Evolutionary tuning of a biosynthetic gene cluster drives furanocoumarin accumulation and diversification

Abstract

Deciphering evolutionary drivers of biosynthetic pathways could enhance bioactive compound production. In Angelica, interspecific variation in furanocoumarins (FCs) accumulation reflects divergent pathway evolution. Here, we conduct comparative genomics between high-FC Angelica sensu stricto (s.s.) and low-FC Angelica sensu lato (s.l.) species. We reveal an FC biosynthetic gene cluster (BGC) comprising core enzymes (p-coumaroyl-CoA 2’-hydroxylases (C2’Hs), prenyltransferases (PTs)) and peripheral O-methyltransferases (OMTs). The ancestral Angelica s.l. clade retains an FC BGC configuration with OMTs on separate chromosomes and PTs performing only C-prenylation. In contrast, Angelica s.s. evolves an FC BGC, where core enzymes and OMTs co-localise on the same chromosome, with C2’H copy number expansion correlating with elevated expression and PTs enabling both C- and O-prenylation, collectively enhancing FC production and structural diversity. These findings elucidate how BGC architecture, gene copy number and functional innovation collectively drive phytochemical innovation, providing a blueprint for engineering medicinal FC biosynthesis.

Most Angelica species possess pharmaceutical value due to furocoumarin (FC) accumulation. Here, the authors compile the genomes of five species (three high-FC species and two low-FC species) and reveal the underlying mechanisms leading to the differentiated patterns of FC accumulation.

Introduction

Plant secondary metabolites (SMs) play a crucial role as biochemical signals in response to abiotic and biotic stress¹. These SMs often exhibit lineage-specific profiles among phylogenetically related species. Specifically, structurally diverse SMs are derived from the same skeleton but modified through different chemical processes, such as methylation, hydroxylation, and prenylation^2,3. For example, cucurbitacins, characteristic SMs of Cucurbitaceae species, exhibit structural variation among cucumbers, melons, and watermelons, which generate cucurbitacin C (CuC), CuB, and CuE, respectively². Similarly, the Symphyomyrtus subgenus in Eucalyptus exhibits higher levels of cyanogenic glycosides than other subgenus species⁴, which also produce a wide array of glycosides.

Such lineage-specific chemical diversity is produced and maintained through divergent evolution. The divergent evolution of SM diversity could be correlated with selection imposed on genes and regulatory elements involved in the SM biosynthetic pathway. Dissecting genetic signals of divergent evolution of biochemical diversity will not only enhance our understanding of the assembly and evolution of the SM biosynthetic pathway, but also supply innovative strategies for improving metabolic engineering of crucial SMs^5,6. On the one hand, tracing natural variations in gene sequences responsible for SM biosynthesis enables the rational optimization of enzyme performance and fine-tuning of pathway expression in engineered systems^7,8. On the other hand, the systematic identification of critical branch point enzymes and tailoring steps facilitates the rational design of biosynthetic routes, as well as the mix-and-match of enzymes from different species into a single chassis, expanding the chemical space accessible in an engineered organism^5,9. Thus, the genetic underpinnings of divergent evolution hold substantial potential for enhancing the production of valuable plant SMs.

Biosynthetic gene cluster (BGC), i.e., multiple genes involved in a single metabolic pathway are clustered together in close proximity on the genome, possibly enhances efficiency of SM production^10–12. BGCs have been detected for biosynthesis of multiple bioactive compounds, including 2,4-dihydroxy-1,4-benzoaxin-3-one (DIBOA) in maize and other Poaceae species¹³, benzylisoquinoline alkaloids (BIAs) in Papaver¹⁴, and cucurbitacins in Cucurbitaceae². BGCs usually contain at least two types of functional genes with a minimum of three, including signature and tailoring enzymes. Based on the spatial arrangement of genes coding signature enzymes or tailoring enzymes, BGCs can be categorized into archetypal clusters, core clusters plus peripheral genes involving a single tailoring gene family, and core clusters plus satellite subgroups involving multiple tailoring gene families (Supplementary Fig. 1)^10,12. The structural organization of BGCs is crucial for SM diversity across different lineages. For example, cucumbers, melons, and watermelons share a conserved six-gene core cluster but showed species-specific variation of gene arrays coding tailoring enzymes². Cytochromes P450 CYP88A60 and basic helix-loop-helix (bHLH) transcription factors are on separate chromosomes in cucumbers and melons, whereas in watermelons they co-localize on one chromosome². Nevertheless, bHLH in both melons and watermelons is located within the domestication sweep region, which resulted in chemotype differences associated with bitterness loss in domesticated varieties². Another distinct evolutionary pattern characterizes the DIBOA BGCs in the Poaceae family. DIBOA biosynthetic genes (BX1-BX5) in Zea mays form a cluster on chromosome 4, while those in rye and wheat, which exhibit lower benzoxazinoid diversity, are arranged into two subclusters on separate chromosomes, with BX1 and BX2 on one chromosome and BX3-BX5 on another^13,15. Although BGCs have been widely recognized to be crucial for SM structural diversity and content, how the architecture, composition and evolutionary trajectories of BGCs collectively shape lineage-specific chemical diversity remains elusive.

Furocoumarins (FCs), with a fused structure of coumarin and furan rings, exhibit diverse biological and pharmaceutical properties such as cytotoxicity, photosensitivity, and antimicrobial and insecticidal activities^16,17. FC biosynthesis is initiated by the phenylpropanoid pathway, where phenylalanine is converted to p-coumaroyl CoA via an enzymatic cascade of phenylalanine ammonia-lyase (PAL), cinnamate 4-hydroxylase (C4H) and 4-coumarate-CoA ligase (4CL)^18,19. The signature enzyme p-coumaroyl CoA 2’-hydroxylase (C2’H) then catalyzes hydroxylation followed by subsequent cyclization, forming umbelliferone, the coumarin scaffold²⁰. Subsequent prenyltransferase (PT) mediates C6/C8 prenylation of umbelliferone, generating demethylsuberosin (linear FC precursor) or osthenol (angular FC precursor) respectively, establishing structural divergence²¹. CYP-catalyzed furan ring formation followed by tailoring reactions (prenylation, O-methylation, hydroxylation, glycosylation) yields diverse FC derivatives including imperatorin, isoimperatorin, phellopterin, cnidilin and isopimpinellin²². Despite significant progress in elucidating FC biosynthesis, the pathways for several FCs, such as alloisoimperatorin, phellopterin, and cnidilin, as well as the multifunctionality of key enzymes like PTs and O-methyltransferases (OMTs), remain unclear. FCs are sporadically distributed among angiosperms, with prominent occurrences in Fabaceae, Moraceae, Apiaceae, and Rutaceae, where evidence suggests convergent evolution of the biosynthetic pathway^23,24. Previous study in Peucedanum praeruptorum revealed one FC BGC, including C2’Hs, PTs and CYP cyclases²⁵. However, the composition, gene arrangement, structural variations and evolutionary divergence of the FC BGC are largely unknown.

