Integration of metabolomics and proteomics reveals the mechanisms of seasonal accumulation and key regulatory networks of polysaccharides in Polygonatum odoratum

Abstract

Background

Polysaccharides are the primary active components responsible for the immunomodulatory and other pharmacological effects of Polygonatum odoratum. The seasonal dynamics of this medicinal herb directly influence its quality assessment and harvesting strategies. However, the accumulation patterns of polysaccharides across different growth stages and their underlying molecular regulatory mechanisms remain poorly understood. This study aims to systematically decipher the spatiotemporal dynamics of polysaccharide accumulation in Polygonatum odoratum rhizomes via broad-target metabolomics and proteomics technologies combined with machine learning algorithms, thereby determining the optimal harvest period and revealing the underlying biological mechanisms involved.

Results

The polysaccharide content of Polygonatum odoratum rhizomes peaked in spring (13.3%) before declining through summer and fall. Multiomics correlation analysis revealed a distinct “source‒sink” conversion mechanism: spring presented significant upregulation of α-amylase (AMY) and invertase (INV), facilitating the conversion of overwintering starch reserves into polysaccharides and resulting in markedly elevated polysaccharide levels. In contrast, although summer presented the highest metabolic enzyme activities (e.g., significant upregulation of sucrose synthase (SUS) and UDP-N-acetylglucosamine pyrophosphorylase (UAP1)), metabolic flux was primarily directed toward cell wall construction (glycosaminoglycan synthesis) and putative immune defense, presenting a “high synthesis, high consumption” characteristic. This rationally explains the “spatiotemporal mismatch” between peak enzyme activity and peak polysaccharide accumulation. Random forest (RF) analysis prioritized UDP-glucuronic acid decarboxylase (UXS1-3) and AMY-2 as highly potential candidate proteins influencing polysaccharide synthesis. UXS1-3 expression was negatively correlated with polysaccharide content, suggesting that nucleotide sugar conversion is putatively associated with a key rate-limiting step. Furthermore, weighted correlation network analysis (WGCNA) revealed the transcription factor C3H-2 as a putative pivotal regulator highly associated with UXS1-3, while the AP2/ERF and NAC families participated in the environmental signal response and coregulation.

Conclusion

This study confirms that spring is the peak period for polysaccharide accumulation in Polygonatum odoratum, establishing scientific grounds for its optimal harvest timing. This study elucidates Polygonatum odoratum’s adaptation to seasonal changes through a metabolic trade-off strategy of “spring reserve mobilization and summer growth consumption”. The identified key enzymes (UXS1-3, AMY-2) and transcription factor (C3H-2) provide genetic resources for future molecular breeding targeting high polysaccharide contents and theoretical support for establishing biomarker-based precision harvesting strategies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12870-026-08796-0.

Introduction

Polygonatum odoratum (Mill.) Druce is a perennial herb that is widely used in traditional Chinese medicine for its various health benefits, including immune regulation and anti-inflammatory and antioxidant properties [1, 2]. The rhizomes of Polygonatum odoratum are highly valued as the primary medicinal component and contain various bioactive components, such as polysaccharides, steroidal glycosides, dipeptides, flavonoids, amino acids, and trace mineral elements. Among these components, polysaccharides are considered one of the primary active components responsible for their medicinal effects [3]. Polysaccharides play crucial roles in plant growth, development, and stress responses, with their contents varying significantly depending on the plant’s growth stage and environmental conditions [4]. As an important bioactive component in Polygonatum odoratum, polysaccharides exhibit immunomodulatory, antidiabetic, antiaging, antitumor, and antioxidant effects [5–7]. These pharmacological properties make them highly promising for applications in modern medicine and health supplements.

Polysaccharides are a class of complex carbohydrate compounds with high molecular weights formed by the dehydration of multiple monosaccharide molecules through condensation reactions. All carbohydrates and their derivatives that meet the definition of macromolecules can be classified as polysaccharides [8]. On the basis of the types of their constituent units, polysaccharides can be divided into two major categories: homopolysaccharides and heteropolysaccharides. Homopolysaccharides are composed of identical monosaccharide repeating units, such as starch and β-glucan; heteropolysaccharides, on the other hand, are composed of two or more different monosaccharides, such as glucuronic acid and pectin. Polygonatum odoratum polysaccharides are a typical example of heteropolysaccharides and are primarily composed of various monosaccharides, including mannose, galactose, glucose, fructose, rhamnose, arabinose, and galacturonic acid [9].

Research has shown that the synthesis and accumulation of Polygonatum odoratum polysaccharides involve the action of multiple key enzymes, such as UDP-glucose pyrophosphorylase (UGPase), sucrose synthase (SUS), and glucose transferases (GTs), which play crucial roles in the polysaccharide polymerization process [10]. Polysaccharide biosynthesis occurs through three main processes [5, 11]. First, sucrose is converted into glucose-6-phosphate and fructose by β-fructofuranosidase, and fructose is converted into fructose-6-phosphate by hexokinase and fructokinase [12]. Glucose-6-phosphate isomerase catalyzes the isomerization of glucose-6-phosphate to glucose-1-phosphate, while UDP-glucose (UDP-Glc) and GDP-mannose (GDP-Man) are produced from glucose-1-phosphate and fructose-6-phosphate precursors, respectively [12, 13]. Second, several NDP-sugar interconversion enzymes catalyze the conversion of UDP-Glc or GDP-Man into other NDP sugars [14]. Finally, different glycosyltransferases remove monosaccharides from sugar nucleotides. The donor is bound to growing polysaccharide polymers, and these repeating units polymerize and output to form plant polysaccharides [15, 16].

There are various methods for extracting Polygonatum odoratum polysaccharides, including hot water extraction, enzymatic extraction, and ultrasonic-assisted extraction [17, 18]. The traditional Polygonatum odoratum processing methods primarily include the “sugar-rubbing” method and steaming and roasting, both of which yield high contents of Polygonatum odoratum polysaccharides [19]. The modern processing method employed is ultrafine grinding, which significantly increases the polysaccharide content compared with traditional methods, providing new avenues for the extraction and application of Polygonatum odoratum polysaccharides [20, 21]. Subsequently, techniques such as ion exchange chromatography and gel permeation chromatography can be used to separate and purify Polygonatum odoratum polysaccharides further, yielding polysaccharide components with specific molecular weights and structures [22, 23].

This study aimed to investigate the seasonal dynamic changes in polysaccharide content in Polygonatum odoratum rhizomes and explore the metabolic and proteomic changes associated with different growth stages. By integrating metabolomics and proteomics methods, we seek to comprehensively understand the changes in polysaccharide biosynthesis and accumulation [24, 25]. This study also aimed to investigate the synthesis and regulatory mechanisms of secondary metabolites in Polygonatum odoratum at different growth stages, which may help optimize the harvest time to maximize the yield of polysaccharides and other major active components. To our knowledge, previous studies have rarely utilized multiomics integration methods to investigate changes in secondary metabolites in Polygonatum odoratum at different growth stages.

The results of this study lay the foundation for future genetic and metabolic engineering research on Polygonatum odoratum and other medicinal plants. By identifying key metabolic pathways and proteins involved in polysaccharide biosynthesis, this study opens new avenues for improving the quality and efficacy of Polygonatum odoratum as a medicinal resource.

