A chromosome-level genome assembly of Lycoris radiata reveals the evolutionary origin of Amaryllidaceae alkaloids and elucidates the complete galanthamine biosynthetic pathway

Abstract

Amaryllidaceae alkaloids (AmAs) comprise a structurally diverse group of specialized metabolites produced almost exclusively by species of the Amaryllidoideae subfamily and are of substantial pharmacological importance. However, the limited availability of high-quality genomes from Amaryllidoideae plants has constrained systematic investigations of the genes and evolutionary processes underlying AmA biosynthesis. Here, we present a chromosome-level genome assembly of Lycoris radiata, which enabled the discovery of key downstream enzymes in the galanthamine biosynthetic pathway and uncovered reversible reactions between two critical metabolite pairs. These findings provide new mechanistic insight into pathway architecture and enable reconstruction of the galanthamine biosynthetic pathway in Yarrowia lipolytica. Comparative genomic analyses indicate that several core genes for AmA biosynthesis originated in ancestral angiosperms, whereas the complete pathway was likely assembled in the Amaryllidoideae subfamily through gene duplication and neofunctionalization. Furthermore, integrated metabolomic and transcriptomic analyses suggest that roots contribute actively to AmA metabolism in Lycoris. Together, these findings provide a genomic and biochemical framework for understanding the evolution and engineering of AmA biosynthesis.

This study provides a chromosome-level reference genome for Lycoris radiata and clarifies the evolutionary origin of Amaryllidaceae alkaloid biosynthesis. Functional characterization of key enzymes resolves downstream steps in the galanthamine biosynthetic pathway, including previously unknown reversible reactions, and enables its successful reconstitution in Yarrowia lipolytica.

Introduction

Amaryllidaceae alkaloids (AmAs) are a distinctive class of plant specialized metabolites largely confined to the Amaryllidoideae subfamily and encompass more than 600 structurally diverse compounds (Rodríguez-Escobar et al., 2025). Many AmAs possess notable therapeutic properties, with demonstrated activities including antitumor, antimalarial, antiviral, and insecticidal effects (Cahlíková et al., 2020). Notably, galanthamine is a potent acetylcholinesterase inhibitor that has been widely adopted as an approved symptomatic treatment for Alzheimer’s disease (Olin and Schneider, 2002; Demir, 2014; Lazarova et al., 2021; Basagni et al., 2023) and has also shown promise in the treatment of schizophrenia (Norén et al., 2006). Despite their pharmacological importance, the broader use and biotechnological production of AmAs are constrained by their limited natural abundance (Liu et al., 2020). For example, galanthamine typically accumulates at only 0.04% of plant dry weight (DW) in Lycoris species (Yeo et al., 2021). Commercial production still depends largely on plant extraction, creating a pronounced gap between supply and demand (Berkov et al., 2014). These limitations highlight the urgent need for synthetic biology strategies that can support efficient heterologous production. A prerequisite for such approaches is a comprehensive understanding of AmA biosynthetic pathways, including systematic identification and functional characterization of key enzyme-encoding genes.

Biosynthetically, AmAs are derived from the precursors tyramine and 3,4-dihydroxybenzaldehyde (3,4-DHBA), which can be converted into norbelladine through a condensation–reduction sequence catalyzed by norbelladine synthase (NBS) and/or noroxomaritidine/norcraugsodine reductase (NR) (Singh et al., 2018; Majhi et al., 2023). Norbelladine is then methylated by norbelladine 4′-O-methyltransferase (N4OMT) to form 4′-O- methylnorbelladine (Kilgore et al., 2014). Subsequently, different enzymes from the CYP96T subfamily catalyze stereoselective oxidative coupling of 4′-O-methylnorbelladine, generating three distinct structural scaffolds that further diversify into numerous AmAs (Figure 1A and 1B) (Kilgore et al., 2016; Mehta et al., 2024). The para–ortho′ coupling product, N-demethylnarwedine, represents the structural backbone of galanthamine, and recent work in a Nicotiana benthamiana expression system identified a minimal gene set from Narcissus cv. ‘Tête-à-Tête’ sufficient for galanthamine biosynthesis, implicating an aldo-keto reductase (NtAKR) and an N-methyltransferase (NtNMT) as downstream enzymes and offering critical insights into pathway reconstruction for heterologous production (Mehta et al., 2024).

Figure 1.

Distribution of AmAs in and the biosynthetic framework of AmAs.

(A) Representative AmAs detected in L. radiata and their distribution patterns across developmental stages and tissues. The heatmap shows relative AmA abundances. For each metabolite, values were log_10-transformed as log₁₀(peak area + 1), followed by row-wise Z-score normalization. Data are presented as mean values from three biological replicates. This heatmap is intended to illustrate relative within-metabolite distribution patterns across samples rather than absolute abundance differences between metabolites. Schematic representations of plant morphology at the flowering stage (Stage 1), fruiting stage (Stage 2), and leafing stage (Stage 3) are shown on the left.

(B) Schematic representation of AmA biosynthesis. 4′-O-methylnorbelladine, generated through sequential catalysis by NBS and N4OMT, is subsequently transformed by CYP96 family enzymes into three key scaffold intermediates.

(C) Tissue-specific accumulation of AmAs across developmental stages in L. radiata. For each developmental stage, the bar plot shows the percentage of AmAs that reached their highest accumulation in each tissue relative to the total number of detected AmAs (refer to Supplemental Table 15).

The Amaryllidoideae subfamily is the only known source of AmAs (Berkov et al., 2020; Huaylla et al., 2020). Among Lycoris species, Lycoris radiata accumulates a relatively high level of galanthamine, reaching 0.48 mg/g DW (Yeo et al., 2021). In addition, L. radiata produces diverse pharmacologically valuable alkaloids, including lycorine, haemanthamine, and narciclasine (Figure 1A and 1B) (Chen et al., 2016). L. radiata is widely distributed across East Asia and exhibits broad environmental adaptability, making it a suitable model organism for investigating AmA biosynthesis and evolution (Liu et al., 2019; Wang et al., 2023). However, genomic resources for Amaryllidaceae have so far focused primarily on crop species such as onion and garlic (Sun et al., 2020; Hao et al., 2023). By contrast, the lack of genomic data for Amaryllidoideae plants has substantially hindered the systematic elucidation of AmA biosynthetic pathways.

In this study, we report a high-quality reference genome of L. radiata, systematically identify and functionally validate key enzymes involved in galanthamine biosynthesis, elucidate their catalytic mechanisms and expression patterns, and investigate the evolutionary dynamics of AmA biosynthetic pathways. Integrative expression analyses further reveal roots as a major site of AmA metabolism, highlighting a previously underexplored tissue context for alkaloid production in Amaryllidaceae. These findings advance our understanding of specialized metabolism in Amaryllidaceae while providing genomic resources and mechanistic insights to support future efforts in medicinal plant breeding and synthetic biology applications.

