Integrated multi-omics analyses reveal that p-coumaroyl-CoA 2′-hydroxylases act upstream of stilbene synthases to mediate oxyresveratrol biosynthesis in mulberry (Morus alba)

Abstract

Oxyresveratrol is a bioactive stilbenoid with strong antioxidant, anti-inflammatory, and tyrosinase-inhibitory activities that accumulates in mulberry (Morus alba) tissues. Despite its importance, the biosynthetic origin of oxyresveratrol has remained unclear, with competing hypotheses proposing either hydroxylation of resveratrol or synthesis from a distinct precursor. Moreover, resveratrol and oxyresveratrol predominantly accumulate in non-renewable parts of mulberry trees, limiting their efficient extraction. To overcome these spatiotemporal constraints, we established cell suspension cultures from mulberry twigs and demonstrated that combined treatment with methyl jasmonate and methyl- or hydroxypropyl-β-cyclodextrins elicits high levels of both resveratrol and oxyresveratrol, which accumulate both intra- and extracellularly. Using this system, we investigated the molecular basis of oxyresveratrol biosynthesis in mulberry. We first improved the structural and functional annotation of the mulberry genome by integrating short- and long-read sequencing data derived from elicited cell suspension transcriptomes. By combining these resources with integrative transcriptomic, proteomic, and metabolomic analyses, we identified coordinated induction of multiple stilbene synthases (STSs) and a group of p-coumaroyl-CoA 2′-hydroxylases (C2′Hs) that were strongly co-expressed with resveratrol and oxyresveratrol accumulation. Functional validation in Nicotiana benthamiana, grapevine cell cultures, and in vitro enzyme assays demonstrated that C2′Hs catalyze the hydroxylation of p-coumaroyl-CoA upstream of the STS reaction, generating 2′,4′-dihydroxycinnamoyl-CoA as an alternative substrate for STSs. These findings demonstrate that oxyresveratrol is produced via a biosynthetic pathway parallel to resveratrol formation rather than through post-synthetic hydroxylation. In addition, we provide genomic and transcriptomic resources within the context of jasmonate-mediated elicitation, facilitating the discovery of novel phenylpropanoid structural and regulatory genes in the Morus genus. Together, our work establishes a new biosynthetic paradigm for stilbenoid diversification in plants and delivers molecular tools and resources for the biotechnological production of oxyresveratrol.

Oxyresveratrol is a high-value stilbenoid with notable health-promoting properties, yet its biosynthetic origin has long remained unresolved. By integrating multi-omics analyses with functional validation, this study demonstrates that p-coumaroyl-CoA 2′-hydroxylases act upstream of stilbene synthases to mediate oxyresveratrol biosynthesis in mulberry. This work establishes a new paradigm for stilbenoid diversification and provides a foundation for future biotechnological applications.

Introduction

Plants produce a myriad of specialized metabolites, among which phenylpropanoids represent one of the most widespread and abundant classes. These phenolic compounds have drawn increasing attention due to their multiple biological activities, including the prevention of cardiovascular and neurodegenerative disorders, cancer, and inflammatory processes, as well as their strong antioxidant effects (Xu et al., 2017).

Within the phenylpropanoid pathway (PPP), stilbenes comprise a small group of phenolic compounds produced by the activity of stilbene synthases (STSs), which evolved independently from chalcone synthases (CHSs) in at least 72 unrelated plant species (Rivière et al., 2012; Dubrovina and Kiselev, 2017). Stilbenes accumulate in response to biotic stresses (Chong et al., 2009) and play important defensive roles as phytoalexins against plant pathogens (Albert et al., 2011; Chalal et al., 2014; Blanco-Ulate et al., 2015). They are also produced in response to ultraviolet (UV) radiation (Adrian et al., 2000; Kiselev et al., 2017; Z. Liu et al., 2019) and mechanical wounding (Chitarrini et al., 2017) through hormone signaling pathways that mainly involve jasmonates (Billet et al., 2018). The synthesis of resveratrol by STSs represents the first committed step of the pathway. Resveratrol can be stored inside the cell after being glycosylated into piceid (Hall and De Luca, 2007), but it can also be transported out of the cell (Martínez-Márquez et al., 2023). Depending on the plant species, resveratrol can be further modified (i.e., methylated, acylated, hydroxylated, or halogenated) or oligomerized (Valletta et al., 2021), with each derivative exhibiting different biological activities in humans (Navarro-Orcajada et al., 2022).

Plants from the Moraceae family have long been used in traditional Chinese medicine due to their bioactive properties, which are mainly attributable to the presence of phenolic compounds, including stilbenes. Species of the Morus genus, including mulberry trees (Morus alba), are known to accumulate several stilbenoids, such as resveratrol, oxyresveratrol (oxyR; i.e., a hydroxylated form of resveratrol), and mulberroside A (mulA; an oxyR diglucoside), in different plant organs, especially in heartwood and root bark (Zhou et al., 2013). Owing to its considerable repertoire of pharmacological applications, M. alba is currently the most extensively studied species for oxyresveratrol production (Lim et al., 2015). OxyR exhibits strong tyrosinase-inhibitory activity (Kim et al., 2012), along with anti-inflammatory (Thaweesest et al., 2022) and antioxidant (Heo et al., 2016) effects. Its protective properties against neurological diseases have also been reported (Ban et al., 2006).

The pathway for resveratrol biosynthesis in mulberry can be inferred from current knowledge in grapevine (Vitis vinifera), a well-characterized high-stilbenoid producer with over 40 STS genes identified in its genome (Vannozzi et al., 2018). In V. vinifera, the enzymes responsible for resveratrol methylation and glycosylation have been fully characterized (Hall and De Luca, 2007; Schmidlin et al., 2008; Härtl et al., 2017). However, whereas the hydroxylation of resveratrol to form piceatannol has only been demonstrated using a human CYP450 enzyme (Martínez-Márquez et al., 2016), the native enzyme that catalyzes this at a radical other than in oxyR remains unknown. Although several STS genes have been isolated in Morus spp. (Li et al., 2016; Wang et al., 2017), whether oxyresveratrol is produced via direct hydroxylation of resveratrol, as proposed by Zhou et al. (2013), or through an alternative pathway, as suggested by Liu et al. (2022), remains unresolved. In addition, the pathway genes responsible for the modification (e.g., monooxygenases, methyltransferases, and glucosyltransferases) or cellular export of resveratrol and oxyresveratrol have not yet been identified in Morus spp. Beyond stilbenoids, other related compounds, such as benzofuran derivatives (i.e., moracins), also accumulate in mulberry tissues, yet their biosynthetic pathways have likewise not been elucidated.

