in maize.

RFOs serve dual functions in seed physiology, acting both as protectants against abiotic stress and as reserve carbohydrates that support germination. As osmoprotectants, they stabilize cellular membranes and macromolecules during dehydration, thereby safeguarding seed viability across diverse taxa, including Arabidopsis, legumes, and cereals. This conserved function is attributed to their ability to replace water molecules and mitigate oxidative damage during desiccation. Concurrently, RFOs function as energy reserves that can be rapidly hydrolyzed by α-galactosidase to release sucrose and galactose, meeting the metabolic demands of germinating seeds, as demonstrated in species such as pea [8]. However, the relationship between RFO content and seed vigor is not straightforward. For example, overexpression of ZmRAFS in Arabidopsis significantly elevated raffinose levels but reduced seed germination percentage, while co-overexpression of ZmGOLS2 and ZmRAFS led to both enhanced RFO accumulation and improved completion of germination following controlled deterioration [25, 26]. These findings indicate that optimal RFO thresholds likely vary across species, reflecting a delicate balance between their protective roles during quiescence and catabolic functions during germination.

Previous studies have consistently highlighted AGA1 as a crucial regulator of seed vigor [13, 14, 40]. In pea, inhibition of PsAGA1 activity results in a fourfold reduction in germination rate [11]. Similarly, knockout of ClAGA2 in watermelon increases seed raffinose content and delays germination, although the final germination percentage remains unaffected [13]. Soybean seeds exhibit a similar pattern, with delayed germination but no significant reduction in total germination percentage [12, 26]. In contrast, constitutive overexpression of ZmAGA1 in Arabidopsis reduces raffinose content, accelerates germination, and maintains a high germination percentage [14]. Collectively, these findings position α-galactosidase enzymes, especially AGA, as metabolic gatekeepers of germination efficiency. In the present study, we found that ZmAGA1-KO lines accumulated raffinose in kernels, especially in the embryo, along with delayed germination and a reduced final germination percentage (Fig. 2). These phenotypes differ from previously reported results in watermelon and Arabidopsis, suggesting a phylogenetically restricted mechanism underlying RFO-mediated regulation of seed vigor in maize. Furthermore, high-vigor seeds, defined by their high final germination percentages and rapid germination speeds, play a critical role in agricultural production. Knockout of ZmAGA1 results in a > 10% reduction in germination percentage and delayed germination (Fig. 2), suggesting that overexpression of ZmAGA1 may enhance seed vigor, which holds significant implications for agricultural production. Phylogenetic and expression analyses of the maize AGA gene family indicate that, in addition to ZmAGA1, ZmAGA4 exhibits the highest transcript abundance among paralogs and clusters phylogenetically with ZmAGA1 in the same clade (Additional file 1: Fig. S3a, b). Although ZmAGA4 is co-expressed and evolutionarily closely related to ZmAGA1, its expression level in germinating embryos remains substantially lower than that of ZmAGA1. Furthermore, no upregulation of ZmAGA4 was detected in zmaga1 knockout seeds under our experimental conditions. Thus, while ZmAGA4 represents the most plausible candidate for partial functional redundancy, its insufficient expression likely limits compensatory capacity in the absence of ZmAGA1.

Comparative analyses of RFO profiles across species reveal further mechanistic divergence. In dicots such as Arabidopsis, soybean, and cucumber, RFOs predominantly exist as stachyose and verbascose, with raffinose playing a relatively minor role [41]. In contrast, monocots such as maize and rice (Oryza sativa) primarily accumulate raffinose, with stachyose and verbascose being nearly undetectable, because their genomes do not encode stachyose or verbascose synthases [25, 41]. In Arabidopsis, seed vigor correlates more strongly with total RFO content and the RFO-to-sucrose ratio than with raffinose levels alone. Indeed, lower raffinose levels are associated with higher germination percentages [14], and stachyose and verbascose appear to contribute more significantly to seed vigor than raffinose [26]. Taken together, prior studies and our current findings highlight fundamental differences in the role of raffinose and its hydrolyzing enzyme ZmAGA1 in monocot seed physiology, suggesting a maize specific regulatory framework for seed vigor.

