Comparative transcriptomic and metabolomic profiling of forest tea and terrace tea in Sandu, Guizhou, China

Abstract

Ancient forest tea is locally valued in Guizhou, but the transcriptomic and metabolomic differences associated with forest and terrace tea cultivation systems remain unclear. We conducted an exploratory integrated transcriptomic and untargeted metabolomic comparison across three shoot types (single buds, one-bud-one-leaf, and one-bud-two-leaves). Transcriptomic analysis showed the largest number of DEGs at the one-bud-two-leaves stage, suggesting substantial transcriptional variation under the present sampling conditions. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses indicated that phenylpropanoid and flavonoid biosynthesis pathways were over-represented. Untargeted liquid chromatography-mass spectrometry (LC–MS) detected 864 differential metabolite features with Metabolomics Standards Initiative (MSI) Level 2 putative annotations, including 374 features with higher relative abundance in forest tea. The largest number of differential metabolite features occurred at the one-bud-one-leaf stage, and these features were enriched in flavonoid biosynthesis. Variable importance in projection (VIP) ranking prioritized discriminant features from false discovery rate (FDR)-filtered data, including four MSI Level 2 features putatively annotated as polyphenol-related compounds that showed higher relative abundance in forest tea. In addition, the expression of genes encoding phenylalanine ammonia-lyase (PAL), a key enzyme in the phenylpropanoid biosynthesis pathway, differed significantly between forest tea and terrace tea. These results suggest that secondary-metabolism-related pathways are associated with molecular differentiation between forest and terrace tea. Because this study lacked sensory evaluation, targeted metabolite quantification, and functional validation, these findings require future validation. This work provides candidate pathways and metabolite features for future validation of tea quality differences.

Graphical abstract

Highlights

• •Forest and terrace teas in Guizhou showed distinct molecular profiles.
• •Flavonoid and phenylpropanoid pathways were enriched in forest tea.
• •Forest tea had higher polyphenol content and superior taste and flavor than terrace tea.

1. Introduction

Tea plant [Camellia sinensis (L.) O. Kuntze] is one of the most widely consumed beverage crops worldwide, and the commercial and sensory value of tea is closely related to the composition of bioactive secondary metabolites in young shoots (Xia et al., 2020; Xu et al., 2022). Catechins and other flavonoids are associated with bitterness and astringency, amino acids contribute importantly to umami and freshness, and volatile metabolites participate in aroma formation; therefore, these compounds and metabolite features are frequently examined in food chemistry studies of tea quality (Li et al., 2025; Moreira et al., 2024; Zhang et al., 2020). Because young shoots are the principal raw materials for tea manufacturing, describing how their gene-expression and metabolite profiles vary under different production contexts can provide useful background for understanding quality-related variation.

Tea plants are considered to have originated from forest ecosystems and are naturally associated with subtropical evergreen broad-leaved forests (Drew, 2019; Wu, Long, & Geng, 2023; Wu, Wang, et al., 2023). Compared with intensively managed terrace plantations, forest tea systems generally retain a multilayer vegetation structure and are often characterized by relatively low-input ecological management (Liu et al., 2013; Pei et al., 2022). However, forests and terraces differ not only in management form but also in many environmental and biological conditions, including habitat and climatic conditions, altitude-related topographic characteristics, and soil nutrient status, all of which can influence the aroma and taste of tea leaves (Luo et al., 2001; Niu et al., 2014). Therefore, forest-versus-terrace comparisons are intrinsically confounded by environmental heterogeneity, plant age, genotype, and management history. This issue is central to the interpretation of such studies: observed molecular or metabolic differences should be treated as cultivation-system-associated patterns rather than direct effects caused by the cultivation system alone.

Previous studies have reported sensory and biochemical differences between ancient or forest tea and terrace tea. For instance, investigations in Yunnan indicated that tea produced from ancient tea gardens may show more coordinated taste, stronger aftertaste and different bitterness and astringency characteristics than terrace tea (Chen et al., 2011; Liang et al., 2006). Differences in phytochemical profiles and drinkers' sensory perceptions between agro-forest and terrace production systems have also been reported (Ahmed et al., 2010). These studies provide useful evidence that tea quality-related traits may vary among production systems, but most such comparisons are observational and cannot fully separate cultivation-system effects from environmental background, tree age, genetic resources, or local management practices.

Integrated transcriptomic and metabolomic analyses have become increasingly common in tea research and have contributed to understanding quality formation, developmental variation, stress responses, and metabolite accumulation. For example, Li et al. (2023) reported transcriptomic and metabolomic divergence among tea plants of different ages, with enrichment of flavonoid biosynthesis, phenylpropanoid biosynthesis, amino acid biosynthesis, and plant-pathogen interaction pathways (Li et al., 2023). Similarly, differential metabolite features related to flavonoids and amino acids have been associated with flavor differences among wild ancient tea, cultivated ancient tea, and terrace-planted tea (Wang et al., 2010). These studies demonstrate the value of omics approaches, but they also illustrate common limitations: many analyses are based on observational materials, integrate omics mainly through pathway overlap, and often provide limited control over environmental, age-related, or genotype-related variation. Consequently, omics results in this context are best used to generate candidate associations rather than to establish coordinated regulation or causal links from cultivation system to gene expression, metabolite accumulation, and sensory quality.