Materials and methods

Plant materials

In a standardized experimental field in Nanling County, Wuhu city, Anhui Province (31°33′N, 118°38′E), the rhizomes of Polygonatum odoratum were planted for three years. Fresh Polygonatum odoratum rhizome samples were collected on February 2, 2022 (winter, abbreviated as WI), May 2 (spring, abbreviated as SP), August 2 (summer, abbreviated as SU), and November 2 (fall, abbreviated as FA). The samples were identified by Professor Hongwei Fu of Zhejiang University of Technology and stored at the School of Life Sciences and Medicine, Zhejiang University of Technology. The experimental field was managed with uniform fertilization, irrigation, and pesticide application measures. Polygonatum odoratum rhizomes were harvested in the third year after planting. At each sampling, 20 fresh rhizomes of Polygonatum odoratum were excavated from five fixed points (four corners and the central intersection point) in the experimental field. To ensure biological representation, minimize individual plant variation, and maintain statistical robustness, a composite sampling strategy was employed. The 20 collected rhizomes from each season were randomly divided into three independent groups (each group contained approximately 6 to 7 rhizomes). The rhizomes within each independent group were washed with purified water, cut into 0.5 cm pieces, pooled together as a single composite sample, rapidly frozen in liquid nitrogen, and stored at -80 °C.

Prior to subsequent metabolomic, proteomic, and biochemical assays, the pooled pieces from each respective group were ground into a homogenized powder in liquid nitrogen. Consequently, one true biological replicate was defined as the independent composite sample derived from these 6 to 7 pooled rhizomes. Each biological replicate was processed, extracted, and analyzed individually (without pooling across replicates).

To quantify the environmental conditions during the sampling periods, the monthly average temperature and precipitation data for the 2022 sampling site were retrieved from the Anhui Statistical Yearbook (2023). Furthermore, the specific photoperiod (day length) for each sampling date was astronomically calculated on the basis of the latitude of the experimental field (31°33′N) (Table S4).

Broad-target metabolomics analysis

Biological samples were first dried via a freeze dryer (Scientz-100 F) and then ground into powder via an MM 400 grinder (Retsch, China) [26]. Fifty milligrams of powder (using an MS105DM electronic balance, Sartorius, Germany) was accurately weighed, and 1.2 mL of prechilled 70% methanol aqueous solution containing an internal standard was added to reach − 20 °C. The sample was subjected to six vortex treatments followed by centrifugation at 12,000 rpm for 3 min at 4 °C, and the supernatant was collected and filtered through a 0.22 μm microporous membrane for subsequent UPLC‒MS/MS analysis. Metabolomic analysis was performed by Met-ware Biotechnology Co., Ltd. (Wuhan, China). The sample extracts were analyzed via a UPLC‒ESI‒MS‒MS‒MS system. The analytical conditions were as follows: UPLC column, Agilent SB-C18 (1.8 μm, 2.1 mm × 100 mm). The mobile phase consisted of solvent A (pure water with 0.1% formic acid) and solvent B (acetonitrile with 0.1% formic acid). Sample measurements were performed with a gradient program starting at 95% A and 5% B. Within 9 min, a linear gradient to 5% A and 95% B was programmed and maintained for 1 min. Subsequently, a composition of 95% A and 5.0% B was adjusted within 1.1 min and maintained for 2.9 min. The flow rate was set to 0.35 mL/min, the column oven temperature was 40 °C, and the injection volume was 2 µL. For mass spectrometry, the effluent was connected to an ESI-triple quadrupole-linear ion trap (QTRAP)-MS, and scans were acquired as multiple reaction monitoring (MRM) experiments with the collision gas (nitrogen) set to medium. To ensure data reliability and correct for instrument drift, quality control (QC) samples—prepared by pooling equal aliquots of all sample extracts—were analyzed repeatedly. Metabolite peak area integration was conducted for all the samples via mass spectrometry. Quantitative normalization was performed via internal standards, and metabolites with a coefficient of variation (CV) greater than 0.5 in the QC samples were strictly excluded from the final dataset to guarantee robust quantification.

Metabolite qualitative identification was achieved by matching exact mass spectrometry parameters (precursor ion Q1, product ion Q3, retention time RT, declustering potential DP, and collision energy CE) and MS/MS spectral matching scores against the self-built MetWare Database (MWDB) constructed from authentic standards. The identification confidence was strictly categorized into three levels analogous to the Metabolomics Standards Initiative (MSI) guidelines: Level 1 (equivalent to MSI Level 1) featured MS/MS fragmentation patterns and RT perfectly matching authentic standards with a spectral matching score > 0.7; Level 2 (equivalent to high-confidence MSI Level 2) exhibited a spectral matching score between 0.5 and 0.7; and Level 3 was assigned when Q1, Q3, RT, DP, and CE precisely matched the database records. To strictly control for false positives, differentially abundant metabolites (DAMs) were screened via a combination of multivariate and univariate statistical criteria: variable importance in projection (VIP) ≥ 1.0 (extracted from the partial least squares discriminant analysis (PLS-DA) / orthogonal partial least squares discriminant analysis (OPLS-DA) model), absolute Log2 fold change (FC) ≥ 1.0, and false discovery rate (FDR) < 0.05 (Table S5). Additionally, metabolic pathways associated with DAMs were retrieved from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database.

Determination of total polysaccharide content

The total polysaccharide content in the dried Polygonatum odoratum rhizomes was determined via the classic phenol‒sulfuric acid method combined with targeted ethanol precipitation [27, 28]. Briefly, approximately 0.2 g of the dried sample powder was accurately weighed and extracted twice with 5 mL of distilled water under ultrasonication for 1 h each time. The extracts were filtered, combined, concentrated, and adjusted to a constant volume of 20 mL. To isolate the crude polysaccharides, a 2 mL aliquot of this extract was precipitated by adding 10 mL of 95% ethanol. This classical ethanol precipitation step, which strictly follows the guidelines of the Chinese Pharmacopoeia for evaluating the medicinal quality of Polygonatum, is crucial for selectively precipitating macromolecular polysaccharides while effectively removing free low-molecular-weight carbohydrates (e.g., free monosaccharides and disaccharides) from the supernatant. After centrifugation, the purified polysaccharide precipitate was redissolved in distilled water and scaled to a final volume of 50 mL.

For the colorimetric assay, 0.4 mL of the diluted polysaccharide solution was thoroughly mixed with 0.2 mL of 4% phenol solution, followed by the rapid addition of 1.75 mL of concentrated sulfuric acid. The reaction mixture was incubated in a 40 °C water bath for 30 min and subsequently cooled in an ice bath for 5 min to terminate the reaction. The absorbance was measured at 490 nm via a UV spectrophotometer (JASCO Corporation, Japan). D-glucose was utilized as the reference standard. The calibration curve exhibited excellent linearity over the calibration range of 0.0024 ‒ 0.0072 mg, with the linear regression equation y = 65.833x ‒ 0.042 (where x represents the absolute mass of D-glucose in mg), and a highly robust correlation coefficient (R² = 0.9971). All samples from the four seasons (WI, SP, SU, and FA) were analyzed in independent biological triplicates. The data are expressed as the means ± standard deviations (SDs). Statistical significance across different seasons was determined via one-way ANOVA followed by Tukey’s post hoc test (p < 0.05).

Determination of starch content

To assess carbon reserve mobilization directly, the absolute starch content of the rhizomes was quantified via a starch content assay kit (Cat# BC0705; Beijing Solarbio Science & Technology Co., Ltd., Beijing, China) according to the manufacturer’s instructions. Briefly, approximately 0.03 g of the dried sample powder was homogenized and extracted to effectively separate and remove soluble sugars. The remaining precipitate was then gelatinized in a boiling water bath and strictly hydrolyzed into glucose. The resulting glucose concentration was determined via the classic anthrone colorimetric method, and the absorbance of the reaction mixture was measured at 620 nm via a microplate reader. The final starch content was calculated on the basis of a D-glucose standard curve and converted to a specific conversion constant via the kit. All samples across the four seasons were analyzed in independent biological triplicates, and the data are expressed as mg/g dry weight.

