Multi-omics insights into the woolly trait of Saussurea medusa and the plant’s coordinated regulation of flavonoid biosynthesis

Abstracts

The woolly plant Saussurea medusa is an endangered traditional Tibetan medicinal species and an ideal model to study high-altitude adaptation. Despite its ecological and pharmacological significance, the molecular mechanisms underlying its survival under harsh environments and its production of bioactive flavonoids remain poorly characterized. In this study, we assembled a high-quality chromosome-level genome of S. medusa (~3.8 Gb) with low heterozygosity and high repeat content. Comparative genomic analyses showed expansion of gene families related to DNA repair and hypoxia-adapted TCA cycle metabolism, and strong positive selection on light-harvesting complex pathways. Structural and functional analyses of the woolly trait demonstrated its roles in thermal insulation and light attenuation. Correspondingly, 48 key trichome-related genes were identified, reflecting a regulatory network dominated by the conserved GL1-GL3/(EGL3)-TTG1 transcriptional activator complex and linked to hormonal pathways. Metabolomics identified 1792 metabolites, including 18 bioactive metabolites with tissue-specific accumulation, most of which are flavonoids. Integrated metabolomics and transcriptomics analysis further revealed 27 flavonoid biosynthetic genes highly expressed in flowers and leaves, correlating strongly with the accumulation of flavonoids, indicating coordinated regulation of secondary metabolism. Our study provides multi-omic resources for S. medusa and provides insights into the genetic basis of its high-altitude adaptation, woolly trait, and medicinal properties.

Multi-omics analysis of the woolly alpine plant Saussurea medusa provides insights into its genome evolution, trichome-associated environmental adaptation, and coordinated regulation of flavonoid biosynthesis.

Introduction

Abiotic stresses, such as extreme temperatures, drought, and soil salinity, are major environmental constraints that limit plant growth, productivity, and quality^1–3. Plants have evolved sophisticated mechanisms to perceive and respond to multiple and fluctuating environmental cues³. While research on individual stress has advanced significantly, increasing attention is now being paid to the crosstalk among multiple concurrent stresses^4–6. High-altitude regions (e.g., Himalaya-Hengduan Mountains) present extreme environmental conditions, including persistent low temperatures, intense UV-B radiation, hypoxia, nutrient scarcity, and dramatic diurnal temperature fluctuations, making them typical representatives of multiple stresses^7,8. These challenges are especially severe in scree habitats above 4500 meters, where mean annual temperatures fall below 0 °C and UV-B intensity is markedly elevated. Frequent freeze–thaw cycles destabilize surface soils, further intensifying abiotic stress and complicating plant survival and reproduction^9,10.

The convergence of a complex geological history and monsoon dynamics on the Qinghai-Tibet Plateau has engendered pronounced altitudinal gradients and habitat fragmentation¹¹. High mountains, deep valleys, screes, and glacier retreat terrains are intricately interspersed, with major rivers and profound gorges forming natural barriers. This has fostered a quintessential “sky island” effect¹², driving allopatric differentiation. The interplay of vertical climate zonation and diverse microhabitats has given rise to the alpine subnival belt, the plateau’s most iconic extreme environment¹². Consequently, plants inhabiting this zone face the compounded stresses of low temperatures, hypoxia, intense radiation, and pollinator limitation, and have evolved diverse morphological and physiological adaptations¹³. For example, the woolly plant Saussurea medusa develops dense trichome-covered inflorescences for efficient heat absorption⁸, while greenhouse plants like Rheum nobile and Rheum alexandrae utilize translucent bracts for UV filtration, thermal regulation, and precipitation protection^10,14. Cushion plants (e.g., Arenaria polytrichoides, Potentilla articulata) buffer microclimates and enhance soil nutrient uptake^15,16, and Cremanthodium campanulatum employs nodding capitula to minimize UV damage through ground radiation absorption⁹.

Increasing evidence indicates that trichomes are not merely developmental traits but function as crucial adaptive structures in alpine environments¹⁷. Dense trichome layers can enhance the reflectance of harmful UV-B radiation, reduce boundary layer conductance to limit transpirational water loss, buffer rapid temperature fluctuations, and mitigate photoinhibition under high irradiance^18,19. Trichome pattern, defined by the distribution and density of trichomes on aerial organs, exhibits substantial inter- and intra-specific variation. In the model plant Arabidopsis thaliana, trichome development is governed by a conserved transcriptional network centered on the conserved MYB-bHLH-WD40 (MBW) transcriptional complex²⁰. Trichome cell fate is specified through regulation of the homeodomain transcription factor GLABRA2 (GL2), whose expression is directly activated by the trimeric GL1-GL3/EGL3-TTG1 complex²¹. This MBW complex, composed of the R2R3-MYB GL1, the bHLH factors GL3/EGL3, and the WD40-repeat protein TTG1, binds promoters of downstream targets such as GL2 to initiate differentiation. GL1 acts as a master regulator; its knockout abolishes trichome formation, while its constitutive expression can suppress initiation²², indicating a dosage-dependent dual role. MYB23 functions redundantly with GL1 to promote the formation of the activator complex and enhance expression of downstream targets like TTG2 and GL2²³. Furthermore, the MBW complex also activates the expression of single-repeat R3-MYB repressors, such as TRY and CPC. These mobile inhibitors form a lateral inhibition loop by competing with GL1 for binding to GL3/EGL3 in neighboring cells, thereby preventing trichome formation and ensuring proper spacing. For instance, mutation in TRY leads to trichome clustering²⁴, while overexpression of CPC suppresses GL3 activity²⁵, underscoring their essential roles in spatial patterning. Ecological studies in alpine species have demonstrated that trichome density often positively correlates with elevation and radiation intensity, highlighting their adaptive significance in extreme high-altitude habitats^26,27. Therefore, deciphering the genetic regulation of trichome development provides crucial insights into how alpine plants integrate intrinsic developmental programs with extrinsic environmental adaptation.

The woolly alpine plant Saussurea medusa, a perennial species of the subgenus Eriocoryne within Saussurea (Asteraceae), inhabits scree slopes and ice-marginal debris at elevations of 3600-5600 m on the Qinghai-Tibet Plateau. Its dense epidermal trichomes enhance light reflectance, reduce transpirational water loss, buffer temperature fluctuations, and promote early reproduction, thereby improving survival under persistent multi-stress alpine conditions⁸. Concurrently, alpine species such as Agriophyllum squarrosum^28,29 and Fagopyrum tataricum³⁰ exhibit elevation-associated upregulation of flavonoid biosynthesis genes, leading to flavonoid accumulation that functions in UV shielding, antioxidant defense, and stress mitigation. These findings suggest that structural protection (trichomes) and metabolic regulation (flavonoids) represent coordinated adaptive strategies to intense UV radiation, low temperature, and hypoxia. Beyond ecological adaptation, flavonoids possess significant pharmacological activities, including enhancement of antioxidant capacity, reduction of hypoxia-induced lactate accumulation, neuroprotection³¹, and inhibition of osteosarcoma (SAOS2) cell proliferation and migration by compounds such as rutin³². Extracts of S. medusa, widely used in Tibetan and Chinese medicine, exhibit diverse bioactivities³³: arctigenin and arctiin show antitumor and anti-inflammatory effects^34,35, while supercritical CO₂ extracts and polysaccharides attenuate edema and regulate AQP^-3 to improve skin hydration³⁶. Thus, the integration of extreme-environment adaptation and enriched secondary metabolism positions S. medusa as a valuable natural system for investigating the genetic coordination of structural and metabolic strategies underlying alpine survival.

Genomic and multi-omics approaches have systematically elucidated the adaptation mechanisms of high-altitude plants by dissecting genotype-phenotype-environment interactions³⁷. At the structural level, genomic and transcriptomic analyses of R. nobile revealed coordinated upregulation of genes involved in cuticle biosynthesis and epidermal development, contributing to the formation of translucent “greenhouse” bracts that provide thermal insulation and UV protection³⁸. Similarly, Saussurea obvallata adapts to low-pathogen, high-altitude environments through the contraction of disease-resistance gene families, potentially redirecting genetic resources toward abiotic stress responses³⁹. Metabolically, the medicinal plant safflower (Carthamus tinctorius) employs tandem duplications of FAD2/CHS genes to co-regulate linoleic acid and flavonoid biosynthesis pathways⁴⁰, a mechanism also observed in tartary buckwheat (F. tataricum)³⁰. Metabolomic profiling of A. squarrosum demonstrates a positive correlation between flavonoid accumulation and elevation gradients¹⁸, highlighting its role in UV protection and stress mitigation. Reproductively, some alpine plants, such as Dasiphora fruticose, may utilize inbreeding to purge genetic load under pollinator-limited conditions⁴¹. These multi-omics findings delineate complete adaptive cascades in alpine plants, spanning genomic variations (gene family expansion/contraction, duplication), phenotypic adaptations (morphological structures, metabolic traits, reproductive strategies), and environmental responses (UV radiation, low temperature, pollination constraints). Collectively, these findings underscore the power of multi-omics approaches in dissecting complex interactions among genotype, phenotype, and environment.

Here, we present a high-quality chromosome-level genome of the woolly alpine plant S. medusa and integrate comparative genomics, transcriptomics, and metabolomics to elucidate its woolly trait and coordinated regulation of flavonoid synthesis. Our study reveals the genetic basis of trichome development and tissue-specific bioactive compound biosynthesis, offering insights into alpine plant evolution and providing a valuable genomic resource for ecological conservation and medicinal research.