Despite the high value of stilbenoid compounds in mulberry, their natural production in plant tissues is limited and highly variable. Chemical synthesis or extraction from raw materials, including non-renewable tissues, is technically complex, inefficient, and unsustainable. Cell suspensions and root cultures arise as biotechnological solutions and can be combined with chemical or physical elicitation and metabolic engineering approaches. Stilbenes have been previously elicited and engineered in cell cultures of pine (Kim et al., 2022), peanut (Vornam et al., 1988), pelargonium, and grapevine berries (Lijavetzky et al., 2008; Martinez-Esteso et al., 2011; Belchí-Navarro et al., 2012; Martínez-Márquez et al., 2016). Inyai et al. (2021) achieved high amounts of oxyR and mulA in M. alba hairy root cultures (HRs) using methyl jasmonate (MeJA) and yeast extract, while Fang et al. (2022) combined MeJA, methyl-β-cyclodextrins, and magnesium chloride as elicitors in HRs. To date, this combined elicitation has not been accomplished using fully suspended cell cultures of mulberry; instead, it has only been performed using plantlets (Sabater-Jara et al., 2023) or hydroxypropyl-β-cyclodextrin-treated calli (Komaikul et al., 2019).

In addition to plant cell and tissue culture, metabolic engineering of microorganisms represents a powerful means for industrial-scale production of oxyR, overcoming the limitations of plant-based production. This approach is gaining increasing interest in stilbene production (Jeandet et al., 2018; Shrestha et al., 2019). In this context, identification and functional characterization of the genes responsible for oxyR production become crucial. A first hypothesis is that oxyR synthesis involves hydroxylases, some of which correspond to cytochrome P450 (CYP450) monooxygenases. Whether these act before or after resveratrol is produced remains unknown. Hydroxylases have undergone extensive duplication events throughout the evolution of plant genomes (e.g., 174 CYP450 genes have been identified in Morus notabilis; Ma et al., 2014). Selecting genes for subsequent characterization through biochemical or biotechnological methods becomes extremely laborious and requires genome-wide approaches to narrow down the list of potentially relevant genes. Here, we established mulberry cell suspension cultures for elicitation experiments and generated time-course multi-omics datasets to identify and further characterize enzymes involved in oxyresveratrol biosynthesis. Our data point to the involvement of p-coumaroyl-CoA 2′-hydroxylase enzymes, members of the 2-oxoglutarate (2OG)-dependent dioxygenase family, acting upstream of STSs.

Results

MeJA and cyclodextrins elicit oxyresveratrol and resveratrol production in mulberry cell suspensions

Oxyresveratrol and piceatannol correspond to the meta- and ortho-hydroxylated derivatives of resveratrol, respectively. This high structural similarity makes their chromatographic separation difficult. To determine whether mulberry was able to produce both positional isomers, we first explored analytical methods that could discriminate between them. Following the MRM methodology developed by Hurtado-Gaitán et al. (2017), we identified a qualifier fragment for oxyresveratrol of [M + H⁺] = 161 (Figure 1A; Supplemental Figure 1A), which allowed differentiation between the two isomeric compounds.

Figure 1.

Combined treatment with MeJA and MBCD promotes the accumulation of resveratrol and oxyresveratrol in mulberry cell suspensions.

(A and B) (A) Mass-spectrometry–based distinction of mono-hydroxylated derivatives of resveratrol (i.e., piceatannol and oxyresveratrol). Mirror plot showing fragmentation mass spectra of oxyresveratrol (top) and piceatannol (bottom) at a collision energy (CE) of 20eV. m/z 107 and m/z 161 correspond to the quantifier and qualifier ion fragments, respectively. P denotes the precursor ion at m/z 245. (B) Bright-field image of untreated cells grown in liquid medium. Scale bar, 50 μm.

(C) Blue fluorescence emitted after UV-C irradiation of control and elicited cell suspensions (top left) and time-series accumulation of extra- and intracellular stilbene aglycones (and intracellular glycosides) in response to methyl jasmonate and cyclodextrin elicitation. Intracellular concentrations are given in μg/g fresh weight (FW), whereas extracellular accumulation is measured in mg/l. Data represent mean ± SD.

Cell suspensions of M. alba (Figure 1B) were initiated from undifferentiated calli established from twigs of specimens belonging to white- and pigmented-fruit cultivars (Supplemental Figure 1B). These cultures were treated with MeJA and methyl-β- (MBCD) or 2-hydroxypropyl-β- (HPBCD) cyclodextrins. Stilbenes emit violet-blue fluorescence under UV light irradiation (Poutaraud et al., 2007). As expected, elicited mulberry cell cultures exhibited this fluorescence when exposed to UV light (312 nm; Figure 1C). We observed an increase in the accumulation of resveratrol and oxyresveratrol at 4–5 days post-elicitation (Supplemental Figures 1C and 1D; Supplemental Tables 1 and 2), whereas piceatannol was not detected. Treatment with MeJA alone had a negligible effect on the production of these compounds, whereas treatment with MBCD alone showed a moderate effect. Compared with the single treatments, the combined application of MBCD and MeJA led to a substantial increase in the synthesis of both resveratrol and oxyresveratrol. Suspensions derived from the pigmented cultivar produced slightly less resveratrol compared with white cultivar–derived cells; however, a greater accumulation of oxyresveratrol was observed (Supplemental Figure 1D). Therefore, cell suspensions generated from the pigmented-fruit cultivar were chosen for further analysis.

A time-course experiment was performed using a combined treatment of elicitors (MBCD–MeJA), with sampling at 0.5, 6, 12, 24, 48, and 72 h. Both intra- and extracellular production of oxyresveratrol, which was considerably higher than that of resveratrol, increased steadily until reaching a maximum at 48 h, after which it remained constant (Figure 1C; Supplemental Table 3). In contrast, extracellular resveratrol increased at a constant rate and did not appear to reach a plateau at any of these time points. Quantification of stilbene glycosides revealed that mulberroside A decreased in response to elicitation, particularly at later time points. Accumulation of piceid was minimal, although slightly higher levels were detected in elicited cells at most time points.

Reannotation of the mulberry genome reveals genes involved in stilbenoid biosynthesis

Despite the well-documented accumulation of resveratrol and oxyresveratrol in mulberry, the current genome annotation (20,386 genes; Jiao et al., 2020) fails to include any STS genes. This annotation therefore appears incomplete, supporting the need for a more accurate annotation to enable functional characterization of genes involved in elicited pathways. We followed a reannotation workflow to define a more complete set of protein-coding genes using an automated pipeline that integrated protein homology, short- and long-read transcript evidence, and ab initio predictions. A total of 85,569 transcripts were obtained by combining the official and updated annotations. After removing redundancy, 31,401 gene models remained. Overall gene space completeness was improved, with BUSCO analysis showing that 95.7% of universal single-copy genes from the eudicot clade were recovered, compared with 92.1% in the official annotation (Figure 2A). Although the total number of genes increased substantially, the proportion of genes present in more than one copy decreased from 24.78% to 5.35%.

Figure 2.

Improved annotation of the M. alba genome reveals two STS gene clusters located on chromosome 8.

