Ten genes in the diosmin biosynthetic pathway were selected from different plants and used for module assembly: PAL, C4H, 4CL, CHS, CHI, FNS, F3′H, F4′OMT, F7GT, and 1,6RhaT ( Supplementary Table 1 ). Type IIS restriction enzyme sites (e.g., BsaI and BpiI) within the genes were mutated to remove restriction sites without introducing changes in the encoded amino acids. The coding sequences (CDSs) of nine genes (AtPAL, AtC4H, Sh4CL, OsCHS, OsFNS, OsF3′H, PaF4′OMT, CsUGT76F1, and Cs1,6RhaT) were synthesized by Bionics Inc. (Korea). Four genes (AtC4H, OsCHS, OsFNS, and OsF3′H) were codon-optimized for efficient expression in N. benthamiana plants using the GenScript GenSmart™ Codon Optimization tool. The CDS of BrCHI was isolated from total RNA extracted from Brassica rapa L. leaves by reverse-transcription PCR using amfiRivert cDNA Synthesis Mastermix (GenDEPOT, Barker, TX, USA), PrimeSTAR HS DNA Polymerase (Takara bio Inc., Japan), and gene-specific primers ( Supplementary Table 1 ) under the following conditions: 98°C for 2 min; 33 cycles of 98°C for 10 s, 58°C for 15 s, and 72°C for 1 min; and a final extension at 72°C for 3 min. The PCR products were purified using a QIAquickGEL Extraction kit (Qiagen, Germany). The ten genes (AtPAL, AtC4H, Sh4CL, OsCHS, OsFNS, OsF3′H, PaF4′OMT, CsUGT76F1, and Cs1,6RhaT) were subcloned into the level 0 (plCH41308) module using a Modular Cloning (MoClo) system (Addgene, USA) (Weber et al., 2011; Engler et al., 2014; Gantner et al., 2018) and verified by Sanger sequencing ( Supplementary Table 1 ).

Three types of modules (level 0, basic modules; level 1, transcription units; level M, multigene constructs) were constructed using the MoClo system (MoClo Tool Kit #1000000044; MoClo Plant Parts Kit# 1000000047; MoClo Plant Parts II and Infrastructure Kit #1000000135). Transcriptional units (level 1 modules) were constructed by assembling the level 0 (Promoter, CDS, Terminator) modules ( Supplementary Tables 2 , 3 ). Level M modules [PCFF (P, AtPAL; C, OsCHS; F, OsFNS; F, OsF3′H), PC4 (P, AtPAL; C, AtC4H; 4, Sh4CL), CCFF (C, OsCHS; C, BrCHI; F, OsFNS; F, OsF3′H), MGR (M, PaF4′OMT; G, F7GT; R, Cs1,6RhaT)] were constructed using the level 1 modules and an end-linker to build multigene expression vectors. Golden Gate reaction mixtures were prepared as follows: 40 fmol of each DNA insert or plasmid, 1.5 μl of T4 DNA ligase (400 U/μl, NEB), 1 μl of restriction enzyme, 2.5 μl of 10× ligation buffer, and sterile water to bring the final volume to 25 μl. BsaI-HF v2 (NEB, USA) or BpiI (Thermofisher, USA) type IIS restriction enzymes were used to assemble modules (Weber et al., 2011). After incubation in a thermal cycler, Escherichia coli was transformed with the entire reaction, and the resulting transformed bacteria were plated onto LB medium with the appropriate antibiotics and X-gal. White colonies were resuspended in 20 μl of sterile water, and 5 μl of the suspension was used for colony PCR (Marillonnet and Grützner, 2020).