In Guizhou Province, forest tea systems represent important local tea resources and are part of a broader landscape of tea plant diversity and historical tea production. Guizhou Province is also a key region along the Ancient Tea-Horse Road and harbors numerous ancient tea trees and relatively well-preserved forest tea ecosystems. However, under pressures from urbanization and agricultural expansion, some forest tea systems have become degraded or neglected (Wu et al., 2025). Given this background, comparative molecular and metabolite profiling of forest and terrace tea from Guizhou is useful, provided that the results are interpreted within the constraints of site-specific sampling and environmental confounding rather than as general or causal conclusions.

Based on these considerations, the present study conducted an exploratory comparison of transcriptomic and untargeted metabolomic profiles of forest and terrace tea shoots from Sandu County, Guizhou Province. Three harvestable shoot types, namely single buds, one-bud-one-leaf shoots, and one-bud-two-leaves shoots, were analyzed. The objective was not to test a strong mechanistic hypothesis or to confirm coordinated regulation between genes and metabolites, but to describe cultivation-system-associated transcriptomic and metabolomic differences within the sampled context, evaluate whether secondary-metabolism-related pathways were repeatedly represented in the datasets, and prioritize candidate genes, pathways, and putatively annotated metabolite features for future targeted validation under more controlled experimental designs.

2. Materials and methods

2.1. Sample collection and experimental design

A total of 18 tea shoot samples were collected from forest tea and terrace tea systems in the Jiuqian area of Sandu County, Guizhou Province, China, on May 6, 2024. Detailed sample information is provided in Table 1. For each cultivation system, three field-level plots were selected and treated as biological replicates. Within each plot, three shoot maturity stages were sampled: single buds, one-bud-one-leaf shoots, and one-bud-two-leaves shoots. Thus, the sampling design included two descriptive factors, cultivation system and shoot maturity stage, with three biological replicates for each cultivation-system × shoot-stage combination. The field-level plot, rather than each individual shoot sample, was treated as the replication unit for cultivation-system comparisons within each shoot maturity stage. For forest tea, the samples were coded as A01–A03, B01–B03, and C01–C03, corresponding to single buds, one-bud-one-leaf shoots, and one-bud-two-leaves shoots, respectively. For terrace tea, the corresponding samples were coded as E01–E03, F01–F03, and G01–G03.

Table 1.

Sample information.

Tea Type	Tenderness Type	Sample ID	Preservation Method	Collection Site	Date
Forest tea	Single bud	A01 A02 A03	Stored in liquid nitrogen	Sandu County, Guizhou Province (25°43′9″N, 108°0′15″E)	May 6, 2024
	One bud and one leaf	B01 B02 B03
	One bud and two leaves	C01 C02 C03
Terrace tea	Single bud	E01 E02 E03
	One bud and one leaf	F01 F02 F03
	One bud and two leaves	G01 G02 G03

All shoots were collected manually using the same shoot maturity criteria, sampling procedure, and post-harvest handling. Samples were immediately frozen in liquid nitrogen after harvest and stored at −80 °C until further analysis. All samples were collected from the same geographic region and on the same sampling date to reduce variation caused by sampling time and broad geographic differences. However, detailed environmental and biological variables, including soil properties, microclimatic conditions, canopy structure, light intensity, management history, plant age, and genotype were not quantitatively measured or fully controlled. Therefore, the present study was treated as an exploratory comparison within the sampled region and time point, and the cultivation system was interpreted as an associated grouping factor rather than as a fully isolated causal factor.

2.2. Transcriptomic analysis

Total RNA was extracted from the collected tissue samples using the MJZol total RNA extraction kit (Shanghai Meiji Biotechnology Co., Ltd., China) and further purified using the RNA Purification Kit according to the manufacturer's instructions. The concentration and purity of the extracted RNA were measured using a NanoDrop 2000 spectrophotometer, and the integrity of RNA was assessed by agarose gel electrophoresis using Biowest Agarose (Biowest, Spain). For library construction, each RNA sample met the following requirements: total RNA amount ≥ 1 μg, concentration ≥ 30 ng·μL⁻¹, RQN > 6.5, and OD260/280 between 1.8 and 2.2.

A total of 18 cDNA libraries were generated. Transcriptome libraries were constructed using Illumina Stranded mRNA Prep, Ligation (Illumina, USA) from forest tea and terrace tea samples, including single bud (A, E), one bud with one leaf (B, F), and one bud with two leaves (C, G). The libraries were sequenced on the NovaSeq platform using the NovaSeq Reagent Kit (NovaSeq X Plus) (Illumina, USA) with a paired-end strategy. All 18 libraries were prepared using the same library preparation protocol and sequenced on the same NovaSeq platform. To reduce potential technical bias, samples from different cultivation systems and shoot maturity stages were processed together rather than as separate biological groups. Potential batch-related structure was examined using PCA and sample correlation analysis before differential expression analysis. No explicit batch covariate was included in the DESeq2 model because no separate sequencing batch was recorded. Raw sequencing data were evaluated using fastp, and a total of 127.3 Gb of clean data were obtained after quality filtering, with no less than 6.3 Gb of clean data per sample and Q30 values above 95%.