(A) BUSCO completeness analysis comparing the official (Jiao et al., 2020) and updated structural annotations based on the eudicots dataset. Colors represent different BUSCO assessment categories. n = total number of gene models in the BUSCO dataset; N = total number of gene models in the annotation.

(B) Comparison of the number of enzymes belonging to the shikimate (ShP), early phenylpropanoid (EPP), and stilbenoid (StP) pathways in V. vinifera and M. alba. Chalcone synthases (CHSs) from the flavonoid pathway (FP) are also included.

(C) Dot plot of chromosome 8 and microsynteny analysis of STS clusters showing segmental duplications forming paralogous gene sets (top). Genomic organization of STS genes annotated in the M. alba reference genome (bottom panel). Some genes are classified as pseudogenes due to the presence of premature stop codons in their coding sequences (CDSs).

(D) Maximum-likelihood phylogenetic tree of the polyketide synthase III (PKSIII) family, including STS, CHS, PKSIII-like, and PKSIIIa/b proteins from V. vinifera and M. alba. The tree supports independent evolutionary origins of stilbenoid synthases in the two species. Evolutionary distances are shown as amino acid substitutions per site. The tree was constructed using 1,000 bootstrap replicates (values are shown in the uncondensed tree, Supplemental Figure 4).

The structural annotation was followed by a gene function prediction pipeline based on the MapMan ontology. This allowed the identification of gene families belonging to the shikimate pathway and early PPPs, with copy numbers broadly comparable to those in grapevine (Figure 2B). We observed a few STS gene models incorrectly annotated as CHSs, whereas others were entirely missing from the official annotation (Supplemental Figure 2). STS genes were located on chromosome 8 in two distinct genomic clusters (Figure 2C). Evidence for such a large duplication event was unique to this chromosome. In addition, the two regions contained similar numbers of STS genes, suggesting that one cluster arose through a segmental duplication of the other (Supplemental Figure 3). A total of 22 STS genes were identified on this chromosome, with 12 in the first cluster and 10 in the second. Approximately half of these genes appear to be pseudogenes, based solely on the presence of premature stop codons or the absence of start or stop codons, independent of expression patterns.

Several genes were manually curated using the Apollo interface, guided by both the short- and long-read sequencing data generated in this study. For this purpose, we also incorporated publicly available Illumina-sequenced datasets from fruit and leaf tissues. Phylogenetic reconstruction of the polyketide synthase (PKS) family, using the manually curated mulberry gene models together with grapevine sequences, confirmed that several genes previously annotated as CHSs in M. alba clustered separately from V. vinifera CHSs. STS, CHS, PKSIIIA/PKSIIIB, and PKSIII-like proteins formed distinct clades, with CHS sequences from all species grouping together, whereas STS sequences clustered separately by species (Figure 2D). This pattern supports the hypothesis of convergent evolution of STS genes, as proposed by Tropf et al. (1994). PKSIIIA and PKSIIIB proteins also segregated by species; these proteins correspond to anther-specific enzymes involved in the biosynthesis of phenolic components of sporopollenin and are conserved across land plants (Dobritsa et al., 2010). Notably, no PKSIII-like proteins were detected in M. alba.

Multi-omics integration identifies and validates C2′Hs involved in oxyR synthesis

The gene expression profiles of M. alba cell cultures treated with MBCD–MeJA were analyzed using weighted gene co-expression network analysis (WGCNA), which clusters genes into modules based on expression similarity. These modules were further correlated with metabolite abundances. WGCNA resulted in a scale-free topology network comprising a total of 44 modules, with most STSs assigned to module ME39, followed by modules ME7 and ME11 (Figure 3A).

Figure 3.

Co-expression analysis and metabolomic/transcriptomic integration to identify potential oxyresveratrol biosynthetic genes.

(A) Time-related clustering of the elicited transcriptome of M. alba by weighted gene co-expression network analysis (WGCNA) (far left). Correlation of each module eigengene (ME) with the time-course accumulation of resveratrol and oxyresveratrol (far right). Values inside each colored box represent the correlation coefficients and p values (in parentheses) obtained between each module and the respective metabolite. Bar plots depict the total number of CHS and STS genes per module.

(B) Individual gene expression profiles for selected modules showing activated transcript profiles in response to elicitation. FPKM data are shown scaled (Z score). Dark-colored trajectories within each module represent the mean gene expression profiles for each condition. Red and blue lines correspond to elicited and control samples, respectively.

The selected modules shared similar expression patterns, characterized by up-regulation in response to elicitation (Figure 3B). Notably, several CHSs were also clustered within these modules. Module ME11, which contains multiple STS and CHS genes, showed a strong and significant correlation with the accumulation of resveratrol and oxyresveratrol. Module ME39 showed a weaker, yet still significant, correlation. In contrast, module ME38, despite lacking STS genes, exhibited the strongest correlation with both resveratrol and oxyresveratrol.

Functional enrichment analysis of these four modules revealed several additional significantly enriched MapMan categories comprising genes distinct from those encoding CHS/STS activities (Supplemental Table 4). This evidence suggests their involvement in the stilbenoid pathway according to the guilty-by-association principle. Among these categories, the term feruloyl-CoA 6′-hydroxylase (F6′H) was overrepresented in modules ME39 and ME7, which also contained STSs (Figure 4A). As observed for MalSTSs, several mulberry genes from this group were also revealed through genome re-annotation (Supplemental Figure 5). By contrast, the CYP450 hydroxylase category was enriched in modules ME38 and ME11.

Figure 4.

C2′H enzymes, initially identified as F6′Hs, are involved in oxyresveratrol production and respond to MeJA+CD elicitation.

(A) Selection of significantly enriched MapMan terms in selected WGCNA modules. The number of genes intersecting with each ontology term is represented by dot size.

(B)STS and C2′H gene expression profiles compared with resveratrol and oxyresveratrol abundances.

(C) Spearman’s rank correlation coefficient (ρ) values of mRNA–protein pairs according to their stage-specific differential expression (DE) and differential abundance (DA): DEG–DAPs, DEG–NVPs, NVG–DAPs, and NVG–NVPs (NV, non-variant). Red lines represent the median correlation for each subset. Dashed lines indicate the thresholds used to define positive and negative correlations. Red dots below the x-axis refer to proteins belonging to the phenylpropanoid pathway identified in Figure 2B.

(D) Z-scored abundances of stage-variant mRNA/protein pairs clustered by WGCNA.

(E) Abundances and correlations of C2′H mRNA/protein pairs putatively involved in oxyresveratrol synthesis (additional related pairs are shown in Supplemental Figure 9).