ZmAGA1 modulates sugar metabolism and α-amylase activity during germination

Seed germination is an energetically demanding process, encompassing water uptake by the seed, mobilization of stored reserves, and active cell division and elongation. This complex cascade is regulated by intrinsic genetic programs in concert with environmental signals such as light, temperature, and nutrient availability. Of these, sugar signaling stands out as a critical mediator, functioning both as an energy substrate and as a regulatory cue. While starch hydrolysis is generally regarded as the primary energy source for germinating seeds, inhibition of amylase activity has been shown to impair germination percentage and reduce germination rates in starch-endosperm seeds [42]. In our study, ZmAGA1 knockout led to significant disruption of sugar metabolism (Fig. 3). Specifically, glucose, fructose and maltose levels in the embryo were markedly reduced in ZmAGA1-KO lines compared with WT plants (Fig. 3b, d), which raises the possibility that these sugars or their downstream metabolic consequences could serve as signals. Furthermore, the expression of sugar metabolism-related genes such as ZmαAMY2 and ZmαAMY3 genes were downregulated, and corresponding enzymatic activity was decreased in knockout lines. These findings indicate that the inhibition of seed germination in maize zmaga1 mutant is likely attributable to an impaired capacity to mobilize sufficient raffinose via ZmAGA1 and starch via α-amylase, both processes providing essential energy for the completion of seed germination.

The role of amylase in seed germination has been well established. In cereal seeds, α-amylase produced in AL cells plays a pivotal role in converting endosperm starch into soluble sugars, which fuel early seedling growth [29]. Physiological and biochemical studies have shown that this process begins with the biosynthesis of active gibberellins (GAs) in the embryo, which are subsequently transported to the AL. Upon arrival, GA signaling activates transcription factors that induce the expression of α-amylase genes. The synthesized α-amylase is then secreted into the endosperm, where it hydrolyzes stored starch to release soluble sugars, which are then taken up by the scutellum for utilization by the nascent seedling [29, 43]. Consistent with this model, our snRNA-seq data revealed that the transcripts of ZmαAMY2/3 were predominantly localized in AL cells (Fig. 4e).

A raffinose metabolism-triggered signaling cascade couples sugar sensing with starch mobilization during germination

SnRK1 is an evolutionarily conserved central regulator of cellular energy metabolism [6, 44–50], functioning both as a direct regulator, by phosphorylating cytosolic metabolic enzymes (e.g., SPS, NR, TPS5, HMGR) to modulate their activity or stability [51–53], and as an indirect regulator of gene expression by phosphorylating nuclear transcription factors (e.g., bZIP63, MYC2, MYBS1, FUS3, IDD8) to alter transcript levels of metabolic genes [35, 54–57]. In this study, ZmSnRK1α interacted with ZmAGA1 in the cytosol (Fig. 5); however, mass spectrometry-based phosphoproteomic analysis detected no phosphorylation sites on ZmAGA1 or its orthologs [36], indicating that ZmSnRK1α does not directly phosphorylate ZmAGA1. We therefore speculate that an auxiliary protein may act as a molecular scaffold or stabilizer, facilitating the ZmSnRK1α–ZmAGA1 interaction and thereby modulating ZmAGA1 enzymatic activity or protein stability in the cytosol. Consistent with this hypothesis, multiple candidate proteins, distinct from ZmSnRK1α, were co-identified in our unpublished Y2H screening and TurboID-based proximity-labeling experiments, both using ZmAGA1 as bait (unpublished data). Similarly, ZmSnRK1α interacted with ZmMYBS1 in the nucleus (Additional file 1: Fig. S6), yet no phosphorylation sites were detected on ZmMYBS1 or its isoforms [36], suggesting that ZmSnRK1α also does not directly phosphorylate ZmMYBS1. Thus, we speculate that an auxiliary regulatory factor may be required to bridge or stabilize the ZmSnRK1α–ZmMYBS1 complex, thereby enabling context-specific transcriptional regulation, including activation of α-amylase genes during germination.

Seed germination is an energy-intensive process during which starch mobilization, primarily via α-amylase-mediated hydrolysis, supplies the bulk of the required energy. In rice, OsSnRK1A is post-transcriptionally regulated by sugars and is required for glucose starvation induced OsMYBS1 activation and OsαAMY3 expression; thus, it functions upstream of the OsMYBS1–OsαAMY3 sugar response complex to regulate germination and seedling growth [58]. In maize, we found that ZmAGA1 expression was positively correlated with that of ZmαAMY2 and ZmαAMY3 (Fig. 4e, f, g); furthermore, ZmMYBS1 was shown to directly bind to the TATCC cis-element of ZmαAMY3 promoter and activate its transcription (Additional file 1: Fig. S7). Collectively, these results suggest the existence of a raffinose metabolism-triggered signaling cascade that functionally couples sugar sensing, transcriptional reprogramming, and starch catabolism to sustain energy provision for maize seed germination.