Results

Sequencing, assembly, and annotation of the high-quality L. radiata genome

A total of 1,222.50 Gb of short reads was generated and used in a k-mer-based genome survey analysis, which estimated the genome size of L. radiata at 23.62 Gb (Supplemental Figure 1A). Subsequently, 646.22 Gb of PacBio HiFi reads (27.36×) were generated and assembled into a high-quality L. radiata genome assembly with a total size of 23.74 Gb and a contig N50 of 22.08 Mb. These contigs were further scaffolded using 1.86 Tb of high-throughput chromatin conformation capture (Hi-C) data (78.75×), anchoring 94.15% of the assembled sequence onto 11 pseudochromosomes, yielding a scaffold N50 of 1,003.40 Mb and a longest pseudomolecule of 2.58 Gb (Figure 2A, Supplemental Figure 1B, and Supplemental Table 2). The presence of telomeric repeats at 12 of 22 chromosomal termini, together with 95% completeness as assessed by benchmarking universal single-copy orthologs (BUSCO) and a base-level consensus quality value of 64.36, indicates that the chromosome-scale assembly is highly complete and accurate (Supplemental Tables 2 and 3).

Figure 2.

Genome features, whole-genome duplication, and evolutionary history of L. radiata.

(A) Genome characteristics of L. radiata. The tracks from outside to inside show (I) chromosomes, (II) GC content, (III) density of LTR-Gypsy transposons, (IV) density of LTR-Copia transposons, (V) repeat sequences, and (VI) gene number.

(B) Paralogous and orthologous Ks distributions of L. radiata were analyzed using ksrates to identify potential whole-genome duplication (WGD) events and rate-adjusted speciation times. The anchor-pair Ks distribution of L. radiata is shown in gray, with putative WGD components identified by lognormal mixture modeling. Vertical dashed lines indicate the modes of these components, which are interpreted as Ks-based age estimates of the inferred WGD events. Rate-adjusted mode estimates of ortholog Ks distributions between L. radiata and other species, corresponding to speciation events, are represented by numbered vertical long-dashed lines. Colored boxes associated with these lines denote ±1 standard deviation around the mean mode estimate.

(C) Maximum-likelihood phylogenetic tree of 10 species reconstructed from single-copy orthologous genes. The black numbers next to the nodes indicate the estimated divergence times in million years ago (Mya). Numerical values in red and green indicate the number of expanded and contracted gene families, respectively. WGD events are marked by black stars on phylogenetic branches.

(D) Long terminal repeat (LTR) insertion analysis of the L. radiata genome.

Figure 3.

Identification and functional characterization of LrNMTs and LrAKR in galanthamine biosynthesis.

(A) Candidate LrNMT genes showing significantly higher expression in L. radiata roots and bulbs at different developmental stages. The heatmap shows gene expression levels as Z scores calculated from log₁₀(FPKM + 1), where FPKM represents fragments per kilobase of transcript per million mapped reads. .

(B)LrNMT1/2 catalyze the N-methylation of compound 2 to generate product 4.

(C)LrAKR catalyzes the reversible redox interconversion between compound 1 and compound 2.

(D) HPLC chromatogram corresponding to the reaction shown in (B).

(E) HPLC chromatogram corresponding to the reaction shown in (C).

(F) Candidate carbonyl reductases in L. radiata were screened based on Pearson correlation coefficients calculated between the transcript abundance of each candidate and that of LrNMT1 or LrNMT2 (r1, with LrNMT1; r2, with LrNMT2). The heatmap shows gene expression levels as Z scores calculated from log₁₀(FPKM + 1).

(G) Compound 1 is ultimately converted to compound 4 via compound 2 or compound 3 through reactions catalyzed by LrAKR and LrNMT1/2.

(H) HPLC chromatogram corresponding to the reaction shown in (G), showing the LrNMT1/2-catalyzed conversion of compound 1 to compound 3.

(I) HPLC chromatogram corresponding to the reaction shown in (G), showing the LrAKR-catalyzed conversion of compound 3 to compound 4.

The L. radiata genome is dominated by repetitive sequences, which constitute 97% of its total length. Long terminal repeat retrotransposons (LTR-RTs) represent the largest component (80.28%), with Gypsy elements (37.58%) substantially more abundant than Copia elements (11.58%), whereas DNA transposons account for another 16.70% (Supplemental Tables 5 and 6). A pronounced burst of LTR-RT activity between 0 and 2 million years ago (Mya) likely contributed to the recent expansion of the L. radiata genome (Figure 2D). Based on de novo predictions, homology alignment, and RNA sequencing (RNA-seq) data, a total of 41,232 protein-coding genes were predicted in the L. radiata genome, with 91.2% of these genes being functionally annotated (Supplemental Tables 7 and 8). Notably, these gene models are comparatively long, with an average length of 20,940 bp, largely owing to long introns, which have a mean length of 3,720 bp (Supplemental Table 9 and Supplemental Figure 1C).

Genome evolution underlying the origin of AmA biosynthesis

To investigate the history of whole-genome duplication (WGD) events in L. radiata, we plotted a rate-adjusted mixed Ks distribution, which revealed two recent peaks: one at Ks = 0.19, corresponding to a recent WGD event at ∼20.79 Mya, and another at Ks = 0.63, consistent with an older WGD shared among Amaryllidaceae (∼70 Mya). This older event occurred after the divergence between L. radiata and Asparagus officinalis (∼89.80 Mya) but before the divergence of L. radiata and Agapanthus africanus (∼65.30 Mya) (Figure 2B). Although clear collinear blocks corresponding only to the most recent WGD were evident in the intragenomic dot-plot of L. radiata (Supplemental Figure 2A), the 1:8 deep syntenic blocks detected in synteny analyses of L. radiata and Amborella trichopoda supported three rounds of WGD, including the ancient τ WGD event shared by monocots (Supplemental Figure 2C).

To further trace the evolutionary history of L. radiata, we conducted a phylogenomic analysis of 10 angiosperm species, including L. radiata, Allium fistulosum, Allium cepa, Allium sativum, A. africanus, A. officinalis, Elaeis guineensis, Zostera marina, Solanum lycopersicum, and A. trichopoda. A total of 407,790 genes were clustered into 34,823 gene families, including 162 single-copy gene families. Phylogenetic reconstruction and divergence dating based on single-copy orthologs indicated that L. radiata (Amaryllidoideae) diverged from Allioideae, represented here by three Allium species, around 41.4 Mya. Synteny analysis between L. radiata and representative species from Allioideae and Agapanthoideae revealed extensive collinear blocks, supporting the conservation of genome structure across Amaryllidaceae subfamilies (Supplemental Figure 4). Based on this phylogeny, we carried out gene family expansion analysis using CAFÉ and identified 3,356 expanded gene families in L. radiata (Figure 2C). Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis of these expanded families highlighted pathways related to isoquinoline alkaloid biosynthesis, a category that includes AmA biosynthesis (Supplemental Figures 5A and 6), while Gene Ontology (GO) enrichment analysis revealed enrichment of O-methyltransferase activity, transferase activity involving the transfer of acyl groups other than amino-acyl groups, tryptophan synthase activity, and oxidoreductase activity (Supplemental Figure 5B).