The raw reads were filtered to obtain high-quality clean reads and then aligned to the reference genome of Camellia sinensis (GCF_004153795.1) using HISAT2 (http://ccb.jhu.edu/software/hisat2/index.shtml) in the orientation mode. Transcripts were assembled using StringTie, and gene-level expression abundance was estimated using RSEM. TPM values were generated for descriptive presentation and exploratory visualization, including expression distribution, heatmaps, and sample-level comparisons. Because TPM values normalize for gene length and sequencing depth but may still be affected by compositional differences among samples, they were not used for statistical testing of differential expression. Raw read counts were used for DESeq2-based differential expression analysis, in which size factors and dispersion parameters were estimated internally. Sequencing data quality metrics, including raw reads, raw bases, clean reads, clean bases, error rate, Q20, Q30, and GC content, were calculated for each sample and are summarized in Table 2.

Table 2.

Sequencing data summary.

Sample	Raw reads	Raw bases	Clean reads	Clean bases	Error rate (%)	Q20 (%)	Q30 (%)	GC content (%)
A01	49,833,340	7,524,834,340	49,320,462	7,349,552,235	0.01	98.8	96.2	43.7
A02	43,106,966	6,509,151,866	42,684,114	6,367,700,300	0.01	98.8	96.2	43.9
A03	52,191,572	7,880,927,372	51,699,204	7,712,164,343	0.01	98.8	96.2	43.5
B01	50,380,906	7,607,516,806	49,888,464	7,455,697,515	0.01	98.8	96.2	43.6
B02	47,256,160	7,135,680,160	46,838,724	7,001,885,296	0.01	98.7	96.0	43.7
B03	47,624,948	7,191,367,148	47,170,114	7,058,943,775	0.01	98.8	96.2	44.2
C01	47,370,804	7,152,991,404	46,969,732	7,018,555,616	0.01	98.7	96.0	43.8
C02	45,835,544	6,921,167,144	45,382,182	6,762,949,596	0.01	98.7	96.0	43.8
C03	45,506,186	6,871,434,086	45,091,726	6,723,899,830	0.01	98.8	96.2	43.7
E01	51,803,234	7,822,288,334	51,313,920	7,664,063,551	0.01	98.8	96.3	43.9
E02	42,909,814	6,479,381,914	42,483,456	6,342,512,015	0.01	98.8	96.2	43.9
E03	45,234,378	6,830,391,078	44,595,204	6,650,691,228	0.01	98.7	95.9	44.1
F01	50,883,892	7,683,467,692	50,411,010	7,514,468,130	0.01	98.8	96.2	43.8
F02	46,189,060	6,974,548,060	45,743,978	6,829,309,542	0.01	98.7	96.0	43.7
F03	47,832,012	7,222,633,812	47,387,996	7,095,877,649	0.01	98.8	96.1	43.6
G01	51,981,968	7,849,277,168	51,497,268	7,678,221,698	0.01	98.8	96.2	43.8
G02	47,880,442	7,229,946,742	47,440,972	7,062,603,318	0.01	98.8	96.2	43.9
G03	47,413,296	7,159,407,696	46,988,656	7,013,312,286	0.01	98.8	96.2	43.9

Note: A, B, C, E, F, and G represent the single bud, one bud with one leaf, and one bud with two leaves of forest-grown tea, as well as the single bud, one bud with one leaf, and one bud with two leaves of terrace-grown tea, respectively.

Differentially expressed genes (DEGs) were identified through DESeq2 (http://bioconductor.org/packages/stats/bioc/DESeq2/) using a multifactor design, in which cultivation system and shoot maturity were treated as fixed factors and their interaction was included in the model (cultivation_system + shoot_maturity + cultivation_system:shoot_maturity). For transcriptomic analysis, the interaction term was included to account for potential stage-dependent forest–terrace differences. However, because the number of field-level replicates was limited, interaction-related results were interpreted cautiously and were not used as the primary basis for biological conclusions. The main DEG results were therefore presented as forest–terrace contrasts within each shoot maturity stage.

Genes with |log₂(fold change)| ≥ 1 and an adjusted FDR ≤ 0.05 were considered significantly differentially expressed. TPM values were used for normalized expression presentation across samples, whereas raw counts were used for differential expression testing. Because TPM values are affected by compositional differences and are not appropriate as input for DESeq2 differential expression testing, raw read counts were used for DEG analysis. DESeq2 internally estimates size factors and dispersion parameters for count-based differential expression analysis.