Subcellular localization is uncoupled from canonical SnRK1α activation during germination

Phosphorylation of the conserved Thr¹⁷⁰ residue, a hallmark of SnRK1 activation, within the T-loop is essential for its protein kinase activity [7, 44, 59]. The dual localization of SnRK1α in both the cytosol and nucleus suggests that its subcellular partitioning serves as a regulatory mechanism to mediate distinct physiological functions. In Arabidopsis, SnRK1α dynamically shuttles between these compartments, and this nucleocytoplasmic trafficking is indispensable for its functional roles, including ABA-dependent regulation of root meristem size and cell proliferation [60], as well as BR-regulated meristemoid differentiation and stomatal development [61]. Specifically, ABA-induced repression of root meristem cell proliferation in Arabidopsis critically depends on the swift nuclear-to-cytoplasmic relocalization of SnRK1: under favorable conditions, SnRK1 accumulates in the nucleus, whereas ABA stimulation promotes its cytoplasmic accumulation [60]. In parallel, KIN10, the Arabidopsis SnRK1α ortholog, is predominantly nuclear and directly phosphorylates and stabilizes the transcription factor SPEECHLESS to promote stomatal development; BR signaling disrupts this by sequestering KIN10 in the cytoplasm via KINβ2 phosphorylation [61]. In this study, ZmSnRK1α showed enhanced nuclear accumulation in the zmaga1 mutant compared with its preferential cytosolic localization in WT (Additional file 1: Fig. S8). Furthermore, no significant differences in either total ZmSnRK1α protein abundance or phosphorylation at the conserved Thr¹⁷⁰ residue were observed between zmaga1 mutant and WT kernels during early seed germination (Additional file 1: Fig. S5 and S9). These results suggested that ZmSnRK1α’s subcellular localization is likely uncoupled from its canonical activation status, which is defined by Thr¹⁷⁰ phosphorylation, during maize seed germination, highlighting the need for spatially resolved phosphoproteomics to quantify Thr¹⁷⁰ phosphorylation specifically within nuclear and cytosolic subcompartments in future studies.

Conclusions

ZmAGA1 is essential for maize seed germination and early seedling establishment by modulating sugar metabolism through the hydrolysis of raffinose and via ZmSnRK1α-dependent regulation of starch mobilization, with both processes providing essential energy for seed germination and early seedling establishment in maize. This work provides valuable insights into the genes and metabolic mechanisms that control seed germination.

Methods

Plant materials and growth conditions

The CUBIC population [24], including 24 parents and 1,404 progenies, were planted in Sanya (109°51′E, 18°25′N) in winter 2015 for mature kernel harvest. Each line was grown in a randomized design in a single 2.5 m row plot. Each plant was self-pollinated, and three ears were randomly selected from each line for mature kernel collection. All samples were stored at –80 °C prior to metabolite extraction. Gene-edited plants (ZmAGA1-KO lines) were also grown in Sanya in the winters of 2020, 2021 and 2022, with sampling performed in the same manner.

GWAS analysis

Single-variant-based GWAS (sGWAS) and identity-by-descent GWAS were performed as described by Liu et al. [24]. Briefly, for sGWAS, high-quality SNPs with minor allele frequency (MAF) ≥ 2% were used as genotypic data and a mixed linear model was applied (TASSEL 3.0; [62]) to control for population structure. The significance threshold for associations was set to 1/N, where N is the number of independent tests. To define candidate loci, significant SNPs for each trait were first grouped into a locus if consecutive significant SNPs were < 20 kb apart and at least two significant SNPs were present. Adjacent loci were merged into one locus if any significant SNPs between them were in linkage disequilibrium (LD; r^2 ≥ 0.2). Loci in very close proximity were merged into a single locus, with the lowest p-value SNP designated as the lead SNP. Ultimately, the significant loci were considered sQTL; the lead SNP determined the significance of each sQTL, and the span of significant SNPs defined the sQTL interval.

Seed germination experiment

Maize seeds with uniform shape and size were surface-sterilized in 70% (v/v) ethanol for 1 min, then in 2% (v/v) sodium hypochlorite plus 0.05% (v/v) Tween-20 for 5 min. Seeds were rinsed five times with sterile water and air-dried on a clean bench for 30 min. Germination was carried out on two layers of wet filter paper in transparent plastic boxes (19 cm × 13 cm × 12 cm) with 30 mL of water with or without 5 mM raffinose. Boxes were incubated in a chamber at 28 ± 1 °C under a 16-h light/8-h dark photoperiod (10,000 lx). Seeds with protruding radicles were counted as germinated at 12-h intervals from 18 to 120 h. The cumulative germination percentage at each time point was calculated, and values are reported as mean ± SE for three biological replicates (50 seeds per replicate). Additionally, 30 sterilized seeds were placed in germination bags (30 cm × 25 cm; CYG-98LB, PhytoTC Co., Ltd., China) with 30 mL water and incubated in a chamber at 28 ± 1 °C under a 16 h light/8 h dark photoperiod (10,000 lx) as well. After 10 days, shoot and root lengths of 30 seedlings were measured; the experiment was performed in triplicate.