Angelica is a large and taxonomically complex genus comprising approximately 110 species^26,27. Most Angelica species hold significant cultural and economic value as traditional herbs, fragrances, ornamentals, and functional foods, with certain species historically revered as “angel herbs” for their role in combating the plague in Europe and their recognized usage in treating amenorrhea, dysmenorrhea, menopausal disorders, and anemia in Asian traditional medicine²⁸. Angelica is well-known for both the content and variety of FCs²⁸. Angelica is phylogenetically polyphyletic, with Archangelica, Sinodielsia, and Ostericum (satellite genera of Angelica) clustered within the previously defined Angelica clade, which is defined as Angelica sensu lato (s.l.)^26,27; whereas the core Angelica clade includes some well-known medicinal plants, such as A. keiskei, A. biserrata, and A. dahurica, which are referred to as Angelica sensu stricto (s.s.) hereafter. The Sinodielsia clade, named Angelica s.l. Hereafter, includes A. sinensis and Levisticum officinale (also known by its taxonomic synonym ‘Angelica levisticum’), representing a phylogenetically distant group in Angelica and exhibiting significant differences in morphology and biochemical compositions, particularly with low FC content compared to Angelica s.s. clade^28,29. The contrast of the two groups provides an ideal system to explore the evolutionary shift of BGC responsible for FC biosynthesis.

Here, we de novo assemble the genomes of A. keiskei, A. biserrata, and L. officinale, complementing previously reported genome sequences of A. sinensis²³ and A. dahurica³⁰. We construct metabolomic maps to elucidate metabolic profiles of FCs between Angelica s.s. and Angelica s.l. clade. Through integrated multi-omics and in vitro enzymatic assays, we functionally verify nearly all enzymatic steps in the downstream modification cascade of FC biosynthesis, uncovering the biosynthetic routes to alloisoimperatorin, phellopterin, and cnidilin. Moreover, we reveal gene compositions and arrangements of the FC BGC, crucial for the divergent evolution of FCs between the two Angelica clades. These findings offer insights into the molecular and evolutionary strategies shaping FC diversity and provide a valuable foundation for future research in metabolic engineering and crop improvement.

Results

Lineage-specific divergence in two phylogenetically related Angelica clades

We employed PacBio high-fidelity (HiFi) long-read sequencing to generate contigs for three Angelica species, A. biserrata, A. keiskei and L. officinale (Supplementary Table 1; Supplementary Method 1). These contigs were further scaffolded with high-throughput chromosome conformation capture (Hi-C) data to yield chromosome-level genome assemblies, resulting in 11 chromosomes for each Angelica species (Fig. 1a and Supplementary Fig. 2), consistent with the base chromosome number (n = 11) estimated via karyotype experiments³¹ and Chromosome Counts Database (CCDB)³². The anchoring rates were 96–98% for the three species (Supplementary Data 1). The final genome assemblies were 3.98 Gb with a scaffold N50 of 370.15 Mb for A. biserrata, 4.78 Gb with a scaffold N50 of 429.00 Mb for A. keiskei, 4.73 Gb with a scaffold N50 of 442.92 Mb for L. officinale, closely matching the estimated genome size by k-mer analysis (Fig. 1a; Supplementary Tables 2 and 3; Supplementary Method 2). The assemblies achieved 96–99% completeness of the viridiplantae gene set using Benchmarking Universal Single-Copy Ortholog (BUSCO)³³, and LTR Assembly Index (LAI)³⁴ and Merqury QV³⁵ retrieved over 15 and 36, respectively, for the three species (Supplementary Tables 3–5 and Supplementary Data 2). We assembled three large genome sequences, a significant achievement given that less than 4% of published plant genome sequences surpass this size³⁶ (Supplementary Fig. 3), highlighting their potential contribution for understanding the genetic and evolutionary dynamics of large genomes. Furthermore, a total of 67,192, 64,383 and 58,081 protein-coding genes were annotated in A. biserrata, A. keiskei and L. officinale, respectively, with 2954-3005 bp of average length of genes and an average of four exons per gene in three species (Supplementary Data 1; Supplementary Method 3). 90–96% of complete BUSCO genes of viridiplantae had been identified³³ (Supplementary Table 3).

Fig. 1. Genome assembly and evolution of Angelica genomes.

a Overview of three de-novo assembled genome sequences. Track A, B and C corresponded to chromosome length, gene density and transposable element density, respectively. The inner images showed their morphological characteristics. b Phylogenetic tree of 17 angiosperm species (13 Apiaceae species; three Araliaceae species; one outgroup). Numbers on the nodes represented the divergence time of the species (million years ago, MYA). The circle and pentagram represented whole genome duplication events and fossil nodes, respectively. Source data are provided as a Source Data file.

We resolved the phylogenetic relationships among the five Angelica species, eight other Apiaceae species, three Araliaceae species and one outgroup species (Lonicera japonica) by constructing a maximum-likelihood phylogenetic tree (Fig. 1b). The five Angelica species were grouped into two distinct clades: Angelica sensu stricto (s.s.) and Angelica sensu lato (s.l.) (Fig. 1b). Within the Angelica s.s. clade, A. dahurica, A. biserrata and A. keiskei were closely related, diverging approximately 4.75 million years ago (MYA) (Fig. 1b). In contrast, A. sinensis and L. officinale formed a monophyletic clade, designated as the Angelica s.l. clade, which diverged from the Angelica s.s. clade approximately 22.20 MYA (Fig. 1b). The phylogenetic topology of these five Angelica species was consistent with prior studies^26,27. This divergence underscored significant evolutionary differentiation between Angelica s.s. and Angelica s.l., denoting the absence of monophyly of the Angelica genus. As the focus of our study is not on taxonomy, we will not dive deep into taxonomic treatment of the genus. The divergence of the two groups is potentially reflected in their genomic sequences and metabolic profiles.

LTR insertions affect genome size in Apioideae

Even though Apioideae species underwent the same polyploidization events (Supplementary Fig. 4; Supplementary Note 1; Supplementary Method 4)²³, they exhibited a wide range of genome sizes, from 0.43 Gb to 5.38 Gb, as recorded in the Plant DNA C-values Database (Supplementary Data 3). To date, most Apioideae genome sequences are smaller than 3 Gb, with only a few exceptions, such as Apium graveolens (3.32 Gb) and Ligusticum chuanxiong (3.43 Gb) (Fig. 2a; Supplementary Data 4)^36–38. The assembled Angelica genome sequences offered opportunities to explore genetic factors driving genome size evolution in Apioideae.