(F) LC–MS chromatograms obtained from methanol extracts of agroinfiltrated N. benthamiana leaves. Leaves were co-infiltrated with Agrobacterium tumefaciens harboring P19, STS, or C2′H genes, in different combinations. Resveratrol and oxyresveratrol infiltrations are also shown, together with Tris buffer, which was used as a negative control. Ion current data were extracted for each ion transition (channels) of resveratrol, oxyresveratrol, and their glycosylated forms in MRM-positive mode. The standard mix includes resveratrol, oxyR, piceid, and mulberroside A. Each chromatogram is representative of three to six independent infiltrations performed in different plants (see Supplemental Figure 10). Selected peaks: 1, mulberroside A; 2, trans-piceid; 2′, trans-piceid in-source fragmentation (SF), 2'', 2''', resveratrol glycosides (SF); 3, oxyresveratrol; 4, resveratrol; 5', cis-piceid (SF); 6′ and 6'', oxyresveratrol glycosides (SF).

(G) MRM analysis of oxyresveratrol in intracellular extracts of grapevine cell suspensions transiently transformed with HcPro, STS, and/or C2′H, at 6 days after transfection. Left: extracted ion chromatograms for oxyresveratrol transitions (245 → 107; additional details in Supplemental Figure 11). Center: composite MS/MS spectrum of the recorded transitions, with the precursor ion m/z indicated by a diamond. Right: normalized chromatographic traces of the monitored oxyresveratrol transitions.

To further investigate the putative F6′H gene family that was highly co-expressed with STS genes, we conducted a phylogenetic analysis including F6′Hs from A. thaliana and V. vinifera, as well as a group of enzymes from sweet potato (Ipomoea batatas), designated Ib2, which are derived from F6′H genes and catalyze the ortho-hydroxylation of p-coumaroyl-CoA (C2′H activity) (Matsumoto et al., 2012). Inclusion of C3′H and C4′H proteins from A. thaliana, V. vinifera, and their homologs identified in M. alba revealed two distinct clades (Supplemental Figure 6). One clade clustered F6′H proteins from Arabidopsis and grapevine together with a single mulberry sequence, whereas the second clade contained the remaining six mulberry F6′H sequences that were co-expressed with STSs.

Umbelliferone (UMB; 7-hydroxycoumarin) is one of the end products of C2′H activity in sweet potato (Matsumoto et al., 2012), produced through the reduction of hydroxylated p-coumaroyl-CoA. By contrast, F6′Hs are only able to produce scopoletin following the hydroxylation of feruloyl-CoA. If any of the elicited MalF6′H-like genes encode intrinsic C2′H enzymatic activity, UMB should be detectable in elicited cells. We indeed observed a unique accumulation of UMB upon elicitation, as evidenced by gas chromatography–mass spectrometry (GC–MS) analysis of the extracellular medium (Supplemental Figure 7). Mulberry sequences within the distinct F6′H clade were therefore hypothesized to act as C2′Hs rather than bona fide F6′Hs, despite retaining the conserved 2OG- and Fe(II)-dependent oxygenase superfamily domains. Specific sequence differences in MalC2′H proteins may be responsible for a substrate switch from feruloyl-CoA to p-coumaroyl-CoA, as previously reported in sweet potato by Sun et al. (2015). Moreover, we hypothesized that the production of hydroxylated p-coumaroyl-CoA under elicited conditions provides a substrate for STSs to catalyze oxyresveratrol formation.

The expression of STS and putative C2′H genes peaked at 12 h post-elicitation and correlated with a general activation of the shikimate pathway and early PPPs (Supplemental Figure 8). Consistent with the modest correlation between oxyresveratrol and resveratrol accumulation and STS-containing modules, we observed a temporal delay in metabolite accumulation relative to gene expression. Considering the full trajectory of gene expression and metabolite accumulation, STS and C2′H transcript levels began to decline at 48 h, despite continued stilbene accumulation (Figure 4B). We further investigated the relationship between protein abundance and mRNA expression in elicited cells to determine whether STS and C2′H mRNAs continued to be translated after 48 h, despite decreased transcript levels. We analyzed the three latest time points from control and elicited samples using a label-free quantitative proteomics approach to capture proteome-level changes. In total, 1,891 protein groups corresponding to 3,158 unique proteins were identified, of which 712 were considered differentially abundant proteins (DAPs) at at least one elicitation time point compared with the control. Among 2,793 matched mRNA–protein pairs, more than 60% showed either positive or negative correlations between transcript and protein levels (Figure 4C). Specifically, a total of 823 mRNA–protein pairs showed positive correlations (Spearman’s rank coefficient ρ ≥ 0.4), with 134 being significant (ρ ≥ 0.8 and p < 0.05), whereas 910 pairs showed negative correlations (ρ ≤ −0.4), with 106 being significant (ρ ≤ −0.8, p < 0.05). Each mRNA–protein pair was classified according to differential expression and abundance, resulting in 246 DEG–DAP (differentially expressed gene–differentially abundant protein) pairs, 522 DEG–NVP (non-variant protein) pairs, 376 NVG (non-variant genes)–DAP pairs, and 1,649 NVG–NVP pairs. DEG–DAP pairs showed the highest correlation (median ρ = 0.7), indicating a strong positive relationship between protein abundance and mRNA expression. A second WGCNA of mRNA–protein pairs revealed that the most abundant modules corresponded to DEG–DAP pairs induced upon elicitation (Figure 4D). Among these, STS and C2′H proteins showed very high correlations with their corresponding mRNAs in all cases (Figure 4E), particularly when comparing control and elicited conditions. However, while STS and C2′H mRNA levels tended to decrease toward the latest time point, protein abundances were maintained or showed a slight increase. This observation also suggests a stronger correlation between these proteins and oxyresveratrol and resveratrol abundances than between metabolite levels and mRNA abundances.

To validate whether the putative C2′Hs are involved in oxyresveratrol biosynthesis, we transiently expressed C2′H1, C2′H4, and C2′H5, in combination with a previously characterized STS (Santos-Rosa et al., 2008), in Nicotiana benthamiana (N. benthamiana) leaves. The individual and combined effects of C2′H and STS agroinfiltration were assessed by liquid chromatography–mass spectrometry (LC–MS) at 3 and 6 days post-agroinfiltration to quantify stilbenoids. Several peaks were detected in the corresponding ion transition channels of oxyresveratrol and resveratrol, but with retention times matching those of their glucosides (i.e., mulberroside A and piceid, respectively). This strongly suggests that non-specific mono-glycosylated forms of oxyresveratrol are produced in all STS+C2′H combinations, similar to direct infiltration of oxyresveratrol into leaves (Figure 4F), all of which undergo partial in-source fragmentation. In a comparable manner, infiltration with the STS construct alone or STS+C2′H constructs generated peaks in the resveratrol channel matching the retention time of piceid. Neither infiltration of C2′H alone nor C2′H plus resveratrol produced any oxyresveratrol glycosylated forms, suggesting that C2′Hs require STS activity but do not hydroxylate resveratrol to form oxyresveratrol.