Eight genes encoding F4′OMT and 1,6RhaT modification enzymes in the diosmin biosynthetic pathway were selected from various plants and synthesized by Bionics Inc. (Korea): PaF4′OMT, MpOMT4, SOMT2, CreOMT1, CreOMT4, Cs1,6RhaT, and two CiRhaTs ( Supplementary Table 1 ). The CDSs of synthesized F4′OMTs (PaF4′OMT, MpOMT4, SOMT2, CreOMT1, CreOMT4) and 1,6RhaTs (Cs1,6RhaT, CiRhaT-GD4x, CiRhaT-AH2x) were subcloned into the pENTR-D-TOPO vector using a TOPO cloning kit (Thermo Fisher Scientific, USA) and verified by Sanger sequencing. The CDSs of F4′OMT and 1,6RhaT genes were cloned into the pGEX-6P-1 vector linearized by BamHI digestion using an In-fusion Advantage PCR cloning Kit (Clontech, USA) and verified by Sanger sequencing; they were then transformed into E. coli BL21 (DE3) (Novagen, Germany). The transformed E. coli BL21 cells were grown overnight in LB medium with ampicillin (100 mg/L) at 37°C and 200rpm. Cells were then inoculated into liquid LB medium and cultured to OD₆₀₀ = 0.6–0.8 (Park et al., 2023). Production of recombinant F4′OMT (PaF4′OMT, MpOMT4, SOMT2, CreOMT1, CreOMT4) or 1,6RhaT (Cs1,6RhaT, CiRhaT-GD4x, CiRhaT-AH2x) proteins fused to glutathione S-transferase (GST) was induced using 100 μM isopropyl β-D-1-thiogalactopyranoside (IPTG) at 20°C for 3 h. Small aliquots of the induced cultures were set aside for SDS-PAGE analysis. Diosmetin or diosmetin7-O-β-D-glucopyranoside (diosmetin7-O-glucoside) (100 μM) was added as a substrate depending on the experiment. After 3 h of incubation at 28°C, the culture medium was extracted with an equal volume of ethyl acetate. After centrifugation at 13,000 rpm for 10 min at 4°C (MDX-310, Tomy Seiko Co., Ltd., Japan), the upper ethyl acetate phase was transferred to a fresh tube and evaporated under N₂ gas. The residue was dissolved in 80% (v/v) methanol and analyzed by high-performance liquid chromatography (HPLC), ultra-performance liquid chromatography (UPLC), and UPLC-DAD-QToF/MS.

Agrobacterium strain GV3101 harboring the overexpression vectors was transformed using the freeze-thaw method (Weigel and Glazebrook, 2006). A positive colony of GV3101 transformant was picked and pre-cultured in YEP medium with antibiotics at 28°C for 24 h. For the main culture, 50 μl of pre-culture was transferred into a fresh 50 ml of YEP medium containing antibiotics. This cell culture was centrifuged at 5,000 rpm for 10 min (MDX-310, Tomy Seiko Co., Ltd., Japan). The pellet was resuspended in agro-infiltration buffer containing 10 mM MES (pH 5.6), 10 mM MgCl₂, 5% (w/v) D-glucose, and 200 µM acetosyringone (4-hydroxy-3,5-dimethoxyacetophenone, Sigma) to a final OD₆₀₀ = 0.6. Equal volumes of Agrobacterium cells (OD₆₀₀ = 0.6) carrying the overexpression vector and those containing a construct harboring P19, which is a suppressor of RNA-mediated gene silencing, were mixed and incubated at room temperature for 2–3 h. N. benthamiana plants were grown in a greenhouse with a 16-h/8-h light/dark cycle (at 28°C during the day and 25°C at night). Agrobacterium cells in infiltration buffer were mixed thoroughly before co-infiltration. Agrobacterium mixtures were gently infiltrated using a 1-ml blunt-end syringe against the abaxial side of the leaves of 4-week-old N. benthamiana plants. Three individual plants were infiltrated with each construct or GV3101 (GV) cells used as a control. After agro-infiltration, the plants were grown in the same greenhouse for a further six days, and infiltrated leaves were then frozen in liquid nitrogen and stored at −80°C until use.

Flavonoid contents of agro-infiltrated N. benthamiana leaves were quantified as their aglycone forms generated by acid hydrolysis. Ground leaf tissue (100 mg FW) was mixed with 400 μl 80% (v/v) methanol and incubated overnight at 4°C. After centrifugation at 13,000 rpm for 10 min at 4°C, 200 μl of the supernatant was transferred to a new tube and was subjected to acid hydrolysis by adding 600 μl 1 M HCl and incubation at 94°C for 2 h. Flavonoids were then extracted using 800 μl of ethyl acetate, and the upper ethyl acetate phase was evaporated under N₂ gas. The residue was dissolved in 80% (v/v) methanol, filtered through a 0.2-μm Teflon polytetrafluoroethylene (PTFE) hydrophilic syringe filter (Thermo Fisher Scientific, USA), and subjected to analysis performed on a HPLC (High Performance Liquid Chromatography) system (Shimadzu, Japan) with the method described in Park et al. (2021) and an LC-20A UPLC (Ultra Performance Liquid Chromatography) system (Shimadzu) equipped with an Inertsil-ODS3 C18 column (3 μm, 3.0 × 100 mm; Shimadzu). For UPLC analysis, the mobile phase consisted of 0.1% (v/v) formic acid, acetonitrile, and methanol (60:16:24, v/v/v) at a constant flow rate of 0.3 ml/min (Chen et al., 2012). The temperature of the column was maintained at 30°C. A diode array detector was used for real-time monitoring of the chromatograms, and the spectra of the compounds were recorded between 210 nm and 800 nm. Flavonoids, including flavonol (quercetin), flavones (luteolin, apigenin), and methylated flavones (chrysoeriol, diosmetin), were analyzed at 350 nm. The compounds were identified by comparing their retention times and UV spectra to those of flavonoid aglycone standards analyzed by HPLC or UPLC.