Fig. 2. LTR expansion contributed to genome enlargement and secondary metabolite biosynthesis in the subfamily Apioideae.

a Comparison of genome size and TE contents in 12 Apioideae genomes. The schematic tree on the left, plotted based on the topology of the phylogenetic tree in Fig. 1b, shows evolutionary relationships among species. b The linear regression between LTR size and genome size. The circle color corresponds to the color code of different species in Fig. 2a. The blue line indicates the linear regression fit, with the shaded area representing the 95% confidence band. Statistical significance was determined using a two-tailed Pearson correlation analysis (n = 12). c PT orthologous genes with LTR insertions exhibit high gene expression, in contrast with bare expression of PT orthologous genes without LTR insertions in B. chinense and D. carota. The first, second, and third rows indicated the gene structure and genomic location, the counts of mapped reads in RNA-seq, and the number of LTR insertions, respectively. Source data are provided as a Source Data file.

We analyzed the genic and intergenic regions of 12 sequenced Apioideae species (Fig. 2a). In the genic regions, these species exhibited similar average exon lengths (~250 bp) and intron lengths (~590 bp) (Supplementary Data 4). Additionally, no significant correlation was observed between the ratio of total intron/exon length and genome size (Supplementary Fig. 5a), suggesting that genome enlargement occurred primarily in the intergenic regions rather than in the genic regions. In the intergenic regions, transposable element (TE) content varied widely among Apioideae species, ranging from 0.3 Gb to 4.2 Gb (Supplementary Data 4). TEs were categorized into three groups: LTRs, TIRs, and Helitrons. Among these, LTRs were the most abundant, accounting for approximately 23.4-78.2% of the genome sequences (Fig. 2a; Supplementary Data 4). Further analysis revealed that LTRs had the strongest correlation with genome size (P < 0.0001, Pearson correlation coefficient; Fig. 2b; Supplementary Fig. 5b, c), indicating that LTR expansion played a major role in genome enlargement in Apioideae species. In order to test whether LTR elimination contributed to genome size variation, we calculated the ratio of solo LTRs over intact LTRs as a statistic evaluating the level of LTR elimination, as solo LTRs primarily result from DNA removal via unequal homologous recombination³⁹. It showed that the ratios in Angelica species were comparable to those in other Apioideae species (Supplementary Fig. 5d), suggesting that reduction of LTR elimination did not contribute to genome size enlargement. Taken together, the LTR expansion likely contributed to the large genome size in Apioideae species^40–42.

We further explored the patterns of LTR insertions across species with different genome sizes by comparing the two small Apioideae genome sequences (Bupleurum chinense and Daucus carota) with the five large Angelica genome sequences. In the five Angelica species, genes with LTR insertions showed higher expression levels compared to those without LTR insertions (Supplementary Fig. 5e, f). A notable example is the PT orthologous gene, a key gene in the FC biosynthesis pathway (Fig. 2c). PT genes with LTR insertions exhibited relatively high expression levels, while PT genes lacking LTR insertions in B. chinense and D. carota were barely expressed (Fig. 2c).

Evolutionary and functional divergence of prenyltransferases underlie FC diversity in Angelica

We first quantified total FC content across five Angelica species (Supplementary Method 5) and found that Angelica s.s. species (A. dahurica, A. biserrata and A. keiskei) exhibited 5- to 18-fold higher FC concentrations than Angelica s.l. species (A. sinensis and L. officinale) (P < 0.0001, Student’s t test, Fig. 3a), with pronounced enrichment in roots, suggesting clade-specific divergence in FC accumulation.

Fig. 3. Functional divergence of PTs drives FC diversity in Angelica species.

a Total FC contents in three tissues of five species. Asterisks indicate significant differences between Angelica s.s and Angelica s.l. species. Data are presented as mean values +/- standard deviations from biological replicates (n = 3). Each dot represents an individual biological replicate. Statistical analysis is performed by a two-tailed Student’s t test (P > 0.05: ns, P < 0.0001: ∗∗∗∗). b Schematic plot of the FC biosynthetic pathway. Light green, red, orange and blue arrows/names represent C2’H, C-PT, O-PT and OMT reactions or genes, respectively. Metabolites highlighted in bold and asterisked, including alloisoimperatorin, phellopterin, and cnidilin, correspond to the elucidated biosynthetic route newly uncovered in this study. Heatmap below compounds shows relative contents and gene expression across five species. c Phylogenetic tree of PT genes in the clade responsible for secondary metabolism, including functionally characterized PTs from Apiaceae. Bootstrap values are shown below branches. Nodes are colored based on catalyzing activity: red = C-PT, yellow = O-PT, gray = uncharacterized. Circles denote PTs functionally validated in this study; stars indicate PTs reported in previous studies. d Enzymatic assays of Angelica PTs: LC-MS traces illustrate representative product peaks for target prenylation reactions. Source data are provided as a Source Data file.

Umbelliferone (2), the common substrate of FC biosynthesis and produced by C2’H, was 1.8- to 2.6-fold higher in A. biserrata and A. keiskei than in Angelica s.l. species, while the umbelliferone content in A. dahurica showed no significant difference from that in Angelica s.l. species (Fig. 3b; Supplementary Data 5). PT-mediated C6/C8-prenylation of umbelliferone drives structural divergence in coumarin frameworks, yielding demethylsuberosin (linear, 3) and osthenol (angular, 4). Demethylsuberosin levels were 8.7- to 9.3-fold higher in Angelica s.s. than in Angelica s.l. clades, whereas osthenol showed no significant clade-specific variation (Fig. 3b; Supplementary Data 5). CYP-mediated furan-ring cyclization products marmesin (5) and columbianetin (6) exhibited clade-specific enrichment in Angelica s.s. (Fig. 3b, Supplementary Data 5). For downstream hydroxylation modifications, bergaptol (7) and xanthotoxol (8) catalyzed by CYPs, were present in Angelica s.s. but nearly absent in Angelica s.l. (Fig. 3b; Supplementary Data 5). O-prenylation products including isoimperatorin (9) and imperatorin (10) showed pronounced increases in Angelica s.s. (P < 0.05, Student’s t test, Fig. 3b; Supplementary Data 5). While O-methylation products, such as bergapten (11), xanthotoxin (12) and isopimpinellin (15), exhibited no significant inter-clade differences (Fig. 3b; Supplementary Data 5). Collectively, these results established Angelica s.s. as possessing both markedly greater abundance and structural diversity.