GO enrichment analysis was performed using GOATOOLS based on Fisher's exact test. KEGG pathway enrichment was performed as an over-representation analysis using annotated KEGG pathway assignments and a hypergeometric/Fisher exact test implemented in Python/R scripts. For both GO and KEGG enrichment analyses, the background set consisted of all expressed genes with corresponding GO or KEGG annotations. p values were adjusted for multiple testing using the Benjamini–Hochberg false discovery rate (FDR) method, and terms or pathways with FDR-adjusted p < 0.05 were considered significantly enriched. Enrichment results were interpreted as statistical over-representation of annotated genes rather than direct evidence of pathway activity.

2.3. Metabolomic experimental methods

2.3.1. Metabolite extraction

Approximately 100 ± 5 mg of each freeze-dried sample was weighed into a 2 mL centrifuge tube, followed by the addition of a 6 mm grinding bead and 400 μL of methanol–water extraction solvent (4:1, v/v) containing a mixture of four internal standards including 2-Chloro-L-phenylalanine (0.02 mg/mL).

Samples were ground for 6 min at −10 °C and 50 Hz using a Wonbio-96c cryogenic tissue grinder. Subsequently, low-temperature ultrasonic extraction was performed at 5 °C and 40 kHz for 30 min, followed by incubation at −20 °C for 30 min and centrifugation at 4 °C, 13,000 ×g for 15 min. The supernatant was transferred to an autosampler vial with an inner tube for LC–MS analysis. In parallel, 20 μL of supernatant from each sample was pooled to generate a quality control (QC) sample.

2.3.2. LC–MS detection and quality control

Metabolite profiling was performed using a Thermo Scientific Vanquish Horizon UHPLC system coupled to a Q Exactive mass spectrometer (Thermo Fisher Scientific, Bremen, Germany). Chromatographic separation was carried out on an ACQUITY UPLC HSS T3 column (100 mm × 2.1 mm i.d., 1.8 μm, Waters, Milford, USA). Mobile phase A consisted of 95% water and 5% acetonitrile containing 0.1% formic acid, while mobile phase B contained 47.5% acetonitrile, 47.5% isopropanol, and 5% water with 0.1% formic acid. The flow rate was 0.40 mL/min, injection volume was 5 μL, and column temperature was maintained at 40 °C.

Mass spectrometry was performed using electrospray ionization (ESI) in both positive and negative ion modes. The scan range was m/z 70–1050. The sheath gas flow rate was 40 arb, the auxiliary gas flow rate was 10 arb, the capillary temperature was 320 °C, and the auxiliary gas heater temperature was 400 °C. The spray voltage was 3500 V in positive ion mode and − 2800 V in negative ion mode. The S-Lens RF level was set to 50. Full MS resolution was 70,000, MS/MS resolution was 17,500, and normalized collision energies of 20, 40, and 60% were applied. Detailed parameters of the LC–MS system are provided in Table S1.

QC samples were prepared by pooling equal aliquots of all sample extracts and were analyzed using the same procedure as that for the biological samples. One QC sample was injected after every 5–15 analytical samples, and internal standards were used to monitor analytical stability. Features with relative standard deviation (RSD) >30% in QC samples were removed to ensure analytical reproducibility (Ghosh et al., 2021; Liu et al., 2025). No dedicated QC-based signal drift correction (e.g., LOESS or QC-RLSC) was applied beyond peak alignment and normalization; therefore, residual drift cannot be fully excluded.

PLS-DA and OPLS-DA were applied to explore group separation among forest and terrace tea samples and to prioritize candidate metabolite features. These supervised methods are prone to overfitting, especially with high-dimensional metabolomics data; therefore, results should be interpreted cautiously and used primarily for exploratory feature selection rather than definitive biological inference. Model robustness was assessed using permutation testing as described in the experimental workflow. Cross-validation and VIP-based feature selection were applied to mitigate overfitting risk.

2.3.3. Data preprocessing and metabolite identification

Raw LC–MS data were processed using Progenesis QI v3.0 for baseline filtering, peak detection, integration, retention-time correction, and peak alignment. Metabolite features were putatively annotated according to the Metabolomics Standards Initiative (MSI) Level 2 criteria based on accurate mass and MS/MS fragmentation information matching against HMDB, Metlin, Lipidmaps, and a self-built spectral database, with related pathway annotations matched to KEGG (Sumner et al., 2007). The mass error tolerance for annotation was set at <10 ppm. No authentic reference standards were used for retention-time or MS/MS confirmation. For downstream data preprocessing, variables with excessive missing values within groups were removed. Log transformation was applied prior to statistical analysis.

2.4. Integrated transcriptomic and metabolomic pathway analysis

Integrated transcriptomic and metabolomic pathway analysis was performed across the three shoot developmental stages, including single buds, one-bud-one-leaf shoots, and one-bud-two-leaves shoots. For each stage, KEGG enrichment results from differentially expressed genes and differential metabolite features were compared to identify shared pathways. This pathway-level comparison was used as an exploratory integration strategy rather than as correlation-network analysis or joint modeling. The one-bud-one-leaf stage was emphasized in the main text because it is an important harvestable shoot stage in tea production and showed the most prominent metabolomic differentiation under the present sampling conditions. Results from the other developmental stages are provided in Table S6.