In the L. radiata genome, we identified three putative NBS homologs, three putative N4OMT homologs, and three putative CYP96T homologs (Supplemental Figure 7), representing the core enzymatic components responsible for the upstream formation and oxidative coupling steps of AmA biosynthesis. In the galanthamine biosynthetic pathway, N-methylation and carbonyl reduction represent key terminal steps that directly determine the formation of the final bioactive scaffold. Alkaloid biosynthesis in plants relies heavily on redox modifications of aldehydes, ketones, and alcohols, as well as methylation of oxygen and nitrogen atoms (Bhambhani et al., 2021). To investigate these processes in L. radiata, we systematically analyzed predicted genes annotated as reductases and methyltransferases. In total, we identified three reductase gene families and one N-methyltransferase (NMT) gene family: a short-chain dehydrogenase/reductase (SDR) gene family of 245 genes, a medium-chain dehydrogenase/reductase (MDRs)gene family of eight genes, an aldo-keto reductase (AKR) gene family of 33 genes, and an NMT gene family of 359 genes. Among these, the SDR and NMT families both showed significant expansion in L. radiata according to the CAFÉ analysis (Supplemental Figures 8 and 9). The expanded SDR and NMT genes showed an uneven chromosomal distribution, with tandemly duplicated gene clusters observed at multiple genomic loci (Supplemental Figure 10). Transcriptome data analysis revealed that many of the expanded SDR and NMT genes showed consistent tissue-specific expression patterns (Supplemental Figures 8 and 9). Collectively, the expansion and tissue-biased expression of these enzyme families suggest a potential role in the biosynthesis and metabolic diversification of AmAs in L. radiata.

Tissue-specific metabolic allocation and alkaloid accumulation in L. radiata

A total of 1,227 metabolites were detected in L. radiata, including 359 amino acids and derivatives (29.26%), 177 phenolic acids (14.43%), and 64 AmAs (5.21%) (Supplemental Table 13). Hierarchical clustering revealed that metabolic profiles segregated primarily by tissue type rather than developmental stage, with bulbs forming a distinct clade separate from all other tissues. Within the remaining clade, root samples were separated from aerial tissues (Supplemental Figure 11).

Compared with other tissues, bulbs showed reduced levels of 500 metabolites and increased levels of 39 metabolites. Among the depleted metabolites, amino acids and derivatives (24.95%) and lipids (19.89%) ranked among the top classes (Supplemental Figure 12A). KEGG enrichment analysis revealed the depletion of metabolites associated with oxidative phosphorylation and zeatin biosynthesis, both of which are associated with energy metabolism and growth (Supplemental Figure 12D). Consistent with this result, transcriptome-based gene set enrichment analysis supported the downregulation of these two important pathways (Supplemental Figure 13B). In contrast, the enriched metabolites were associated with pathways related to amino acid biosynthesis and ABC transporters (Supplemental Figure 12C). Similarly, gene set enrichment analysis also supported the upregulation of these pathways (Supplemental Figure 13A). Among the top 5% of metabolites (61 metabolites) with the highest absolute loadings in principal-component analysis, 83.61%, including 50 depleted and one enriched metabolite, overlapped with the significantly differentially abundant metabolites identified above (Supplemental Figures 12F and 12G), suggesting that they play a major role in driving the distinct clustering of bulb samples.

Compared with aerial tissues, 188 metabolites were enriched in roots. In addition to a large proportion of amino acid derivatives (37.01%) and phenolic acids (16.37%), lipids (11.03%) and alkaloids (9.61%) were also markedly enriched (Supplemental Figure 14B). The arachidonic acid metabolism pathway was significantly enriched (Supplemental Figure 14C). Meanwhile, 93 metabolites were depleted, with phenolic acids (31.37%) accounting for the largest proportion (Supplemental Figure 14A). Among the top 5% of metabolites (61 metabolites) with the highest absolute loadings in principal-component analysis, 95.1%, corresponding to 58 metabolites, overlapped with the significantly enriched metabolites (Supplemental Figures 14F and 14G). These metabolites collectively drove the distinct clustering of root samples.

The AmAs displayed distinct accumulation patterns across tissues and developmental stages, forming seven K-means clusters. Class 1, which included 17 compounds, was predominantly distributed in roots throughout all developmental stages and included compounds such as narwedine and lycoranine A. Class 2, comprising 13 compounds, accumulated mainly in bulbs across all developmental stages, encompassing important bioactive metabolites such as galanthamine, narciclasine, and haemanthamine. Class 3, which contained 12 compounds, displayed high levels in both aerial tissues and roots across all developmental stages and included compounds such as lycorine (Supplemental Figure 15; Supplemental Table 14). We further examined the tissue distribution of key intermediates involved in galanthamine biosynthesis. Both 4′-O-methylnorbelladine and galanthamine (Class 2) exhibited the highest concentrations in bulbs across all developmental stages, with their levels in roots consistently ranking second. N-demethylnarwedine (Class 5) accumulated predominantly in bulbs, whereas N-demethylgalanthamine (Class 4) showed relatively higher abundance in aerial tissues (Supplemental Figure 16 and Supplemental Table 14). Consistent with the partitioning observed in the galanthamine pathway, roots and bulbs served as the major reservoirs for AmAs at a broader scale. Throughout the three developmental stages, AmAs that exhibited the highest accumulation in roots consistently represented the largest proportion: 53.13% at the flowering stage, 42.19% at the fruiting stage, and 39.06% at the leafing stage. Bulbs also served as a major accumulation site, accounting for 25.00%, 29.69%, and 32.81% of AmAs with peak accumulation at these three stages, respectively (Figure 1C and Supplemental Table 15A). Further quantitative analysis using a 50% content threshold confirmed this trend, showing that 11, 11, and 17 alkaloids exhibited root-dominant accumulation during the flowering, fruiting, and leafing stages, respectively, whereas three, three, and seven alkaloids were bulb-dominant (Supplemental Table 15B). Overall, bulbs and roots were identified as the principal AmA accumulation organs, with roots exhibiting preferential alkaloid accumulation.

Identification and functional characterization of the enzymes constituting the galanthamine biosynthetic pathway

Given the persistently high levels of galanthamine in roots and bulbs, we sought to identify NMT genes potentially involved in galanthamine biosynthesis by screening for those with significantly elevated expression in these tissues compared with other tissues across developmental stages. This analysis yielded a set of NMT gene candidates for individual functional validation (Figure 3A and Supplemental Figure 18A). The identified candidate proteins were heterologously expressed and individually assayed in vitro with N-demethylgalanthamine (compound 2) as the substrate. Among them, two catalyzed the conversion of compound 2 to galanthamine (compound 4), and these two were designated LrNMT1 and LrNMT2 (Figures 3B and 3D and Supplemental Figure 20). Notably, a recent study suggested that N-demethylnarwedine (compound 1) may be converted to galanthamine via narwedine (compound 3) by NtNMT and NtAKR (Mehta et al., 2024). To test whether LrNMTs could also participate in this hypothetical route, we assayed LrNMT1 and LrNMT2 with compound 1 as the substrate. The results demonstrated that both LrNMT1 and LrNMT2 could also catalyze the N-methylation of compound 1 to form compound 3 (Figure 3G and 3H and Supplemental Figure 21). RT-qPCR analysis further revealed that the LrNMT genes are highly expressed in roots (Supplemental Figure 22B).