To unravel the molecular basis for FC diversity between clades, we first investigated PT genes given their significant roles in determining geometric isomerism of FCs and enabling downstream enzymatic diversification. Genome-wide screening identified 86 PTs from five Angelica species (Supplementary Note 2; Supplementary Fig. 6a; Supplementary Method 6), and 25 putative PTs were clustered in the clade responsible for secondary metabolism, possibly involved in coumarin biosynthesis (Fig. 3c). Based on the phylogenetic analysis, we defined three distinct evolutionary clades (A1, A2, and B) (Fig. 3c). Based on the identified PT enzyme functions in published studies^22,24,43, it seems that PTs exhibiting C-prenylation activity (C-PT, catalyzing prenylation at carbon position) all clustered in Clade A, whereas the only PT mediating O-prenylation activity (O-PT, catalyzing covalent attachment of prenyl group to the oxygen atom of substrate) was located in Clade B. Enzymatic assays were set up to test functions of PTs in five Angelica species (Fig. 3d). Clade B PTs (AbPT1, AkPT1, AdPT1) displayed broad-spectrum O-PT activity. Besides catalyzing bergaptol (7) or xanthotoxol (8) to generate isoimperatorin (9) or imperatorin (10), these enzymes were newly identified to convert 8-hydroxybergapten (13) and 5-hydroxyxanthotoxin (14) into phellopterin (17) and cnidilin (18), respectively. Clade A1 PTs, including AsPT1/2, LoPT1, AkPT2/3, AbPT2/4, and AdPT2/4, primarily catalyzed umbelliferone (2) at C6 to produce demethylsuberosin (3) as the major product with trace osthenol (4) formation, confirming C6-PT function. An additional novel function was identified for AsPT1/2, LoPT1 and AkPT3 in clade A1, which were capable of converting bergaptol (7) into alloisoimperatorin (16) (Fig. 3d; Supplementary Figs. 7 and 8), thereby expanding the known substrate specificity of this enzyme subfamily. Conversely, Clade A2 PTs (AbPT3/5) mainly yielded osthenol (4), thus defining their C8-PT activity. Collectively, PTs in Clade A appeared to diverge their functions: PTs in A1 clade preferentially targeted the C6 position of umbelliferone (C6-PT), subsequently producing linear coumarin scaffolds, while PTs in A2 clade showed C8 position specificity (C8-PT), successively generating angular coumarin scaffolds. Cross-species functional profiling revealed genetic differentiation: Angelica s.s. species (A. keiskei, A. biserrata, A. dahurica) encompassed both C-PT and O-PT genes, while Angelica s.l. species (A. sinensis, L. officinale) were solely equipped with C-PT genes responsible for constructing the FC backbone, entirely lacking O-PT genes associated with downstream modifications. This disparity likely serves as a key molecular mechanism underlying the heightened FC structural diversity in Angelica s.s. clade.

Physical arrangement of core cluster and peripheral genes accounts for FC variation between Angelica s.s and Angelica s.l

Based on chromosomal location analysis, we found that functional PT genes in each Angelica species were localized on a single chromosome with physical proximity: PT loci reside on Chr01 in A. dahurica, Chr04 in A. biserrata, Chr05 in A. keiskei, Chr03 in A. sinensis, and Chr07 in L. officinale (Supplementary Fig. 6c). We further identified twelve, six, eight, four, and four candidate C2’H genes in A. dahurica, A. biserrata, A. keiskei, A. sinensis and L. officinale, respectively, all of which were located in close proximity to PT genes within the same syntenic blocks (Supplementary Fig. 6c). Meanwhile, one CYP gene in A. keiskei (AK05G03995) and another in A. sinensis (AS03G00683) were found adjacent to PT genes (Supplementary Fig. 6c). In addition, candidate OMT genes in A. dahurica, A. biserrata and A. keiskei were located on the same chromosome as the C2’H and PT genes, whereas in A. sinensis and L. officinale, OMT genes were located on different chromosomes (Supplementary Fig. 6b, c). These findings corroborated the existence of previously reported FC BGCs in Angelica²⁵, but also highlight clade differentiation with regard to the physical arrangement of the FC BGC between Angelica s.s. and Angelica s.l.

To functionally validate the putative FC BGCs, we performed comprehensive in vitro enzyme assays on C2’H, CYP and OMT candidate genes from the five Angelica species (Fig. 4a, b). Nine (AdC2’H1-9), four (AbC2’H1-4), five (AkC2’H1-5), one (AsC2’H1) and one (LoC2’H1) C2’H orthologs from the five Angelica species possessed the activity to catalyze p-coumaroyl-CoA (1) into umbelliferone (2) (Fig. 4a; Supplementary Fig. 8), establishing their conserved role in the initial step of FC biosynthesis. Enzyme assays of AdOMT, AbOMT, AkOMT, AsOMT and LoOMT1-4 revealed that broad substrates, including osthenol (4), bergaptol (7), xanthotoxol (8), 8-hydroxybergapten (13) and 5-hydroxyxanthotoxin (14), were efficiently O-methylated to produce compounds osthol (19), bergapten (11), xanthotoxin (12) and isopimpinellin (15), respectively (Fig. 4b; Supplementary Fig. 8). These results highlighted the role of OMTs in the structural diversification of FCs. Enzyme assays of AkCYP showed its capacity in the conversion of demethylsuberosin (3) into marmesin (5) and decursinol, as well as the conversion of osthenol (4) into columbianetin (6) and lomatin, indicating its significant roles in FC biosynthesis²⁵ (Supplementary Fig. 9; Supplementary Method 7).

Fig. 4. Divergent evolution of FC BGCs and gene expression shift between the Angelica s.s. and Angelica s.l. clades.

a Enzymatic assays of C2’Hs in Angelica. LC-MS traces illustrate the target product peak for umbelliferone. b Enzymatic assays of OMTs in Angelica. LC-MS traces illustrate representative product peaks for target O-methylation reactions. c The distribution of FC BGC across five species. Light green, red, orange and blue arrows represented C2’H, C-PT, O-PT and OMT genes. respectively. Heatmaps above each gene showed its expression level in roots and leaves, respectively. TPM, Transcripts Per Million. d Expression evolutionary rate of C2’H genes. Color gradient of branches indicates evolutionary rates, ranging from slow (light red) to quick (dark red). The middle heatmap shows gene expression levels in leaf and root tissues. The right arrows highlight genes that exhibit an upshift in expression between different clades. Source data are provided as a Source Data file.