The ground powder (100 mg FW) of agro-infiltrated N. benthamiana leaves was incubated overnight with 80% (v/v) methanol at 4°C. After centrifugation at 13,000 rpm for 10 min at 4°C, half of the extract was used for acid hydrolysis of flavonoids, and the other half was used for analysis of flavonoid glycosides. The extracts were injected into an ultra-performance liquid chromatography diode array detector (UPLC-DAD) system (SCIEX Co., USA) connected to a reverse-phase column (CORTECS UPLC T3, 2.1 × 150 mm I.D., 1.6 μm; Waters Co., Milford, MA, USA) and a CORTECS UPLC VanGuard™ T3 pre-column (2.1 × 50 mm I.D., 1.6 μm; Waters Co.). The mobile phase consisted of water containing 0.5% (v/v) formic acid (Sol. A) and acetonitrile containing 0.5% (v/v) formic acid (Sol. B). A gradient program was used, with the following mobile phase compositions at the times indicated (in v/v): 20 min, 25% B; 25 min, 50% B; 30 min, 90% B; 32 min, 90% B; 35 min, 5% B; 40 min, 5% B, with a flow rate of 0.3 ml/min. Flavonoid aglycones and glycosides were characterized using a quadrupole time-of-flight mass spectrometer (QToF-MS) (SCIEX Co.) in positive or negative ionization mode. Mass spectrometry conditions were maintained as follows: ion source gas, 50 psi; curtain gas, 30 psi; ion source temperature, 450°C; declustering potential (DP), 80 V; collision energy (CE), 15 ± 10 V; spray voltage, 5500 V; scan ranges of m/z 100–1200.

Quercetin, naringenin, eriodictyol, apigenin, luteolin, and diosmetin were purchased from Sigma-Aldrich (USA). Diosmetin 7-O-β-D-glucopyranoside (diosmetin 7-O-glucoside) was purchased from MedChemExpress (USA), and diosmin was purchased from Supelco (USA). The flavonoids were prepared as 100 mM stock solution in DMSO.

The plant phenylpropanoid and flavonoid biosynthetic pathways from phenylalanine to luteolin are well characterized (Ferrer et al., 2008; Muruganathan et al., 2022) ( Figure 1 ). Thus, to reconstitute the biosynthetic pathway that would yield the highest amounts of luteolin, we designed, tested, and compared three different gene expression systems (P+C+F+F, PCFF, PC4+CCFF). PAL and CHS are the rate-limiting enzymes in the phenylpropanoid pathway and flavonoid pathway, respectively, and FNS and F3′H catalyze C2-C3 double bond formation and B-ring hydroxylation of flavonoids, respectively. Therefore, we chose the four key genes PAL (P), CHS (C), FNS (F), and F3′H (F) and all seven genes [PAL (P), C4H (C), 4CL (4), CHS (C), CHI (C), FNS (F), F3′H (F)] in the flavonoid biosynthesis pathway to produce luteolin. We constructed three types of modules, namely PCFF (PAL, CHS, FNS, F3′H), PC4 (PAL, C4H, 4CL), or CCFF (CHS, CHI, FNS, F3′H) in each vector. Then, we co-infiltrated the vectors harboring the individual genes (P+C+F+F) or infiltrated vectors containing multigene expression modules (PCFF or PC4+CCFF) in N. benthamiana leaves ( Figure 2A ).

We analyzed the infiltrated leaves by UPLC-DAD-QToF/MS in positive ionization mode to compare the luteolin productivity of each module ( Figures 2B–D ). We detected and quantified luteolin through extracted-ion chromatograms (XICs) and mass profiles. In the QToF/MS analysis, luteolin (m/z 287.05 [M+H]⁺) had a retention time of 24.1 min in agro-infiltrated leaves for all three combinations of expression constructs (P+C+F+F; PCFF; PC4+CCFF) ( Figures 2B, C ). Luteolin content was 3.6-fold higher in leaves infiltrated with the PCFF (29.8 nmol/g FW) module than in those co-infiltrated with the individual constructs (P+C+F+F [8.4 nmol/g FW]). Furthermore, luteolin content was 4.5-fold higher in leaves co-infiltrated with the PC4+CCFF (134.4 nmol/g FW) modules than in those infiltrated with the PCFF module alone. Apigenin (32.3 nmol/g FW) was only detected in leaves co-infiltrated with the PC4+CCFF modules ( Figure 2D ). Thus, infiltration with the multigene expression modules resulted in higher levels of luteolin than did expression of the individual genes, and the PC4+CCFF modules together yielded higher levels of luteolin than did the PCFF module, indicating that overexpression of all seven genes in the flavonoid biosynthesis pathway resulted in the greatest accumulation of luteolin.