2.5. Statistical analysis

Statistical analyses were structured at two levels. First, the overall sampling design was described considering cultivation system and shoot developmental stage as explanatory factors. Second, stage-specific forest–terrace comparisons were performed for differential analyses, in order to partially account for developmental variation. Given the exploratory nature of the study and limited biological replication, factorial modeling results were interpreted cautiously. For metabolomic data, differential features were identified within each developmental stage. Univariate comparisons were performed using two-sided unpaired Student's t-tests as an initial screening step only. Resulting p-values were corrected using the Benjamini–Hochberg procedure to control the false discovery rate (FDR). Therefore, statistical significance was defined as FDR-adjusted p-value (p_adjust) ≤ 0.05.

Variable importance in projection (VIP) scores were used only as a secondary ranking metric after FDR filtering and were not treated as an independent criterion for statistical significance. Thus, the combined thresholds (FDR ≤ 0.05 and VIP ≥ 1) were applied for exploratory feature prioritization rather than confirmatory biological inference.

Multivariate analysis was conducted using PCA (unsupervised) followed by PLS-DA and OPLS-DA for supervised group discrimination. Model robustness was evaluated using permutation testing and cross-validation. All multivariate outputs were interpreted cautiously given the high-dimensional nature of metabolomics datasets. R²Y (cum) and Q² (cum) values were reported as descriptive model performance indicators rather than predictive validation metrics (Supplementary Table S3).

For pathway analysis, KEGG enrichment p-values were also adjusted using the Benjamini–Hochberg procedure, and results were reported as P_adjust. Pathways with P_adjust <0.05 were considered significantly enriched, and all interpretations were treated as hypothesis-generating.

3. Results

3.1. Transcriptome sequencing quality assessment

For samples of forest tea and terrace tea at different shoot maturities, three biological replicates were established for each group, resulting in a total of 18 cDNA libraries. Sequencing on the NovaSeq platform generated 127.3 Gb of clean data, with no less than 6.3 Gb per sample (Tables 2 and S2). The GC content was approximately 43%, and Q30 values exceeded 95%, indicating acceptable sequencing quality.

3.2. Sample principal component and correlation analysis

Principal component analysis (PCA) and sample correlation analysis were performed to explore overall variation among samples (Fig. 1). QC metrics indicated consistent data quality across samples. PCA showed separation trends between forest tea and terrace tea samples across developmental stages, suggesting underlying differences in gene expression patterns; however, as an unsupervised method, PCA does not provide direct evidence of specific biological mechanisms. Pearson's correlation coefficient (PCC) was used to evaluate the consistency of biological replicates. The r² values were close to 1, indicating high reproducibility among replicates, although high correlation alone does not fully capture data quality in high-dimensional transcriptomic datasets.

Fig. 1.

PCA of transcriptome samples (gene expression distribution). (a) Principal component analysis (PCA) of the transcriptome; (b) Heatmap of the expression correlation between two samples. A, B, C, E, F, and G represent the single bud, one bud with one leaf and one bud with two leaves of forest-grown tea as well as the single bud, one bud with one leaf and one bud with two leaves of terrace-grown tea, respectively.

3.3. Differentially expressed genes

Because shoot maturity strongly affects tea leaf molecular and metabolic profiles, forest–terrace differences were primarily interpreted within each shoot maturity stage rather than by pooling all samples.

Using the criteria of |log₂(fold change)| ≥ 1 and FDR ≤ 0.05, a total of 20,362 DEGs were identified between forest and terrace tea samples, including 11,145 upregulated and 9217 downregulated genes (Fig. 2a). The large number of DEGs should not be interpreted as an effect of cultivation system alone; rather, it likely reflects combined variation associated with shoot maturity, environmental heterogeneity, genotype, management history, and cultivation system, which could not be fully disentangled in the present observational design. Among the shoot maturity stages, the one-bud–two-leaves stage exhibited the highest number of DEGs (15,848; Fig. 2b), indicating substantial transcriptional variation under the present sampling conditions.

Fig. 2.

Comparison of DEGs between forest and terrace teas. (a) forest-grown tea vs. terrace-grown tea; (b) stage-specific comparisons. A, B, C, E, F, and G represent the single bud, one bud with one leaf, and one bud with two leaves of forest-grown tea, as well as the single bud, one bud with one leaf, and one bud with two leaves of terrace-grown tea, respectively.

3.4. Functional and metabolic pathway analysis

3.4.1. Differential gene enrichment analysis between forest and terrace teas

KEGG enrichment analysis revealed that DEGs were over-represented in pathways such as phenylpropanoid biosynthesis, plant–pathogen interaction, and flavonoid biosynthesis (Fig. 3). These pathways are commonly involved in plant secondary metabolism and stress responses, and the enrichment results were therefore summarized as pathway over-representation rather than direct evidence of functional activation.

Fig. 3.

KEGG (metabolic pathway) enrichment analysis. A, B, C, E, F, and G represent the single bud, one bud with one leaf, and one bud with two leaves of forest-grown tea, as well as the single bud, one bud with one leaf, and one bud with two leaves of terrace-grown tea, respectively.