Based on the assembled genome sequences and functional verification experiments, we found a complete FC BGC across Angelica species, including the signature enzyme, C2’H and PT, and the tailoring enzymes, CYP and OMT (Fig.4c). In five Angelica species, C2’H and PT genes were physically adjacent in the core cluster, with A. keiskei uniquely containing one CYP gene in this region. The Angelica s.s. clade exhibited higher copy numbers in the core cluster, with nine C2’H and four PT for A. dahurica, four C2’H and four PT for A. biserrata, and five C2’H and three PT for A. keiskei (Fig. 4c). In contrast, A. sinensis and L. officinale had fewer gene copies in the core cluster, with one C2’H, two PT, and one C2’H, one PT, respectively (Fig.4c), indicating significant copy number variations in the core cluster likely drive evolutionary divergence between the two clades. Furthermore, the positioning of OMTs, the peripheral genes, also highlighted evolutionary divergence between the two Angelica clades. In the Angelica s.s. clade, OMT and the core cluster were located on the same chromosome with a distance of approximately 10-15 Mb, forming an in cis BGC¹⁰, which indicated the physical proximity (on one single chromosome) of genes involved in the cluster facilitated efficient metabolite biosynthesis. In contrast, OMT was located on a different chromosome (Chr09 for A. sinensis and Chr05 for L. officinale) in the Angelica s.l. clade, generating an in trans BGC¹⁰.

To investigate the evolutionary dynamics of FC BGCs, we compared the genomic localization of key FC biosynthetic genes across 11 representative Apiaceae species, including Sanicula chinensis from the Saniculoideae subfamily⁴⁴ and 10 species from the Apioideae subfamily (Supplementary Fig. 10). In S. chinensis, the PT and C2’H homologs were dispersed on different chromosomes, and the PT genes were not expressed or lost their function in FC biosynthesis⁴⁴. In D. carota, L. chuanxiong, A. graveolens, and C. sativum, a tightly linked core cluster composed of PT and C2’H homologs was identified, without collinear OMT genes (Supplementary Fig. 10; Supplementary Data 6). In contrast, P. praeruptorum, a species phylogenetically close to the Angelica s.s. clade, exhibited in cis core cluster plus peripheral OMT genes on the same chromosome, similar with the genomic configuration observed in three Angelica s.s. species (Supplementary Fig. 10; Supplementary Data 6). It suggested that the origin of the FC BGC likely occurred post the divergence of the Apioideae subfamily, and that the two distinct types of FC BGC architectures identified in this study reflect a structural divergence between Angelica s.l. and the common ancestor of Angelica s.s. and its close relatives.

Gene expression shift may underlie FC differentiation between the two Angelica clades

Interestingly, we observed a differentiated tissue-specificity of expression of BGC genes between the two Angelica clades (Fig. 4c; Supplementary Data 7). Specifically, in the Angelica s.s. clade (A. dahurica, A. biserrata, and A. keiskei), C2’H, PT, and CYP genes within the core cluster showed higher expression in roots than in leaves (P < 0.01, Student’s t test), with a few exceptions (AdPT2, AbPT1 and AkPT2). In contrast, in the Angelica s.l. clade (A. sinensis and L. officinale), the core cluster genes exhibited relatively higher expression in leaves than in roots (P < 0.01, Student’s t test). Such distinct tissue-specificity between clades was absent for the peripheral OMT genes (Fig. 4c; Supplementary Data 7).

A key driver of chemical diversity is the evolution of gene expression, as variations in gene expression evolutionary rates among species, reflecting differential evolutionary pressures and constraints⁴⁵, lead to lineage-specific shifts that ultimately drive metabolic divergence⁴⁶. To reconstruct evolutionary trends of FC BGC, we built gene trees and calculated expression evolutionary rate for each gene (Fig. 4d; Supplementary Fig. 11a, b). The C2’H genes can be divided into three clades: Angelica s.l. clade (ASL), Angelica s.s. clade A (ASS-A), Angelica s.s. clade B (ASS-B). Both ASS-A and ASS-B exhibited a faster evolutionary rate than ASL, especially AdC2’H3 in ASS-A and AkC2’H3, AkC2’H5, AkC2’H4 and AkC2’H2 in ASS-B (Fig. 4d). To test if the rapidly evolving C2’H demonstrates lineage-specific expression shift, we quantified gene expression patterns and tested for expression divergence among ASS-A, ASS-B, and ASL. Compared to ASL, the expression level of C2’H in both ASS-A and ASS-B significantly shifted up (P < 0.05). As for PT and OMT genes, expression evolutionary rates and expression shifts were not significant among groups (Supplementary Fig. 11a, b). It indicated that the differences of FC content between Angelica s.s. and Angelica s.l. were to some extent related to changes in gene expression dosage of C2’H, one signature enzyme responsible for umbelliferone production⁴³. We also preliminarily explored conserved non-coding elements (CNEs) and identified an Angelica s.s. Specific CNE within the proximal region of AdPT3 in A. dahurica (Supplementary Note 3; Supplementary Table 6; Supplementary Figs. 12 and 13; Supplementary Method 8–10), underscoring the potential role of CNEs in modulating gene expression⁴⁷.

In summary, FC BGC was conservatively present in the five species, but exhibited different gene copy numbers and organizational types: in cis core cluster with higher copy numbers plus peripheral genes on the same chromosome for Angelica s.s. clade, and in the trans core cluster with lower copy numbers plus peripheral genes on a different chromosome for Angelica s.l. clade (Fig. 5). These genomic distinctions likely modulate both biosynthetic efficiency and structural diversification of FCs, representing key adaptive evolutionary drivers of chemical divergence between the clades.

Fig. 5. Schematic plot of genetic factors underlying differences of FC content and diversity between Angelica s.s. and Angelica s.l. clades.

FC BGC in Angelica s.s. clade exhibits as in cis core cluster plus a peripheral gene. The PT genes exhibit both C- and O-prenylation activities, and C2’H genes show an upshift in expression. Physical arrangement, gene dosage effect and functional divergence collectively contribute to the high content and diversity of FCs in Angelica s.s. clade. In contrast, FC BGCs in Angelica s.l. species display in trans core cluster plus peripheral genes, where PT genes lack O-prenylation function and C2’H genes exhibit copy number and downshift of expression. Consequently, it leads to a low content and reduced diversity of FCs in Angelica s.l. clade.

Discussion

Angelica species have significant medicinal and cultural importance, and they are noted for producing abundant FCs with diverse pharmacological properties. In this study, we compiled five genomes representing divergent Angelica clades: three Angelica s.s. species (A. biserrata, A. dahurica, and A. keiskei) known for their high diversity of FCs, and two Angelica s.l. species (A. sinensis and L. officinale) producing lower levels of FCs. The differentiated pattern of FCs between the two clades was attributed to physical organization of FC BGC genes, shifted gene expression of the signature enzyme C2’H, and multi-functionalization of the signature enzyme PT.