To identify candidate reductases, we used LrNMT1 and LrNMT2 as bait genes in a series of co-expression analyses and screened genes annotated as carbonyl reductases that showed significant Pearson correlations with the two LrNMT genes (Figure 3F). All identified candidates were individually tested in vitro by incubating each recombinant protein with the substrate N-demethylnarwedine (compound 1) and NADPH (Supplemental Figure 18B). Among the candidates, one enzyme catalyzed the production of compound 2 and was designated LrAKR (Figure 3C and 3E and Supplemental Figure 23). We noted that carbonyl-reductase-catalyzed reactions can be either irreversible, such as the reduction of tropinone by tropinone reductase-II (Hashimoto et al., 1992), or reversible, as demonstrated by codeinone reductase in the interconversion between codeinone and codeine (Carr et al., 2021). Given these precedents, we examined whether the reaction catalyzed by LrAKR was also reversible by co-incubating LrAKR with compound 2 and NADP⁺. The result confirmed that the reverse reaction from compound 2 to compound 1 could also be catalyzed by LrAKR (Figure 3C and 3E and Supplemental Figure 24). Furthermore, to test its activity within the alternative route involving narwedine, we assayed LrAKR with compounds 3 and 4 as substrates. These experiments confirmed that LrAKR can also catalyze a reversible reaction between these two compounds (Figures 3G and 3I and Supplemental Figures 25 and 26). RT-qPCR analysis also revealed that LrAKR is highly expressed in roots (Supplemental Figure 22B). Collectively, our findings reveal a minimum metabolic network composed of two complete biosynthetic routes from compound 1 to compound 4, catalyzed by the same set of enzymes, LrAKR and LrNMT1/2, but acting in different orders and generating distinct intermediates.

Building on the in vitro enzymatic activities described above, we sought to reconstitute the complete galanthamine biosynthetic pathway in Yarrowia lipolytica. The coding sequences of LrN4OMT, NtCYP96T6, LrAKR, and LrNMT1 were co-integrated into the genome of the Y. lipolytica W29 strain using a CRISPR–Cas9-mediated approach, generating an engineered strain for functional characterization. When the resulting engineered strain was supplemented with norbelladine in shake-flask culture, galanthamine was detected at approximately 10 μg/l (Supplemental Figure 27).

Catalytic mechanisms of LrNMT and LrAKR

Molecular docking was used to generate a structural model of the ternary complex formed by LrNMT1, S-Adenosylmethionine (SAM), and compound 2. To examine the enzyme–substrate interface, we identified a total of 19 amino acids within 5 Å of the substrate (Supplemental Figure 28). In this model, compound 2 formed hydrophobic interactions with Trp26, Leu30, and Phe203 and a hydrogen bond with Tyr238 (Figure 4A). Beyond these direct interactions, we compared the sequences of active and inactive candidate enzymes to identify residues among these 19 positions whose variation correlated with catalytic activity. This comparison identified Thr19 as an activity-related residue (Figure 4E). Further inspection of the structural model revealed that residue 19 may contribute to a broader hydrogen-bonding network: in the inactive enzyme candidates, Ile19 was predicted to form hydrogen bonds only with Trp15 and Gln22, whereas in active LrNMT1/LrNMT2, Thr19 was predicted to form hydrogen bonds with Trp15, Tyr16, Gln22, and Gln105. This expanded hydrogen-bonding network may stabilize the enzyme’s conformation and enhance SAM binding (Figure 4B and Supplemental Figure 28A). The five selected residues in LrNMT1 were individually mutated to alanine (Ala), and the activities of the mutant enzymes were assayed with compound 2 as the substrate. Compared with wild-type LrNMT1, the W26A mutation almost completely abolished catalytic activity toward compound 2, and F203A markedly reduced catalytic activity. T19A, L30A, and Y238A each reduced he enzyme activity to approximately 30% of the wild-type level (Figure 4G).

Figure 4.

Key catalytic residues underlying LrNMT1/2 and LrAKR activity.

(A) Interaction between LrNMT1 and compound 2.

(B) Amino acids located within 5 Å of compound 2 (gray) in the inactive enzyme L40847.1 (orange) and the active enzymes LrNMT1 (purple) and LrNMT2 (blue). Divergent amino acids are shown as sticks.

(C) Interaction between LrAKR and compound 2.

(D) Amino acids located within 5 Å of compound 2 (gray) in the inactive enzyme L0008.36 (orange) and the active enzymes LrAKR (purple) and NtAKR (PP708064.1) (blue). Divergent amino acids are shown as sticks.

(E) Sequence alignment of NMTs. Light blue triangles represent amino acids forming hydrogen bonds, while light green highlights indicate amino acids involved in hydrophobic interactions.

(F) Sequence alignment of AKRs. Light blue triangles indicate amino acids forming hydrogen bonds, while light green highlights indicate amino acids involved in hydrophobic interactions.

(G) Catalytic activities of LrNMT1 and site-directed variants, shown as stacked bars.

(H) Catalytic activities of LrAKR and variants resulting from targeted mutations are shown with stacked bars.

Similarly, molecular docking was used to generate a structural model of the ternary complex formed by LrAKR, NADP⁺, and compound 2. A total of 20 amino acids with 5 Å of the substrate were identified (Supplemental Figure 29). Interaction analysis indicated that compound 2 formed hydrophobic interactions with Leu53, Tyr54, Ile130, and Gln218 and hydrogen bonds with His117 and Gln218 in LrAKR (Figure 4C). Among these substrate-proximal residues, Ala24, Phe28, Leu53, Met118, and Gln218 were not conserved between active and inactive candidates (Figures 4D and 4F). Further structural comparison suggested that, in the inactive candidates, the residue at position 24 has a larger side chain, which may introduce steric hindrance and distort the substrate-binding conformation, thereby weakening the enzyme–substrate interactions (binding energy: LrAKR, −3.8 kcal/mol; L0008.36, −3.0 kcal/mol). To test the functional importance of selected substrate-proximal residues, the corresponding residues were individually mutated to Ala, and enzymatic activity was assayed. Y54A and H117A almost completely abolished LrAKR activity toward compound 2, whereas L53A, I130A, and Q218A reduced catalytic activity to varying degrees (Figure 4H).