Results

Chromosome-level genome assembly and annotation of S. medusa

The alpine woolly species S. medusa (2n = 34), endemic to the subnival zone, was selected as the focal species for whole-genome sequencing in this study (Fig. 1a). Flow cytometry suggested its genome size at ~4.45 Gb (Supplementary Fig. 1), while K-mer analysis (K = 21) produced an estimate of 3.84 Gb with 0.23% heterozygosity (Supplementary Fig. 2; Supplementary Table 1). Using third-generation sequencing, we generated 93.26 Gb of PacBio HiFi reads, which enabled the construction of a high-quality primary assembly comprising 562 contigs with a total length of 3.8 Gb (contig N50: 116.88 Mb; maximum contig: 196.76 Mb; Table 1, Supplementary Table 2; Supplementary Fig. 3). Chromosomal scaffolding was achieved using 3.5 × 10⁹ Hi-C reads (Supplementary Table 3), resulting in 17 pseudo-chromosomes with a scaffold N50 of 213.91 Mb and a GC content of 40.34%, covering 99.85% of the assembled genome (Supplementary Fig. 4, Supplementary Table 4). Hi-C integration anchored 98.53% of the genome to chromosomal structures, successfully clustering 99 scaffolds into chromosome-level assemblies (Fig. 1b; Table 1, Supplementary Data 1, Supplementary Table 4). Assembly quality was validated using BUSCO analysis, which identified 97.72% of expected genes as complete (Supplementary Table 5; Supplementary Fig. 5). Additionally, remapping of PacBio reads confirmed the assembly’s continuity and accuracy, with a 99.59% alignment rate and 24.39× average coverage, yielding 99.85% genome coverage at ≥1× depth (Supplementary Table 6). These results collectively demonstrate that the S. medusa genome assembly is of high quality, with low heterozygosity and exceptional completeness.

Fig. 1. Genomic features, phylogenetic relationships, and evolutionary dynamics of Saussurea medusa.

a Morphology of wild-grown S. medusa. (i) Wild-grown S. medusa plant showing the characteristic dense, woolly pubescence covering the bracts and leaves. (ii) Excised whole plant, showing the aerial parts and root system. The scale bar in (ii) indicates 5 cm. b Genomic features of S. medusa. The outermost layer shows the 17 pseudochromosomes, with ticks spaced at 50 Mb intervals. Concentric circles, from the innermost to the outermost, represent: GC content, Copia retrotransposon density, Gypsy retrotransposon density, repeat sequence density, gene count, gene density, and chromosome length (Mb). The central colored lines represent syntenic blocks between pseudochromosomes. c Retrotransposition timeline. Kernel density estimation of LTR retrotransposon insertion times. The peak insertion time is estimated at ~0.22 million years ago (MYA). d Time-calibrated phylogeny and gene family dynamics. The phylogeny is dated using MCMCTree, with node ages (MYA) shown at internal branches. Green and red numbers indicate significantly expanded and contracted gene families, respectively (FDR < 0.05). The whole-genome triplication events WGT-γ and WGT-1 are marked. The bar plot on the right shows orthologous gene categories across 10 species: single-copy orthologs (dark blue), multi-copy orthologs (cyan), species-specific paralogs (orange), and unclustered genes (gray). Species abbreviations: Saussurea medusa (Smed), Carthamus tinctorius (Ctin), Erigeron canadensis (Ecan), Helianthus annuus (Hann), Lactuca sativa (Last), Taraxacum mongolicum (Tmon), Cynara cardunculus (Ccar), Taraxacum koksaghyz (Tkok), Cichorium intybus (Cint), Vitis vinifera (Vvin). e Macro-synteny visualization. Collinear blocks between S. medusa (Smed) and two Asteraceae species (S. involucrata [Sinv], C. cardunculus [Ccar]) are shown. Chromosome numbers are labeled above each pseudochromosome.

Table 1.

Statistics of the S.medusa genome assembly and gene annotation

Sequencing	Saussurea medusa	Sequencing	Saussurea medusa
Illumina sequencing		PacBio sequencing
Estimated genome size (by K-mer analysis, Gb)	~3.84	Clean data (Gb)	93.26
Heterozygosity (%)	0.23	Average reads length (bp)	18670
Genome assembly		Genome annotation
Assembled genome size (Gb)	3.8	Annotated protein-coding genes	36469
Contig N50 (Mb)	116.88	Average gene length (bp)	5145.65
Total number of contigs	562	Average exons per gene	5.05
Longest contigs (Mb)	196.76	Mean exon length (bp)	214.2
GC content (%)	38.89	Average CDS length (bp)	1080.99
Coverage depth (X)	24.39	Average intron length (bp)	687.29
scaffold N50 (Mb)	213.91	Annotated ncRNA	54654
BUSCO completeness of assembly (%)	97.72	BUSCO completeness of annotation (%)	97.42

Repeat annotation revealed that 89.86% (3.42 Gb) of the genome comprises repetitive elements, with transposable elements (TEs) accounting for 89.41% of the total assembly. Among these, retrotransposons were the most abundant, representing 63.17% of all TEs. Long terminal repeat (LTR) elements were predominant (62.31%), followed by long interspersed nuclear elements (LINEs, 0.8%) and short interspersed nuclear elements (SINEs, 0.06%) (Supplementary Table 7; Supplementary Fig. 6). Within the LTR category, Gypsy (59.58%) and Copia (37.54%) families dominated, with the majority of insertions estimated to have occurred approximately 0.22 million years ago (Mya) (Fig. 1c, Supplementary Data 2).

Gene annotation resulted in the identification of 36,469 high-confidence protein-coding genes using a combined strategy incorporating homology-based, transcriptome-based, and ab initio prediction methods. These genes had an average length of 5145 bp and contained on average 5.5 exons per gene, with a mean coding sequence (CDS) length of 1080 bp (Supplementary Table 8). In addition, 54,654 non-coding RNAs were annotated, including 25,627 rRNAs, 5415 tRNAs, and 161 miRNAs (Supplementary Table 9). Functional annotation assigned putative roles to 94.92% (34,618/36,469) of protein-coding genes based on homology searches across multiple databases (Supplementary Table 10; Supplementary Fig. 7). BUSCO analysis further supported the completeness of gene annotation, yielding a score of 97.42% (Supplementary Table 11; Supplementary Fig. 8).

In summary, the S. medusa genome is characterized by a large genome size, high repeat content (89.86%), and a well-annotated gene set, providing a valuable genomic resource for exploring the molecular mechanisms underlying alpine adaptation in plants.

Genomic signatures of ancient duplication and divergence in S. medusa

Comparative gene family clustering was conducted among S. medusa and eight other species within the Asterid clade (Carthamus tinctorius, Erigeron canadensis, Helianthus annuus, Lactuca sativa, Taraxacum mongolicum, Cynara cardunculus, Taraxacum koksaghyz, Cichorium intybus), with Vitis vinifera as an outgroup (Fig. 1d). A total of 36,448 genes in the S. medusa genome were classified into 17,890 gene families, including 947 species-specific families. Functional classification was achieved for 93.4% of genes (34,058), with an average of 1.9 genes per family, while 2390 genes remained unassigned (Supplementary Table 12).

Phylogenetic relationships were inferred using maximum likelihood analysis based on 1076 single-copy orthologs across above 10 species (Supplementary Fig. 9, Supplementary Table 13). The divergence between Asterids and Vitis vinifera was dated to approximately 116.9 million years ago (Mya) (95% highest posterior density: 111.8–124.6 Mya), consistent with prior reports (118 Mya)⁴². Notably, S. medusa formed a monophyletic clade with C. tinctorius and C. cardunculus, its closest relatives. The divergence between S. medusa and C. cardunculus was estimated at 21.0 Mya (17.8–24.4 Mya) (Fig. 1d). For comparison, the estimated divergence between S. obvallata and Arctium lappa was ~13.34 Mya³⁹, and between A. lappa and C. cardunculus is 12.8 Mya (6.1–22 Mya)⁴³, suggesting that S. medusa diverged earlier than its congeneric species S. obvallata.

To investigate the polyploidization history of S. medusa, we performed comprehensive analyses integrating synteny-based comparisons, synonymous substitution rate (Ks) distributions and fourfold degenerate third-codon transversion (4DTv) analyses. Intra-genomic synteny dot plots revealed the genome-wide collinearity patterns, while Ks and 4DTv density distributions were analyzed for both paralogous and orthologous gene pairs. Comparative synteny depth analyses were conducted using WGDI⁴⁴ between S. medusa and four representative species, including V. vinifera, L. sativa, C. intybus, and C. tinctorius. Synteny depth patterns showed a 3:1 relationship between S. medusa and V. vinifera and 3:3 relationships with the three Asteraceae species (Supplementary Fig. 10c), indicating the presence of the ancient core eudicot whole-genome triplication (WGT-γ, ~122–164 Mya) and the Asterid-specific WGT-1 (~38–57 Mya) events⁴², consistent with previous reports (Fig. 1d). The intra-genomic synteny patterns and Ks and 4DTv distributions further corroborate this interpretation (Supplementary Fig. 10a, Supplementary Fig. 10b, Supplementary Fig. 11). In addition, comparative synteny analysis among S. medusa and three additional Saussurea species (S. bogedaensis, S. involucrata, and S. obvallata) revealed a high degree of collinearity (Fig. 1e, Supplementary Fig. 12, Supplementary Data 3), indicating strong genomic conservation within this genus. Chromosome 1 of S. medusa displayed conserved gene collinearity with the Asterid species C. cardunculus, suggesting localized genomic stability within this lineage (Fig. 1e, Supplementary Data 3).