Proteomic analysis

For proteomic analysis, total proteins were extracted independently from the three biological replicates (as defined in the Plant Materials section) for each season via the acetone precipitation method. The specific steps were as follows: the composite powder of each independent biological replicate was ground in liquid nitrogen and homogenized, after which an extraction mixture containing 100 mM Tris-HCl (pH 7.6), 1 mM phenylmethylsulfonyl fluoride (PMSF), and 2 mM ethylenediaminetetraacetic acid (EDTA) was added. Next, the protein sample was boiled for 15 min, sonicated in an ice bath for 10 min to lyse the cells, and centrifuged to obtain a clear protein mixture. Four times the volume of frozen acetone was added to the protein mixture, which was then incubated at -20 °C overnight to precipitate the proteins. The precipitate was then collected by centrifugation at 4 °C. The precipitate was washed with cold acetone and dissolved in 8 M urea. Finally, the protein concentration was determined via the BCA method according to the kit instructions.

For each sample, an equal amount of protein was used for trypsin digestion. 8 M urea was added to the supernatant to a volume of 200 µL, which was then treated with 10 mM dithiothreitol (DTT) at 37 °C for 45 min for reduction, followed by treatment with 50 mM iodoacetamide (IAM) at room temperature in the dark for 15 min for alkylation. Next, four volumes of ice-cold acetone were added, and the peptide segments were precipitated at -20 °C for 2 h. After centrifugation, the protein precipitate was air-dried and resuspended in 200 µL of 25 mM ammonium bicarbonate solution. Three microliters of trypsin (trypsin-to-protein mass ratio of 1:50, Promega) was added, and the mixture was digested overnight at 37 °C. After digestion, the peptide segments were desalted via a C18 column (IonOpticks, Australia), dried in a vacuum concentrator, concentrated via vacuum centrifugation, and finally dissolved in 0.1% (v/v) formic acid solution.

The sample extracts were analyzed via a NanoElute UHPLC system coupled to a timsTOF Pro2 mass spectrometer (Bruker). Peptides were separated on an analytical column (IonOpticks, 25 cm × 75 μm, C18, 1.6 μm) at a flow rate of 300 nL/min. The mobile phase consisted of solvent A (0.1% formic acid in water) and solvent B (0.1% formic acid in acetonitrile). The gradient elution program was set as follows: 2–22% B for 0–25 min, 22–35% B for 25–30 min, 35–80% B for 30–35 min, and 80% B for 35–40 min. Mass spectrometry data were collected in diaPASEF mode [29]. For protein identification and label-free quantification (LFQ), the resulting MS/MS data were processed via DIA-NN software (v1.8.1) via the MaxLFQ algorithm. The database used for the search was a customized sequence database (MWXS-20-2311D-04-b_unigene.fa.transdecoder.pep.fasta) containing 25,217 sequences. The search was performed in library-free mode, and neural network-based deep learning was used to predict the theoretical spectral library. Match between runs (MBR) was enabled for protein quantification via the MaxLFQ algorithm. To ensure high-quality identification, both the precursor ion and protein-level FDRs were strictly filtered at < 1%, and a minimum of 1 peptide was required for valid protein identification.

Finally, to minimize potential batch effects and systematic biases, a two-step normalization strategy was employed. Initial cross-run normalization was inherently handled by the MaxLFQ algorithm during quantification. The extracted protein intensities were subsequently subjected to median normalization, where the intensity of each protein was divided by the median protein intensity within its respective sample (R_ij = I_ij/Median(I_i)), thereby yielding the relative quantitative values used for downstream differential analysis. Differentially expressed proteins (DEPs) across all groups were identified on the basis of stringent screening criteria: an FDR < 0.05 (adjusted via the Benjamini‒Hochberg method) combined with a FC > 1.5 or FC < 0.667.

Bioinformatics analysis

Weighted correlation network analysis (WGCNA)

WGCNA was performed on the differentially metabolized compounds and DEPs to identify coexpressed gene modules [30]. To ensure data quality prior to network construction, input features with excessive missing values (> 50% across samples) were excluded, and the remaining missing data were imputed via the k-nearest neighbors (k-NN) method. Furthermore, to reduce baseline noise, features with the lowest 10% variance were filtered out. Network topology analysis was then conducted to determine the optimal soft-thresholding power (β). A power of β = 18 was selected, which yielded a scale-free topology fitting index (R²) of 0.772. This parameter achieved an optimal balance between establishing a biologically robust scale-free network structure and maintaining adequate mean connectivity. This approach explored the association between gene networks and the target phenotype, identified core genes within the networks, and determined the transcription factors corresponding to each gene module. WGCNA was conducted via MetWare Cloud (a free online data analysis platform: https://cloud.metware.cn/) [31].

KEGG pathway enrichment analysis

Bioinformatics analysis of differentially expressed metabolites and proteins was performed via the KEGG biological pathway database to understand the biological pathways enriched with metabolites and proteins [26, 32–34].

Gene Ontology (GO) enrichment analysis

The identification of significantly enriched biological functions, molecular activities, or cellular localizations from DEPs helps explain the potential roles of these genes [34–36]. To address multiple hypothesis testing, all initial P values from the KEGG and GO enrichment analyses for both metabolomics and proteomics were strictly adjusted via the Benjamini‒Hochberg (BH) procedure. The enrichment results were evaluated on the basis of the FDR-adjusted P values (reported as corrected P values or FDRs in the Supplementary Tables).

PLS-DA analysis

The raw metabolic abundance data were normalized, log-transformed, and autoscaled. A PLS-DA model was subsequently constructed to assess metabolic differences among samples [37, 38]. Differentiating metabolites contributing significantly to intergroup discrimination were identified by calculating VIP values. PLS-DA analysis was performed via the MetWare Cloud. To evaluate the risk of model overfitting, a permutation test with 200 iterations was performed, and the R²X, R²Y, and Q² metrics alongside empirical p values were evaluated.

Machine learning analysis

The input dataset for model training and testing comprised 12 biological samples and 3864 features (representing the prefiltered multiomics variables). Although the feature space significantly exceeds the sample size—a common scenario in high-throughput omics studies—ensemble algorithms such as random forest (RF) are specifically suited for such high-dimensional data, aiming to minimize the risk of overfitting. To eliminate the influence of different variable dimensions, all the input features were first standardized via Z score normalization. The dataset was subsequently randomly split into training and test sets at an 8:2 ratio. The training set was used for model construction and parameter optimization, whereas the test set was employed solely to evaluate the model’s exploratory predictive performance.

This study constructed and compared multiple regression models, including decision tree, k-NN, RF, and neural network (NNET) models. To achieve optimal model performance, we employed 5-fold cross-validation combined with a grid search to optimize hyperparameters (e.g., the number of trees, maximum depth, and kernel function parameters) for each model. Model performance was quantified primarily via the root mean square error (RMSE) and coefficient of determination (R²). In addition to numerical metrics, we introduced visual residual analysis to validate the statistical assumptions: residual line plots were used to examine randomness and homoscedasticity, whereas residual boxplots were used to assess bias and outliers in the error distribution.

To enhance model interpretability, we extracted feature importance scores within each model. By comparing feature rankings across different algorithms, we highlighted putative key variables and prioritized potential core drivers influencing Polygonatum polysaccharide synthesis on the basis of feature weights from the optimal model.