Evolutionary origin of AmA biosynthesis genes

Although the biosynthesis of haemanthamine (Mehta et al., 2024) and galanthamine has been elucidated, the evolutionary trajectory of these pathways remains poorly understood. We found that multiple gene families central to AmA biosynthesis, including TYDC, NBS, NR, NMT, CYP71DW1, and ODD, have undergone expansion in L. radiata (Figure 5A). Genes involved in the AmA pathway are distributed across multiple chromosomes, with chromosomes 1, 8, and 10 harboring a relatively higher number of pathway-related genes (Supplemental Figure 7). Two functional LrNMT genes were identified, each embedded within syntenic blocks formed by recent genome duplication. The Ks value between LrNMT1 and LrNMT2 (Ks = 0.20) corresponds to the recent WGD event in L. radiata, suggesting that these genes originated through this genomic duplication.

Figure 5.

Evolutionary expansion and lineage-specific diversification of AmA biosynthetic genes.

(A) Schematic representation of the biosynthetic pathways of galanthamine and haemanthamine along with key associated genes. Genes highlighted in pink represent gene families showing evidence of expansion.

(B) Copy numbers of homologous genes involved in AmA biosynthesis across 16 species. The phylogenetic tree follows the Angiosperm Phylogeny Group IV classification.

(C) Lineage-specific mutations in NBS genes identified in AmA-producing species of the subfamily Amaryllidoideae.

Phylogenetic and synteny analyses across angiosperms, gymnosperms, and bryophytes revealed that most AmA pathway genes can be traced to ancient origins. For instance, OMT1 could be traced to basal angiosperms, whereas NMT and ODD homologs are found in bryophytes, and NBS and SDR2 homologs are found in gymnosperms (Figure 5B and Supplemental Figure 36). Synteny analysis suggested that NBS was lost in eudicots. In general, syntenic blocks containing these AmA-pathway–related genes were strongly conserved across Amaryllidaceae and monocots (Supplemental Figure 30).

The NBS enzyme, responsible for generating the unique precursor of AmAs, is a member of the Bet V1 protein family. The glycine-rich phosphate-binding loop is a conserved ligand-binding structural domain, and this sequence is highly conserved in monocot plants. Conserved residues in NBS, such as Tyr68, Glu71, Lys83, and Ile102 (Singh et al., 2018), are present in AmA-producing plants, including Narcissus sp. aff. pseudonarcissus, Leucojum aestivum, and L. radiata, whereas positions 68 and 102 exhibit extensive variation in non-producing species (Figure 5C). These variations in conserved residues likely contributed to the functional divergence of NBS, enabling it to catalyze the key dedicated step in AmA biosynthesis. The CYP96T subfamily, unique to the Amaryllidoideae, forms a monophyletic group associated with AmA diversification, and AmA-producing species exhibit lineage-specific residue substitutions (Supplemental Figure 32A) (Mehta et al., 2024). Together, these findings indicate that the AmA pathway has ancient evolutionary origins. Its establishment in the Amaryllidoideae lineage was likely driven by the functional specialization of core enzymes, evidenced by conserved residue substitutions and the evolution of the clade-specific CYP96T subfamily. Subsequent gene family expansion likely further reinforced and diversified the pathway in Amaryllidoideae.

Discussion

In this study, we report a reference genome of L. radiata, representing, to our knowledge, the first reference genome for the Amaryllidoideae subfamily. Our analyses indicate that L. radiata has undergone only one recent WGD event, and that its enlarged genome size is primarily attributable to extensive LTR-RT insertions acting in concert with this duplication. The long introns observed in L. radiata, with an average length of 3,720 bp, contribute significantly to its large average gene length. This observation aligns with a broader trend across plants with expanded genomes, as evidenced by recent assemblies of other giant genomes, such as Paris polyphylla (52.75 Gb), with introns averaging 7,524 bp (Zeng et al., 2025), and Lilium sargentiae (35.66 Gb), with introns averaging 9,070 bp (Liang et al., 2025). These cases of even more extreme intron expansion support a positive correlation between genome size and intron length (Supplemental Figure 1C).

Research on galanthamine biosynthesis has spanned several decades (Eichhorn et al., 1998; Kilgore et al., 2014, 2016; Singh et al., 2018; Li et al., 2019; Tousignant et al., 2022). A recent study in Narcissus reported the putative identification of compound 1 ([M + H]⁺ m/z 272.1281), compound 2 ([M + H]⁺ m/z 274.1443), and compound 3 ([M + H]⁺ m/z 286.1437). Metabolic pathways often exhibit plasticity, in which a single precursor can be converted to the same product via multiple parallel routes. Our study elucidates one such example in the galanthamine pathway, showing that compound 1 is channeled to compound 4 through either intermediate compound 2 or compound 3. Notably, while our in vitro enzymatic assays support the existence of these two parallel routes, this does not necessarily indicate that both pathways are equally dominant in planta. Previous studies using isotope labeling have demonstrated that N-demethylgalanthamine (compound 2) serves as the direct precursor of galanthamine (compound 4) in vivo. We therefore speculate that the pathway involving N-demethylgalanthamine (compound 2) as an intermediate likely represents the predominant route in native plants. Nevertheless, verification of the actual in planta pathway remains an important direction for future research. In addition, the two parallel routes observed in vitro form a parallelogram-like metabolic network, a topology that has also been reported in the biosynthesis of other specialized metabolites (Cheng et al., 2022; McClune et al., 2025). We identified three key downstream genes, namely LrNMT1, LrNMT2, and LrAKR, through integrated analysis of metabolite and gene expression profiles. The functions of these enzymes were confirmed by matching liquid chromatography–mass spectrometry (LC–MS) retention times and mass spectra with those of authentic standards, thereby enabling reconstruction of the reaction pathways and refinement of the biosynthetic model for AmAs. Sequence comparison revealed that LrNMT1 and LrNMT2 share 77% and 84% identity with NtNMT, respectively, whereas LrAKR shares 81% identity with NtAKR. Consistently, phylogenetic analysis indicates that the functional genes from L. radiata are most closely related to those from Narcissus (Supplemental Figure 34). Further phylogenetic analysis showed that LrNMT1 and LrNMT2 belong to the γ-tocopherol methyltransferase (γ-TMT) clade (Supplemental Figure 35).

Furthermore, the successful engineering of Y. lipolytica for heterologous production demonstrates the feasibility of synthesizing AmAs in yeast and provides proof-of-concept for functional pathway reconstruction and enzymatic compatibility among the reconstructed pathway components. However, the catalytic efficiencies of several pathway enzymes remain limited, leading to the accumulation of intermediates, and the overall yield remains low. Future efforts focusing on enzyme optimization, pathway balancing, and host metabolic engineering could help overcome these bottlenecks and improve production efficiency.