Adaptive evolution of S. medusa through gene family remodeling and positively selected genes

Comparative genomic analysis across 10 species identified 2333 expanded and 2014 contracted gene families in S. medusa (Fig. 1d, Supplementary Table 14). KEGG pathway enrichment analysis of expanded gene families demonstrated significant amplification in DNA repair mechanisms (including mismatch repair, homologous recombination, DNA replication, and nucleotide excision repair) (P < 0.05, Supplementary Fig. 13), which aligns with adaptive responses to high-altitude stressors such as hypoxia and elevated UV radiation. These findings are consistent with previous reports of expansions of DNA repair systems in alpine plants³⁹. Additional gene family expansions were observed in metabolic pathways such as starch and sucrose metabolism (related to stress resilience), the tricarboxylic acid (TCA) cycle (energy production), and selenocompound and cyanoamino acid metabolism. In contrast, contracted gene families were enriched in pathways such as tryptophan metabolism and cytochrome P450-mediated xenobiotic metabolism (Supplementary Fig. 14).

Gene Ontology (GO) enrichment analysis of expanded gene families (FDR < 0.05) further highlighted roles in DNA biosynthesis, ionic homeostasis (particularly proton transport and osmoregulation), mitochondrial remodeling, and seedling development (Supplementary Fig. 15). These patterns suggest that S. medusa has adapted to alpine conditions through enhanced UV resistance (via reinforced DNA repair), improved osmotic regulation (through ion transport mechanisms), and optimized energy metabolism.

To investigate adaptive evolution at the sequence level, 58 positively selected genes (PSGs) were identified based on pairwise Ka/Ks analysis (likelihood ratio test p < 0.05, Ka/Ks > 1). Functional annotation revealed enrichment in genes encoding components of the light-harvesting complex (LHC) (e.g., D-tyrosyl-tRNA deacylase, BAH domain proteins), chloroplast-associated functions (e.g., CHUP1, asparagine-tRNA ligase), photosystem-related proteins, and epigenetic regulators (e.g., BAH domain-containing proteins, PRMT3 methyltransferase) (Supplementary Table 15). These molecular adaptations may facilitate photoprotection, sustained photosynthesis under hypoxia, and the establishment of epigenetic stress memory in S. medusa.

Ecological adaptive significance and transcriptional network of trichome morphogenesis in S. medusa

In extreme alpine environments, the dense woolly trichomes of S. medusa function as an adaptive structure for thermal insulation, reducing heat loss and mitigating desiccation. These elongated white trichomes create a static air layer that impedes convective heat transfer and suppresses surface airflow-driven water evaporation⁸(Fig. 2a). Here, morphological observations revealed a uniform distribution of trichomes on the bracts, with densities of 0.72 ± 0.12 trichomes /mm² on the upper leaf and 0.79 ± 0.03 trichomes/mm² on the lower leaf. Except for Group 4, trichome density on the lower leaf is slightly higher compared to the upper leaf (Fig. 2b, Supplementary Table 16). Infrared thermographic measurements (n = 6 plants, three points per plant) confirmed significantly higher surface temperatures on the lower leaf (Fig. 2c, Supplementary Table 17), supporting the speculation of trichome thermal insulation. Further analysis of the visible spectrum (380–780 nm) of normal and hairless leaves revealed that trichome-bearing leaves exhibited higher reflectance compared to hairless leaves, indicating that the trichomes in S. medusa reflect a portion of the visible light incident on the leaf surface (Fig. 2d, Supplementary Data 4). Conversely, hairless leaves showed higher absorptivity and transmittance than their trichome-bearing counterparts, further supporting the hypothesis that trichomes reflect most visible light directly and prevent it from penetrating deeply into the mesophyll, thereby reducing potential photodamage (Fig. 2d, Supplementary Data 4). Based on these findings, it is also speculated that this ecological adaptive phenomenon may be linked to the strong positive selection observed in light-harvesting complex-related pathways in the S. medusa genome.

Fig. 2. Trichome morphology, physiological functions, genetic regulation, tissue-specific expression and functional enrichment of trichome related genes in S. medusa.

a Morphology and leaf epidermal trichome micromorphology of the alpine plant S. medusa. (i) Rosette plant of S. medusa in its native high‑altitude habitat. (ii) Morphology of a single detached leaf (scale bar, 1 mm). (iii) Scanning electron micrograph showing dense and interwoven leaf epidermal trichomes (scale bar, 100 μm). b Trichome density quantification. For each of five randomly selected bracts, 20 microscopic fields were sampled from both the upper and lower leaf surfaces. Box plots show trichome density (counts per mm²), with the box representing the 25th–75th percentiles, the line inside the box indicating the median, and whiskers extending to the minimum and maximum values. Orange boxes represent the upper leaf surface, and blue boxes represent the lower leaf surface. Data are presented as mean ± SD (n = 5 bracts, with 20 fields sampled per leaf surface). A paired t-test was used to assess statistical significance between upper and lower leaf surfaces (***p < 0.001). All error bars represent mean ± SD. c Thermal insulation function of leaves. Temperatures of the upper and lower leaf surfaces were directly measured in situ on field-collected plants. Each line plot shows temperature (°C) across sampling points, with green lines and circular markers representing the upper leaf surface, and orange lines and circular markers representing the lower leaf surface. Data are presented as mean values from 6 plants, with 3 measurement points sampled per plant. d Spectral analysis of trichome-covered (normal) and trichome-removed (hairless) bracts. Reflectance (top panel), absorptivity (middle panel), and transmittance (bottom panel) were measured across the 380–780 nm wavelength range using a CI-710 spectroradiometer. Orange, light blue, and pink lines represent the reflectance, absorptivity, and transmittance of normal leaves, respectively. Light blue, green, and purple lines represent the reflectance, absorptivity, and transmittance of hairless leaves, respectively. Data are presented as mean ± SD (n = 5 leaves per group) e A model of trichome initiation regulation in S. medusa. Two pathways are shown: a promoting pathway and an inhibitory pathway. Arrows indicate stimulation, and T-shaped lines indicate inhibition. Rectangles are positive regulators, hexagons are hormones, and ovals are negative regulators. Hormones (GA, CTK) and transcription factors converge on the GL1–GL3(EGL3)–TTG1 complex to activate TTG2/GL2 and promote trichome initiation. Negative regulators suppress trichome initiation by targeting this complex or downstream factors. f Copy number of trichome-related genes in S. medusa (Smed) and related species. Each cell shows the number of gene copies, with color intensity representing copy number (gradient scale from 0 to 21) g Heatmap showing the tissue-specific expression of trichome-related genes in S. medusa across four organs: leaf, stem, flower, and root. Gene expression levels are normalized by Z-score, with red representing higher expression and blue representing lower expression. Each row corresponds to a gene family member, and each column corresponds to a tissue. h Summary model of trichome function and regulation. Schematic integrates trichome-mediated traits (light reflectance/absorption/transmittance, thermal insulation) with the underlying regulatory network. Positive regulators (expanded families) and negative regulators (contracted families) control trichome development, with tissue-specific expression establishing organ-specific trichome patterns.

The model plant Arabidopsis thaliana has served as a cornerstone for elucidating the regulatory mechanisms of trichome morphogenesis, with its transcriptional network systematically deciphered. Specifically, this developmental process is orchestrated by a hierarchical interplay of multiple transcription factor (TF) families, including the HD-ZIP (homeodomain-leucine zipper) transcription factor GL2^45–47; C₂H₂ zinc finger proteins transcription factor ZFP1/ZFP5/ZFP6 /ZFP8/GIS/GIS2/GIS3⁴⁸; the bHLH (basic helix-loop-helix) transcription factor GL3/EGL3^49,50; and MYB transcription factor GL1/MYB23/TRY/CPC/ETC1/ETC2/ETC3^51,52; WRKY transcription factor TTG2; WD40 transcription factor TTG1, and RAV transcription factor TEM⁵³. Building upon previous research on trichome development in A. thaliana, we reconstructed a putative regulatory network for trichome formation in S. medusa, which was primarily divided into two pathways: trichome initiation and trichome inhibition (Fig. 2e).

To further investigate the expansion and contraction of trichome-related genes in S.medusa, we used known trichome-associated genes from A. thaliana as queries to compare gene copy numbers among S. medusa, four related Asteraceae species (C. cardunculus, C. intybus, T. kok-saghyz, and E. canadensis), and three additional Saussurea species (S. bogedaensis, S. involucrata, and S. obvallata). This analysis identified 48 putative trichome-related gene copies in S. medusa (Fig. 2f; Supplementary Table 18), which were classified into three functional modules: positive regulators, core complex components, and negative regulators of trichome development.