To address potential overfitting and feature selection instability caused by the limited sample size and the single 8:2 train‒test split, a dual-layered validation strategy was implemented. First, the RF feature importance was extracted as a global average across hundreds of bootstrapped decision trees, intrinsically minimizing split-induced variance. Second, rather than relying exclusively on nonlinear predictive machine learning (ML) outcomes, a strict multialgorithm consensus approach was adopted. The ML-prioritized features were subjected to orthogonal validation via linear, unsupervised WGCNA and Pearson correlation networks to further support the biological relevance of the prioritized core drivers.

Quantitative real-time polymerase chain reaction (qRT‒PCR) analysis

The qRT‒PCR was employed to experimentally verify the transcript abundance of five critical targets derived from our WGCNA and ML predictions. These targets included four central metabolic enzymes (α-amylase (AMY-2), UDP-glucuronic acid decarboxylase (UXS1-3), SUS-2, invertase (INV-3)) and one hub transcription factor (C3H-2). Assays were conducted using Polygonatum odoratum rhizome tissues collected across four distinct phenological stages (WI, SP, SU, and FA). For RNA isolation, seasonal samples were homogenized, and total RNA was subsequently extracted via the SteadyPure Plant RNA Extraction Kit (Accurate Biotechnology, Hunan, China). Following purification, first-strand cDNA synthesis was carried out via the Evo M-MLV Reverse Transcription Premix Kit Ver. 2, which includes a gDNA removal step (Accurate Biotechnology, Hunan, China). Each 10 µL amplification reaction included 5 µL of 2× SYBR Green Pro Taq HS Premix (Accurate Biotechnology, Hunan, China), 1 µL of synthesized cDNA, 0.5 µL of both forward and reverse primers, and 3 µL of RNase-free water. The thermal cycling protocol was executed on a fluorescent quantitative PCR detection system (Bioer Technology, Hangzhou, China) with an initial denaturation step at 95 °C for 30 s. This step was followed by 40 amplification cycles consisting of 95 °C for 5 s and an annealing/extension step at 60 °C for 30 s. All analytical groups contained three independent biological replicates. To normalize the gene expression data, the P. odoratum Actin gene was used as the endogenous control. Details regarding the specific primer sequences are provided in Supplementary Table S3. The final relative transcription levels were determined via the standard 2^−ΔΔCt approach, with the winter (WI) samples used as the baseline calibrator (Fig. S2).

Results

Total polysaccharide content in the rhizomes of Polygonatum odoratum at different stages

Total polysaccharide extraction and content determination were performed on Polygonatum odoratum root samples collected across different seasons (Fig. 1A; Table 1). The results indicated that the total polysaccharide content was maintained at significantly high levels during both spring (13.31% ± 0.56%) and winter (12.67% ± 0.48%), with no significant difference between the two seasons (p > 0.05). Following this accumulation phase, the contents declined significantly (p < 0.05) during the rapid vegetative growth phases in summer (8.28% ± 0.39%) and fall (7.71% ± 0.30%). Analysis across four seasons (winter, WI; spring, SP; summer, SU; fall, FA) revealed significant seasonal dynamics in Polygonatum odoratum polysaccharide content.

Fig. 1.

Seasonal dynamics of carbohydrate reserves in cultivated Polygonatum odoratum rhizomes. A Polysaccharide content of cultivated Polygonatum odoratum. The x-axis represents seasons, and the y-axis represents polysaccharide content. B Starch content of cultivated Polygonatum odoratum. The x-axis represents seasons, and the y-axis represents starch content. DW, dry weight

Table 1.

Total polysaccharide content in the rhizomes of Polygonatum odoratum

Seasons	Polysaccharide Content(% DW)
WI	12.67 ± 0.48% a
SP	13.31 ± 0.56% a
SU	8.28 ± 0.39% b
FA	7.71 ± 0.30% b

To further validate the source of carbon for polysaccharide synthesis, we quantified the absolute seasonal starch reserves (Fig. 1B; Table 2). Compared with polysaccharide accumulation, starch content exhibited a striking inverse trend. Specifically, starch reserves were maintained at robustly high levels during summer (22.30 ± 3.43 mg/g), fall (22.86 ± 2.64 mg/g), and winter (22.20 ± 2.81 mg/g) but sharply and statistically significantly decreased in spring (13.35 ± 0.57 mg/g) (p < 0.05). This dramatic reduction in starch reserves directly coincided with the peak accumulation of total polysaccharides in spring, strongly supporting source-to-sink carbon reallocation.

Table 2.

Total starch content in the rhizomes of Polygonatum odoratum

Seasons	Starch Content (mg/g DW)
WI	22.20 ± 2.81 a
SP	13.35 ± 0.57 b
SU	22.30 ± 3.43 a
FA	22.86 ± 2.64 a

Polysaccharide analysis of Polygonatum odoratum at different stages

PLS-DA analysis of metabolites in Polygonatum odoratum rhizomes at different growth stages revealed complete separation of the four communities (WI, SP, SU, FA) on the planes of the first principal component (PC1, explaining 58.1% of the variance) and second principal component (PC2, explaining 38.4% of the variance), with no overlap (Fig. 2A) [39]. The figure indicates that the metabolites in the winter and fall Polygonatum odoratum samples are highly similar but differ distinctly from those in the spring and summer samples. To assess the reliability of this multigroup separation and evaluate potential overfitting, a 200-iteration permutation test was conducted (Fig. S1). The model successfully captured 96.5% of the metabolome variance (R²X = 0.965). While the absolute global predictive metric of this 4-class model was relatively low (Q² = 0.0223), both the R²Y (0.334) and Q² values were statistically significant (empirical p < 0.005), and all permuted Q² values were distributed below zero. This finding indicates that the observed distinct spatial separation among the four seasons is mathematically valid and primarily driven by genuine biological differences, suggesting that the model captures a meaningful biological trend despite the low global predictive power (Q²). These findings demonstrate that the metabolite composition of Polygonatum odoratum rhizomes undergoes dramatic and regular changes with seasonal growth. Seasonality (environmental temperature, light exposure, and plant physiological state) is the core factor driving these metabolic differences. The three biological replicates within each group (e.g., FA-1, FA-2, FA-3) clustered tightly, indicating exceptionally high data reliability. Intersample variations primarily stem from seasonal treatments rather than experimental errors. Heatmaps generated from polysaccharide-related metabolite data further revealed differences in the polysaccharide metabolome profiles across growth stages (Fig. 2B, Table S1). These metabolites primarily include monosaccharides, disaccharides, trisaccharides, tetrasaccharides, monosaccharide phosphates, glycosamines, sugar alcohols, glycosides, sugar phosphates, sugar lactones, sugar acids, glycolipids, and sugar esters [40]. Heatmap analysis revealed that most polysaccharide-related metabolites presented peak contents in SP, whereas some metabolites presented elevated levels in other seasons. Individual heatmap analysis of these metabolites revealed that disaccharide, glycoside, sugar ester, and sugar phosphate peaks occurred during WI, whereas sugar amine peaks occurred at high levels during SU (Fig. 2C-G). The group of metabolites comprising disaccharides, glycosides, glycosides, and glycosylphosphate esters peaked during WI, representing a “biochemical armament” for plants to endure severe cold. This includes regulating the intracellular osmotic pressure and accumulating metabolites to withstand the impacts of the winter environment. Glycosamines serve as key components of glycoproteins and cell wall polysaccharides (e.g., hemicellulose precursors). Rapid summer growth demands substantial amounts of glycosamine for the construction of new cellular tissues. During high temperatures and humidity in summer, when soil pathogens become active, glycosamines and their derivatives are hypothesized to function as pathogen-associated molecular patterns (PAMPs) or plant defense signals. Elevated glycosamine levels may help activate the plant immune system.