To directly demonstrate the production of the oxyresveratrol aglycone through the combined activity of C2′Hs and STSs, we transiently expressed these genes in cell suspensions of V. vinifera cv. “Gamay” (Figure 4G), which typically accumulate aglycones upon elicitation (Orduña et al., 2022). Oxyresveratrol was detected in intracellular extracts of cells transiently transformed with STS and C2′H at 6 days after transformation (Supplemental Figure 12).

To further confirm the catalytic activity of the putative C2′Hs as bona fide p-coumaroyl-CoA 2′-hydroxylases, we expressed and purified C2′H5 from Escherichia coli and assessed its enzymatic activity using p-coumarate as a substrate (Figure 5). Fluorescence-based detection of UMB (excitation: 325 nm; emission: 455 nm) revealed a marked increase in product accumulation over time in reactions containing C2′H5, reaching up to 0.5 μM within 6 h, whereas no detectable product was observed in the enzyme-free control (Figure 5B; Supplemental Figure 13A). High-performance liquid chromatography (HPLC)–fluorescence analysis showed clear accumulation of UMB at both 1 and 6 h post-reaction (Figure 5C). HPLC–UV analysis of samples collected at 1 h further validated the enzymatic conversion of p-coumarate to UMB (Figure 5D; Supplemental Figure 13B), and LC–MS confirmed the presence of 2,4-dihydroxycinnamate as an intermediate, consistent with spectra reported by Yin et al. (2017) (Supplemental Figure 13C). Kinetic analysis using varying concentrations of p-coumaroyl-CoA yielded a K_m of 16.5 μM and a V_max of 5.8 nM UMB/min (Supplemental Figure 14). This K_m value falls within the range reported for C2′Hs from I. batatas (Matsumoto et al., 2012). Altogether, these results demonstrate that p-coumaroyl-CoA 2′-hydroxylases catalyze the formation of 2,4-dihydroxycinnamate and are essential for oxyresveratrol biosynthesis in M. alba.

Figure 5.

Purification and enzymatic activity assay of C2′H5.

(A) Expression and purification of the C2′H5 protein in Escherichia coli BL21 (DE3) following IPTG induction. The purified protein (∼43 kDa) is detected in the eluted fractions (E1–E5). The identity of the C2′H5 protein observed on the SDS–PAGE gel was confirmed by LC–MS analysis of the excised protein band (Supplemental Table 5).

(B and C) (B) Fluorescence-based detection of umbelliferone accumulation over time (excitation, 325 nm; emission, 455 nm). Reactions containing active C2′H5 (blue) show a progressive increase in fluorescence, whereas the enzyme-free control (red) shows no detectable change. UMB, umbelliferone standard (black).

(C) HPLC–fluorescence chromatograms showing umbelliferone accumulation in samples collected 1 and 6 h after reaction initiation.

(D) HPLC–UV (324 nm) chromatograms confirming the conversion of p-coumarate to umbelliferone (peak 1), with detection of 2,4-dihydroxycinnamate (peak 3) as a reaction intermediate. Enlarged views of the chromatograms are shown in Supplemental Figure 13B.

Discussion

Cell suspension cultures are a promising approach for obtaining valuable plant natural compounds. This approach is particularly relevant for molecules whose utility is limited by low yields and unsustainable plant extraction methods, or by the reliance on chemical synthesis, which often produces non-specific mixtures, generates toxic waste, and entails high production costs.

In this study, we show that mulberry cell cultures produce large amounts of both resveratrol and oxyresveratrol when elicited with MeJA and cyclodextrins. The amounts generated appear appealing for potential applications in the pharmaceutical and cosmetic industries. These production rates resemble those of hairy roots elicited with the same types of compounds (Fang et al., 2022). Nonetheless, as shown by comparisons between pigmented and white-fruited cultivars, stilbenoid production varies depending on the genetic background of the starting material used to establish the cell suspensions. A synergistic action of jasmonates and cyclodextrins in enhancing stilbenoid production has previously been reported in grape cell suspensions (Almagro et al., 2014; Halder et al., 2019). The constitutive secretion of these compounds into the culture medium may occur through passive or active transport, via transporters such as those identified in grape (Martínez-Márquez et al., 2023), or through the intrinsic capacity of cyclodextrins to act as carriers (Morales et al., 1998). This secretion facilitates compound extraction and purification without affecting the cells in suspension.

We further demonstrate that elicitation boosts the expression of enzyme-coding genes involved in both primary (i.e., the shikimate pathway) and specialized (early phenylpropanoid and stilbenoid synthesis) metabolic pathways. However, the absence of stilbenoid pathway genes in the official M. alba genome annotation (Jiao et al., 2020) prompted us to integrate transcriptomic resources to revise this annotation, particularly to confirm the presence of STS genes. De novo transcriptomes assembled using long-read sequencing data also allowed us to verify the accuracy of the newly annotated gene models. The revised gene set was more complete, especially with respect to specialized metabolism and its transcriptional responses to MeJA.

Two distinct STS clusters were identified on chromosome 8, likely originating from different duplication events. The observed sequence similarity patterns suggest that an initial cluster of STS genes arose through tandem duplication, which was subsequently followed by a segmental duplication of the entire cluster. Such expansion may reflect positive selection pressure, indicating the importance of stilbene accumulation for the fitness and/or survival of M. alba. Resveratrol and its derivatives (e.g., methylated, glycosylated, and acylated forms) are well-characterized phytoalexins capable of protecting plants against reactive oxygen species and pathogens, such as Botrytis cinerea (Adrian et al., 1997). In addition, oxyresveratrol has been shown to be a potent free radical scavenger, surpassing the activity of resveratrol (Lorenz et al., 2003).

CHSs and STSs compete for the common substrates p-coumaroyl-CoA and malonyl-CoA, because both enzymes are type III PKSs and STS activity evolved from a CHS ancestor. Our phylogenetic analysis grouped all CHS sequences together. However, STSs from grapevine and mulberry clustered independently, as expected given the convergent evolution of STSs. STSs have emerged in more than 70 unrelated plant species. Future studies can focus on characterizing the entire STS family in mulberry to confirm their activity and to understand the conformational effects of different mutations in CHS that lead to STS activity in plants. Our transcriptomic analysis showed that both STS and CHS genes were induced upon MeJA-mediated elicitation, suggesting that both stilbene and flavonoid pathways are induced. Other closely related pathways, such as the benzofuran family of compounds (e.g., moracin), are also induced (Fang et al., 2022). Our data and resources can guide gene discovery approaches to identify the genes responsible for these additional pathways.

It has generally been accepted that resveratrol is the first committed compound of the stilbenoid pathway. Under this paradigm, resveratrol derivatives are thought to arise subsequently through the action of resveratrol-modifying enzymes. This model has been demonstrated for resveratrol glycosylation, prenylation, acylation, and methylation (reviewed by Valletta et al., 2021), and it has also been proposed for the hydroxylation of resveratrol into piceatannol (Martínez-Márquez et al., 2016). In the case of oxyresveratrol, it has similarly been hypothesized to originate from the oxidation of resveratrol (Zhou et al., 2013); however, our results demonstrate that this model is incorrect.