Comparative analysis was primarily conducted relative to A. thaliana, in which the genetic regulatory network controlling trichome development has been extensively characterized⁵⁴. Based on this reference framework (Fig. 2e), we observed a pronounced expansion of several positive regulatory genes involved in trichome initiation (Fig. 2f, Supplementary Fig. 16). In particular, members of the ZFP gene family (ZFP1, ZFP5, and ZFP6/GIS3) showed increased copy numbers in S. medusa (8, 2, and 2 copies, respectively) compared with A. thaliana, while ZFP8/GIS/GIS2 retained the same copy number (three copies). MYB23 exhibited a notable expansion to nine copies. These genes are known to participate in cytokinin and gibberellin signaling pathways associated with trichome initiation (Fig. 2e). Among the genes functioning in both initiation and inhibition pathways, TTG2 displayed substantial expansion (15 copies), whereas the core MBW complex components GL2, GL3, and TTG1 showed largely conserved copy numbers relative to A. thaliana (Fig. 2f). Notably, a similar pattern of gene family composition was observed across the three additional Saussurea species analyzed. Several positive regulators, particularly TTG2, MYB23, and members of the ZFP family, consistently exhibited elevated copy numbers relative to other Asteraceae species, as further supported by positive Z-score values (Supplementary Fig. 16). This shared genomic configuration indicates that expansion of trichome-promoting regulators represents a common evolutionary feature within the genus Saussurea. We further analyzed the tissue-specific expression patterns of the aforementioned trichome-related gene copies across four different tissues (root, stem, flower, and leaf) using transcriptomic data. Most positive regulators showed predominant expression in the leaf, stem, and flower, indicating clear tissue specificity (Fig. 2g, Supplementary Table 19).

To integrate the morphological function and genetic regulation underlying the woolly trait, long and dense trichomes in S. medusa, we summarized the entire study into a comprehensive model (Fig. 2h). The upper section illustrates the functional role of trichomes, which provide thermal insulation against rainwater and reflect a portion of visible light. Additionally, the trichomes absorb a small fraction of light and prevent full transmission into the mesophyll, thereby reducing potential photodamage. The lower section depicts the regulatory mechanism of trichome development. Positive regulators, largely expanded and associated with hormone signaling, exhibit distinct tissue-specific expression. In contrast, negative regulatory genes have undergone significant contraction and show lower expression in flowers.

Tissue-specific expression of flavonoid biosynthetic genes and corresponding metabolite accumulation in S. medusa

S. medusa, a key species in Tibetan ethnomedicine, is traditionally employed to treat ailments such as rheumatoid arthritis, altitude sickness, and gynecological disorders³⁴. Flavonoids, including luteolin, quercetin, apigenin, kaempferol, and rutin, have been identified as its major pharmacologically active constituents⁵⁵. Metabolomic profiling identified 18 major medicinal compounds in S. medusa by TCMSP (Fig. 3a, Supplementary Table 20). Among them, seven flavonoids (luteolin, quercetin, apigenin, kaempferol, rutin, dioscin, and hesperidin) accumulated predominantly in flower tissues. Tissue-specific distribution analysis revealed distinct compartmentalization patterns: phenolic acids peaked in stems and leaves; lignans concentrated in roots and stems; phenylpropanoids, coumarins, and apocarotenoids preferentially accumulated in leaves. Importantly, anthocyanins exhibited flower-specific accumulation, consistent with the traditional heat-clearing applications of S. medusa³⁴. Untargeted metabolomics further uncovered differential spatial accumulation of 41 flavonoid derivatives (adjusted P < 0.05, Fig. 3b, Supplementary Table 21), with hierarchical clustering confirming tissue-specific partitioning. Flavonols and flavones were enriched in flowers and leaves, while roots contained metabolites such as hesperidin, artemetin, centaurein, phellamurin, sinensetin, tamoxifen, and moracetin; stems were enriched in taxifolin and sakuranetin (Fig. 3b). These tissue expressions indicate that the flavonoid components in S. medusa accumulated with different patterns in different tissues, presumably with differences in their functions.

Fig. 3. Integrated analysis of flavonoid biosynthesis, phytochemical composition, and transcriptional regulation in S. medusa.

a Tissue distribution of major phytochemicals. Eighteen compounds were detected, including flavonoids (predominantly accumulated in flowers and leaves, e.g., luteolin, quercetin, apigenin, kaempferol, rutin, dioscin, hesperidin), phenolic acids (stems and leaves), lignans (roots and stems), and phenylpropanoids, coumarins, and apocarotenoids (leaves). b Tissue-specific accumulation of flavonoid metabolites. Heatmap shows normalized abundances of 42 flavonoid derivatives across tissues (red: high accumulation; blue: low accumulation). Flavonols (quercetin, kaempferol, rutin) and flavones (luteolin, apigenin) are enriched in flowers and leaves, while roots uniquely accumulate hesperidin, artemetin, centaurein, and related compounds. c Copy number variation of flavonoid biosynthesis genes in S. medusa and related species. Heatmap displays gene copy numbers, with color intensity representing copy number. Genes include key pathway enzymes: PAL, C4H, 4CL, CHS, CHI, F3H, F3′H, FLS, DFR, ANS and ANR. Species abbreviations: Atha (Arabidopsis thaliana), Ccar (Cynara cardunculus), Cint (Cichorium intybus), Ecan (Erigeron canadensis), Tkok (Taraxacum koksaghyz), Smed (Saussurea medusa), Sbog (Saussurea bogongensis), Sinv (Saussurea involucrata), Sobv (Saussurea obvallata). d Tissue-specific expression of genes in the flavonoid biosynthesis pathway. RNA-seq expression profiles of phenylpropanoid (PAL, C4H, 4CL) and core flavonoid biosynthetic genes (CHS, CHI, F3′H, FLS, DFR, ANS) across four tissues (root, stem, flower, leaf). Red indicates high expression; blue indicates low expression. Three biological replicates per tissue are shown.

To dissect the molecular basis of flavonoid biosynthesis in S. medusa, we first identified flavonoid biosynthetic genes in S. medusa by homology-based searches using well-characterized flavonoid biosynthetic genes from A. thaliana and other plants as references. A total of 27 flavonoid-regulatory gene copies in S. medusa were found (Fig. 3c, Supplementary Table 22, Supplementary Fig. 17). Of these, C4H⁵⁶ and CHS⁵⁷ (2 and 7 copies) showed notable expansion relative to Arabidopsiss (Fig. 3c). To decipher the molecular basis of tissue-specific accumulation of flavonoids, we then integrated transcriptomic data with metabolite profiles. These flavonoid-related genes were primarily expressed in flowers and leaves (Fig. 3d). We then focused on these flavonoid biosynthetic genes significantly differentially expressed in flowers and leaves. Core phenylpropanoid pathway genes, PAL⁵⁸, C4H, and 4CL, were significantly upregulated in flowers (Fig. 3d, Supplementary Table 23), indicating enhanced precursor supply for downstream flavonoid biosynthesis⁵⁹. Branch pathway genes, including CHS, CHI, and FLS, were highly expressed in flowers and leaves, consistent with the accumulation of flavonols such as quercitrin, kaempferol, and rutin⁵⁷. Anthocyanin biosynthetic genes, such as DFR and ANS, showed flower-specific upregulation, correlating with cyanidin-3-glucoside accumulation⁶⁰ (Fig. 3d).

To further validate the coordination between gene expression and flavonoid accumulation levels of four tissues of S. medusa, we conducted integrated transcriptomic-metabolomic correlation analyses (Supplementary Fig. 18). The integrated results demonstrated strong positive associations between the expression of ANS CHI, F3’H genes, and the level of their corresponding flavonoids quercitrin, kaempferol, and rutin (Supplementary Data 5). Overall, these findings align with established flavonoid regulatory mechanisms, elucidating the molecular basis of tissue-specific secondary metabolite accumulation in S. medusa.

To more comprehensively explore cross-omics coordination beyond flavonoid-specific associations, we applied the MOFA2 pipeline to jointly model transcriptomic and metabolomic variation. Three latent factors were identified by MOFA2, among which Factor1 (Transcriptome 46.5%, Metabolome 37.2%) and Factor2 (Transcriptome 41.5%, Metabolome 18.2%) together explained the majority of the variance across both transcriptomic and metabolomic datasets (Supplementary Fig. 19a). Factor1, which explained the largest shared variance across both omics layers, represented global transcriptional-metabolic coordination across tissues (Supplementary Fig. 19b). The top contributing genes and metabolites were broadly distributed and reflected general metabolic activity and core cellular processes, indicating that Factor1 captures the fundamental metabolic state shared among tissues (Supplementary Fig. 19c, d). In contrast, Factor2 clearly separated floral tissues from roots, stems, and leaves, revealing a tissue-specific transcriptional-metabolic signature associated with flower differentiation (Supplementary Fig. 19e). The top features of Factor2 included genes involved in secondary metabolism and regulatory processes, together with diverse metabolites extending beyond flavonoids (Supplementary Fig. 19f). Gene-metabolite network analysis based on high-confidence correlations (|r|> 0.7) further demonstrated coordinated variation among these features, suggesting integrated regulation of multiple metabolic pathways in flowers. Collectively, these integrative analyses extend our initial flavonoid-focused observations and demonstrate that transcriptional and metabolic variations in S. medusa are structured along both global and tissue-specific latent axes, providing a broader understanding of its adaptive metabolic capacity.