Fig. 2.

Analysis of dynamic changes in DAMs. C-G Heatmap of nonspring-enriched polysaccharide metabolites in Polygonatum odoratum. A PLS-DA of Polygonatum polysaccharides at different growth stages. B Heatmap of Polygonatum polysaccharides at different growth stages. C Disaccharide. D Glycosamine. E Glycosides. F Sugar ester. G Sugar phosphate

KEGG enrichment analysis of polysaccharide-related metabolites from Polygonatum odoratum revealed that the affected metabolites were significantly enriched in multiple pathways closely associated with carbohydrate metabolism (Fig. 3, Table S7). Specifically, pathways related to galactose metabolism, starch and sucrose metabolism, amino sugar and nucleotide sugar metabolism, and fructose and mannose metabolism not only presented high statistical significance (FDR < 0.05) but also presented elevated enrichment coefficients. Furthermore, the significant enrichment of glycolysis/gluconeogenesis and the pentose phosphate pathway suggested that Polygonatum odoratum polysaccharides may exert their biological effects by profoundly intervening in the core networks of energy metabolism and carbon source conversion processes within the organism.

Fig. 3.

KEGG enrichment analysis of polysaccharide-related metabolites in Polygonatum odoratum

Dynamic analysis of DEPs

On the basis of the screening criteria of FDR < 0.05, FC > 1.5, or FC < 0.667, DEPs across all groups were identified (Fig. 4A) [41, 42]. Quantitative analysis of DEPs from pairwise comparisons across growth seasons revealed significant proteomic dynamics in Polygonatum odoratum during different developmental stages. Among all comparisons, the summer versus spring (SU_vs_SP) group presented the most pronounced differences, with 1,721 DEPs identified—1,339 of which were upregulated—suggesting that summer represents the peak period for physiological metabolism and biosynthesis. In contrast, the spring versus winter (SP_vs_WI) comparison revealed a pronounced downregulation trend (909 downregulated proteins), potentially attributable to the depletion of stress-responsive and reserve-related proteins during the transition from dormancy to growth. Notably, the fall versus summer (FA_vs_SU) comparison yielded the fewest differential proteins, indicating a relatively gradual physiological transition during this phase. Furthermore, both summer and fall presented substantial protein upregulation relative to winter, further confirming the fundamental metabolic differences between active growth and winter dormancy. These findings indicate that Polygonatum odoratum adapts to seasonal phenological changes through protein expression regulation.

Fig. 4.

Dynamic analysis of DEPs. A Number of DEPs at different growth stages. B KOG functional annotation of DEPs. C KEGG enrichment analysis of DEPs. D KEGG enrichment analysis of DEPs

Identification and annotation of DEPs

To better investigate the mechanisms underlying polysaccharide synthesis and changes in Polygonatum odoratum, we performed proteomic profiling and analysis on rhizome samples from different growth stages. Using screening criteria of FDR < 0.05 and FC > 1.5 or FC < 0.667, a total of 3,320 DEPs were identified (Table S6). To further elucidate the functions of these DEPs, functional annotation was performed via the euKaryotic Orthologous Groups (KOG) database [43, 44], which categorizes the identified DEPs into 23 functional groups (Fig. 4B). The most abundant category was general function prediction only (381 DEPs), followed by posttranslational modification, protein turnover, chaperones (334 DEPs), energy production and conversion and translation, ribosomal structure and biogenesis (214 DEPs), carbohydrate transport and metabolism (193 DEPs), signal transduction mechanisms (160 DEPs), intracellular trafficking, secretion, and vesicular transport (156 DEPs), lipid transport and metabolism (117 DEPs), amino acid transport and metabolism, and secondary metabolite biosynthesis, transport and catabolism (114 DEPs). Notably, carbohydrate transport and metabolism accounted for a significant proportion, providing a functional explanation for the molecular basis of the seasonal fluctuations in Polygonatum odoratum polysaccharide content. Furthermore, many DEPs were enriched in pathways related to protein turnover and signal transduction, indicating that Polygonatum odoratum primarily maintains physiological homeostasis through increased protein quality control, energy metabolism restructuring, and the activation of intracellular signaling networks when adapting to seasonal phenological changes.

DEP enrichment analysis

To further elucidate the functional landscape of the DEPs and their involvement in metabolic regulatory networks, this study performed GO functional classification and KEGG pathway enrichment analysis (Fig. 4C, D, Table S8). The GO terms are divided into three categories: molecular function, biological process, and cellular component. The results revealed exceptionally high enrichment of DEPs in the “Metabolic Processes” and “Cellular Processes” categories under Biological Process (BP), as well as in the “Catalytic Activity” and “Binding” categories under Molecular Function (MF). These findings indicate that seasonal evolution has induced intense enzymatic reactions and shifts in biochemical homeostasis within Polygonatum odoratum. KEGG analysis precisely delineated the molecular pathways of this metabolic restructuring: carbon metabolism, which serves as a central hub, synergizes with the tricarboxylic acid (TCA) cycle and oxidative phosphorylation pathways to provide critical bioenergetic support for rapid plant growth. Concurrently, the significant enrichment of endoplasmic reticulum protein processing, starch and sucrose metabolism, and multiple secondary metabolite synthesis pathways collectively formed a putative molecular defense and developmental network through which Polygonatum odoratum responded to seasonal fluctuations. Notably, the biosynthesis pathways for cutin, suberin, and wax presented exceptionally high enrichment factors, suggesting that plants enhance stress resistance during specific growth stages by strengthening physical barrier structures. In summary, the functional enrichment patterns of the proteome profoundly reveal the molecular mechanisms by which Polygonatum odoratum achieves a precise balance between polysaccharide accumulation and environmental adaptability while maintaining energy homeostasis.

WGCNA

To further elucidate the molecular regulatory network underlying seasonal metabolic fluctuations in Polygonatum odoratum, this study employed WGCNA to analyze proteomic data. Through WGCNA, nine coexpression modules were identified and correlated with metabolite levels. The blue and red modules, which are significantly associated with metabolic traits, were isolated, and their constituent transcription factors (TFs) were systematically identified. The results (Fig. 5) indicate that the blue module has extremely significant positive correlations (r > 0.8) with core Polygonatum odoratum metabolite clusters. This module is enriched in core transcription factor families regulating development and stress responses, including NAC, AP2/ERF-ERF, and TCP, suggesting its potential involvement in facilitating polysaccharide backbone synthesis and responding to environmental rhythms. In contrast, the red module was significantly associated with another class of metabolites and contained numerous transcription factors from the C3H, Trihelix, and NF-Y families, which are potentially involved in the precise regulation of photosynthetic product allocation and secondary metabolic flux. By integrating WGCNA module characteristics with transcription factor classifications, this study constructed a “transcription factor–metabolism module” coregulation map. These findings provide insights into the putative molecular mechanisms by which Polygonatum odoratum achieves energy homeostasis and active compound accumulation across different growth stages through the spatiotemporal expression switching of multiple transcription factor families.

Fig. 5.