Vector construction and maize transformation

To analyze tissue-specific expression of ZmAGA1, a 2,000-bp promoter region upstream of the ZmAGA1 start codon (ATG) was amplified from B73 genomic DNA and cloned into pCAMBIA1381Z-Bar (containing a bar selection marker) using a one-step homologous recombination kit (ClonExpress II, Vazyme, China). The resulting pZmAGA1::GUS binary vector was confirmed by sequencing. To create ZmAGA1 knockouts, single-guide RNAs (sgRNAs) targeting ZmAGA1 were designed using the CRISPR-P 2.0 tool [63]. Expression cassettes carrying sgRNAs under the OsU3 promoter were inserted into the binary vector pYLCRISPR/Cas9Pubi-B [64] via Bsa I digestion and ligation, generating the construct pZmUbi::Cas9-ZmAGA1-KO. Primers for vector construction and mutant genotyping are listed in Additional file 2: Table S5. All binary vectors were introduced into Agrobacterium tumefaciens strain EHA105 and transformed into maize immature embryos (KN5585 inbred background) by Wimi Biotechnology Co., Ltd. (China).

Metabolomics analysis using gas chromatography tandem mass spectrometry (GC–MS)

For metabolomic analysis of germinating seeds, embryos and endosperms of maize seeds were dissected at 18, 24, 36, and 48 HAI with six biological replicates per sample. Samples were immediately flash-frozen in liquid nitrogen and stored at –80 °C until use. Freeze-dried embryos and endosperms were ground to fine powder for metabolite extraction. Subsequently, 50 µg of each sample was extracted and then derivatized with 150 µL of N-methyl-N-(trimethylsilyl) trifluoroacetamide following the previous reported procedures in our lab as described in Yin et al. [65]. Quality control samples were prepared by pooling small aliquots of each sample and thoroughly mixing, which were run every 10 samples to monitor system performance and for data normalization. The extract for each sample was analyzed by GC–MS (7890A-5975C, Agilent, USA) using the same method as described before, as well as data analysis including metabolite identification based on our in-house database and statistical analysis [65].

AFADESI-MSI analysis

Maize seed samples were collected at 18, 24, 36, and 48 HAI during germination were used for MSI analysis. The protocol for cryosection, quality inspection, and further air flow-assisted desorption electrospray ionization mass spectrometry imaging (AFADESI-MSI) analysis was adapted from He et al. [66]. Briefly, samples were equilibrated at –20 °C in a Leica CM3050 S cryostat (Leica, Germany) for 20 min, then sectioned longitudinally (10 µm thickness) at –20 °C and thaw-mounted onto slides. After vacuum drying for 15 min to remove moisture, sections were analyzed by a mass spectrometer (Q Exactive, Thermo Fisher Scientific, USA) coupled with an AFADESI ion source (Viktor Technologies Co.,Ltd, Beijing, China) at a spatial resolution of 50 µm in full MS mode (m/z 70–1000; AGC 3E6; max injection time 200 ms). Key MSI parameters: solvent flow rate 5 µL/min; sprayer voltage 7,000 V; transport tube voltage 3,000 V; nebulizing gas flow 45 L/min; capillary temperature 350 °C; scan speed 200 µm/s in x-direction with 200 µm step in y-direction. MSI data were processed with MassImager™ v1.0 [67] for ion image reconstruction, background subtraction, and average mass spectrum generation. Metabolites were annotated using MSImagingAnnotator [68].

Laser capture microdissection and metabolite identification using nano liquid chromatography tandem mass spectrometry (nanoLC–MS/MS)