Consistent with previous findings that transcription factors mediate flavonoid biosynthesis⁶¹, our analysis uncovered significantly higher expression levels of MYB6/7 in flowers and leaves compared to roots and stems (Supplementary Fig. 20, Supplementary Table 24). Notably, bHLH VII/IIIf transcription factors co-localized with FLS-1 exhibited leaf-specific expression patterns and were positively associated with Quercetin accumulation (Supplementary Fig. 21, Supplementary Table 25)⁶². Integrated transcriptomic and metabolomic analyses have further demonstrated that 4 bHLH VII genes (g6840/g32232/g28453/g26229) showed strong correlations with quercetin content in leaves, suggesting a key regulatory role in flavonoid biosynthesis (Pearson’s r > 0.8, p < 0.05, Supplementary Fig. 22, Supplementary Data 6). In addition, MYB-bHLH-WD40 complexes likely activate anthocyanin biosynthesis through the regulation of ANS, ANR, and DFR. The expression level of ANR showed a strong positive correlation with taxifolin levels, indicating a potential role for ANR in modulating taxifolin metabolism. Collectively, these results demonstrate a coordinated transcriptional regulatory framework underlying the tissue-specific biosynthesis and accumulation of flavonoids in S. medusa, providing insights into the molecular basis of its medicinal properties.

Discussion

The pronounced uplift of the Qinghai-Tibet Plateau has triggered drastic climatic transformations that fostered extreme alpine conditions, marked by persistent low temperatures, intense ultraviolet radiation, and recurrent freeze-thaw cycles. Specialized morphological traits in alpine plants, notably pubescent surfaces and bract-enclosed floral structures, have evolved in alpine plants, combined with physiological adaptations such as freeze tolerance and enhanced antioxidant mechanisms, to ensure survival under hypoxic and high-UV stress³⁸. To dissect the molecular basis of these adaptive traits, such as the woolly trait, we generated a chromosome-scale big genome assembly (~3.8 Gb) of the high-altitude endemic species S. medusa. At the same time, comparative genomic analyses and integrated multi-omics profiling (transcriptomic and metabolomic) were employed to systematically elucidate the molecular regulatory mechanisms of alpine survival and medicinal secondary metabolites formation of S. medusa.

Transposable elements (TEs), particularly LTR retrotransposons, are increasingly recognized as important agents of genome evolution, promoting genetic diversity and facilitating species-specific adaptation through insertional mutagenesis and regulatory rewiring^63–65. In S. medusa, LTR retrotransposons occupy 62.31% of the genome - a markedly higher proportion than in related species such as S. obvallata (43.95%) and R. alexandrae (44.31%)^38,39. This large-scale LTR amplification is likely linked to enhanced genomic plasticity, facilitating rapid adaptation to high-elevation stresses.

In parallel, whole-genome duplication (WGD) events represent another key mechanism driving adaptive evolution in angiosperms⁶⁶. Similar to the sunflower (H. annuus), where WGD enabled floral regulatory innovation⁴², S. medusa experienced two WGD events: the ancestral eudicot WGT-γ (122–164 Mya), the Asterid II-specific WGT-1 (38–57 Mya). Phylogenomic analyses across eight Asteraceae species, with V. vinifera as an outgroup, indicate that S. medusa diverged from C. cardunculus around 21 Mya, and that subsequent diversification occurred within the Saussurea subgenus (Fig. 1d). These structural innovations - extensive LTR expansion and multiple WGDs—are likely to have conferred increased genomic redundancy and functional diversification, facilitating the emergence of complex regulatory and metabolic traits essential for alpine adaptation.

Interestingly, while convergent contraction of disease-resistance (R-gene) families was observed in multiple alpine species, possibly as an energy-conserving strategy in low-pathogen environments³⁹. S. medusa displayed no such contraction. This may reflect its occupation of an ultra-low-pathogen niche where costly defensive investments are unnecessary. Additionally, unlike previous findings highlighting positive selection in reproductive and respiratory regulators (e.g., MMD1, NBS1, HPR), our study identified significant enrichment of genes involved in light-harvesting complexes (LHCs), which support photosynthetic efficiency under high-irradiance alpine environments (Supplementary Table 15). Molecular signatures also include the expansion of DNA repair-related pathways, highlighting genomic strategies to mitigate UV-induced damage, as well as remodeling of primary metabolism^67,68. Specifically, S. medusa exhibits expansion of pathways related to starch and sucrose metabolism and the tricarboxylic acid (TCA) cycle, potentially enhancing carbon storage and energy production under cold, hypoxic conditions. Conversely, contraction of tryptophan metabolism, steroid biosynthesis, and cytochrome P450–mediated xenobiotic pathways may reflect reduced investment in defense under low biotic stress (Supplementary Figs. 13, 14). Despite taxon-specific differences in gene family expansion/contraction, convergent themes emerge across alpine plant genomes. The expansion of DNA repair systems represents a shared response to high UV radiation, while enhanced carbohydrate metabolism supports growth in cold environments with limited resource availability. In S. medusa, through these critical metabolic and regulatory pathways (e.g., LHC, TCA), with TE-mediated regulatory rewiring providing further adaptive flexibility.

Collectively, this study demonstrates that the interplay between WGD events (Fig. 1d), transposable element expansion (Supplementary Fig. 6, Supplementary Table 7), and metabolic network remodeling has fueled S. medusa’s adaptation to extreme alpine environments (Fig. 3d). These mechanisms underpin convergent evolution of morphological and physiological traits across distantly related high-altitude species, offering a theoretical insights into the genomic basis of alpine adaptive radiation.

The dense white epidermal trichomes coverage in S. medusa represents a critical morphological adaptation to the extreme alpine environments. We identified 48 trichome development-related genes through orthologous copies in S. medusa, including key transcriptional regulators such as TTG1, TTG2, GL2, GL3, GIS/GIS2/GIS3, ZFP1/5/6/8, MYB23, TRY/CPC, TEM and ETC (Fig. 2f). Among these, the R2R3-MYB transcription factor GL1, a master regulator of trichome initiation and fate determination, exhibited stage-specific expression during early trichogenesis⁶⁹. Functional studies in A. thaliana demonstrated that GL1 knockout abolishes trichome formation⁵¹, while its constitutive expression suppresses developmental initiation⁷⁰, revealing dual regulatory roles. Importantly, MYB23 shares functional redundancy with GL1, particularly in marginal leaf trichome patterning^71,72. In S. medusa, we identified 9 MYB23 gene copies, with tissue-specific expression profiling showing predominant transcriptional activity in stems and flowers. Moreover, MYB23 promotes trichome development by facilitating GL1-GL3/(EGL3)-TTG1 complex formation, which positively regulates downstream TTG2 and GL2 expression. This regulatory module exhibits evolutionary conservation in cotton (Gossypium hirsutum), where the GhMYB2 ortholog governs fiber cell specification through a homologous pathway⁷³, while GhDEL65 modulates trichome density via MYB-mediated regulation^74,75. In our study, the upstream regulatory framework integrates GA-CTK phytohormone crosstalk via C₂H₂ zinc finger proteins (GIS/GIS2/GIS3, ZFP1/5/6/8) that fine-tune MBW complex activity. Notably, a total of 16 gene copies of GIS/GIS2/GIS3/ZFP1/5/6/8 were identified in S. medusa, with expression detected across root, stem, flower, and leaf tissues. Comparative studies in tobacco (Nicotiana tabacum) revealed parallel mechanisms: NbGIS orchestrates glandular trichome initiation through GA signaling⁷⁶, whereas JcZFP8 overexpression drives ectopic trichome differentiation⁷⁷.

Trichome development is also subject to negative regulation by R3-MYB repressors (CPC /TRY, ETC) and RAV (TEM), which impose lateral inhibition by competitively binding GL3/EGL3 and disrupting MBW complex assembly^78–80. Functional studies in A. thaliana show that TRY mutants develop trichome clusters⁸¹, while CPC overexpression inhibits GL3 activity⁸⁰. In S. medusa, we identified only four genes encoding negative regulators, indicating a significant contraction of the repressive gene repertoire. This numerical imbalance between the substantial expansion of positive regulators and the pronounced contraction of negative regulators may reflect an evolutionary shift favoring trichome proliferation, contributing to the characteristic woolly phenotype.

Furthermore, trichome development in S. medusa appears to be tightly linked to stress-responsive /hormonal pathways. For example, jasmonic acid (JA) has been shown to induce GL3 expression in response to stress in A. thaliana^82,83, and JA-dependent MYB–WD40 signaling regulates cotton fiber development via GhJAZ2⁸⁴. These insights support the hypothesis that trichome formation in S. medusa is regulated by a conserved but lineage-tuned network of transcription factors and phytohormone signaling components, facilitating its adaptation to the harsh high-altitude environment.

While our study focused on gene copy number and tissue-specific expression of known trichome regulators, future investigations using emerging technologies could greatly deepen our understanding of woolly trait formation. For example, single-cell transcriptomics could resolve cell-type-specific gene expression dynamics during trichome differentiation, while spatial transcriptomics would allow mapping of regulatory networks in their native tissue context. Integration of these approaches with live-cell imaging and functional genomics could uncover regulators, delineate the temporal sequence of trichome initiation and maturation, and clarify how these processes are coordinated with adaptive responses to alpine environmental stresses.