Heatmap of traits from WGCNA, with metabolites on the x-axis and WGCNA modules on the y-axis

Analysis of polysaccharide biosynthesis pathways

To analyze the molecular regulatory mechanisms underlying polysaccharide accumulation in Polygonatum odoratum in detail, this study integrated proteomics, metabolomics, and WGCNA correlation analysis to reconstruct the polysaccharide biosynthesis pathway (Fig. 6) [45]. Within this pathway, 44 DEPs associated with polysaccharide synthesis were identified, exhibiting distinct phased enrichment patterns following phenological rhythms (Table S2). Specifically, trehalose 6-phosphate phosphatase (otsB), α-amylase (AMY), invertase (INV), and 1,4-beta-D-xylan synthase (E2.4.2.24) were significantly upregulated in spring to mobilize stored starch for germination support; hexokinase (HK), INV, 1,4-alpha-glucan branching enzyme (GBE1), SUS, glutamine-fructose-6-phosphate transaminase (glmS), and UDP-N-acetylglucosamine pyrophosphorylase (UAP1) were significantly upregulated in summer to drive efficient conversion of photosynthetic products, providing ample precursors for polysaccharide synthesis; trehalose 6-phosphate synthase/phosphatase (TPS), starch synthase (glgA), GBE1, 4-alpha-glucanotransferase (malQ), and UDP-glucuronic acid decarboxylase (UXS1) were significantly upregulated in fall, marking a metabolic shift from “carbon consumption” to “structural polymerization and component modification”; and phosphoglucomutase (pgm), alpha-glucosidase (malZ), UDP-apiose/xylose synthase (AXS), and glucose-6-phosphate isomerase (GPI) were significantly upregulated in winter, ensuring physiological homeostasis and cold tolerance during dormancy.

Fig. 6.

Biosynthetic pathway of Polygonatum odoratum polysaccharides. The average expression is displayed in the heatmap, with solid lines representing single-step reactions and dashed lines indicating multistep reactions. treP, alpha-alpha-trehalose phosphorylase; otsB, trehalose 6-phosphate phosphatase; TPS, trehalose 6-phosphate synthase/phosphatase; pgmB, beta-phosphoglucomutase; pgm, phosphoglucomutase; glgC, glucose-1-phosphate adenylyltransferase; glgA, starch synthase; GBE1, 1,4-alpha-glucan branching enzyme; AMY, alpha-amylase, alpha-glucosidase; malQ, 4-alpha-glucanotransferase; HK, hexokinase; INV, beta-fructofuranosidase; SUS, sucrose synthase; UGDH, UDPglucose 6-dehydrogenase; UXS1, UDP-glucuronate decarboxylase; AXS, UDP-apiose/xylose synthase; E2.4.2.24, 1,4-beta-D-xylan synthase; galE, UDP-glucose 4-epimerase; GAUT, alpha-1,4-galacturonosyltransferase; USP, UDP-sugar pyrophosphorylase; GPI, glucose-6-phosphate isomerase; glmS, glutamine-fructose-6-phosphate transaminase; glmM, phosphoglucosamine mutase; glmU, bifunctional UDP-N-acetylglucosamine pyrophosphorylase / glucosamine-1-phosphate N-acetyltransferase; UAP1, UDP-N-acetylglucosamine/UDP-N-acetylgalactosamine diphosphorylase; CHS1, chitin synthase

Machine learning

This study screened and identified key proteins in the polysaccharide synthesis pathway of Polygonatum odoratum via decision tree, k-NN, RF, and NNET models. To evaluate the predictive accuracy and robustness of each model, a visual analysis of the model residuals was conducted (Fig. 7A, B). The residual boxplot clearly demonstrates that the RF model outperforms the other models (Fig. 7B). The RF model has the narrowest box height (interquartile range) and a median close to the zero line. This indicates that the prediction errors of the RF model are concentrated within a small range and show no significant systematic bias (overestimation or underestimation). Further analysis via the residual line chart revealed that the RF model’s residuals were tightly clustered around the zero line and exhibited good randomness without significant heteroscedasticity (Fig. 7A). In summary, owing to its comprehensive advantages in terms of error control and goodness-of-fit, the RF model was identified as the optimal model for feature prioritization for this study. A quantitative evaluation of the independent test set (20% of the dataset) further confirmed this selection, with the RF model achieving an RMSE of 0.288 and an R² of 0.556. Given the complexity and high dimensionality of the multiomics data paired with a limited sample size (n = 12), these metrics suggest that the model captured biologically relevant patterns. However, we strictly position this RF model primarily as an exploratory prioritization tool to filter high-confidence candidates, rather than as a robust generalized predictive framework. Consequently, the RF model was selected for subsequent core gene mining.

Fig. 7.

Error analysis and feature importance evaluation of machine learning models. A Residual scatter plot illustrating model fitting deviation and randomness. B Residual box plot comparing error distribution ranges and outliers across different models. C Feature importance map across multiple models validating the optimal model. D Feature importance ranking for the optimal model (RF), revealing key proteins influencing Polygonatum polysaccharide synthesis

Given the superior fitting performance and robustness of the RF model, as demonstrated by residual analysis, it was adopted as the benchmark for identifying key proteins influencing Polygonatum odoratum polysaccharide synthesis. To mitigate potential biases from a single algorithm, we first compared feature importance across all the models (Fig. 7C). The results revealed that proteins in the RF model presented higher feature importance scores than those in the other three models did.

Further analysis of the feature importance outputs from the RF model revealed a pronounced long-tail distribution in the contribution of each variable to the prediction results (Fig. 7D). The top three features were UXS1-3, AMY-2, and UXS1-1. Their importance scores significantly exceeded those of the other variables, suggesting their potential critical role in polysaccharide synthesis. The contribution of lower-ranked variables rapidly diminished, demonstrating the model’s reliance on a small number of key features for decision-making.

Correlation analysis

To further investigate the intrinsic relationships between key enzyme proteins and metabolic products, this study performed Pearson correlation analysis on 44 DEPs related to polysaccharide synthesis and 67 core metabolites to construct an interactive regulatory network (Fig. 8). Correlation analysis revealed that key enzymes, such as AMY, E2.4.2.24, HK, and GPI, were positively correlated with polysaccharide accumulation, suggesting that these enzymes might promote the biosynthesis of Polygonatum odoratum polysaccharides. However, key enzymes such as glgA, UXS1, TPS, and INV were negatively correlated with polysaccharide accumulation, suggesting that these enzymes may inhibit polysaccharide synthesis or participate in metabolic pathways opposing polysaccharide accumulation.

Fig. 8.

Pearson correlation analysis between DAMs and DEPs. Red circles represent proteins, and green circles represent metabolites. Red lines indicate positive correlations, and blue lines indicate negative correlations

To further elucidate the potential regulatory mechanisms of the key modules identified via WGCNA in Polygonatum odoratum polysaccharide synthesis, we performed a Pearson correlation analysis between the transcription factors (TFs) in the red and blue WGCNA modules and the 10 key DEPs selected via the RF model (Fig. 9). The network diagram results revealed distinctly different coexpression patterns among transcription factors in different modules. In the blue module, the transcription factors AP2/ERF-ERF and NAC were positively correlated with the proteins E2.4.2.24 and AMY-2, suggesting that these TFs might positively regulate their expression. Conversely, AP2/ERF-ERF, C2C2-LSD, and NAC presented negative correlations with the proteins INV-3, glgA, galE-3, and UXS1-4, suggesting potential negative feedback regulation or inhibitory effects to control polysaccharide synthesis. In the red module, TFs were positively correlated with most key proteins, suggesting that these TFs might positively regulate their expression to promote Polygonatum odoratum polysaccharide synthesis. Notably, we identified a potential hub regulator—C3H-2. This transcription factor not only belongs to the red module highly correlated with the trait but also exhibits an extremely strong correlation (r = 0.91, p < 0.01) with UXS1-3, the top-ranked key protein in the RF model.

Fig. 9.