To validate the MSI results, we conducted laser capture microdissection (LCM) and followed by nanoLC–MS/MS on specific tissues from seeds collected at 18, 24, 36 and 48 HAI. Regions of interest (e.g., SC, PL, RA) were selected on cryosections (10 µm) and isolated using a PALM MicroBeam LCM system (Zeiss) with laser pressure catapulting (PALM MicroBeam, Zeiss, Germany). Cutting and capture were executed automatically using a 10 × objective. Laser settings were: cutting energy 43, focus 55; capture energy 20, focus –15. Dissected tissues were collected into 50 µL of methanol:water (3:1) in the caps of 200 µL PCR tubes, with 4 µg/mL L-2-chlorophenylalanine as an internal standard. Tubes were then gently shaken (300 rpm, 10 min, 4 °C). Metabolite extraction in PCR tubes was conducted following the above-mentioned procedure for metabolomic analysis. The dried extract was dissolved in 50 µL of 90% acetonitrile, and transferred to glass vials for nanoLC–MS/MS analysis, using a reversed-phase nano-liquid chromatography system (UltiMate™ 3000 RSLC nano, Thermo Fisher Scientific, USA) and a high-resolution MS system (Orbitrap Fusion, Thermo Fisher Scientific, USA). The extract was eluted by a lab-made nano column packed with ChromCore HILIC Imidazole, and the elute was separated by an elution gradient: 87% B (acetonitrile with 0.1% ammonium hydroxide) and 13% A (water with 0.1% ammonium hydroxide) was initially firstly maintained for 5 min, then linearly decreased to 60% B from 5 to 20 min, and maintained at 60% B from 20 to 30 min, then linearly increased to 87% B from 30 to 31 min followed by equilibration at 87% B for 24 min. The eluents were subjected to MS analysis operating in the heated ESI mode. The spray voltage was set to −3000 V in negative ion modes. The ion-source parameters were set as described in our previous study [69]. Xcalibur version 4.6.67.17 (Thermo-Fisher Scientific, San Jose, CA, USA) was used for data analysis. The in-house library was established using authentic standards for metabolite identification.

snRNA-seq analysis

Five whole germinating seeds per sample were harvested at 18, 24, 36 and 48 HAI and immediately frozen in liquid nitrogen. Nuclei were isolated from seeds according to Sunaga-Franze et al. [70]. Briefly, approximately 1 g of germinating seeds was ground into fine powder under liquid nitrogen and transferred to a 50 mL centrifuge tube containing 10 mL of pre-cooled nucleus isolation lysis buffer (composition: 2.5% Ficoll 400, 5% Dextran T40, 0.4 M sucrose, 10 mM MgCl₂, 1 mM DTT, 0.5% Triton X-100, 1 tablet/50 mL Complete Protease Inhibitor Cocktail). The mixture was gently swirled and incubated on ice for 10 min, followed by centrifugation at 500 × g for 10 min at 4 °C. The resulting suspension was filtered through a 70-μm cell strainer (BD Falcon) and centrifuged at 1000 × g for 6 min at 4 °C. The pellet was carefully resuspended in 5 mL of lysis buffer and re-filtered through a 40-μm strainer. After centrifugation at 1000 × g for 5 min at 4 °C, the pellet was resuspended in 10 mL of wash buffer (composition: 10 mM 1 × PBS, 0.2 U/μL RNase inhibitor, 1% BSA). The nuclei solution was stained with 3 × staining buffer (composition: 12 µM DAPI, 0.4 U/µL RNase inhibitor in 1 × PBS) and analyzed using a flow cytometer (BD FACSAria II) equipped with a 70 μm nozzle (20 psi). The gate selection for DAPI-stained nuclei sorting was based on a small particle gate (FSC-A vs SSC-A mode) to remove macroaggregated impurities, and the final nucleus extraction concentration was adjusted to 1000 nuclei/μL. Sorted nuclei were additionally stained with NucBlue using the Invitrogen Ready Probes Cell Viability Imaging Kit (Thermo Fisher Scientific, USA), and their integrity was assessed and counted in Neubauer counting chambers. The quality of RNA extracted from sorted nuclei was evaluated using an Agilent Bioanalyzer 2100 system (Agilent Technologies Inc., USA). Nucleus suspensions co-stained with NucBlue and DAPI were loaded onto a 10 × Genomics Chromium controller to generate single-nucleus Gel Beads-in-Emulsion (GEMs). snRNA-seq libraries were constructed using the Chromium Single Cell 3’ GEM, Library & Gel Bead Kit v3 (10 × Genomics) per the manufacturer’s instructions. Libraries were sequenced paired-end on an Illumina NovaSeq X Plus platform (by Gene Denovo, Guangzhou, China). Two independent biological replicates were performed. The 10 × Genomics Cell Ranger software (v3.1.0) was used to convert raw BCL files to FASTQ, perform alignment, and quantify gene expression. Reads with low-quality barcodes or UMIs were filtered out, and only reads uniquely mapped to the transcriptome and overlapping an exon by ≥ 50% were counted. The maize B73 reference genome (Zm-B73-REFERENCE-NAM-5.0) was used for alignment. Prior to quantification, UMI sequences were corrected for sequencing errors, and cell barcodes were identified using the EmptyDrops method [71]. Cell-by-gene count matrices for each sample were generated and imported into Seurat v3.1.1 [72] for downstream analysis. The Harmony package [73] was applied to correct batch effects prior to clustering, followed by UMAP construction. Principal component analysis (PCA) on the scaled, integrated expression matrix was used for dimensionality reduction, and t-SNE visualization was generated after batch correction. Log-normalized matrices were loaded into the SingleR package for cell type annotation, correlating single-nucleus data with reference profiles of known nucleus types. Unsupervised nucleus clustering was performed using the Louvain algorithm (resolution 0.15; [74]), and marker genes were identified using Seurat’s FindAllMarkers (log2FC > 0.36, adjusted p < 0.01, pct.1 > 0.25). To refine cluster identities, clusters of the same cell type were merged for re-UMAP analysis, graph-based subclustering, and marker identification. Bioinformatic analyses were performed using the OmicShare platform [75].