3.4.2. Differential gene enrichment analysis across bud maturities

Across different shoot maturities, enriched pathways included photosynthesis, cutin/suberin/wax biosynthesis, phenylpropanoid biosynthesis, and flavonoid biosynthesis (Fig. S2). These pathways were consistently over-represented across comparisons, suggesting shared metabolic patterns across developmental stages.

3.5. KEGG pathway enrichment analysis of upregulated and downregulated DEGs

KEGG pathway enrichment analysis indicated that genes upregulated in forest tea were enriched in pathways such as phenylpropanoid biosynthesis, MAPK signaling (plant), plant–pathogen interaction, and flavonoid biosynthesis, whereas downregulated genes were associated with pyrimidine metabolism, ribosome, sesquiterpenoid/triterpenoid biosynthesis, and glutathione metabolism (Tables S4 and S5). Phenylpropanoid and flavonoid biosynthetic pathways are known to be involved in the production of tea polyphenols. The observed enrichment patterns therefore suggest potential associations with secondary metabolism.

3.6. Metabolome analysis and differential metabolite features

To explore metabolomic variation between forest and terrace tea, PCA and clustering analysis were first performed. PCA showed separation trends between forest-grown and terrace-grown tea samples (Fig. S3a). The clustering heatmap further indicated that samples grouped primarily according to developmental stage and cultivation system (Fig. S3b), suggesting structured variation in metabolite profiles.

Untargeted LC–MS analysis detected a total of 2352 metabolite features, all putatively annotated at MSI Level 2. Based on FDR-controlled statistical analysis, 864 differential metabolite features were identified, including 374 with higher relative abundance in forest tea. Stage-specific comparison showed that the one-bud–one-leaf stage exhibited the highest number of differential metabolite features (Fig. S4), indicating that metabolomic differentiation varied across developmental stages under the present sampling conditions. KEGG enrichment analysis showed that differential metabolite features were over-represented in pathways such as flavonoid biosynthesis, flavone and flavonol biosynthesis, and phenylalanine, tyrosine, and tryptophan biosynthesis (Fig. 4).

Fig. 4.

KEGG pathway enrichment of differential metabolite features.

(a) Forest-grown tea vs. terrace-grown tea; (b) single bud (forest-grown tea) vs. single bud (terrace-grown tea); (c) one bud with one leaf (forest-grown tea) vs. one bud with one leaf (terrace-grown tea); and (d) one bud with two leaves (forest-grown tea) vs. one bud with two leaves (terrace-grown tea).

For feature prioritization, VIP ranking was applied following FDR-based filtering and used only as a supplementary criterion. Among the top-ranked features, several polyphenol-related and nitrogen-containing putatively annotated features showed higher relative abundance in forest tea (Fig. 5a, Table 3). Because these features were based on MSI Level 2 annotations, they may represent isomeric or co-eluting compounds and require confirmation using authentic standards.

Fig. 5.

Key differential metabolite features. (a) VIP plot of metabolite features: left panel (scatter plot): the x-axis represents VIP values (reflecting the contribution of metabolite features to inter-group differences; VIP ≥ 1 is used as the screening threshold). Each dot represents one metabolite feature, and colors distinguish abundance change trends (red for up-regulation, blue for down-regulation); right panel (heatmap): corresponding to the metabolite features in the scatter plot, it shows their relative abundances in each sample group, simultaneously presenting the importance and abundance differences of metabolite features. (b) Heatmap of differential metabolite features: Shows the relative abundance distribution of metabolite features across different sample groups. Each row represents one metabolite feature, and each column corresponds to one sample; the blue-red color gradient indicates the standardized abundance value of metabolite features, with blue representing lower abundance and red representing higher abundance. Labels on the right indicate metabolite classes, and labels at the top represent sample groups, intuitively presenting the differential expression patterns of metabolite features between groups. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Table 3.

Selected putatively annotated metabolite features with higher relative abundance in forest tea than in terrace tea.

Metabolite category	Metabolite name
Tea polyphenols	3-O-Galloylhamamelitannin
	3,5-Digalloylpyrocatechin
	Taxifolin 3-arabinoside
	Hyperin 6″-glucoside (1 → 3)-rhamnoside
	Comosin
Nitrogen-containing	4’-O-Methylisodelphinidin 3-rutinoside
	Avenanthramide L
	6”-O-Acetyllycorine
	1-(2-Hydroxyphenylamino)-1-deoxy-beta-D-gentiobiose 1,2-acetal
	S-(2,2-Dichloro-1-hydroxyethyl) glutathione

Note: The VIP plot is provided solely as an exploratory visualization of MSI Level 2 putatively annotated metabolite features; only selected features with biological plausibility were retained in the candidate metabolite table and interpreted in the main text.

The top metabolite features included 3-O-Galloylhamamelitannin, 3,5-Digalloylpyrocatechin, Taxifolin 3-arabinoside, Hyperin 6″-glucoside (1 → 3)-rhamnoside, and Comosin, which showed higher relative abundance in forest tea. These results suggest a pathway-level association between phenylpropanoid-related gene expression and putatively annotated polyphenolic features.