Pearson correlation analysis between transcription factors corresponding to the blue and red WGCNA modules and DEPs selected by the RF model. A represents the blue module, and B represents the red module. Red circles denote proteins, and green circles denote transcription factors. Red lines indicate positive correlations, and blue lines indicate negative correlations

Validation of core predictive targets by qRT‒PCR

To validate the reliability of the core regulatory hubs identified by the RF model and WGCNA, qRT‒PCR was performed on five critical genes across the four seasonal developmental stages (Fig. S2). These included key metabolic enzymes (AMY-2, UXS1-3, SUS-2, INV-3) and the predicted hub transcription factor (C3H-2).

Consistent with our multiomics observations of massive starch mobilization, the transcript level of AMY-2 was significantly upregulated 2.58-fold in spring (SP) compared with that in winter, followed by a sharp decline in summer and fall. Similarly, SUS-2 and INV-3, which are vital for sucrose metabolism and carbon flux allocation, exhibited prominently elevated expression in spring (2.18-fold and 8.72-fold, respectively), further supporting early-stage “source-to-sink” carbohydrate conversion. Interestingly, INV-3 displayed a massive secondary peak in the fall (20.08-fold), likely reflecting a metabolic shift toward the accumulation of simple soluble sugars for overwintering osmotic protection.

Crucially, the top-ranked key enzyme identified by the ML model, UXS1-3, and its WGCNA-predicted highly coexpressed hub transcription factor, C3H-2, demonstrated remarkably synchronized expression trajectories. Both genes reached their absolute peak expression in the fall (FA), with UXS1-3 and C3H-2 significantly upregulated by 4.41-fold and 6.71-fold, respectively. This synchronized transcriptional burst perfectly corroborates our bioinformatics prediction that the fall season marks a critical metabolic transition toward structural polysaccharide modification. Overall, the in vivo transcriptional dynamics highly agreed with the in silico multiomics predictions, supporting the reliability of the inferred regulatory network.

Discussion

The accumulation of plant secondary metabolites exhibits remarkable spatiotemporal specificity, reflecting not only intrinsic developmental programs but also adaptive strategies for surviving environmental fluctuations [46, 47]. Metabolomic analysis in this study revealed dramatic seasonal restructuring of the metabolic profile of Polygonatum odoratum rootstocks, revealing a metabolic trade-off mechanism between “survival defense” and “growth accumulation.”

The distinct separation of the winter metabolic profile is driven primarily by the accumulation of small-molecule soluble sugars (such as disaccharides, glycosides, and sugar phosphates). This phenomenon aligns closely with the physiology of plant cold tolerance: under low-temperature stress, Polygonatum odoratum accumulates high concentrations of osmotic regulators to lower the cellular water potential and freezing point, thereby protecting the integrity of the membrane system. This “biochemical armament” strategy is crucial for surviving severe cold [48]. Conversely, during winter dormancy and the subsequent spring germination, the polysaccharide content is maintained at its highest annual level (12.67% and 13.31%, respectively), accompanied by significant increases in monosaccharides and polysaccharide-related metabolites. This indicates that the plant breaks dormancy, shifting its metabolic focus from “defense maintenance” to “substance storage” and “germination initiation”. Importantly, our quantified meteorological data revealed that the sampling site experienced an extreme regional heatwave and severe drought in August 2022 (monthly average 31.8°C; precipitation 12.0 mm; Table S4). Under such severe environmental stress, Polygonatum odoratum was forced to drastically redirect its carbon flux toward the synthesis of glycosamine and its derivatives. As a precursor for cell wall construction and a stress-signaling molecule, the significant increase in glycosamine content suggests that the plant might adopt a putative dual strategy of ‘structural consolidation’ and ‘immune activation’ during summer to prevent excessive water loss and defend against stress-induced damage. As a precursor for cell wall construction and a signal for recognizing PAMPs, the significant increase in glucosamine content suggests that Polygonatum odoratum might adopt a putative dual strategy of “structural consolidation” and “immune activation” during summer [49, 50]. This approach may help defend against pathogen invasion while supporting rapid above-ground growth. Thus, carbon allocation in Polygonatum odoratum rhizomes follows a precise seasonal logic: winter prioritizes osmotic protection, spring emphasizes polysaccharide storage, and summer focuses on structural development and immune defense.

A core finding of this study is the significant temporal mismatch between the peak phenotypic accumulation of Polygonatum odoratum polysaccharides (spring) and the peak protein expression of synthase enzymes (summer and fall). Proteomic data revealed that the greatest number of DEPs occurred during summer, and these DEPs were predominantly upregulated and covered core pathways such as carbon metabolism and starch and sucrose metabolism. This seemingly contradictory phenomenon of “high enzyme activity, low accumulation” profoundly reveals the dynamic relationship between “metabolic flux” and “metabolic pools.”

The high polysaccharide content in spring is essentially the result of seasonal shifts in the “source-pool” relationship. The significant upregulation of AMY and INV in spring indicates that starch stored over winter is extensively mobilized and degraded [51–53]. This omics-based hypothesis is now directly supported by our physiological quantitative data, which demonstrated a drastic and statistically significant depletion of starch reserves from winter (22.20 ± 2.81 mg/g) to spring (13.35 ± 0.57 mg/g) (Fig. 1b; Table 2). The resulting intermediates, such as glucose-1-phosphate, rapidly enter the sugar nucleotide pathway and flow toward pharmacologically active nonstarch polysaccharides. Thus, the spring polysaccharide peak arises from “reserve mobilization” rather than solely from “seasonal synthesis.”

In contrast, the high expression of enzymes such as SUS, HK, and UAP1 detected in summer reflects the high-throughput transport of photosynthetic products to roots and stems. However, as summer is a period of rapid biomass accumulation, these carbon sources are rapidly metabolized toward cell wall synthesis (structural polysaccharides) and energy consumption, maintaining a dynamic equilibrium of “high synthesis and high consumption” [54–56]. Consequently, the static accumulation of polysaccharides is actually lower than that in spring. The activation of UXS1 and glgA in fall signifies a metabolic shift from “quantitative expansion” to “qualitative modification,” potentially facilitating structural modification and polymerization of polysaccharide chains. This phased regulation of production, storage, and mobilization constitutes the survival mechanism by which Polygonatum odoratum adapts to its perennial biocyclic rhythm.

To decipher the core drivers of polysaccharide synthesis within complex enzymatic networks, this study innovatively employed an RF algorithm as an exploratory prioritization tool. While the relatively small sample sizes inherent to multiomics studies raise legitimate concerns regarding overfitting and feature selection stability via standard train‒test splits, our study employed a dual-layered defense approach to enhance methodological rigor. First, the intrinsic ensemble nature of the RF algorithm—which aggregates feature importance across hundreds of bootstrapped decision trees—effectively neutralizes the variance from any single data split, yielding highly stable feature rankings. The key features identified by the model—UXS1-3, AMY-2, and UXS1-1—provide precise molecular targets for understanding polysaccharide accumulation. Second, and most importantly, to strictly prevent overfitting, the robust identification of these targets was validated orthogonally: the nonlinear RF predictions perfectly converged with the linear, unsupervised WGCNA and Pearson correlation networks. This strict multialgorithm consensus provides further support that UXS1-3 and AMY-2 are highly likely to be biologically relevant rather than statistical artifacts.

Furthermore, targeted qRT‒PCR provided the ultimate orthogonal validation for these ML-selected nodes. The transcript level of AMY-2 was upregulated 2.58-fold in the spring to drive massive starch degradation, whereas the expression of UXS1-3 peaked dramatically in the fall, with a 4.41-fold increase in the redirection of nucleotide sugar conversion. This profound consistency between the in silico predictions and in vivo transcriptional dynamics supports the prioritization of our feature selection.