TTC staining

Tetrazolium (TTC) staining was used to assess seed viability. Seeds at 48 HAI were stained following Li et al. [76]. Based on staining intensity, seed viability was categorized as strong, weak, or non-viable. The proportion of seeds in each category was calculated from three replicates of 50 seeds each.

GUS staining assay

GUS (β-glucuronidase) staining of pZmAGA1::GUS transgenic plants was carried out as described previously [77]. Briefly, germinating seeds and various tissues were incubated in GUS staining buffer (100 mM phosphate buffer pH 7.0, 0.5 mM K₃Fe(CN)₆, 0.5 mM K₄Fe(CN)₆, 10 mM EDTA, 0.1% Triton X-100, 1 mM X-Gluc) at 37 °C for 2 h. Samples were then destained with 70% ethanol (repeated 2–3 times) until control samples were completely clear. Stained samples were photographed with a Zeiss SteREO Lumar V12 stereomicroscope.

Amylase activity and starch content assays

For amylase activity and starch content assays comparing WT and ZmAGA1-KO lines, whole seeds at 18, 24, 36 and 48 HAI were collected and ground in liquid nitrogen. For each sample, 100 mg of powder was used to measure α-amylase activity using a commercial kit (D799324-0100, Sangon Biotech, China) and starch content assay kit (BC0705, Solarbio, China) according to the manufacturer’s protocols, respectively. Three biological replicates were performed per sample.

Western blot analysis of ZmAGA1 protein expression

A peptide antigen (ENTKFQTKPDGDG, corresponding to a segment in ZmAGA1; Fig. 2a) was synthesized and used to raise a Oryctolagus cuniculus (rabbit) polyclonal antibody against ZmAGA1 (Sangon Biotech, China). A plant actin antibody (D110007-0200, Sangon Biotech, China) was used as a loading control. Total protein was extracted from 100 mg of 48-HAI seed embryo tissue (WT and KO lines) using a plant total protein extraction kit (PE0230-1KT, Sigma-Aldrich, USA). Protein concentration was determined by the Bradford assay [78] on a Thermo Varioskan Flash multimode reader. For each sample, 40 µg of total protein was separated by 12% SDS-PAGE (after boiling in loading buffer) and transferred to a PVDF membrane. The membrane was blocked with 5% (w/v) nonfat milk in TBST (20 mM Tris–HCl pH 7.5, 150 mM NaCl, 0.1% Tween-20) for 1 h at room temperature, then briefly rinsed in TBST. The blot was incubated overnight at 4 °C with primary antibodies (anti-ZmAGA1, 1:5000 dilution; anti-Actin, 1:10,000 dilution). After three 5-min washes in TBST, the membrane was incubated for 1 h at room temperature with HRP-conjugated goat anti-rabbit secondary antibody (C650202-0100, Sangon Biotech; 1:5000 dilution). Signals were detected using a Chemistar High-sig ECL kit (Tanon, China) and imaged by using the BIO-RAD ChemiDoc XRS Imaging system. Actin served as a loading control.

qRT-PCR analysis

To characterize amylase gene expression profiles, five whole seeds at 18, 24, 36, and 48 HAI were collected and ground in liquid nitrogen. Total RNA was extracted from 100 mg of powder using the Column Plant RNAout 2.0 kit (TIANDZ, Beijing). cDNA synthesis and qPCR were performed as described [76, 79]. Gene-specific primers (Additional file 2: Table S5) were designed to avoid conserved regions, introns, and exon–exon junctions. Glyceraldehyde-3-phosphate dehydrogenase 1 (ZmGAPDH1; Zm00001eb173410) was used as an internal control. Relative expression levels were calculated by the 2^^–ΔΔCt method [80]. Three biological replicates were analyzed per gene. Primer sequences are provided in Additional file 2: Table S5.