Conserved genomic signatures provide important clues to the evolutionary history of AmA biosynthesis. Genes involved in AmA biosynthesis formed well-supported clades in phylogenetic trees, and their conserved syntenic blocks further supported clear orthologous relationships. Homologs of several core biosynthetic genes, including NBS, NMT, SDR2, and ODD, were detected in basal angiosperms, suggesting that these gene families originated before the diversification of the Amaryllidaceae. However, the complete biosynthetic pathway likely arose later through lineage-specific neofunctionalization of key enzymes, such as CYP96Ts and NBS, from these ancient gene families. Consistent with this scenario, members of the CYP96T family and CYP71DW1 show more restricted phylogenetic distributions, reflecting lineage-specific specialization. The amplification of key enzyme families through WGD, coupled with the fixation of conserved catalytic residue variants, likely played a central role in shaping the biosynthetic capacity for alkaloid production.

The metabolic profiles of L. radiata were primarily segregated by tissue type, reflecting functional differentiation among organs. As a major storage organ (Jiang et al., 2021), the bulb is generally characterized by predominant storage and primary metabolic functions, whereas the root shows features associated with active secondary metabolism. The distribution pattern of alkaloids further supports this distinction, with alkaloid levels consistently higher in roots than in other tissues across all developmental stages, followed by the bulb. At the transcriptional level, L. radiata AmA biosynthesis genes, including TYDC, C4H, C3H, PAL, NBS, and the galanthamine-related AKR and NMT, exhibited higher expression levels in roots, while haemanthamine-related genes (ODD, CYP71DW1, SDR2, and OMT1) showed elevated expression in either bulbs or roots (Figure 5A and Supplemental Figure 22A). A report on Narcissus indicated that the leaf base serves as a major biosynthetic center for AmAs; however, that study primarily focused on aerial tissues (Mehta et al., 2024). Our findings extend this view by highlighting the metabolic contribution of underground organs to AmA biosynthesis. In other plant species, roots have also been reported as important sites for secondary metabolite synthesis. In Solanaceae plants, tropane alkaloids such as scopolamine, as well as pyridine alkaloids such as nicotine, are predominantly synthesized and accumulated in the roots (Hashimoto and Yamada, 1994; De Luca and St Pierre, 2000; Ziegler and Facchini, 2008; Yang et al., 2023; Zhang et al., 2023). In Senecio species, pyrrolizidine alkaloid N-oxides are mainly produced in the roots before being transported to the aerial parts (Hartmann and Dierich, 1998). Similarly, nicotine biosynthesis in tobacco occurs exclusively in roots (Dawson, 1941). Together, these observations indicate that roots may serve as important sites of AmA metabolism and may provide key tissue-level targets for future targeted exploration and exploitation of the medicinal potential of L. radiata.

Overall, our study not only resolves key enzymatic steps of galanthamine biosynthesis but also illuminates the evolutionary trajectory of AmA pathways. The L. radiata genome provides a robust foundation for future investigations into the diversification of specialized metabolism and offers a valuable genetic resource for synthetic biology efforts aimed at sustainable alkaloid production.

Methods

Genome sequencing

Plants were grown in a greenhouse at Northwestern Polytechnical University. Young leaves were collected during the active leaf growth stage and used for genome sequencing. High-quality genomic DNA was extracted from fresh young leaves using the CTAB method. Following the standard PacBio protocol, size-selected SMRTbell libraries with target insert sizes of 10 or 20 kb were constructed for sequencing. Genomic DNA was sheared using G-tubes, and single-stranded overhangs were removed before DNA damage repair and A-tailing. The DNA fragments were then ligated to sequencing adapters to generate PacBio-compatible libraries. After polymerase reads were obtained, they were processed with SMRT Link to generate subreads, which were subsequently used to generate circular consensus sequencing HiFi reads.

Genome assembly and assessment

To estimate the genome size and heterozygosity of L. radiata, we first generated k-mer frequency distributions from Illumina short reads using Jellyfish (v2.2.0) (Marçais and Kingsford, 2011). The resulting k-mer spectra were analyzed with GenomeScope 2.0 (Ranallo-Benavidez et al., 2020) to estimate genome size, heterozygosity, and repeat content. PacBio HiFi reads were used for contig assembly with HiFiasm (v0.21.0) using default parameters (Cheng et al., 2021). The assembly results were processed to remove redundant haplotigs using purge_haplotigs (Roach et al., 2018). The completeness and accuracy of the genome assembly were assessed using BUSCO (Seppey et al., 2019) with 1,614 core genes from the embryophyta_odb10 dataset. Merqury (v1.3) was used to evaluate the base-level accuracy (Rhie et al., 2020).

Chromosome assignment and assembly using Hi-C

For Hi-C library construction, cells were fixed with formaldehyde to cross-link DNA with associated proteins. The cross-linked DNA was then digested with the restriction enzyme HindIII to introduce breaks at restriction sites, followed by end repair and biotin labeling. Subsequently, proximity ligation was performed to generate chimeric DNA molecules, which were then purified and sheared into fragments ranging from 300 to 700 bp. Biotin-labeled DNA fragments were enriched by pull-down and used for library construction. The library concentration and insert-size distribution were assessed using Qubit 2.0 and an Agilent 2100 Bioanalyzer to ensure library quality. The library was then sequenced as paired-end 150-bp reads (PE150) on an Illumina platform. After the raw sequencing data were filtered, clean reads were mapped to the genome assembly using Juicer (v1.6) (Durand et al., 2016b). Next, a chromosome-level assembly was generated using 3D-DNA (v180922) with default parameters (Dudchenko et al., 2017). Finally, Juicebox (v2.20) was used to manually review and refine the candidate assembly based on Hi-C contact frequencies (Durand et al., 2016a).

Transcriptome analyses

For transcriptome profiling, tissues were collected from L. radiata plants at three developmental stages: flowers, rachises, bulbs, and roots at the flowering stage (Stage 1); fruits, rachises, bulbs, and roots at the fruiting stage (Stage 2); and leaves, bulbs, and roots at the leafing stage (Stage 3). For each sample, three biological replicates were collected. The samples were flash-frozen in liquid nitrogen for RNA extraction. Total RNA was extracted using the RNAprep Pure Plant Kit (Tiangen, China). RNA quality was assessed using the Qubit RNA Assay Kit and the Agilent 2100 Bioanalyzer (Agilent Technologies). After quality assessment, eukaryotic mRNA was enriched using oligo(dT)-attached magnetic beads. The mRNA was fragmented using fragmentation buffer, reverse transcribed into cDNA, and then subjected to end repair, A-tailing, and sequencing adapter ligation. The resulting cDNA fragments were amplified, and the PCR products were purified using AMPure XP beads to construct the sequencing library. Qualified libraries were sequenced on an Illumina platform. RNA-seq reads were aligned to the reference genome using HISAT2 (v2.1.0) (Kim et al., 2015), and transcript assemblies were generated with StringTie (v2.2.1) (https://github.com/skovaka/stringtie2) (Pertea et al., 2015; Shumate et al., 2022). Protein-coding sequences were subsequently predicted from the assembled transcripts using TransDecoder (v5.7.1) (https://github.com/TransDecoder/TransDecoder). Transcripts with an average count of < 10 across all samples were removed before downstream analyses. Differentially expressed genes were identified using the R package DESeq2 (Love et al., 2014). Genes with |log₂FC| > 1 and adjusted p < 0.05 were considered significantly differentially expressed.

