The field experiment was conducted in Xixiangtang District, Nanning City, China, at a precise location of 22°50′48″N, 108°7′16″E, and an altitude of 80 m. The purplish soil in the study area exhibited an organic matter content of 22.57 g kg⁻¹, an alkaline hydrolysis nitrogen content of 80.42 mg kg⁻¹, an available phosphorus content of 15.70 mg kg⁻¹, and an available potassium content of 149.50 mg kg⁻¹, with a pH of 7.84. During planting, the mean temperature was from 24°C to 32°C, and the mean rainfall was 87.2 mm. The organic fertilizers were applied at the dosages of 2250, 4500, and 6750 kg hm⁻², designated as YJF150, YJF300, and YJF450, respectively. The compound fertilizers were applied at the dosages of 300, 450, and 600 kg hm⁻², designated as FHF20, FHF30, and FHF40, respectively. Each treatment was with three replicates, and the control group (CK) had no fertilization treatment (Fig. 1). The organic fertilizer was purchased from Yuantaifeng (Baotou) Biotechnology Co., Ltd., Baotou, China (N + P₂O₅ + K₂O ≥ 4%, organic matter ≥ 30%). The compound fertilizer was bought from Stanley Agricultural Group Co., Ltd. (N + P₂O₅ + K₂O ≥ 45%). On 22 April 2022, the P. palustre cutting seedlings purchased from a commercial company (Guangxi Runda Biotechnology Co., Ltd., Lingshan County, Qinzhou City, China) were transplanted into the field at a planting density of 45,000 plants per hectare. Throughout the growth period, routine field management practices were implemented.

The whole P. palustre plants were sampled (collected) on 20 June 2022 (about two months after transplanting), to estimate the yield and quality. For each treatment, we collected the data of the fresh weight, shade dry weight, and dry weight. Five plants were randomly collected per treatment. 5 fresh plants were randomly harvested and weighed to record their fresh weight. Randomly harvested 5 fresh plants were dried in the shade for 7 days, and then weighed their shade dry weight. 5 fresh plants were placed in an oven set at a temperature of 60°C until a constant weight was achieved to accurately measure and record their dry weight. The dried plants were crushed into powder with a mill for subsequent metabolome analysis.

Metabolite extraction was performed by combining 50 mg samples with a 6 mm diameter grinding bead, 400 μL of extraction solution (methanol:water = 4:1 (v:v)) that included 0.02 mg/mL of internal standard (L-2-chlorophenylalanine) in a 2 mL centrifuge tube. The sample solution was ground using a frozen meat grinder at a temperature of −10°C and a frequency of 50 Hz for 6 min, followed by extraction using cryogenic ultrasound at 5°C and 40 kHz for 30 min. Subsequently, the processed samples were stored at −20°C for 30 min and then centrifuged at 4°C and 13,000 g for 15 min to separate the fractions. The resulting supernatant obtained after centrifugation was carefully transferred to sample vials to undergo machine analysis.

We used a Q Exactive HF-X ultra-high-performance liquid chromatography-tandem Fourier transform mass spectrometry system, manufactured by Thermo Fisher (USA) for LC-MS analysis. LC-MS analysis was performed following the method of Liu et al. [26]. The 2 μL sample was separated by the HSS T3 chromatography column (100 mm × 2.1 mm i.d., 1.8 µm) and then subjected to mass spectrometry detection. Mobile phase A composed of a mixture of water and ethyl acetate in a ratio of 95:5 (v/v), while mobile phase B consisted of a mixture of ethyl acetate, isopropanol, and water in a ratio of 47.5:47.5:5 (v/v). Both mobile phases A and B were supplemented with 0.1% formic acid. The flow rate was set at 0.40 mL/min, while the column temperature was controlled at a temperature of 40°C.

The Thermo UHPLC-Q Exactive HF-X Mass Spectrometer, equipped with an electrospray ionization (ESI) source and operating in positive/negative ion scanning mode with a mass range of 70–1050 m/z, was utilized to analyze the samples. Mass spectrometry conditions were set as follows: the sheath gas and auxiliary gas flow rates were maintained at 50 and 13 psi, respectively, while the temperature of the auxiliary gas was heated to 425°C. The positive ion spray voltage and negative ion spray voltage were adjusted to 3500 and −3500 V, respectively. The auxiliary gas temperature and ion transfer tube temperature were 425°C and 325°C, respectively. The normalized collision energy was adjusted to 20, 40, and 60 V in a normalized manner during the experiment. The primary mass spectrum was obtained with a high resolution of 60,000, while the secondary mass spectrum had a lower resolution of 7500. Both spectra were acquired using a data-dependent acquisition (DDA) mode.