Coumarins are formed through 2'/6′-hydroxylation of the p-coumaroyl phenyl ring, followed by trans/cis isomerization of the side chain and subsequent lactonization. The first step in the synthesis of the 2H-1-benzopyran-2-one structural core of coumarins has been demonstrated to be enzymatically catalyzed, whereas the downstream steps may occur spontaneously under light conditions or via enzyme catalysis (Stringlis et al., 2019; Vanholme et al., 2019). In A. thaliana, feruloyl-CoA 6′-hydroxylase (AtF6′H) catalyzes the committed step in coumarin scopoletin biosynthesis and displays high substrate specificity (Kai et al., 2008). By contrast, a p-coumaroyl-CoA 2′-hydroxylase from Ruta graveolens (RgC2′H) catalyzes the committed step in UMB biosynthesis (Vialart et al., 2012). Interestingly, RgC2′H is also capable of hydroxylating feruloyl-CoA at the equivalent position, similar to F6′Hs, thereby promoting the synthesis of scopoletin. The induction of C2′H in elicited M. alba cells is thus consistent with the observed production of UMB in response to MeJA and cyclodextrins. Scopoletin, however, was not detected, suggesting either that feruloyl-CoA is absent in M. alba cells or that MaC2′H displays high specificity toward p-coumaroyl-CoA. In agreement with this interpretation, we demonstrate UMB production through enzymatic assays using recombinant Morus C2′H, which exhibits kinetic properties similar to those of C2′Hs from Ipomoea batatas. The reaction begins rapidly, within the first hour of incubation. The immediate product, 2′,4′-dihydroxycinnamate, is detected early on and precedes its spontaneous conversion to UMB. Unexpectedly, the use of the C2′H product 2′,4′-dihydroxycinnamoyl-CoA by STS provides a plausible pathway for the production of oxyresveratrol in M. alba.

Our findings suggest that C2′H enzymes synthesize 2′,4′-dihydroxycinnamoyl-CoA from p-coumaroyl-CoA, which in turn serves as a substrate for STSs to produce oxyresveratrol (Figure 6). Under this model, STS enzymes can accept p-coumaroyl-CoA substrates with different hydroxylation states. We further propose that the enzymatic activity of C2′Hs, in terms of velocity and overall performance, must be high, as oxyresveratrol production rates exceed those of resveratrol under elicitation. The spontaneous production of UMB is likely slow and therefore has a minor effect on oxyresveratrol levels.

Figure 6.

Oxyresveratrol biosynthetic pathway revealed by integrative multi-omics and functional analyses.

Schematic representation of the revised stilbenoid biosynthetic pathway in M. alba. Integrative multi-omics analyses (genomics, transcriptomics, proteomics, and metabolomics) identified a group of p-coumaroyl-CoA 2′-hydroxylases (C2′Hs) that are transcriptionally and translationally co-induced with stilbene synthases (STSs) upon elicitation. C2′Hs catalyze the 2′-hydroxylation of p-coumaroyl-CoA to generate 2′,4′-dihydroxycinnamoyl-CoA, which is subsequently utilized by STSs to synthesize oxyresveratrol. In parallel, STSs can also act on non-hydroxylated p-coumaroyl-CoA to produce resveratrol. This model refutes previous hypotheses proposing that stilbenoid derivatives arise exclusively from post-synthetic modification of resveratrol and demonstrates that oxyresveratrol biosynthesis proceeds through an upstream, dedicated branch of the phenylpropanoid pathway. Umbelliferone formation from the hydroxylated intermediate is also indicated. CHS, chalcone synthase; STS, stilbene synthase; C2′H, p-coumaroyl-CoA 2′-hydroxylase. Blue segments represent the portion of the molecule derived from the malonyl-CoA backbone, whereas the red segment highlights the additional hydroxyl group relative to resveratrol.

We provide experimental evidence from heterologous expression systems demonstrating that C2′H expression is required for oxyresveratrol production. The well-established observation that the stilbene core can be synthesized from substrates other than p-coumaroyl-CoA, such as cinnamoyl-CoA or caffeoyl-CoA (Raiber et al., 1995), together with the presence of STSs as multigene families (Vannozzi et al., 2012; Figure 2D), which may confer differential substrate specificity, supports a biosynthetic pathway for oxyresveratrol that operates in parallel with that of resveratrol, rather than a linear pathway involving resveratrol hydroxylation in M. alba.

We cannot infer that the production of other hydroxylated stilbenes, such as piceatannol, occurs in the same manner as oxyresveratrol. Indeed, grapevine does not appear to possess C2′H activity, and Martínez-Márquez et al. (2016) showed that a human CYP450 enzyme was capable of producing piceatannol when agroinfiltrated together with STS genes in N. benthamiana. Nevertheless, our work contributes to a new paradigm in which STS enzymes exhibit substrate promiscuity. Consistent with this, STSs from pine (Pinus spp.) can use cinnamoyl-CoA to produce pinosylvin (Schanz et al., 1992).

The expression and abundance patterns of STS and C2′H genes and proteins correlate well with the accumulation of resveratrol and oxyresveratrol, suggesting that these genes are co-regulated and act in a coordinated manner to produce these compounds. The tight co-expression of these genes further suggests that they may be controlled by the same transcriptional regulators. In grapevine, subgroup 2 R2R3-MYBs are able to control the expression of almost the entire STS gene family while also directly regulating genes involved in the shikimate pathway and early PPP (Orduña et al., 2022).

Our open-access resources serve as repositories of novel genes for studying their relationship to stilbene production in mulberry. We have centralized all the data generated in this work. Specifically, the updated gene annotation and transcriptomic data can be downloaded or explored from MorusalbaDB (http://plantaeviz.tomsbiolab.com:8000/easy_gdb/) and the Mulberry Visualization Platform (MulViz; https://plantaeviz.tomsbiolab.com/mulviz/morus_meja_atlas/), respectively. These are the first open-access databases created for this species. In these resources, users can browse the genome, explore functional and structural information for annotated genes, examine gene models, and visualize co-expression data. Despite these advances, some gene families may still require manual curation; to facilitate this, we have also created an Apollo-based curation interface that allows the community to access and contribute to high-quality mulberry genome annotation.