Since its inception in 2010, the Medicinal Plant Genomics Program (Herb Genome Program) has enabled the elucidation of biosynthetic pathways for bioactive compounds such as triptolide in Tripterygium wilfordii⁸⁵ and notoginsenosides in Panax notoginseng⁸⁶. Multi-omics integration strategies have further accelerated the mechanistic understanding of medicinal properties^87–89. In this context, we conducted a comprehensive multi-omics analysis of the Tibetan alpine medicinal species S. medusa, integrating genomic, transcriptomic, and metabolomic data. Untargeted metabolomics profiling revealed tissue-specific accumulation patterns of major classes of medicinal metabolites: flavonoids predominantly in flowers/leaves, phenolic acids in stems/leaves, lignans in roots/stems, and phenylpropanoids/coumarins/apocarotenoids mainly in leaves (Fig. 3a). These widespread metabolite distributions support ethnobotanical knowledge of S. medusa’s whole-plant medicinal utility. Among the 18 prioritized bioactive compounds were flavonoids such as luteolin, quercetin, apigenin, kaempferol, rutin, diosgenin, and hesperidin (Fig. 3a, Supplementary Table 20). Flavonoids from Saussurea species are well documented for their antitumor, anti-inflammatory, antioxidant, and lipid-modulatory effects^90–93. Given the pharmacological overlap among species, we identified 41 flavonoid-related metabolites in S. medusa and reconstructed their regulatory networks using multi-omics integration (Fig. 3b, Supplementary Table 21). The concordance between tissue-specific expression of flavonoid biosynthesis and regulatory genes with tissue-resolved metabolite accumulation patterns substantiates the functional specialization of flavonoid metabolism in S. medusa.

The alpine habitat of S. medusa imposes extreme abiotic stressors such as hypoxia and high UV radiation, leading to oxidative damage at the cellular and molecular levels. Flavonoids play a critical protective role under these conditions by scavenging reactive oxygen species and mitigating UV-induced genotoxic stress⁹⁴. Comparative studies have shown that R. nobile accumulates UV-absorbing flavonoids in its translucent bracts⁹⁵, a strategy paralleled in S. medusa, where flavonoid enrichment likely serves both adaptive and medicinal functions. Altogether, our integrative analysis underscores flavonoid biosynthesis as a dual-function trait in S. medusa - contributing to both high-altitude environmental adaptation and its ethnopharmacological utility. These findings lay a molecular foundation for targeted bioengineering and bioprospecting of flavonoid pathways in alpine medicinal plants.

While our study provides a comprehensive multi-omics characterization of S. medusa, including high-quality genome assembly, tissue-specific transcriptomics, and metabolomic profiling, it is important to note that direct functional validation of candidate genes involved in trichome development and flavonoid biosynthesis is currently lacking. The identification of candidate genes such as trichome positive regulators (e.g., GL1, GL3) and key flavonoid biosynthetic enzymes (e.g., PAL, CHS, FLS, ANS) is based on integrated correlative analyses between gene expression and metabolite accumulation, as well as comparative genomics. Although these approaches provide strong indirect evidence, causal relationships between specific genes and phenotypic traits remain to be experimentally verified.

Future research should aim to functionally characterize these candidate genes using molecular genetics approaches such as gene knockout, RNA interference, CRISPR-Cas9-mediated mutagenesis, or overexpression studies. These experiments will be crucial to confirm the regulatory roles of these genes in conferring the woolly trait, flavonoid accumulation, and other adaptive phenotypes. Moreover, combining functional assays with physiological and environmental measurements will help link gene function directly to ecological adaptation in extreme alpine environments.

By explicitly acknowledging this limitation and outlining clear future directions, we provide a framework for subsequent experimental validation that will complement the multi-omics insights presented here. This approach ensures that the genomic and metabolomic resources generated in this study can serve as a robust foundation for mechanistic studies on high-altitude adaptation and medicinal trait evolution in S. medusa.

Methods

Plant materials

Plant tissue samples (root, stem, leaf, and flower) of S. medusa were collected on August 4, 2023, from the scree zone at Bulunkhanbuda Pass in the Eastern Kunlun Mountains near Lake Arak, Dulan County, Qinghai Province, China (35°41′3″ N, 97°19′13″ E; elevation: 4,914 m). Immediately following collection, the fresh specimens were flash-frozen in liquid nitrogen and subsequently stored at –80°C in an ultra-low temperature freezer to ensure the preservation of biomolecular integrity for comprehensive multi-omics analyses, including genomic sequencing, transcriptome profiling, and metabolomic characterization.

DNA extraction, quality assessment, and sequencing

High molecular weight genomic DNA was extracted from 50–100 mg of S. medusa tissue using a CTAB-based method. Tissue was incubated in lysis buffer with proteinase K and SDS at 56 °C for 1–2 h, followed by purification with phenol/chloroform/isoamyl alcohol (25:24:1). DNA was precipitated with isopropanol at –20 °C for 2 h, washed with 75% ethanol, and dissolved in TE buffer. DNA purity was assessed using a NanoDrop™ One spectrophotometer (Thermo Fisher Scientific, USA), with OD260/280 ratios of 1.8–2.0 and OD260/230 ratios of 2.0–2.2. DNA concentration was quantified using a Qubit® 4.0 Fluorometer (Invitrogen, USA), and integrity was monitored by electrophoresis on 1% agarose gels. For sequencing, DNA was further purified using the QIAGEN® Genomic DNA Kit (Cat#13343). Genomic DNA (1–1.5 μg) was fragmented using a Covaris ultrasonicator (Covaris, USA), and fragments of 200–400 bp were size-selected using Agencourt AMPure XP-Medium Kit (Beckman Coulter, USA). Selected fragments underwent end repair, A-tailing, adapter ligation, and PCR amplification. Amplified products were purified, denatured, and circularized to generate single-strand circular DNA libraries. Libraries were quality-validated and sequenced on the MGISEQ2000 platform (MGI, China). Genome size estimation of S. medusa was performed via flow cytometry using tomato (Solanum lycopersicum) as an internal reference standard. Samples were placed in 500 ul Nuclei Extraction buffer, chopped with a sharp blade, then filtered through a 50 um filter after 60 seconds. Followed by the addition of 2000 ul of staining buffer with RNase for 30 minutes in the dark. Nuclei suspension was analyzed by CyFlow Cube6 Flow Cytometer (Sysmex Partec, Muenster, Germany) and the corresponding FCSExpress software.

Data quality control, genome size, and heterozygosity estimation

Raw reads were filtered using fastp (v0.21.0)⁹⁶ to remove adapters, low-quality reads (Q < 5), reads with ≥10% N, and PCR duplicates with default parameters for other options⁹⁷. To assess contamination, 100,000 clean reads were aligned against the NT database. To estimate genome size, repetitive content, and heterozygosity rate, k-mer frequency analysis was outsourced to Xi’an Haorui Biotechnology Co., Ltd. Following quality control of the paired-end sequencing reads, Kmerfreq (GCEv1.0.2)⁹⁸ was used to count the frequencies of 21-nt k-mers, generating a frequency distribution file. Subsequently, the main program of GCE was run with the frequency file and the k-mer length parameter (-k 21). GCE employs a continuous model to account for coverage biases (e.g., GC preference). The observed k-mer species curve was fitted by superimposing dense discrete Poisson distributions, and a Bayesian iterative algorithm was used to estimate the expected coverage depth (c) of single-copy regions and the proportion (a_k) of each depth component. During iteration, low-frequency erroneous k-mers were automatically filtered out, and the influence of repetitive sequences on the main peak was corrected. Finally, based on the formula G = n_k-mer/c, where n_k-mer is the total number of k-mers and c is the corrected coverage depth of unique regions, the pipeline outputs the estimated genome size and generates a histogram of the k-mer frequency distribution⁹⁸. Heterozygosity and repeat content were evaluated by analyzing 21-mer frequency peak distributions, with reference to simulated Arabidopsis data⁹⁹.

PacBio HiFi sequencing, genome assembly and quality assessment

PacBio sequencing data were processed using SMRT Link to generate subreads, followed by circular consensus sequencing (CCS) to produce HiFi reads with >99% accuracy (Q20 threshold). The nuclear genome was assembled by Xi’an Haorui Biotechnology Co., Ltd. using hifiasm (v0.19.9)¹⁰⁰ with default parameters, generating a primary assembly by randomly selecting one allele per bubble at heterozygous sites supported by at least three reads per allele¹⁰¹. Assembly completeness was evaluated using BUSCO (v5.2.2)¹⁰² with the embryophyte lineage database and default parameters. For sequence consistency, Illumina short reads were aligned to the assembly using BWA (v0.7.17)¹⁰³ with default parameters; single-nucleotide variants were called with samtools (v1.9) and bcftools (v1.9), and the homozygous SNP rate was calculated as the assembly error rate with default parameters. HiFi reads were aligned to the assembly using minimap2 (v2.24)¹⁰⁴ to compute per-base depth, and GC-depth scatter plots were generated in 50-kb sliding windows to assess potential contamination¹⁰⁴ with parameter ‘-x map-hifi -t 30 --secondary=no’.

Hi-C library preparation and chromosome-level scaffolding

Hi-C libraries were prepared using DpnII digestion followed by biotin labeling, proximity ligation, and purification, then sequenced on the Illumina HiSeq or MGI platform (PE150). Raw reads were filtered using fastp (v0.21.0) to remove adapters, reads containing N bases, and low-quality reads (>20% bases with Q < 20) with default parameters. Clean reads were processed with HiC-Pro¹⁰⁵ using bowtie2 (v2.3.5)¹⁰⁶ for alignment with default parameters; reads spanning ligation junctions were trimmed and re-aligned¹⁰⁷. Uniquely mapped read pairs located on different restriction fragments were retained as valid interactions, while self-ligation and dangling-end pairs were discarded¹⁰⁸. Based on Hi-C interaction patterns, contigs were clustered, ordered, and oriented using 3D-DNA (v180922)¹⁰⁹ to generate a chromosome-level assembly with default parameters.