The MS and MSMS spectra were matched against the Human Metabolome Database (HMDB, http://www.hmdb.ca/, accessed on 04 November 2022), Metlin (https://metlin.scripps.edu/, accessed on 04 November 2022), and Majorbio’s proprietary database to retrieve metabolite information. After querying the database, the data matrix is uploaded to the Majorbio Cloud Platform (https://cloud.majorbio.com, accessed on 04 November 2022) for subsequent data analysis.

We preprocessed the raw data and validated the overall data applicability by estimating the relative standard deviation (RSD) of a certain ion in the quality control (QC). The Spearman correlation between samples was calculated using the ‘cor’ function in the R package “ropls” (Version 1.6.2) and presented in a heat map. Furthermore, we utilized the R package “ropls” (Version 1.6.2) to perform Principal Component Analysis (PCA) and Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA). The model’s stability was evaluated by using a 7-fold cross-validation. The selection of DMs was determined according to the Variable Importance in the Projection (VIP) scores and p-values from the Student’s t-test within the OPLS-DA model. Metabolites with a VIP value greater than 1 (VIP > 1) and a p-value less than 0.05 (p < 0.05) were identified and designated as DMs. The significance of the difference was assessed using analysis of variance (ANOVA) with SPSS 8.0 software (version 22, IBM).

It can be seen from Fig. 2 that the utilization of compound fertilizer and organic fertilizer could significantly promote the fresh weight, dry weight, and shade dry weight of P. palustre compared to the control (p < 0.05). Among the compound fertilizer treatments, the fresh weight, shade dry weight, and dry weight of P. palustre treated with 450 kg hm⁻² compound fertilizer were the highest, which were 71.0%, 91.0%, and 61.5% higher than the control, respectively (Figs. 2A–2C). Among the organic fertilizer treatments, the fresh weight of P. palustre treated with 6750 kg hm⁻² organic fertilizer presented the highest, while the highest shade dry weight and dry weight were found in the 4500 kg hm⁻² treatment. Among the organic fertilizer treatments, the fresh weight, shade dry weight, and dry weight of P. palustre were no significant differences (p > 0.05) (Figs. 2D–2F). In summary, it could be concluded that the 450 kg hm⁻² compound fertilizer and 4500 kg hm⁻² organic fertilizer were the two optimal fertilization treatments. The fresh weight of P. palustre exhibited a statistically significant difference between the two treatments (p < 0.05), whereas no significant difference was observed in either the shade dry weight or the dry weight, as depicted in Fig. 3.

To conduct a comparative analysis of the impact of organic fertilizer and compound fertilizer on the quality and metabolites of P. palustre, we performed untargeted metabolomics on the two selected treatments: 450 kg hm⁻² compound fertilizer (FHF30) and 4500 kg hm⁻² organic fertilizer (YJF300). The metabolomics analysis identified a total of 1096 metabolites, with 506 detected in negative ion mode and 590 in positive ion mode.

The RSD of the overall data was less than 30%, and data with a cumulative proportion of peaks greater than 70% were qualified and could be used for subsequent analysis (Fig. S1A). Correlation analysis revealed that the correlation coefficients were consistent within the FHF30 and YJF300 treatment groups, but varied between the groups. The correlation coefficients of the samples were approaching 1, suggesting a strong similarity in metabolite expression across samples, indicating a positive correlation (Fig. S1B).

PCA revealed that the two principal components PC1 (50.30%) and PC2 (8.96%) could well indicate the main characteristic information of the samples (Fig. S2A). Venn analysis showed 590 common metabolites in positive ion mode and 506 in negative ion mode between FHF30 and YJF300 (Fig. S2B). The PLS-DA intuitively revealed that the classification effect of the model was significant (Fig. S2C). Permutation testing uncovered that the original model demonstrated strong effectiveness and reliability. Additionally, the PLS-DA model exhibited excellent predictive ability without suffering from over-fitting issues (Fig. S2D).

After data processing, 885 metabolites were obtained and annotated into the HMDB, and these metabolites were matched and classified into 14 categories, including 352 lipids and lipid-like molecules, 119 organic acids and derivatives, 111 organoheterocyclic compounds, 108 organic oxygen compounds, 86 phenylpropanoids and polyketides, 66 benzenoids, 18 organic nitrogen compounds, 16 nucleosides, nucleotides and analogues, 3 hydrocarbons, 2 alkaloids and derivatives, 1 homogeneous non-metal compound, 1 hydrocarbon derivative, 1 lignans, neolignans and related compound, 1 organosulfur compound (Fig. S3).