What we demonstrate is, to our knowledge, the most comprehensive characterization of the FC biosynthetic pathway in any plant. Prior studies typically examined individual enzymes or gene families, e.g., CpPT1 in Citrus paradisi²⁴ and CmOMT1 and CmOMT2 in Cnidium monnieri⁴⁸. Whereas we have systematically identified and functionally validated three key enzymes (C2’H, PT, OMT) recruited to produce a suite of FC compounds, including umbelliferone, demethylsuberosin, osthenol, imperatorin, isoimperatorin, alloisoimperatorin, bergapten, xanthotoxin, isopimpinellin and osthole (Fig. 3b). In doing so, we discovered PT activities that generated alloisoimperatorin, phellopterin and cnidilin, compounds with promising anti-osteoporotic potential⁴⁹. By elucidating both known and novel steps of FC biosynthesis and filling the gaps of FC biosynthetic pathway, our work lays a foundation for revealing the evolutionary innovations that have shaped the complex pathway and targeted metabolic engineering to enhance production of these FCs.

Multi-omics data among two Angelica clades enabled a complete exploration of the genetic basis underlying differential FC production, providing a model of divergent evolution of SM biosynthesis. The first genetic factor of FC difference between Angelica clades is attributed to the physical arrangement of core and peripheral BGC genes. Based on comprehensive gene function identification and genomic resources, we found a conserved FC BGC (including C2’Hs, PTs and OMTs) in Angelica genus (Fig. 4c), echoed with a previous report of BGC in P. praeruptorum²⁵. Nevertheless, the Angelica s.s. clade exhibited an in cis arrangement of the core cluster and the tailoring OMTs approximately 10-15 Mb away on the same chromosome, while the core cluster genes and OMTs reside on different chromosomes (in trans) in the Angelica s.l. clade (Fig. 4c). Owing to the absence of a genetic map of species in the Angelica s.s. clade, it is unknown whether the tailoring OMTs and the core cluster genes are in a genetic linkage group or not. If they are linked, in cis arrangement of BGC would more likely maintain these catalytic genes together. Conversely, in trans arrangement would render the pathway genes more susceptible to recombination-mediated disruptions. Given that the Angelica s.l. clade is more basal on the phylogenetic tree (Fig. 1b), we infer that the in cis arrangement is a derived genotype, which might be selected owing to its adaptive advantage. Our hypothesis that the physical proximity of BGC core and peripheral genes facilitates efficient metabolite production was supported in multiple previous studies^50,51. In plants, the DIMBOA biosynthetic genes (Bx1-Bx8) of maize were organized as a BGC on the same chromosome, whereas the corresponding ortholog genes in rye were unlinked and located on different chromosomes^13,51. It is worth noting that although the content and diversity of DIMBOA is higher in maize than in rye⁵², a direct causal relationship between the physical arrangement of BGC and metabolite accumulation has not been established. In engineered systems, bringing astaxanthin pathway enzymes originally located in three separate compartments in Yarrowia lipolytica into close physical proximity accelerates the conversion of their precursor, β-carotene, into astaxanthin⁵³. Taken together, naturally proximal BGC genes ensure that pathway components are co-produced, effectively creating an assembly line for SMs.

The second genetic underpinning of FC divergence between the two clades lies in the copy number variation of core cluster genes. Angelica s.s. species harbored four to nine C2’H and three to five PT gene copies, in contrast to only one C2’H and one to two PT gene copies in Angelica s.l. (Fig. 4c). Our research supported that gene copy number in the core cluster among species is likely a powerful driver in FC production, subsequently contributing to adaptive evolution⁵⁴. As one signature enzyme in FC BGC, C2’H initiates the FC biosynthetic pathway, producing the basic 7-hydroxycoumarin backbone of FCs⁴³. The expression level of the C2’H genes in the Angelica s.s. The clade is markedly upshifted compared to that in the Angelica s.l. clade (Fig. 4d). An increase in gene copy number is often associated with elevated transcript levels, a phenomenon known as the gene dosage effect⁵⁵. Such a case has been reported in the Rosa genus. Species that possess increased copy numbers of Nudix hydrolase 1 (NUDX1) genes involved in geraniol biosynthesis exhibit elevated transcript levels of NUDX1, and thus produce more geraniol, directly linking gene dosage to enhanced metabolite accumulation and species-specific adaptation⁵⁶. Thus, evolutionary tuning of C2’H expression within the FC BGC through gene duplication can boost the supply of umbelliferone, the common precursor of all FCs, and ultimately increase total FC content. As another signature enzyme in FC BGC, PT serves as a molecular switch that not only dictates the configuration of linear or angular FCs, but also enables downstream enzymatic diversification, ultimately giving rise to the structurally diverse and biologically active FCs^24,25. Phylogenetic analysis (Fig. 4c) and prior studies²⁵ revealed that the ancestral PT functioned as an O-prenylase (O-PT). Evolutionary and experimental evidence suggest that Angelica s.s. species exhibit both ancestral O-PT and the derived C-PT functions, while Angelica s.l. species lost O-PT in the BGC during evolution. It underscores the PT functional diversification as a key driver of metabolic innovation in Angelica. In sum, the elevated copy number of C2’H genes likely contributes to the high content of FCs, and functional diversity of PTs ensures the diversity of FCs in the Angelica s.s. clade. The synergistic effect of both signature enzymes serves as a powerful driver of metabolic innovation, enabling plant lineages to evolve novel SMs for local adaptation and contributing to the chemical diversity across the plant kingdom. Such knowledge provides a blueprint for rational metabolic engineering, enabling efficient pathway reconstruction, transfer, and novel compound biosynthesis, as signature enzymes serve as key control points for metabolic flux, and their overexpression can drive more precursors into a desired pathway.

In conclusion, our study provides valuable genomic resources of Angelica, an economically significant genus, fills enzymatic gaps in the FC biosynthetic pathway, explores the evolution of the FC BGC genes, and dissects the genetic factors modulating the SM diversity and content between Angelica s.s. and Angelica s.l. clades. It provides valuable strategies for metabolic production of key bioactive compounds, and sets up a paradigm for related research aiming to investigate the genetic bases and evolutionary history of biochemical diversity in plants.