4′-O-methylation of luteolin to generate diosmetin is catalyzed by F′4OMT ( Figures 1 , 3A ). To select the optimal F4′OMT gene, we compared the enzyme activities of five candidate F4′OMTs (PaF4′OMT, MpOMT4, SOMT2, CreOMT1, CreOMT2) previously reported (Kim et al., 2005; Liu et al., 2017; Zohra et al., 2020; Liu et al., 2022). To examine the efficiency of diosmetin production, we conducted a substrate-feeding assay with luteolin as the substrate. Successful production of the recombinant proteins following induction by IPTG was demonstrated by SDS-PAGE analysis ( Figure 3B ). Bacterial cultures expressing these F4′OMTs fed with luteolin were extracted and analyzed by HPLC, which revealed that PaF4′OMT or MpOMT4 produced diosmetin (RT: 27.3 min), whereas those three OMTs (SOMT2, CreOMT1, CreOMT4) did not. As PaF4′OMT produced the most diosmetin from luteolin ( Figure 3C ), we decided to co-express PaF4′OMT with the PC4+CCFF modules in N. benthamiana.

We infiltrated N. benthamiana leaves with the PC4+CCFF or PC4+CCFF+PaF4′OMT constructs, and analyzed their acid-hydrolyzed extracts by UPLC. Luteolin produced by expression of the PC4+CCFF modules was methylated to chrysoeriol, but diosmetin was newly synthesized in the leaves co-infiltrated with PC4+CCFF+PaF4′OMT, indicating that PaF4′OMT successfully catalyzed the 4′-O-methylation of luteolin ( Figure 3D ). QToF/MS analysis of the leaves co-infiltrated with the PC4+CCFF+PaF4′OMT constructs identified diosmetin (m/z 301.07 [M+H]⁺) at a retention time of 25.6 min ( Figures 3E, F ). Diosmetin was produced only in the PC4+CCFF+PaF4′OMT combination (48.7 nmol/g FW) as much as luteolin content decreased (52.7 nmol/g FW) ( Figure 3G ). These results indicate that luteolin is successfully converted to diosmetin by PaF4′OMT expression in agro-infiltrated leaves but two-thirds of the luteolin (99.2 nmol/gFW) is still present without methylation.

Flavonoid 7-O-glucosyltransferase (F7GT) is well-characterized in Citrus species, and diosmetin 7-O-glucoside is reported to be produced by CsUGT76F1 in vitro (Liu et al., 2018) ( Figure 4A ).

Thus, we transiently co-expressed CsUGT76F1 with the PC4+CCFF+PaF4′OMT constructs in N. benthamiana leaves and assessed the ability of this combination of constructs to produce diosmetin 7-O-glucoside. The infiltrated leaves were extracted with 80% methanol and analyzed by UPLC-DAD-QToF/MS in positive ionization mode. Diosmetin 7-O-glucoside (m/z 463.12 [M+H]⁺) was detected in leaves co-infiltrated with PC4+CCFF+PaF4′OMT (50.1 nmol/g FW) or PC4+CCFF+PaF4′OMT+CsUGT76F1 (33 nmol/g FW) ( Figure 4B ). Unexpectedly, diosmetin 7-O-glucoside production was lower in leaves co-infiltrated with PC4+CCFF+PaF4′OMT+CsUGT76F1 than in those co-infiltrated with PC4+CCFF+PaF4′OMT ( Figures 4B, D ). These results indicate that the 7-O-glycosylation of diosmetin can be carried out effectively by endogenous F7GT activity in N. benthamiana. Even though these results suggest that CsUGT76F1 expression is unnecessary to catalyze the production of diosmetin 7-O-glucoside in N. benthamiana, we decided to include CsUGT76F1 in the set of constructs used for diosmin biosynthesis, so that our expression system could be used in other hosts with different levels of endogenous F7GT activity.