Liquid chromatography–MS analysis of metabolites

Metabolome profiling was performed using a widely targeted metabolomics approach by Wuhan MetWare Biotechnology Co., Ltd. (Wuhan, China). The sampling strategy was consistent with that used for transcriptome analysis. Plant tissue samples were freeze-dried under vacuum using a lyophilizer (Scientz-100F) and ground into a fine powder. Approximately 50 mg of powdered tissue was weighed, mixed with 1,200 μl of prechilled (−20°C) 70% aqueous methanol, and thoroughly homogenized. The supernatant was collected and filtered through a 0.22-μm microporous membrane, and data were acquired using ultra-performance liquid chromatography (ExionLC AD) coupled with tandem MS (Applied Biosystems 6500 QTRAP). Chromatographic separation was performed on an Agilent SB-C18 column (1.8 μm, 2.1 × 100 mm). The mobile phase consisted of solvent A (ultrapure water containing 0.1% formic acid) and solvent B (acetonitrile containing 0.1% formic acid) with gradient elution. The flow rate was set at 0.35 ml/min, the column temperature was maintained at 40°C, and the injection volume was 2 μl. Metabolite identification was performed based on database matching and MS/MS spectral information, and metabolite quantification was conducted in multiple-reaction-monitoring mode using a triple quadrupole mass spectrometer. MS data were processed using Analyst 1.6.3 software. Characteristic precursor–product ion pairs for each metabolite were selected, and the signal intensity of the selected ions was recorded by the detector as counts per second. The raw mass spectral data files were further analyzed using MultiQuant software for chromatographic peak integration and correction. The integrated peak area of each chromatographic signal was used to represent the relative abundance of the corresponding metabolite.

Metabolomic analysis

Differential metabolite analysis was performed using the R package MetaboAnalystR (v4.0) (Pang et al., 2024). Differential metabolites were identified using the following criteria: variable importance in projection scores > 1, |log₂FC| > 2, and p < 0.05. KEGG enrichment analysis of differential metabolites was conducted using the R package clusterProfiler (Yu et al., 2012), with a significance threshold of p < 0.05. Heatmaps were generated using the R package pheatmap (https://github.com/raivokolde/pheatmap).

Genome annotation

Genome-wide repetitive sequences and transposable elements were identified using a combination of homology-based and de novo prediction approaches. For homology-based annotation, RepeatMasker (v4.1.0) (Tarailo-Graovac and Chen, 2009) and ProteinMask (v4.0.5, included in the RepeatMasker package; p-value threshold, 1 × 10⁻⁴) were used to search the genome assemblies against the Repbase database and to identify and classify known repetitive elements based on sequence similarity. For de novo prediction, species-specific repeat libraries were constructed using RepeatModeler (v1.0.5) (Sato et al., 2016) and LTR_FINDER (v1.07) (Xu and Wang, 2007). The resulting de novo repeat libraries were then used as input for RepeatMasker (v4.1.0) to detect repetitive sequences across the four genomes. Finally, repeat annotations obtained from homology-based and de novo analyses were integrated according to their genomic coordinates, and redundant entries occupying identical positions were removed to generate a non-redundant repeat annotation dataset. Tandem repeats were identified using Tandem Repeats Finder (v4.07) (Benson, 1999), with the maximum period size parameter set to 2,000.

Protein-coding genes in L. radiata were predicted using an integrative strategy combining ab initio gene prediction, homology-based gene prediction, and transcriptome evidence. Ab initio gene structure prediction was performed using AUGUSTUS (v3.4.0) and GlimmerHMM (v3.0.4) (Majoros et al., 2004; Stanke et al., 2006; Sato et al., 2016). For homology-based gene prediction, protein sequences from Arabidopsis thaliana, A. cepa, A. fistulosum, and A. sativum were aligned to the L. radiata genome using TBLASTN. TBLASTN (Camacho et al., 2009) was employed with an E-value cutoff of 1 × 10⁻⁵ and a minimum query coverage of 25% to identify candidate homologous regions, and GeneWise (v2.4.1) (Birney et al., 2004; Birney and Durbin, 2000) was subsequently used to delineate protein-coding regions based on the alignment results. Finally, gene models generated from ab initio prediction, homology-based prediction, and RNA-seq evidence were integrated using EvidenceModeler (EVM; v1.1.1) (Haas et al., 2008). The predicted gene products were functionally annotated by searching them against the GO, KEGG, TrEMBL, and InterPro databases using BLAST where applicable (Kanehisa and Goto, 2000; O'Donovan et al., 2002; Zdobnov and Apweiler, 2001).

Gene family and phylogenomic analysis

Gene family identification was performed using OrthoFinder (v2.4.0; Emms and Kelly, 2015) with protein sequences from L. radiata, A. fistulosum, A. sativum, A. cepa, A. africanus, A. officinalis, E. guineensis, Z. marina, S. lycopersicum, and A. trichopoda. Single-copy orthologous gene sequences were used to construct a phylogenetic tree with IQ-TREE (v2.2.6) (Minh et al., 2020). Divergence times were estimated using the MCMCtree program in PAML (v4.9) (Yang, 2007) based on the phylogenetic tree. Gene family expansion and contraction were analyzed using CAFÉ (De Bie et al., 2006). The calibration points were set as follows: 173–199 Mya for the divergence between A. trichopoda and S. lycopersicum; 139–156 Mya for the divergence between S. lycopersicum and Z. marina; and 55–71 Mya for the divergence between A. officinalis and A. africanus. GO and KEGG enrichment analyses of gene families were performed using the R package clusterProfiler (Yu et al., 2012), with a significance threshold of p < 0.05.

WGD analysis

MCScanX was employed to identify intragenomic syntenic blocks in the L. radiata genome (Wang et al., 2012). The Ks values for paralogous gene pairs within these blocks were calculated using wgdi (v0.4.1) (Sun et al., 2022). To calibrate the Ks distribution and accurately infer WGD events in L. radiata, we used the established phylogenetic relationships among A. fistulosum, A. officinalis, L. radiata, E. guineensis, Oryza sativa, Dioscorea rotundata, and A. thaliana, together with ksrates (v1.1.1) and wgd (v1.2) (Zwaenepoel and Van de Peer, 2019; Sensalari et al., 2022).