Cluster analysis further indicated that flavonoid-related metabolite features (e.g., quercetin and kaempferol) and L-theanine showed higher relative abundance in forest tea (Fig. 5b). As no targeted metabolite quantification using authentic standards was performed, these observations were used to prioritize candidate secondary-metabolism-related features for future validation.

3.7. Integrated transcriptomic and metabolomic pathway analysis

Integrated transcriptomic and metabolomic pathway analysis was conducted across all three shoot developmental stages to explore potential pathway-level associations between gene expression and metabolite variation. Shared KEGG pathways were examined separately for single buds, one-bud-one-leaf shoots, and one-bud-two-leaves shoots. The detailed results for all three stages are summarized in Table S6. In the main text, we focused on the one-bud-one-leaf stage as this stage is an important harvestable shoot stage and showed the highest number of differential metabolite features under the present sampling conditions. Therefore, the one-bud-one-leaf stage was used as a representative stage for detailed presentation, while the other stages were retained in the supplementary results to avoid selective interpretation.

KEGG pathway-based integration analysis was performed to explore potential associations between transcriptomic and metabolomic changes. Flavonoid biosynthesis was enriched in both datasets, suggesting a possible pathway-level association between transcriptomic and metabolomic variation. Linoleic acid metabolism was jointly enriched, further indicating shared molecular patterns across omics layers. Phenylpropanoid biosynthesis was enriched in the transcriptomic dataset and also observed in the metabolomic dataset, although the metabolomic enrichment did not reach statistical significance (Table 4). These pathway overlaps were used to prioritize candidate secondary-metabolism-related pathways for later targeted metabolite quantification and functional assays.

Table 4.

Joint KEGG pathway enrichment analysis of transcriptomic and metabolomic datasets.

KEGG ID	Pathway	Transcriptome P_adjust	Metabolome P_adjust	Integrated interpretation
map00941	Flavonoid biosynthesis	8× 10−4	3 × 10−7	jointly enriched
map00591	Linoleic acid metabolism	0.04	0.01	jointly enriched
map00940	Phenylpropanoid biosynthesis	6 × 10−4	0.2	transcriptome-significant, metabolome-supported but not jointly enriched

At the one-bud-one-leaf stage, phenylalanine ammonia-lyase (PAL), the first rate-limiting enzyme of the phenylpropanoid pathway, showed differential expression among four homologous genes, with three upregulated and one downregulated in forest tea (Table 5). These trends were broadly consistent with the higher relative abundance of downstream flavonoid and phenolic metabolite features, but they were not used to infer coordinated regulation.

Table 5.

Phenylpropanoid pathway key enzyme expression (one-bud-one-leaf, forest vs. terrace tea).

Gene ID	Gene set	Regulate	KO name	Description
gene-LOC114273057	B_vs_F	up	PAL	Phenylpropanoid biosynthesis
gene-LOC114269233	B_vs_F	up	PAL	Phenylpropanoid biosynthesis
gene-LOC114262497	B_vs_F	up	PAL	Phenylpropanoid biosynthesis
gene-LOC114309513	B_vs_F	down	PAL	Phenylpropanoid biosynthesis

4. Discussion

4.1. Transcriptomic analysis of differentially expressed genes between forest tea and terrace tea

Transcriptomics serves as a bridge linking genetic information to gene-expression variation, thereby helping to identify biological functions and regulatory pathways in which genes are involved (Conesa et al., 2016; Wang et al., 2009; Wörheide et al., 2021; Yang et al., 2019). In this study, transcriptome sequencing was conducted to compare forest tea and terrace tea, leading to the identification of 20,362 significant DEGs, including 11,145 upregulated and 9217 downregulated genes. The high DEG count likely reflects the combined effects of shoot maturity, environmental heterogeneity, genotype, management history, and cultivation system, rather than cultivation system alone. Among the developmental stages examined, the one-bud–two-leaves stage showed the most pronounced differences in gene expression.

KEGG pathway enrichment analysis indicated that the DEGs were mainly enriched in phenylpropanoid biosynthesis, stilbenoid, diarylheptanoid, and gingerol biosynthesis, plant–pathogen interaction, and flavonoid biosynthesis pathways, involving 165, 53, 258, and 75 DEGs, respectively. Flavonoid biosynthesis is a branch of the phenylpropanoid metabolic pathway (Wu, Long, & Geng, 2023; Wu, Wang, et al., 2023). In addition, a considerable number of upregulated DEGs were concentrated in these pathways, with 113 and 40 DEGs annotated to them, respectively (Table S4). These results suggest that transcriptional differences may exist in flavonoid-related pathways between forest tea and terrace tea.

The phenylpropanoid biosynthetic pathway is one of the major secondary metabolic routes in plants and is closely associated with the biosynthesis of flavonoids and other phenolic compounds involved in stress responses. As major downstream products of this pathway, flavonoids participate in multiple biological processes, including ultraviolet protection, defense against pathogens and herbivores, regulation of auxin transport, and plant–microbe interactions (Jaakola, 2013; Yi, 2020). The enrichment of this pathway, together with the observed gene expression patterns, suggests that secondary metabolism may be associated with transcriptomic variation between the two tea sample groups.