OPLS-DA (Component 1 was 65.00% and Orthogonal Component 1 was 4.54%) could be better able to distinguish between groups (Fig. 4A). The validity of the OPLS-DA model was checked through the permutation tests (Fig. 4B). These results indicated that over-fitting phenomenon was not presented in the model. Based on the fold change of relative content ≥1 (up-regulation) or ≤1 (down-regulation), p-value <0.05, and VIP value ≥1, a total of 318 significant DMs were screened out. Among these, 161 DMs were detected in positive ion mode, with 74 up-regulated and 87 down-regulated, while 157 DMs were found in negative ion mode, comprising 101 up-regulated and 56 down-regulated DMs (Fig. 4C,D). HMDB compound classification revealed that a total of 263 DMs were identified and annotated into HMDB 4.0. These DMs were matched and classified into 9 categories, including 116 lipids and lipid-like molecules, 33 organic oxygen compounds, 32 organic acids and derivatives, 29 phenylpropanoids and polyketides, 24 orrganoheterocyclic compounds, 16 benzenoids, 6 organic nitrogen compounds, 6 nucleosides, nucleotides, and analogues, 1 homogeneous non-metal compounds (Fig. S4, Table S1). Among these significantly different metabolites, a total of 33, 14, 5, and 5 secondary metabolites were identified as organic oxygen compounds, flavonoids, phenolic compounds, and cinnamic acids and derivatives (Table 1), respectively, which were associated with essential active ingredients (polysaccharides, phenols, and flavonoids) of P. palustre.

P. palustre is rich in polysaccharides, which are a very important quality evaluation indicator [48]. The polysaccharides of P. palustre contained 8 kinds of monosaccharides (glucose, galacturonic acid, ribose, rhamnose, glucuronic acid, galactose, arabinose, and xylose) [49]. In this study, we found 33 significantly differential organic oxygen compounds (12 up-regulated in YJF300 and 21 down-regulated in FHF30) (Fig. 5A). Among these, the abundance of 2-Deoxy-D-Ribose significantly increased in the YJF300 treatment relative to the FHF30 treatment (Fig. 5B). In addition, oligosaccharides refer to the compounds containing 2–10 monosaccharides polymerized by glycosidic bonds. Here, 4 kinds of differential oligosaccharides (galactose-beta-1,4-xylose, sucrose, raffinose, stachyose) and 1 polysaccharide (levan) were identified between the two treatments. The abundance of these five saccharides was significantly increased in the FHF30 treatment compared to the YJF300 treatment (Figs. 5C–5G).

A total of 14 significantly differential flavonoids were detected in this study. As shown in Fig. 6A, the FHF30 treatment exhibited a higher abundance of the 14 differential flavonoids compared to the YJF300 treatment. Of these, quercetin, quercetin 3-O-glucoside, kaempferol, and rutin are the flavonoids that have been identified in P. palustre, and these four kinds of flavonoids had a significantly high abundance expression in the FHF30 treatment (Fig. 6B–E). This indicated that FHF30 treatment was a suitable treatment to improve the pharmacological effects of P. palustre.

In the present study, 5 significantly differential phenolic compounds and 5 significantly differential cinnamic acids and their derivatives were detected between the FHF30 and YJF300 treatments. The abundance of [7]-Paradol, sinapyl alcohol, and isoeugenol in YJF300 treatment was higher than in FHF300 treatment, while the abundance of grevilline B and (5a,6a,8a,11a)-8-Hydroxy-2-oxo 1(10),3-guaiadien-12,6-olide-15-al 8-(4-hydroxyphenylacetate) in FHF300 treatment was higher (Fig. 7A). Meanwhile, in the FHF30 treatment, there was a higher abundance of five cinnamic acids and their derivatives compared to the YJF300 treatment (Fig. 7B). These two phenolic compounds sinapyl alcohol and isoeugenol were found to be a higher abundance in the YJF300 treatment (Fig. 7C,D). In addition, caffeic acid and ferulic acid are essential active ingredients of P. palustre, and the abundance of these two compounds in FHF30 was significantly higher (Fig. 7E,F).

Based on the selected 33 organic oxygen compounds, 14 flavonoids, 5 phenolic compounds, and 5 cinnamic acids and derivatives, there were 5 significantly enriched metabolic pathways, including flavone and flavonol biosynthesis, flavonoid biosynthesis, galactose metabolism, starch and sucrose metabolism, and phenylpropanoid biosynthesis (Fig. 8A). In addition, a metabolic map was developed based on these five metabolic pathways (Fig. 8B), in which a total of 14 DMs were involved, including 6 organic oxygen compounds (dulcitol, raffinose, sucrose, levan, stachyose, and chlorogenic acid), 4 flavonoids (kaempferol, quercetin, quercetin 3-O-glucoside, and rutin), 2 phenolic compounds (sinapyl alcohol and isoeugenol), and 2 cinnamic acids and derivatives (caffaric acid and ferulic acid). In particular, flavone and flavonol biosynthesis emerged as the most significantly differential metabolic pathway, which was associated with compounds such as quercetin, quercetin 3-O-glucoside, kaempferol, and rutin.