Subcellular location analysis of ZmAGA1

The coding sequence (CDS) of ZmAGA1 was synthesized based on the B73 reference transcript and cloned into the binary vector pCAMBIA1301-ZmUbi::eGFP to generate pZmUbi::ZmAGA1–eGFP. Cloning was performed using the ClonExpress II kit (Vazyme, China) by homologous recombination. After sequence confirmation, pZmUbi::ZmAGA1–eGFP and the control pZmUbi::eGFP were introduced into Agrobacterium strain EHA105 and transiently expressed in Nicotiana benthamiana leaves, with nuclear localization directed by NLS-localization signaling, following the method of Zhou et al. [28]. GFP signals were visualized with a Leica LAS X confocal microscope after co-cultivation at 26 °C in darkness for 24 h followed by 48 h under 16 h light/8 h dark (10,000 lx). Primers for constructing pZmUbi::ZmAGA1–eGFP are listed in Additional file 2: Table S5.

Y2H assay for the interaction between ZmAGA1 and ZmSnRK1α

The full-length and truncated CDS of ZmSnRK1α, as well as the full-length CDS of ZmAGA1, were cloned into the pGADT7 and pGBKT7 vectors using ClonExpress II (Vazyme) and verified through DNA sequencing, respectively. Subsequently, the bait and prey constructs were co-transformed into Yeast Y2HGold Chemically Competent Cells following the protocol outlined in the Yeastmaker Yeast Transformation System 2 User Manual (Clontech). The plasmid pair pGADT7-T and pGBKT7-p53 was used as a positive control, while the pair pGADT7-T and pGBKT7-Lam served as a negative control. The transformed yeast cells were cultured on DDO (SD/-Leu/-Trp), TDO (SD/-Leu/-Trp/-His), and QDO (SD/-Leu/-Trp/-His/-Ade + X-α-Gal) selective media, respectively. Primers used for Y2H construct validation are provided in Additional file 2: Table S5.

BiFC assay for the interaction between ZmAGA1 and ZmSnRK1α

For bimolecular fluorescence (BiFC) complementation assays, full-length CDS of ZmAGA1 and ZmSnRK1α were cloned into pUC-SPYNE and pUC-SPYCE vectors (terminal YFP fragments), respectively, using ClonExpress II. The resulting constructs pCaMV 35S::ZmAGA1–VN and pCaMV 35S::ZmSnRK1α–VC were confirmed by sequencing. As a positive control, pCaMV 35S::bZIP63–VN and pCaMV 35S::bZIP63–VC (interacting pair) were constructed similarly. All vectors were transiently transformed into maize leaf protoplasts to test ZmAGA1–ZmSnRK1α interaction, as described by Coy et al. [81]. YFP fluorescence was observed using a Leica LAS X confocal microscope. Primers for BiFC constructs are listed in Additional file 2: Table S5.

LCA assay for the interaction between ZmAGA1 and ZmSnRK1α

Luciferase complementation assay (LCA), a sensitive method for detecting protein interactions in planta, was used to test ZmAGA1–ZmSnRK1α interaction. Full-length ZmAGA1 and ZmSnRK1α CDS were cloned into pCaMV 35S::NLuc and pCaMV 35S::CLuc vectors (N- or C-terminal luciferase halves) using ClonExpress II. The resulting pCaMV 35S::ZmAGA1–NLuc and pCaMV 35S::ZmSnRK1α–CLuc constructs were confirmed by sequencing. All binary vectors were introduced into Agrobacterium EHA105 and infiltrated into N. benthamiana leaves as described [28]. After incubation (26 °C, 24 h dark then 48 h 16 h light/8 h dark), luciferase signals were detected using a NightSHADE LB 985 imaging system (Berthold, Germany). Primers for LCA constructs are listed in Additional file 2: Table S5.

Co-IP assay for the interaction between ZmAGA1 and ZmSnRK1α

For co-immunoprecipitation (Co-IP) assays, the full-length coding sequences (CDS) of ZmAGA1 and ZmSnRK1α were cloned into the pSAT6-CaMV 35S-Flag and pSAT6-CaMV 35S-eGFP plasmids, respectively, using ClonExpress II (Vazyme). The resulting constructs, pCaMV 35S::ZmAGA1-Flag and pCaMV 35S::ZmSnRK1α–eGFP, were verified by DNA sequencing. The pCaMV 35S::ZmSnRK1α–eGFP construct was co-transfected into maize leaf protoplasts with either pCaMV 35S::ZmAGA1-Flag or the pSAT6-CaMV 35S-Flag empty vector, according to the protocol described by Coy et al. [58]. After 16 h, the protoplasts were harvested, and total proteins were extracted using 1 mL of protein lysis buffer with the Anti-Flag Nanobody IP Kit (Magarose) (KTSM1361, AlpaLifeBio Co. Ltd., China). The immunoprecipitated and input protein samples were separated by SDS–PAGE and analyzed using anti-Flag (D110005-0100, Sangon Biotech; 1:5000 dilution) and anti-GFP (D110008-0100, Sangon Biotech; 1:5000 dilution) antibodies. Protein signals were visualized using the Chemistar High-sig ECL Kit (Tanon, China) and captured using the BIO-RAD ChemiDoc XRS Imaging System. Primers used in the co-immunoprecipitation assay are listed in Additional file 2: Table S5.