Methods

Plant materials, DNA extraction and sequencing

Angelica biserrata and A. keiskei plants were respectively collected in Juxian, Shandong Province and Enshi, Hubei Province, China. L. officinale plants were obtained from Beijing Botanical Garden. Fresh young leaves were used for DNA extraction and sequencing. First-year seedlings were cultivated in pots (20 cm × 20 cm; one plant per pot) filled with a nutrient matrix (soil relative water content maintained at 60-70%) and grown in a greenhouse under a 12 h light/12 h dark photoperiod at 20 °C and 60% relative humidity. Fresh tissues were harvested approximately two months after planting, when plants had reached about 20 cm in height and before flowering. To ensure consistency, all samples were collected at a uniform time of day (6 hours after the start of the photoperiod) to minimize circadian effects. The collected samples were immediately flash-frozen in liquid nitrogen and stored at −80 °C until RNA extraction and transcriptome sequencing (Supplementary Data 8). Additional details were provided in Supplementary Method 1.

Genome assembly

After genome size estimation (Supplementary Method 2), the draft genome sequence was assembled by PacBio CCS using Hifiasm v0.15.5-r350 with default parameters⁵⁷. Clean Hi-C paired-end reads were subsequently mapped to the genome sequence using Juicer v1.6⁵⁸ to improve the draft assembly. Then, a candidate chromosome-level assembly was generated automatically using the 3D-DNA v180114 pipeline⁵⁹ to correct mis-joins, order and orientation, and to organize the contigs from the draft chromosome assembly. Finally, manual review and refinement of the candidate chromosome-level assembly were performed by Juicebox Assembly Tools⁶⁰ to further improve the chromosome-scale assembly and quality control. To assess the completeness and accuracy of the final genome sequences, we utilized BUSCO v5.1.2³³ with parameter ‘-m geno --offline’ employing the viridiplantae_odb10 database and LTR Assembly Index (LAI)³⁴ with default parameters.

Repeat and gene annotation

Transposable element (TE) annotations were first derived using the Extensive De Novo TE Annotator (EDTA) v1.9.3 pipeline with parameter ‘--sensitive 1’⁶¹. TE-sorter v1.2.5⁶² was then employed to reclassify TEs initially annotated as “LTR/unknown” by EDTA. Full-length and solo LTRs were identified using the LTR_FINDER_parallel⁶³ and LTR_retriever⁶⁴ implemented with the EDTA pipeline. For comparative analysis, the same methodology was applied to the other 10 Apioideae species described below in the “Phylogenetic analyses” section.

For gene prediction of newly genome assembly, we first used RepeatMasker v4.10⁶⁵ to mask the genome sequences with the TE libraries constructed by EDTA⁶¹. An integrated strategy that combined homology-based, transcriptome-based and ab initio prediction was then used to predict the protein-coding genes based on the masked genomic sequences. For homology-based prediction, we used the annotated proteome data of Arabidopsis thaliana, Coriandrum sativum⁶⁶ and A. graveolens³⁷ and the Swiss-Prot database as the protein data. For transcriptome-assisted prediction, transcript data were retrieved from a de novo assembled transcriptome generated in Trinity v2.11.0⁶⁷. For ab initio gene prediction, Augustus⁶⁸ was used with default parameters incorporating the homology- and transcript-based evidence for gene model training. Finally, all the gene models generated by the above three approaches were integrated into a comprehensive gene set using EvidenceModeler v1.1.1⁶⁹, and the resulting gene models were further updated using PASA for three rounds of iteration⁷⁰. Furthermore, BUSCO v5.1.2³³ with parameter “-m prot --offline” employing the viridiplantae_odb10 database was used to evaluate the gene annotations. Functional annotations of protein-coding genes were conducted via GFAP⁷¹ and EGGNOG-MAPPER v2⁷². TBtools⁷³ was used for downstream gene set enrichment analysis. The circos plot illustrating genome sequence features was drawn using Circos (https://circos.ca/).

Phylogenetic analyses

Paralogs and orthologs were identified among 17 plant species: Lonicera japonica, Aralia elata⁷⁴, Panax stipuleanatus⁷⁵, P. notoginseng⁷⁵, Centella asiatica⁷⁶, B. chinense⁷⁷, D. carota⁷⁸, L. chuanxiong sub A and sub B³⁸, Heracleum sosnowskyi⁷⁹, A. graveolens³⁷, C. sativum⁶⁶, P. praeruptorum⁸⁰, A. keiskei, A. biserrata, A. dahurica³⁰, A. sinensis²³ and L. officinale using the OrthoFinder v2.5.2 (parameters: -S blast -M msa)⁸¹. Protein sequences of 320 single-copy orthologous genes were used to construct a phylogenetic tree. These sequences were aligned with MAFFT v7.271 (parameters: --auto)⁸² and trimmed by trimAI v1.4.rev22 with default parameters⁸³. A maximum likelihood phylogenetic tree was constructed using RAxML v8.2.12 of a PROTGAMMAJTT model with 1000 bootstrap replicates⁸⁴, and L. japonica was used as the outgroup. The species tree was then used as an input to estimate divergence time in the MCMCTree of the PAML package⁸⁵ with three constraints for calibration (Supplementary Data 9). The phylogenetic tree was visualized and annotated using iTOL (https://itol.embl.de/). Syntenic blocks within one species or between two species were defined by MCscanX with parameters: -m = 25, -s = 5⁸⁶ based on homologous gene sets using BLASTP v2.10.0 (E-value < 1e-5; the number of genes required to call a syntenic block ≥ 5).

Determination of coumarins by LC-MS analysis

Chromatographic separation was performed on an ACQUITY UPLC BEH C18 column (2.1 mm × 100 mm, 1.7 μm, Waters) using a gradient of water containing 0.1% formic acid (A) and acetonitrile (B) at a flow rate of 0.3 mL/min: 5–25% B (0–2 min); 25% B (2–6 min); 25–40% B (6–6.5 min); 40% B (6.5–18 min); 40–95% B (18–21 min); 95% B (21–24 min). The column temperature was maintained at 40 °C. Mass spectrometry was conducted using a Thermo Q-Exactive Focus Orbitrap mass spectrometer operated in positive electrospray ionization (ESI + ) mode. Key parameters included: spray voltage 3500 V, sheath gas 30 arb. units, auxiliary gas 10 arb. units, capillary temperature 320 °C, and normalized collision energy (NCE) of 30 eV. Full-scan MS spectra were acquired over m/z 100-1200 at a resolution of 35,000, and data-dependent MS² spectra were obtained at a resolution of 17,500. For qualitative identification, peaks were annotated using Thermo Scientific Xcalibur by matching retention times, exact precursor m/z (mass error <5 ppm), and MS² fragmentation patterns to those of authentic reference standards.