Methods

Initiation, growth conditions, and elicitation of Morus cell suspensions

M. alba undifferentiated cell growth in the form of calli was achieved using young stems (twigs) cut into 2 cm–long pieces as explants by Mr. Juan J. Montesinos. These were collected in 2012 from two specimens located on the campus of the University of Alicante (Spain), one with intensely pigmented fruits and another with non-pigmented fruits, from which we obtained lines named “red” and “white,” respectively (Supplemental Figure 1B). Both were assessed to be M. alba based on ITS (internal transcribed spacer) length and ploidy, according to Zeng et al. (2015). After thorough surface sterilization (sequential rinses in sterile water 3 times, once in 70% ethanol for 30 s, sterile water 3 times, once in 1% sodium hypochlorite for 3 min, and final rinses in sterile water), explants were wiped with sterile filter paper, cut longitudinally into two halves, and placed with the cutting surface onto solid medium with different hormonal balances. The medium composition per liter was 4.3 g Murashige and Skoog basal medium, 30 g sucrose, 0.25 g casein hydrolysate, 1 mg calcium pantothenate, 100 mg myo-inositol, 0.01 mg biotin, 1 mg pyridoxine, 1 mg thiamine, 1 mg nicotinic acid, 7.5 mg FeSO₄.7H₂O, and 5.6 mg Na₂EDTA.2 H₂O, adjusted to pH 5.8, with 8 g agar added for solidification. The hormonal composition that supported callus development was the Murashige and Skoog3 variant for the white line, containing 0.2 mg naphthalene acetic acid and 1 mg kinetin, and Murashige and Skoog6 for the red line, containing 1 mg naphthalene acetic acid and 0.2 mg kinetin. The obtained calli were maintained on their respective solid Murashige and Skoog media at room temperature in darkness and subcultured every 30 days. Cell suspensions were initiated by placing 1.25 g of calli into 100 ml flasks containing 25 ml of either liquid Murashige and Skoog3 or Murashige and Skoog6 medium prepared without agar and grown in darkness at 25°C under orbital shaking at 110 rpm. The culture volume was gradually scaled up to 1 l flasks containing 400 ml of suspension and subsequently maintained by subculturing every 14 days by doubling the suspension volume, adding fresh liquid medium, and dividing it between two new sterile flasks.

Elicitation treatments with MBCD and MeJA were carried out in quadruplicate as previously described for grapevine cells (Lijavetzky et al., 2008), with slight modifications. Fresh cells were prepared in an aseptic environment by filtration of a 14-day-old M. alba cell suspension through a sterile glass filter (90–150 μm pore size), followed by washing with sterile water under gentle vacuum. A weighed amount of cells was transferred into shaking flasks and suspended in fresh growth medium (4 g/g cell FW) supplemented with elicitors (50 mM MBCD, 100 μM MeJA, or 50 mM MBCD + 100 μM MeJA) and maintained at 25°C on a continuous rotary shaker (110 rpm) in the dark for up to 120 h. MBCD was added to the medium before autoclaving. MeJA was prepared as a stock solution of 47.5 mM in ethanol and filter-sterilized through a 0.2 μm PTFE (polytetrafluoroethylene) filter before being added to the sterile medium. Each replicate consisted of 8 g of cells and 32 g of elicitor-supplemented MS3 or MS6 liquid medium in 100 ml flasks. Control and MBCD treatments received the same volume of ethanol as the MeJA and MBCD–MeJA treatments (2.1 μL/mL cell suspension). Each flask was maintained in darkness. The total number of samples was 24, corresponding to a single time point (5 days after elicitation) and three biological replicates per condition (control, MBCD, MeJA, and MBCD–MeJA). After incubation, each sample was individually filtered to separate the cells from the medium.

Four 14-day-old cell suspension cultures grown in 1 l flasks (containing approximately 500 ml of cell suspension) were pooled and filtered using a sterile glass filter (90–150 μm pore size) and a vacuum pump to separate the cells from the growth medium. Each replicate consisted of 8 g of cells and 32 g of liquid medium in 100 ml flasks. Control cells were grown in MS6 (pigmented-fruited cultivar) or MS3 (white-fruited cultivar) medium, whereas elicited cells were grown in MS6 or MS3 medium supplemented with MBCD at a final concentration of 50 mM prior to autoclaving. Ethanol-diluted MeJA was added after medium sterilization. MeJA was filter-sterilized using a 0.2 μm PTFE filter and added to the MS3/MS6 + MBCD medium to a final concentration of 100 μM (84 μl of a 47.5 mM MeJA stock). Each control flask received 84 μl of 100% ethanol. Cells were maintained under orbital agitation at 110 rpm and 24°C. Each flask was protected from light using aluminum foil.

The total number of samples in the time-series experiment was 48, corresponding to six time points (30 min, 6, 12, 24, 48, and 72 h after elicitation) and four biological replicates per condition (control and MBCD–MeJA). After incubation, each sample was individually filtered to separate the cells from the medium. The medium was collected, and the cells were rinsed with cold sterile water to remove residual medium and filtered again. The weight of both cells and medium was recorded, and samples were rapidly frozen in liquid nitrogen and stored at −80°C. Cell samples were lyophilized for 48 h and then stored at −80°C until metabolite, protein, or RNA extraction. Media samples were stored at −20°C for subsequent metabolite analysis.

Extraction and LC–MS quantification of stilbenoids

For the extraction of fresh or frozen material, 2 g of cells were mixed with 4.6 ml of HPLC-grade methanol, vortexed, and shaken at 1800 rpm for 75 min. The remaining debris was filtered using a Pasteur pipette packed with cotton wool. For lyophilized material, 10 mg of dry cells was extracted with 1 ml of 80% methanol and incubated at 220 rpm and 4°C overnight. Extracts were centrifuged at 14,000 × g for 10 min, and the supernatant was collected for LC–MS analysis (Hurtado-Gaitán et al., 2017). For extracellular stilbenoids, a 1 ml aliquot of filtered cell-free medium was extracted with 250 μl of HPLC-grade ethyl acetate, vortexed, agitated at 1800 rpm for 2 min, and centrifuged at maximum speed for 5 min. The organic phase was collected into a separate tube, and a second round of ethyl acetate extraction was performed. The pooled organic phase was dried using a SpeedVac at room temperature. The dried residue was resuspended in 500 μl of 80% MS-grade methanol, centrifuged at 14,000 × g for 10 min, and the supernatant was collected for LC–MS analysis. LC–MS analyses were performed on an Agilent 6490 Triple Quadrupole mass spectrometer coupled to an Agilent 1290 Infinity UHPLC (ultra high performance liquid chromatography) system, as described in Hurtado-Gaitán et al. (2017). In addition to previously established MRM methods for the quantitative analysis of major V. vinifera stilbenoids (pterostilbene, resveratrol, piceatannol, viniferin, and piceid), we developed new MRM methods for two additional stilbenes found in Morus species, namely oxyresveratrol and mulberroside A, using authentic analytical standards. Piceatannol and oxyresveratrol, being regioisomers, could not be reliably distinguished based solely on retention times. However, their fragmentation spectra revealed a diagnostic fragment at m/z = 161 for oxyresveratrol—representing the second most intense fragment after m/z = 107.1—and nearly absent in piceatannol (Figure 1A). Based on this observation, the transitions 245.1/107.1 and 245.1/161 were used as the quantifier and qualifier ions, respectively, to detect and quantify oxyresveratrol, which eluted at a retention time of 1.2–1.4 min.