Gene annotation, repetitive sequence and non-coding RNAs analysis

Repeat annotation was first performed. Tandem repeats were detected using GMATA (v2.0)¹¹⁰ and TRF (v4.09)¹¹¹ with default parameters. Concurrently, de novo repeat libraries were constructed with MITE-Hunter (v1.0) and RepeatModeler (v2.0.3), followed by classification and annotation of repetitive elements using RepeatMasker (v4.0.9)¹¹² with default parameters. Gene structure prediction integrates multiple lines of evidence. An integrated gene set was constructed using EVM (v1.1.1) by combining ab initio predictions from Augustus (v3.3.3)¹¹³, homology-based evidence from GeMoMa (v1.8)¹¹⁴, and transcriptome-supported data. To validate the robustness of the annotation, an integrated Helixer + BRAKER3 pipeline¹¹⁵ was also applied; the results were highly consistent with the EVM-based annotation. Therefore, the original EVM gene set was retained for subsequent analyses. Functional annotation of protein sequences was performed by searching against multiple databases, including SwissProt, TrEMBL, NR, KEGG, KOG, and GO. Protein domain analysis was conducted using InterProScan (v5.52-86.0), and functional validation was executed via BLASTp (v2.12.0) with an E-value cutoff of ≤1e. Non-coding RNAs were characterized using dedicated tools: tRNAscan-SE (v2.0.9)¹¹⁶ for tRNA identification, Infernal cmscan (v1.1.4)¹¹⁷ for other ncRNA searches, and RNAmmer (v1.2)¹¹⁸ for rRNA prediction with default parameters. Genome structural and functional annotation was completed by Xi’an Haorui Biotechnology Co., Ltd.

Phylogenetic analysis, gene family expansion/contraction, and positive selection analysis

Protein sequences from Carthamus tinctorius, Erigeron canadensis, Helianthus annuus, Lactuca sativa, Taraxacum mongolicum, Cynara cardunculus, Taraxacum koksaghyz, Cichorium intybus and Vitis vinifera were aligned using OrthoFinder (v2.4.0)¹¹⁹ with BLASTP (E-value < 1e–5). Single-copy orthologous genes were aligned with MUSCLE (v3.8.31)¹²⁰ and filtered using Gblocks (v0.91b)¹²¹ with default parameters. A maximum likelihood phylogenetic tree was constructed with RAxML (v8.2.12) under the JTT + F + GAMMA model with 1000 bootstrap replicates. Divergence times were estimated using MCMCtree in PAML (v4.9)¹²² with fossil calibrations and default parameters from TimeTree. Gene family expansion and contraction were analyzed using CAFÉ (v4.2)¹²³ under a birth-death model (likelihood ratio test [LRT], p < 0.05). To identify statistically significant expansions or contractions, we applied a Z-transform approach, calculating Z-scores for each gene family relative to the mean and standard deviation across all species (|Z|>2). Positively selected genes were identified by calculating Ka/Ks ratios for orthologous pairs using Codeml in PAML (v4.9) (LRT p < 0.05; Ka/Ks > 1). GO and KEGG enrichment analyses were performed using clusterProfiler (v4.0) with a hypergeometric test. P-values were adjusted for multiple testing using the Benjamini–Hochberg procedure, and terms with adjusted P < 0.05 were considered significantly enriched. All comparative genomic analyses, including synteny, gene family clustering, and divergence time estimation, were performed by Xi’an Haorui Biotechnology Co., Ltd.

Whole-genome duplication and genomic synteny analysis

Whole-genome duplication events were detected via the fourfold synonymous third-codon transversion (4DTv) method using the calculated_4DTV_correction.pl script with default parameters (https://github.com/JinfengChen/Scripts). Homologous gene pairs spanning chromosomes were identified with JCVI (v1.3.8)¹²⁴ with default parameters, and syntenic blocks were annotated for gene positional information extraction. Synteny depth analysis was also performed with JCVI (v1.3.8) with default parameters. Synteny dot plot of the intra-genomic synteny pattern of S. medusa was performed with WGDI (v0.75)⁴⁴ with default parameters.

Transcriptome sequencing and analysis

Total RNA was extracted from different tissues (root, stem, leaf, and flower) of Saussurea medusa using TRIzol reagent (Invitrogen, USA) to investigate tissue-specific expression patterns of trichome and flavonoid-related genes. RNA integrity and genomic DNA contamination were assessed by electrophoresis on 1.5% agarose gels. RNA purity was determined using a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific, USA) with OD260/280 and OD260/230 ratios measured. RNA integrity was precisely quantified using a Qsep100 system (Bioptic, Taiwan) and verified on an Agilent 2100 Bioanalyzer (Agilent Technologies, USA) with the RNA 6000 Nano Kit. Only high-quality RNA samples (RNA integrity number ≥ 7.0) were used for subsequent library construction. mRNA was enriched from total RNA using poly-T oligo-attached magnetic beads (Thermo Fisher Scientific, USA). Sequencing libraries were prepared using the VAHTS Universal V10 RNA-seq Library Kit (MGI Tech, USA) with unique dual-index adapters following the manufacturer’s protocol. Briefly, the enriched mRNA was fragmented into short fragments. First-strand cDNA was synthesized using random hexamer primers and M-MuLV reverse transcriptase (RNase H⁻). Second-strand cDNA synthesis was performed using DNA polymerase I and RNase H. The double-stranded cDNA was purified, subjected to end repair, 3’ adenylation, and ligation to adapters. Adapter-ligated fragments were size-selected (250–300 bp) using AMPure XP beads (Beckman Coulter, USA), followed by PCR amplification. Library quality was validated on the Qsep100 system, and Agilent 2100 Bioanalyzer, and library concentration was quantified using a Qubit 4.0 Fluorometer (Invitrogen, USA). Qualified libraries were pooled and sequenced on the Illumina NovaSeq 6000 platform (Illumina, USA) with paired-end 150 bp reads. Raw fluorescence signals were converted to sequence data by Illumina’s base-calling software (bcl2fastq v2.20) with default parameters. Raw sequencing data were processed with fastp (v0.19.7)⁹⁶ to remove low-quality reads (Q < 20). Clean reads were aligned to the S. medusa reference genome using HISAT2 (v2.2.1)¹²⁵ with default parameters. Transcript abundance was quantified as fragments per kilobase of transcript per million mapped reads (FPKM) values¹²⁶ using StringTie (v2.1.6)¹²⁷. Differentially expressed genes (DEGs) between tissue comparison groups were identified using DESeq2 (v1.30.1)¹²⁸ with criteria of |log₂ fold change|≥ 1 and adjusted P-value < 0.05. K-means clustering analysis of expression patterns was performed using the clusterGVis (v1.0.0) R package with default parameters. RNA library construction and transcriptome sequencing were commissioned to Xi’an Haorui Biotechnology Co., Ltd.

Metabolite extraction, LC-MS/MS analysis, data preprocessing and annotation

For metabolite extraction, tissue samples from different organs (root, stem, leaf, and flower) of S. medusa were processed separately. Frozen plant tissues (20 mg ± 1 mg) were lyophilized and homogenized with beads in 1000 μL of extraction solution (methanol:acetonitrile:water, 2:2:1, v/v) containing deuterated internal standards¹²⁹. The mixture was vortexed for 30 s, homogenized (35 Hz, 4 min), and sonicated in a 4 °C water bath for 5 min; this cycle was repeated three times¹³⁰. Samples were incubated at –40 °C for 1 h to precipitate proteins, followed by centrifugation at 12,000 rpm (13,800×g, R = 8.6 cm) for 15 min at 4 °C. The supernatant was transferred to a glass vial for analysis. A quality control (QC) sample was prepared by pooling equal aliquots of all supernatants¹³¹.LC-MS/MS analysis was performed using a Vanquish UHPLC system (Thermo Fisher Scientific, USA) coupled to an Orbitrap Exploris 120 mass spectrometer (Thermo Fisher Scientific, USA). Chromatographic separation was achieved on a Phenomenex Kinetex C18 column (2.1 mm × 50 mm, 2.6 μm) with mobile phase A (0.01% acetic acid in water) and mobile phase B (isopropanol:acetonitrile, 1:1, v/v)¹³¹. The autosampler temperature was set to 4 °C, and the injection volume was 2 μL. The mass spectrometer was operated in information-dependent acquisition (IDA) mode controlled by Xcalibur software (Thermo Fisher Scientific, USA). ESI source parameters were set as follows: sheath gas flow rate, 50 Arb; auxiliary gas flow rate, 15 Arb; capillary temperature, 320 °C; full MS resolution, 60,000; MS/MS resolution, 15,000; collision energy, stepped NCE 20/30/40; spray voltage, 3.8 kV (positive ion mode) or –3.4 kV (negative ion mode)¹³⁰. Raw data files were converted to mzXML format using ProteoWizard (v3.0.2) with default parameters. Peak detection, extraction, alignment, and integration of the mass spectrometry data were outsourced to Nanjing Jisihuiyuan Biotechnology Co., Ltd. The company performed these analyses using its proprietary R script based on the XCMS package (v3.12)¹³² with default parameters. Metabolites were annotated by matching MS/MS spectra against spectral libraries and accurate mass measurements against public databases¹³³. Integrated univariate and multivariate statistical analyses to interrogate the data from multiple perspectives. Univariate analysis was performed using Student’s t-test¹³⁴, while multivariate analysis was based on an orthogonal partial least-squares discriminant analysis (OPLS-DA) model with default parameters. Differential metabolites were screened by combining the variable importance in projection (VIP > 1) from the first principal component of the OPLS-DA model with p-values (p < 0.05) from the t-test and fold change¹³⁵. Specifically, metabolites with VIP > 1, p < 0.05, and fold change > 1 were classified as upregulated, whereas those with VIP > 1, p < 0.05, and fold change < 1 were classified as downregulated. These statistical analyses were conducted using R software (v4.x) and its ropls package (v1.26.4)¹³⁶. The identified differential metabolites were subsequently subjected to pathway annotation using the KEGG database to elucidate the underlying biological processes and metabolic pathways involved¹³⁷. The identified compounds were classified into four levels according to the confidence of annotation: level 1 corresponds to metabolites that match reference standards in MS1, MS2, and retention time (RT); level 2 refers to those that match public databases in both MS1 and MS2; level 3 includes metabolites that match compounds in MS1, MS2, and RT; and level 4 represents unknown compounds.