Functional characterization for candidate PT in yeast

To functionally characterize the candidate PT genes, we designed specific primers for each gene (Supplementary Data 10) to clone them into the pESC-Ura vector. Subsequently, the recombinant plasmids were transformed into the Saccharomyces cerevisiae strain DD104. As a negative control, the DD104 strain was separately transformed with the empty pESC-Ura vector. Single colonies were picked and cultured in 250 mL of uracil-deficient synthetic dropout (SD-Ura) medium at 28 °C with shaking until OD600 reached 1.0-1.2. Subsequently, cells were washed twice with sterile water and resuspended in yeast extract peptone dextrose (YEPD) medium supplemented with 2% galactose to induce protein expression for 24 h. Yeast cells were then harvested by centrifugation at 5000 × g 4 °C for 15 min. The pellet was washed once with TEK buffer (50 mM Tris-HCl, 1 mM EDTA, 100 mM KCl, pH 7.5), incubated for 5 min, then re-centrifuged. Subsequently, it was resuspended in 20 mL TESB buffer (50 mM Tris-HCl, 1 mM EDTA, 600 mM sorbitol, pH 7.5), and then 0.5 mm glass beads were added to lyse the cells by vigorous shaking for 10 min (30 s shaking followed by 30 s on ice). The lysate was collected and centrifuged at 23,000 × g 4 °C for 30 min. The supernatant was diluted 2-fold with TESB buffer, and 150 mM sodium chloride and 0.1 g/mL PEG-4000 were added. The mixture was incubated on ice for 3 h with periodic swirling. Microsomes were collected by centrifugation at 10,000 × g, 4 °C for 30 min, and the pellet was resuspended in 1 mL of TEG buffer (50 mM Tris-HCl, 1 mM EDTA, 20% [v/v] glycerol, pH 7.5). Using the Bradford assay, the concentration of microsomal protein was determined.

Each in vitro enzymatic reaction mixture of PTs (total volume: 200 μL) contained 10 μg of microsomal protein, 200 µM DMAPP, 10 mM MgCl₂, 100 µM one substrate (umbelliferone, xanthotoxol, bergaptol, 8-hydroxy-bergapten or 5-hydroxy-xanthotoxin) in a total reaction volume of 200 µL. The reactions were incubated at 30 °C for 3 h and terminated by adding an equal volume of ethyl acetate. Following centrifugation, the ethyl acetate layer was collected for subsequent LC-MS analysis using the chromatographic and mass spectrometric conditions detailed in the “Determination of coumarins by LC-MS analysis” section above.

Functional characterization for candidate C2’H and OMT

To functionally characterize the candidate C2’H and OMT genes, we cloned them into the His-tag-based bacterial expression vector pET-28a. Specific primers for each gene (Supplementary Data 10) were used following the manufacturer’s protocol. After purification, the plasmids were transformed into Escherichia coli BL21 (DE3) via heat-shock transformation. Transformed cells were selected overnight on Lysogeny broth (LB) plates with kanamycin at 37 °C. The positive colonies were screened by colony PCR and further cultured in 200 mL LB containing kanamycin at 200 rpm and 37 °C until the OD600 absorbance reached 0.6-0.8. Then, 0.5 mM isopropyl-β-D-thiogalactopyranoside was added to induce protein expression. The assay for C2’H enzyme activity was carried out in a 250 μL assay buffer (5 mM 2-oxoglutarate, 5 mM dithiothreitol, 5 mM sodium ascorbate, 1 mM FeSO₄, and 25 mM Tris-HCl buffer pH 7.5) containing 10 μg purified proteins and 100 μM p-coumaroyl-CoA. The reaction was incubated at 20 °C for 3 h and then extracted with 250 μL ethyl acetate. The assay for OMT enzyme activity was performed in a 250 μL assay buffer (100 μM S-adenosyl-L-methionine and 25 mM Tris-HCl buffer pH 7.5) containing 10 μg purified proteins and 100 μM each substrate (osthenol, xanthotoxol, bergaptol, 8-hydroxy-bergapten and 5-hydroxy-xanthotoxin). The reaction was incubated at 37 °C for 3 h and then extracted with 250 μL ethyl acetate. The negative control of protein was prepared from E. coli harboring the vector pET-28a. LC-MS analysis was conducted to detect the reaction products using the chromatographic and mass spectrometric conditions detailed in the “Determination of coumarins by LC-MS analysis” section above.

Gene expression analysis

To compare gene expression level between different species (Supplementary Method 3), we normalized the TPM by firstly calculating normalisation factor for each species using ‘calcNormFactors.default’ fuction in R package evemodel⁸⁷. Furthermore, we calculated the normalized TPM (TPMn) using the following equation:

$$TPMn=\frac{TPM}{NF}$$ 1

where NF was each specie’s own normalisation factor. Next, CAGEE⁴⁵ was used to calculate the evolution rate of gene expression within each orthologous gene tree. To determine which orthologous genes demonstrate lineage-specific expression shifts, R package evemodel⁸⁷ based on the Expression Variance and Evolution (EVE) model, was used to test for shifts in gene expression levels between Angelica s.s. and Angelica s.l. within each gene tree. The normalized TPM of each gene and phylogenetic tree were used as input files. For every ortholog, a likelihood ratio test (LRT) score was calculated, representing the likelihood of the alternative hypothesis over the null hypothesis. LRT scores were compared to a chi-squared distribution with one degree of freedom, and scores above the 95% quantile were considered to be significant.

Reporting summary

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

Author contributions

L.W. conceived and designed the study. X.H., M.G. and P.Y. analyzed the data. M.G., P.Y., D.H. and Y.C. performed the experiments. X.H. and M.G. wrote the manuscript. L.W., G.M., Y.J., A.T., Z.L., E.N., J.T. and H.P. revised the manuscript. All authors read and approved the final manuscript. X.H., M.G. and P.Y. contributed equally to this work.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the National Key Research and development Program of China, grant no. 2023YFA0915800; the National Natural Science Foundation of China, grant no. 32470245, 32570274, 32400192, and 32300223; Shenzhen Science and Technology Program, grant no. JCYJ20241202130723030; the Agricultural Science and Technology Innovation Program, grant no. CAAS-ZDRW202601.

Data availability

The raw sequence data for the PacBio HiFi reads, Hi-C reads, RNAseq data of Angelica keiskei, A. biserrata and Levisticum officinale generated in this study have been deposited in the CNGBdb database under Project number CNP0009487. The genome assembly and annotation files are available at Figshare: A. keiskei[10.6084/m9.figshare.29977309], L. officinale [10.6084/m9.figshare.29977321], and A. biserrate [10.6084/m9.figshare.29334650]. Source data are provided with this paper.

Code availability

The code used in this study is publicly available at GitHub [https://github.com/XXboop/Angelica-genus-project].

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74337-w.