Dual-luciferase reporter assay to assess the transcriptional regulatory activity of ZmMYBS1 on the ZmαAMY3 promoter

For dual-luciferase assays, A 2000-bp fragment upstream of the ZmαAMY3 transcription start site was cloned into the pGreenII 0800-Luc vector to generate the reporter construct pGreenII 0800-pZmαAMY3::LUC. Full-length ZmMYBS1 CDS was cloned separately into the pAN580 and pGreenII 62-SK vectors—both driven by tandem CaMV 35S promoters—to yield the effector constructs pAN580-2 × 35S::ZmMYBS1–GFP and pGreenII 62-SK-2 × 35S::ZmMYBS1. All vectors were transiently transformed into maize leaf protoplasts or introduced into Agrobacterium EHA105 and infiltrated into N. benthamiana leaves to evaluate ZmMYBS1-dependent regulation of the ZmαAMY3 promoter, as described previously [81]. Luciferase activity was quantified with Dual-Glo® Luciferase Assay System kit (Promega, Cat. No. E1910), and luminescence signals were measured with a NightSHADE LB 985 imaging system (Berthold Technologies, Germany). Primers for dual-luciferase assay assays constructs are listed in Additional file 2: Table S5.

Electrophoretic mobility shift assay (EMSA)

The full-length coding sequence of ZmMYBS1 was cloned into the pET28-MBP-TEV vector and expressed in Escherichia coli BL21(DE3). MBP and MBP-ZmMYBS1 fusion proteins were purified from the soluble fraction of bacterial lysates using the His-tag Protein Purification Kit (Beyotime, Cat. No. P2247S). The oligonucleotides to predicted ZmMYBS1-binding motifs were chemically synthesized (Sangon Biotech) with HPLC purification: M1: 5′-AGGAATATATCCATCCTATCCCTATAAAC-3′; M2: 5′-TACCTTTTTATCCCCCTCTTTAATG-3′; M3: 5′-CAACGGAGCTTATCCATCGTTAG-3′; and M4 (M1 mutant, TATCC → ATTCC): 5′-AGGAATAATTCCATCCATTCCCTATAAAC-3′. EMSAs were performed following our recently published laboratory protocol [82].

Statistical analysis

Data are presented as mean ± SE of three biological replicates. Two-tailed Student’s t-tests were used for comparisons between two groups, and one-way ANOVA for comparisons among multiple groups.

Authors' contributions

W. L. contributed to the conception of the study, designed and performed experiments, analyzed the data, and wrote the manuscript. W. H. contributed to untargeted and targeted metabolomics experiments, and writing parts of the manuscript. M. J. contributed to mGWAS, dual-luciferase reporter assay and writing parts of the manuscript. M. Y. performed MSI, LCM, and NanoLC-MS/MS analyses. Z. Z. performed the CoIP and LCA assay. Z. C. performed parts of Y2H experiments. Z. C. participated in subcellular localization assays, and W. Z. performed phosphoproteomic analysis. Y. C. participated in mGWAS. J. Y. provided the materials of maize CUBIC population. A. R. F. revised the manuscript. S. Y. contributed to the conception and design of the study, performed part of experiments and data analysis, data interpretation, organize and wrote the manuscript. All authors read and approved the final manuscript.

Funding

This work was jointly supported by the Natural Science Foundation of Guangdong Province (2023A1515011973, 2024A1515012901, 2023A0505090005, 2021TQ06N115, and GHZX2025-XZ01), the Special Fund for Scientific Innovation Strategy-construction of High-Level Academy of Agriculture Science (XT202213, GHZX2025-XZ01, and NYQS202628).

Data availability

All snRNA-seq and metabolomics datasets have been publicly deposited in the National Genomics Data Center (NGDC; https://www.cncb.ac.cn/) [83], with corresponding accession IDs available under the GSA accession number CRA034325 [84] and the OMIX accession number OMIX013362 [85], respectively. MS–based phosphoproteomics data generated in our laboratory were also used in this study, and its raw mass spectrometry data are available at the iProX repository under project accession number IPX0015314000 [36, 86]. Publicly available bulk RNA-seq data from the ZEAMAP database (https://db.cngb.org/zeamap/) [27].

Declarations

Ethics approval and consent to participate

Not relevant to our study.

Consent for publication

Not relevant to our study.

Competing interests

The authors declare no competing interests.