Accordingly, the DEG results should be interpreted as cultivation-system-associated transcriptomic patterns within the sampled field context. Pathway enrichment indicates over-representation of annotated gene sets, not direct evidence of pathway activation or metabolic flux.

4.2. Metabolomic analysis and integrated multi-omics analysis

The economic and sensory value of tea (Camellia sinensis) depends in part on metabolite profiles in fresh young shoots, which serve as important precursors related to taste, aroma, and potential bioactive properties (Wang et al., 2025). However, metabolite composition is jointly modulated by multiple factors, including genetic background, cultivation practices, and environmental conditions (Ahmed et al., 2019; Mukhopadhyay et al., 2016; Owuor, 2018). Previous studies have suggested that tea plants grown under contrasting forest and terrace cultivation systems may exhibit distinct metabolic characteristics.

In this study, an untargeted metabolomic approach was used to compare forest and terrace tea. PCA showed separation trends between the two cultivation groups, indicating structured differences in metabolite profiles. KEGG pathway enrichment analysis further showed that differential metabolite features at the same shoot maturity stage were enriched in the flavonoid biosynthesis pathway. This result is consistent with previous reports that flavonoid biosynthesis in tea is affected by environment and cultivation practices (Liu et al., 2018; Xiang et al., 2021).

Among the top-ranked differential metabolite features screened by VIP and false discovery rate (FDR) criteria, several showed higher relative abundance in forest tea, including putatively annotated features related to tea polyphenols and theanine-related compounds. Polyphenol-related features may be associated with taste attributes and antioxidant potential (Meng et al., 2019). Amino acids, particularly theanine, may contribute to umami and freshness attributes (Chen et al., 2021; Dai et al., 2022; Jia et al., 2018).

These metabolite features were therefore treated as candidates for future sensory and functional validation rather than as confirmed quality markers.

Integration of transcriptomic and metabolomic data was performed at the one-bud–one-leaf stage, where metabolite differences were most pronounced. Flavonoid biosynthesis and linoleic acid metabolism pathways were enriched in both datasets, indicating potential pathway-level associations between transcriptomic variation and metabolite features related to flavonoid metabolism in forest and terrace tea. Within the phenylpropanoid biosynthesis pathway, four homologous genes encode PAL, the first rate-limiting enzyme of this pathway. Among these genes, three were upregulated and one was downregulated in forest tea. The expression trends of these genes were broadly consistent with the enrichment patterns of multiple differential metabolite features in the same pathway, but this correspondence does not establish coordinated regulation between transcriptional and metabolic changes.

All metabolites in this study were putatively annotated at MSI Level 2 without structural confirmation using authentic standards. Consequently, each detected feature may represent isomers, co-eluting compounds, or adducts, and their precise identities remain uncertain. In addition, targeted metabolite quantification, sensory evaluation, and functional validation experiments were not performed. Therefore, associations between the observed metabolite differences and tea quality should be considered tentative. Overall, these findings highlight candidate pathways and metabolite features for future investigation. Future studies integrating sensory evaluation, odor/taste activity values, and targeted metabolite quantification are necessary to establish direct links between metabolite features and perceived tea quality.

5. Conclusion

This study presented an exploratory comparison of transcriptomic and untargeted metabolomic profiles of tea shoots from forest and terrace cultivation systems in Sandu, Guizhou, China. Differences were detected in both gene expression patterns and metabolite features, with many associated with secondary metabolism, particularly phenylpropanoid and flavonoid biosynthesis pathways. Developmental stage effects were also evident, with transcriptomic differences most pronounced at the one-bud–two-leaves stage and metabolomic differences peaking at the one-bud–one-leaf stage.

Pathway enrichment analyses indicated that differentially expressed genes and metabolite features were enriched in several overlapping metabolic pathways. At the transcriptional level, genes encoding PAL, a key enzyme in the phenylpropanoid pathway, showed differential expression patterns between forest and terrace tea. These observations are broadly consistent with patterns observed at the metabolite level; however, such correspondence should be interpreted cautiously and does not demonstrate coordinated regulation, mechanistic linkage, or causality between transcriptional activity and metabolite accumulation.

Several limitations should be acknowledged. Metabolites were putatively annotated at MSI Level 2 without authentic standards and may include isomeric or co-eluting compounds. No targeted quantification, sensory evaluation or functional validation was performed. In addition, environmental and genetic factors were not fully controlled, which may have contributed to the observed variation. Overall, this study provides an exploratory multi-omics dataset and identifies candidate pathways and features for further investigation. Future work should include targeted validation, controlled experimental designs, and functional assessments to clarify the biological relevance of these molecular differences.

CRediT authorship contribution statement

Dani Wei: Writing – original draft, Visualization, Validation, Software, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Huan Wu: Investigation. Mengxue Yang: Formal analysis, Data curation. Chaya Sarathchandra: Writing – review & editing. Yanfei Geng: Writing – review & editing, Supervision, Funding acquisition.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary material 1

Supplementary material 2

Data availability

Data will be made available on request.