Verification of enzyme activity

The coding sequences of LrAKR and LrNMT genes from L. radiata were amplified and cloned into the pET28a expression vector. The primers used for vector construction are listed in Supplemental Table 20. The constructed plasmids were transformed into the Escherichia coli expression strain BL21 (DE3). The transformed cells were cultured in 800 ml of 2× YT liquid medium (containing 100 μg/ml kanamycin sulfate) with shaking at 37°C and 220 rpm. When the culture reached an OD600 of approximately 0.8, isopropyl β-D-thiogalactopyranoside was added to a final concentration of 0.5 mM, and the culture was incubated at 16°C to induce protein expression. After induction, the bacterial cells were harvested and lysed, and the supernatant was collected. The target proteins were purified by nickel-affinity chromatography and eluted using a stepwise imidazole gradient of 50, 100, and 300 mM. The eluted protein solution was concentrated using Millipore centrifugal filter units, and the protein concentration was determined using a BCA assay.

The enzymatic reaction mixture contained 50 mM Tris, 2 mM substrate, 2 mM cofactor (S-adenosyl-L-methionine for NMT reactions or NADPH for AKR reactions), and 20 μg of purified protein in a final volume of 200 μl. The reaction mixture was incubated at 35°C with shaking for 12 h, terminated with an equal volume of methanol, and ultrasonicated for 30 min. Sample analysis was performed using high-performance liquid chromatography (HPLC)–MS on an Agilent 1200 system equipped with a Waters C18 column (4.6 × 250 mm, 5 μm). The mobile phase consisted of 0.1% formic acid in water (solvent A) and acetonitrile (solvent B) with the following gradient elution program: 0–9 min, 2% B; 9–19 min, linear increase to 25% B; 19–23 min, linear increase to 95% B; 23–25 min, 95% B; and 25–29 min, return to 2% B. The flow rate was 1 ml/min, and the injection volume was 50 μl. Ultraviolet detection was performed at 289 nm, and mass spectrometric detection was performed in full-scan mode, with the total ion current recorded.

Reconstruction of the galanthamine biosynthetic pathway in yeast

Y. lipolytica exhibits substantial metabolic potential for natural product biosynthesis. Because it does not exhibit the Crabtree effect and can provide key metabolic resources, including acetyl-CoA, NADPH, and ATP, it is a promising host for metabolic engineering (Blank et al., 2005; Wasylenko et al., 2015; Ma et al., 2020; Wang et al., 2022). The genes encoding LrN4OMT (GenBank: QEP29044.1), NtCYP96T6 (GenBank: PP708068), ATR2 (GenBank: NP_194750.1; P450 reductase from A. thaliana), LrAKR (GenBank: PZ049380; this study), and LrNMT1(GenBank: PZ049378; this study) were synthesized by TsingKe (Beijing, China) after codon optimization for expression in Y. lipolytica. The integrative vectors were constructed using pMD-19T as the backbone and were designed to target three chromosomal integration sites: KU80, E-14, and IntE_3. For the construction of engineered Y. lipolytica strains, single gRNA plasmids and linearized homologous donor plasmids were introduced into cells by lithium acetate transformation, enabling CRISPR-Cas9-mediated chromosomal integration through homologous recombination. The detailed construction process and integration methods followed a previously described protocol (Wang et al., 2022). LrN4OMT was integrated into the KU80 locus, NtCYP96T6 and ATR2 were integrated into the E-14 locus, and LrAKR and LrNMT1 were integrated into the IntE_3 locus, with each gene pair assembled using head-to-head promoters. Positive transformants were selected on YPDAN solid medium containing 250 μg/ml nourseothricin and 400 μg/ml hygromycin. The engineered strain seed culture was inoculated at 1% (v/v) into 30 ml of fermentation medium supplemented with 1 mM substrate and cultured at 30°C with shaking at 220 rpm for 4 days. After fermentation, 500 μl of culture was mixed with an equal volume of methanol, ultrasonicated for 30 min, and centrifuged at 12 000 rpm for 15 min. The supernatant was collected and analyzed by HPLC–MS using the conditions described above.

Molecular docking and site-directed mutagenesis

The structures of target proteins in complex with their corresponding small molecules, including substrates or cofactors, were predicted using AlphaFold 3 (Krokidis et al., 2025), and the model with the highest-ranking score was selected as the receptor structure for molecular docking. Docking was performed using AutoDock (v4.2.6) (Morris et al., 2009). Protein and ligand structures were preprocessed with AutoDock Tools (v1.5.6), including the addition of hydrogens, charge assignment, and file format conversion. During docking, a grid box encompassing the entire protein was defined with dimensions of 15 × 15 × 15 Å, while all other parameters were kept at their default values. Docking results were screened to select binding poses with plausible conformations and favorable binding energies for subsequent analysis. Protein–ligand interactions, including key residues and interaction types, were further analyzed using PLIP (Adasme et al., 2021). Substitutions at the selected residues in LrNMT1 and LrAKR were constructed by overlap-extension PCR, and the activities of the resulting mutants were assayed using compound 2 as the substrate with three biological replicates.

Identification of AmA-biosynthesis genes and microsynteny analysis

Homologous genes involved in AmA biosynthesis were identified from multiple species using BLASTP searches with an E-value cutoff of 1 ×10⁻⁵ and Hidden Markov Model profiling using HMMER (v3.3). For phylogenetic analysis, amino acid sequences were aligned using MAFFT (v7.245) (Katoh and Standley, 2013), and poorly aligned regions were removed with trimAl (v1.4.rev22) (Capella-Gutiérrez et al., 2009). Maximum-likelihood phylogenetic trees were constructed using IQ-TREE (v.2.2.6) with 1,000 bootstrap replicates. An all-versus-all BLASTP search with an E-value cutoff of 1 × 10⁻⁵was performed using the protein sequences of 14 annotated genomes to generate a protein similarity database. MCScanX was subsequently used to identify syntenic blocks across these genomes. Using AmA biosynthetic genes from L. radiata as anchor genes, homologous genes from other genomes were identified, and microsyntenic regions were examined within the 50 genes upstream and downstream of each target gene to detect conserved collinear segments.

Data and code availability

The genome assembly and annotations, together with all raw sequencing data and transcriptome reads of L. radiata, were deposited in the China National GeneBank Database (CNGBdb, https://db.cngb.org/) under accession number CNP0008599. The sequences generated in this study have been deposited in GenBank under accession numbers PZ049378 (LrNMT1), PZ049379 (LrNMT2), and PZ049380 (LrAKR).

Funding

This study was supported by the National Natural Science Foundation of China (32370667 and 32470272), the National Key R&D Program of China (2022YFC3400300), the Practice and Innovation Funds for Graduate Students of Northwestern Polytechnical University (ZX2025026), and the Postdoctoral Fellowship Program of CPSF (GZC20242255).

Acknowledgments

This work was supported by the computational resources provided by the Computing Center in Xi’an. No conflict of interest is declared.

Author contributions

J.C., H.J., and J.H. conceived, designed, and supervised the entire study. Y.Z. conducted most of the data analyses and experiments, and prepared the manuscript. P.Z., Y.H., and P.L. analyzed the genomic data. X.L., H.C., F.F., and Q.L. assisted with experimental work. D.Z. and Y.P. contributed to manuscript preparation . J.H., H.J., and J.C. revised the manuscript.

Published: June 5, 2026