Diosmin is biosynthesized from diosmetin 7-O-glucoside in a reaction catalyzed by 1,6-rhamnosyltransferase (1,6 RhaT) ( Figure 5A ). We examined the ability of three previously reported 1,6RhaT genes, i.e., Cs1,6RhaT, CiRhaT-GD4x, and CiRhaT-AH2x (Frydman et al., 2013; Wu et al., 2022), to produce diosmin when heterologously expressed in N. benthamiana. To this end, we conducted a substrate-feeding assay with the enzymes encoded by these three genes and diosmetin 7-O-glucoside as a substrate. SDS-PAGE analysis revealed that two recombinant RhaT proteins (Cs1,6RhaT and CiRhaT-AH2x) were produced upon IPTG induction, whereas recombinant CiRhaT-GD4x was poorly induced ( Figure 5B ).

The bacterial cultures expressing 1,6RhaTs supplied with diosmetin 7-O-glucoside were extracted and analyzed by UPLC-DAD-QToF/MS in negative ionization mode to detect diosmin formation. QToF/MS analysis showed that the cells expressing Cs1,6RhaT produced the most diosmin (m/z 607.16 [M-H]⁻) ( Figures 5C, D ). To produce diosmin in N. benthamiana, Cs1,6RhaT was transiently co-infiltrated with the PC4+CCFF+PaF4′OMT+CsUGT76F1 constructs ( Figures 5E, F ). QToF/MS analysis showed that the leaves co-infiltrated with the ten selected genes (i.e., PC4+CCFF+PaF4′OMT+CsUGT76F1+Cs1,6RhaT) successfully produced diosmin (38.1 nmol/g FW) ( Figure 5F ). By determining the enzyme catalyzing the final stage, the 6′′-O-rhamnosylation of diosmetin 7-O-glucoside, we completed the selection of 10 genes (from PAL to RhaT) to reconstitute the diosmin biosynthesis pathway.

For optimum production of diosmin in N. benthamiana leaves, we constructed three modules [module 1, PC4; module 2, CCFF; module 3, MGR (PaF4′OMT+CsUGT76F1+Cs1,6RhaT)] that together contained the ten genes selected above ( Figure 6A ). We transiently co-infiltrated the three modules (PC4+CCFF+MGR) into N. benthamiana leaves and evaluated the capacity of these leaves to produce diosmin ( Figures 6B–D ). UPLC-DAD-QToF/MS analysis in positive ionization mode detected a protonated ion at m/z at 609.18 [M+H]⁺ corresponding to that of diosmin in leaves co-infiltrated with the PC4+CCFF+MGR modules ( Figures 6B, C ). The MS/MS spectra showed a protonated molecular ion (m/z 609.18 [M+H]⁺) and fragment ions at m/z 463.12 ([M+H-rham]⁺), m/z 301.07 ([M+H-rham-Glu]⁺), and m/z 286.04 ([M+H-rham-Glu-CH₃]⁺), suggesting the loss of a rhamnose; a rhamnose and a glucose; and a rhamnose, a glucose, and a methyl group, respectively, and these corresponded to the fragment ions of the diosmin standard. Thus, we identified diosmin based on its fragmentation pattern ( Figure 6C ). By co-infiltrating the PC4+CCFF+MGR modules in N. benthamiana leaves, a substantial amount of diosmin was produced (61.9 nmol/g FW), while diosmetin 7-O-glucoside was barely detected (4.8 nmol/g FW) as it was mostly converted to diosmin ( Figure 6D ). The diosmin content of N. benthamiana leaves co-expressing the three modules (PC4+CCFF+MGR) was 1.7-fold higher than that of leaves co-expressing two modules and the genes encoding the enzymes that catalyze the last three steps of the pathway individually (PC4+CCFF+PaF4′OMT+CsUGT76F1+Cs1,6RhaT) ( Figures 5D , 6D ). Thus, diosmin production is more efficient when all the genes in the pathway are grouped into modules than when some of them are expressed individually.

Other flavone dihexosides accumulated in leaves co-infiltrated with the PC4+CCFF+MGR modules. XIC and MS/MS spectra suggested these dihexosides were apigenin 7-O-rutionoside (12.2 nmol/g FW) and luteolin 7-O-rutinoside (33.4 nmol/g FW), respectively ( Supplementary Figure 1 ), and accounted for approximately 18% and 50% of the diosmetin glycoside contents ( Figure 6D ). Thus, F3′H and F4′OMT activities are not sufficient to convert most luteolin to diosmin in this system. However, the glucosyltransferase and rhamnosyltransferase activities appeared to completely cover the metabolic flow, since all detected flavones underwent rutinosylation. These results show that we successfully constructed a three-module expression system that can be used to produce diosmin in heterologous plants.