The high feature importance of AMY-2 further supports the critical contribution of carbon flux from starch degradation to polysaccharide accumulation. More importantly, the isoenzyme of UXS1-3 was identified by the RF model as a critically important node and was significantly negatively correlated with total polysaccharide yield. UXS1 catalyzes the conversion of UDP-glucuronic acid (UDP-GlcA) to UDP-xylose (UDP-Xyl) in the nucleotide sugar pathway. While our current phenol‒sulfuric acid assay only quantifies total macromolecular carbohydrate yield and cannot resolve monosaccharide composition, the omics-driven identification of UXS1-3 allows us to formulate a novel hypothesis. We hypothesize that the peak expression of UXS1-3 in the fall represents a metabolic shift prioritizing the synthesis of specific UDP sugars (such as UDP-Xyl) for structural heteropolysaccharides over bulk storage polysaccharide accumulation. This substrate competition may explain the observed decrease in total yield, suggesting a potential trade-off between polysaccharide “quantity” and “structural complexity”.

The operation of metabolic pathways is governed by precise upstream transcriptional regulatory networks. WGCNA was used to construct a cascade regulatory model linking “environmental signals–transcription factors–metabolic enzymes.” The blue module revealed the molecular pathway through which plants respond to environmental signals: the positive correlation between AP2/ERF and NAC transcription factors with E2.4.2.24 and AMY-2 indicates that these transcription factors, after sensing seasonal signals (light and temperature), may synergistically activate genes involved in starch mobilization and cell wall polysaccharide synthesis, achieving synchronization between energy release and structural material synthesis.

Notably, the pivotal regulator C3H-2 identified in the red module exhibited an exceptionally strong positive correlation (r = 0.91) with the key enzymes UXS1-3. The C3H zinc finger protein family is known to participate in the synthesis of cell wall components [57]. The high coexpression of C3H-2 with UXS1-3 suggests a putative regulatory linkage potentially involved in modulating the conversion efficiency of nucleoside sugars and the structural modification of polysaccharides. Furthermore, the negative correlations observed between certain TFs and key enzymes suggest the potential presence of negative feedback regulatory mechanisms within Polygonatum odoratum [58]. These putative mechanisms might prevent the depletion of essential carbon sources required for survival during extreme conditions (such as summer heat) by inhibiting excessive polysaccharide synthesis. This balance maintains energy homeostasis between growth and stress resistance.

While this study systematically delineates the seasonal metabolic shifts in Polygonatum odoratum, several limitations must be acknowledged.

First, regarding environmental and geographical constraints, the observed metabolic dynamics are a synergistic outcome of the intrinsic developmental phenology of plants—which demonstrates interannual stability, as supported by our previous 3-year tracking study [59]—and specific external climatic variations (such as the extreme summer heatwave quantified in this study). Given our single-location experimental design, future multisite validations are necessary to decouple universal phenological rhythms from regional climate effects. Second, concerning analytical and biochemical limits, the phenol‒sulfuric acid method can be used to evaluate the total macromolecular carbohydrate yield but cannot be used to resolve exact monosaccharide compositions or fine structural variations. Furthermore, while our integrated omics, static starch measurements, and qRT‒PCR establish robust molecular evidence for this seasonal carbon reallocation model, we acknowledge that transcript and protein abundances do not strictly equate to real-time in vivo enzymatic flux. The current study lacks real-time dynamic measurements, such as in vitro kinetic assays on fresh tissues, in vivo 13 C-isotopic labeling, and continuous respiration monitoring. Since respiration is the primary driving force for carbon consumption during rapid vegetative growth, the absence of these dynamic data limits our ability to precisely quantify the exact partitioning ratio and kinetic fluxes between energy metabolism and polysaccharide accumulation. Future targeted physiological studies employing real-time fluxomics will be essential to definitively map the precise kinetic rates of this source‒sink conversion. Finally, it is imperative to clarify that the WGCNA and Pearson correlation networks serve as heuristic filters to prioritize high-confidence candidates from a massive omics background. These networks represent putative coexpression associations susceptible to temporal confounding rather than proven causal regulatory events.

Future investigations will incorporate targeted monosaccharide composition analysis and advanced structural characterization to precisely resolve specific polysaccharide compositions and fine structures (e.g., branching degree, molecular weight) and employ real-time fluxomics to quantify dynamic carbon partitioning. Furthermore, in vivo functional assays (such as promoter binding assays, transient expression, and CRISPR/Cas9 technologies) will be utilized to biologically validate and definitively confirm the causal regulatory relationships of the identified hub genes (e.g., the C3H-2 and UXS1-3 axes). Additionally, combining climate simulation experiments will help investigate the spatiotemporal mechanisms by which light‒temperature environments regulate key enzymes (AMY-2, UXS1-3) through the AP2/ERF and NAC transcription modules, revealing the ecological basis for authentic quality formation. Ultimately, by deciphering the temporal misalignment between metabolic flux and substance accumulation, this study provides unprecedented molecular targets (AMY-2, UXS1-3, and C3H-2) for the precise molecular breeding of high-polysaccharide germplasms. The integration of these multiomics insights establishes a crucial theoretical and practical foundation for developing rapid detection kits based on identified biomarkers (e.g., AMY-2 and glycosamines) to establish precision harvesting strategies, thereby achieving synergistic improvements in both the yield and authentic medicinal quality of P. odoratum.

Conclusion

This study employed broad-target metabolomics and proteomics technologies to systematically elucidate the seasonal mechanisms governing polysaccharide accumulation in Polygonatum odoratum rhizomes. The results indicate that both winter and spring represent peak accumulation periods (maintained at approximately 13%), confirming that late winter to early spring is the optimal harvest season. Multiomics correlation analysis revealed that polysaccharide accumulation follows a “source‒sink” conversion and seasonal metabolic trade-off strategy. The elevated polysaccharide content in spring primarily stems from the mobilization of overwintering starch reserves, where significant upregulation of AMY and INV enzymes facilitates the conversion of starch into medicinal polysaccharides. In contrast, summer presented the highest metabolic enzyme activities (e.g., SUS, UAP1) but directed metabolic flux toward cell wall construction (glycosaminoglycans) and putative immune defense, presenting a “high synthesis, high consumption” pattern. This rationally explains the “spatiotemporal misalignment” between peak enzyme activity and peak substance accumulation. Through machine learning and WGCNA, this study prioritized key targets regulating polysaccharide synthesis. UXS1-3 and AMY-2 were identified as the most decisive key enzymes. Specifically, UXS1-3 exhibited a negative correlation with polysaccharide accumulation, suggesting a putative association where nucleotide sugar conversion acts as a potential rate-limiting step in polysaccharide synthesis. Concurrently, the transcription factor C3H-2 was identified as a highly correlated putative regulator of UXS1-3, while the AP2/ERF and NAC families are hypothesized to synergistically respond to environmental cues. Ultimately, this study not only elucidates the spatiotemporal molecular mechanisms underlying polysaccharide accumulation in P. odoratum but also provides unprecedented molecular targets for precise molecular breeding and establishes a robust theoretical foundation for biomarker-based precision harvesting strategies.

Authors’ contributions

Conceptualization, YZ and SL; methodology, YZ; software, SL and DF; validation, YZ, SL and DF; formal analysis, YZ; investigation, YZ and SL; resources, HF; data curation, YZ and SL; writing—original draft preparation, YZ; writing—review and editing, HF and HZ; visualization, YZ; supervision, HF and HZ; project administration, HF and HZ. All the authors read and approved the manuscript.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data availability

All of the datasets supporting the results of this article are included within the article and its additional files. The original contributions presented in the study are publicly available. These data can be found here: https://proteomecentral.proteomexchange.org/cgi/GetDataset?ID=PXD058887.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.