Integrated metabolomic and transcriptomic analysis, and flavonoid biosynthesis-related gene and metabolites identification

Differentially expressed genes were identified through six pairwise comparisons among the four tissue types (root, stem, leaf, and flower). Metabolic pathways of S. medusa were annotated based on the KEGG database, with a particular focus on the flavonoid biosynthesis pathway (map00941). Characteristic metabolites with reported pharmacological activities (e.g., flavonoids and phenolic acids) were prioritized from the untargeted metabolomic data by screening against the TCMSP database (https://old.tcmsp-e.com/tcmsp.php) and published bioactive components from medicinal plants. Core flavonoid constituents were determined via KEGG pathway annotation, and their contents in each tissue were quantified and visualized for tissue-specific distribution patterns using heatmaps generated on the Qiantang Life Science Cloud Platform (https://www.rnaseq.top). To elucidate the molecular basis of flavonoid biosynthesis, the genome data of E. canadensis, C. cardunculus, T. koksaghyz and C. intybus from NCBI, the S. involucrate and S. bogedaensi from the NGDC, and the S. obvallata from the CNGBdb, and reference sequences of flavonoid transcription factor families (bHLH and MYB) from A. thaliana were retrieved from NCBI. Homologous gene copies in S. medusa and its closely related species were identified via BLASTP (v2.11.0) with an E-value threshold < 1e^-5. Genes from the same family were grouped and used to construct a phylogenetic tree together with Arabidopsis genes; those clustering on the same branch as Arabidopsis genes were considered orthologous copies. Conserved protein domains were validated using the NCBI Conserved Domain Database (CDD, v3.20) (https://www.ncbi.nlm.nih.gov/cdd), retaining only sequences with intact functional domains. Tissue-specific expression patterns (root, stem, leaf, and flower) of these genes were analyzed using transcriptomic data, and hierarchical clustering heatmaps were generated on the Qiantang Life Science Cloud Platform (https://www.rnaseq.top). Finally, correlation coefficients between the expression levels of flavonoid-related genes and the accumulation levels of corresponding metabolites were calculated using R software (v4.x), and correlation heatmaps were generated using TBtools (v2.0)¹³⁸ to reveal the synergistic regulatory relationships between genes and metabolites.

Ecological and spectroscopic experiments

Leaves of S. medusa were dehydrated through an ethanol series, thoroughly dried using a critical point dryer, and mounted on metal stubs with conductive adhesive. The leaf surface morphology was examined and photographed under the SEM. Six individual plants were selected in field experiments, with three leaves per plant randomly marked. Instantaneous temperatures of upper leaf/lower leaf surfaces at identical plants were measured using infrared thermometers. Absorption, reflection, and transmission rates of bracts (380–780 nm) were quantified with a CI-710 fiber optic spectrometer (Zealquest, Shanghai). Upper leaf/lower leaf epidermises were stained with 1% safranin (Sigma-Aldrich, USA) for 3 min, followed by gradient ethanol dehydration (50%, 70%, 80%, 90%, 95%, 100%), and trichome density was quantified within 1 mm² microscopic fields with 20 fields counted per leaf. Statistical analysis was performed using GraphPad Prism (v11.0). Trichomes were mechanically removed from treatment groups using precision forceps, while intact trichomes served as controls. Fresh leaves were collected, and five randomly selected leaves underwent epidermal peeling.

Trichome-related gene copies identification and analysis of S. medusa and closely related Species

Using protein sequences of A. thaliana trichome-related genes as queries, comparative analyses of gene copy number variation were conducted using uniform criteria between A. thaliana and S. medusa, and closely related species E. canadensis, C. cardunculus, T. koksaghyz, C. intybus, S. involucrate, S. bogedaensi and S. obvallata. Through BLASTP (v2.11.0) alignment with this species¹³⁹, genes from the same family were grouped and used to construct a phylogenetic tree with Arabidopsis genes. Genes clustering on the same branch as Arabidopsis genes were identified as orthologous copies. Leveraging established A. thaliana trichome research frameworks, we constructed a regulatory network and performed functional enrichment analysis on the Qiantang Life Science Cloud Platform (https://www.rnaseq.top). Tissue-specific expression patterns (root, stem, leaf, flower) were analyzed using transcriptomic data, with hierarchical clustering heatmaps generated to visualize spatiotemporal expression profiles.

Statistics and Reproducibility

Reproducibility and replicates: Biological replicates were defined as independent samples from different individual plants. For transcriptomics analysis, three biological replicates were used for each tissue (root, stem, leaf, and flower). For metabolomics analysis, six biological replicates were applied per tissue. For trichome density quantification, five bracts were examined with 20 microscopic fields detected per leaf surface. For leaf temperature measurement, six individual plants were measured at three independent points on each leaf surface. For spectral analysis, five leaves were measured per group. For descriptive experiments including SEM imaging and Hi-C heatmap analysis, representative results were presented, and all experiments were independently repeated twice to ensure consistent and reproducible outcomes.

Statistical analysis: All statistical analyses were performed using R software (v4.0 or later) with the ropls package (v1.26.4) and GraphPad Prism (v11.0). Two-tailed paired or unpaired Student’s t-tests were applied for two-group comparative analyses. Specifically, paired t-tests were used to compare trichome density and temperature between the upper and lower leaf surfaces.

For transcriptomic data, differentially expressed genes (DEGs) were identified using DESeq2 (v1.30.1) with the threshold of |log₂(fold change)| ≥ 1 and adjusted P < 0.05 based on the Benjamini–Hochberg correction. For metabolomic data, univariate statistical analysis was conducted via Student’s t-test, and multivariate analysis was performed using the orthogonal partial least-squares discriminant analysis (OPLS-DA) model. Differential metabolites were screened by integrating OPLS-DA variable importance in projection (VIP > 1), t-test P values (P < 0.05), and fold change. Two-tailed Pearson’s correlation coefficient was used for correlation analyses between gene expression levels and metabolite accumulation. All quantitative data are presented as mean ± standard deviation (SD). Detailed sample sizes (n) for each analysis are provided in the figure legends and Methods.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

G.J.H. and R.J.D. initiated and designed the research. G.J.H. and L.M.Y.J. performed the experiments and analyzed the data. G.J.H., M.Y.D. M.X.W., and L.M.Y.J. contributed to materials collection. G.J.H. wrote the manuscript. T.H., X.Q.J. and R.J.D revised the manuscript. All authors read and approved the final manuscript.

Peer review

Peer review information

Communications Biology thanks Heng-Chang Wang, Yongzhi Yang, and the other anonymous reviewer(s) for their contribution to the peer review of this work. Primary Handling Editors: Zhaoming Qi and David Favero. A peer review file is available.

Data availability

All data supporting the findings of this study are publicly available as described below. Raw genome sequencing reads and raw RNA-seq data have been deposited in the NCBI Sequence Read Archive (SRA) under BioProject accession PRJNA1278884 and in the NGDC GSA database under accession CRA037854. The chromosome‑level genome assembly and corresponding gene annotation files are available in the NGDC Genome GWH database under accession GWHIGTP00000000.1. Processed transcriptomic and metabolomic datasets are hosted in the Figshare repository (https://doi.org/10.6084/m9.figshare.31527019.v2), and raw metabolomics data have been submitted to MetaboLights under accession MTBLS14630. All numerical source data underlying the main text figures have been organized into Excel‑formatted supplementary files and fully uploaded. The source data for each figure are explicitly specified as follows: Fig. 1b in Supplementary Data 1; Fig. 1c in Supplementary Data 2; Fig. 1d in Supplementary Tables 13, 14; Fig. 1e in Supplementary Data 3; Fig. 2b in Supplementary Table 16; Fig. 2c in Supplementary Table 17; Fig. 2d in Supplementary Data 4; Fig. 2f in Supplementary Table 18; Fig. 2g in Supplementary Table 19; and Figs. 3a–3d in Supplementary Tables 20–23.

Code availability

Custom scripts for k-mer-based genome size estimation were developed by Xi’an Haorui Biotechnology Co., Ltd. through commercial technical cooperation. Mass spectrometry data processing pipelines (peak detection, extraction, alignment, and data integration for metabolomic datasets) were developed by Nanjing Jisihuiyuan Biotechnology Co., Ltd. All these custom scripts and analytical pipelines are proprietary intellectual properties of the corresponding service companies and cannot be shared or deposited in public repositories. All essential parameters and full analytical workflows needed for result replication have been comprehensively described in the Methods section of this manuscript.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s42003-026-10521-4.