In-source fragmentation of glycosylated compounds, characterized by the loss of the entire sugar moiety and resulting in detection of the aglycone, has previously been described for piceid (Hurtado-Gaitán et al., 2017), in which a secondary resveratrol peak appears at the retention time of the glycosylated form. We observed a similar phenomenon for mulberroside A, the diglycosylated form of oxyresveratrol, which produces an aglycone peak at the retention time of the glycosylated compound. Consequently, the presence of additional, unknown glycosylated derivatives in the sample may be inferred from the appearance of aglycone-associated peaks at earlier retention times.

GC–MS analyses of cell suspensions for umbelliferone detection

GC–MS analyses were carried out using an Agilent 6890N gas chromatograph equipped with a 30 m × 0.25 mm ID, 0.25 μm film thickness Agilent 19091S-433 HP-5MS (5% phenyl methyl siloxane) column, coupled to an Agilent 5973N mass spectrometer, as previously reported (Martínez-Márquez et al., 2018). The column was operated with hydrogen as the carrier gas (1 ml/min) at an initial temperature of 40 °C, which was then increased at a rate of 12 °C/min to 290 °C and held for 6 min, followed by an increase at 20 °C/min to 320 °C and a final hold at this temperature for 10 min. The injector temperature was set to 250 °C. Mass spectra were recorded from m/z 33 to 250 at 70 eV. Chromatographic peaks were initially identified by searching against the NIST Standard Reference Database (NIST11.L) and subsequently confirmed by comparison with the retention times and mass spectra of authentic compounds.

Microscopy

Morus cells were imaged using a confocal microscope (Leica TCS SP2; Leica Microsystems, Wetzlar, Germany). Cell suspensions were placed onto microscope slides in MS3 medium, and bright-field images were acquired in a single focal plane with a thickness of 1 μm. Brightness and contrast were adjusted using Adobe Photoshop 7.0.

RNA extraction and sequencing

The Spectrum Plant Total RNA Kit (Sigma–Aldrich, SKU: STRN250-1KT) was used to extract total RNA for Illumina sequencing from 34 lyophilized cell samples collected during the time-course experiment. For PacBio sequencing, a four-sample pool of different elicited samples was generated. DNase-treated RNA integrity was assessed using an Agilent Technologies 2100 Bioanalyzer, with high-quality RNA defined as having an RNA integrity number (RIN) ≥ 7. For PacBio sequencing, cDNA synthesis was performed using the Takara SMARTer PCR cDNA Synthesis Kit for Iso-Seq. Molecules under 4 kb were captured for library preparation. Library quality was assessed using an Agilent Technologies 2100 Bioanalyzer and Qubit. Long-read sequencing was performed using a PacBio SMRT Cell on the RS II platform (P6-C4 chemistry). For Illumina sequencing, libraries were generated using the TruSeq Stranded mRNA Library Construction Kit. Library quality was assessed using an Agilent Technologies 2100 Bioanalyzer and Qubit. Finally, short-read sequencing was performed on an Illumina NovaSeq 6000 platform, producing 150 bp paired-end reads.

Genome re-annotation pipeline

A comprehensive method was implemented to (i) revise the protein-coding genes annotated by Jiao et al. (2020) in M. alba, evaluate their structures, and reannotate them if necessary, and (ii) generate new gene models. Supplemental Figure 15 provides a comprehensive compilation of the datasets employed, along with a detailed workflow of each step in the bioinformatics pipeline (see Supplemental Table 6 for the parameters used).

MAKER v2 (Campbell et al., 2014) was used to generate a new structural annotation integrating data from de novo transcriptomes (based on Illumina and PacBio reads obtained in this study), ab initio gene prediction, and database gene models. The de novo transcriptome was constructed using Trinity (Grabherr et al., 2011) in strand-specific mode, utilizing both Illumina and PacBio evidence. Illumina reads were trimmed using fastp (S. Chen et al., 2018). PacBio sequencing errors were corrected using LoRDEC (Salmela and Rivals, 2014) with Illumina reads. Filtering involved the use of CD–HIT to retrieve the most representative sequences, Kallisto (Bray et al., 2016) to remove misassembled transcripts while retaining only expressed transcripts (TPM: transcripts per million > 0), and TransDecoder (https://github.com/TransDecoder/TransDecoder) to retain only protein-coding transcripts. For ab initio gene prediction, the GeneMark software (Besemer and Borodovsky, 2005) was trained using introns annotated by aligning Illumina reads from cell cultures to the M. alba genome assembly with STAR (Dobin et al., 2013). The AUGUSTUS software (Stanke et al., 2006) was also employed, trained using 2,117 genes from the previous annotation. Finally, homology evidence was generated using Moraceae protein sequences downloaded from NCBI and clustered with CD–HIT (Fu et al., 2012) to reduce redundancy. These proteins were then mapped using BLASTx within MAKER after masking the genome. Masked repetitive regions were identified by MAKER using RepeatMasker (v4.1.2) (N. Chen, 2004) together with Dfam database sequences (release 3.5) to locate both previously known and novel transposable elements.

In addition, Minimap2 (Li, 2018) and StringTie (Pertea et al., 2015) were used to generate an additional annotation file based solely on PacBio data. The three annotations—the v0 official annotation from Jiao et al. (2020), the MAKER-derived annotation, and the StringTie-derived annotation—were merged to create a file containing all possible gene structures.

To reduce redundancy in the final annotation, a modified version of TransDecoder, customized to output only complete protein open reading frames, was used to filter and retrieve the best possible gene structure for each gene in the merged file. A single protein was then selected for each transcript. The best protein was chosen using DIAMOND (Buchfink et al., 2015) and InterProScan (Paysan-Lafosse et al., 2023) based on the hit scores from a series of databases, prioritized in the following order: reviewed Viridiplantae proteins from UniProt with ≥ 90% identity, Viridiplantae proteins from NCBI, unreviewed Viridiplantae proteins from UniProt with ≥ 90% identity, and InterPro domains. Reviewed UniProt hits were considered the most reliable, and subsequent databases were only considered in cases of tied hit scores. Overlapping transcripts with the same BLAST hit, as well as those with more than 80% overlap, were filtered out to reduce redundancy, with the aim of retaining one transcript per gene. The best transcript was selected according to the same database priority described above. If multiple transcripts for a gene had identical hit scores (based on E-value and bit score), the transcript was then selected based on the highest FPKM expression or, if this information was unavailable, the longest transcript.

The annotation was evaluated at the protein level against the annotated proteins from Jiao et al. (2020) using BUSCO with the eudicots_odb10 dataset (Simão et al., 2015). We introduced a new gene ID that contains information about the species, cultivar, assembly, and annotation versions, with the fixed prefix “Mal_HE_154X_1.”

Functional annotation

Functional annotation of genes was performed using the eggNOG-mapper (Cantalapiedra et al., 2021), KAAS: KEGG automatic annotation server (Moriya et al., 2007), and Mercator4 (Schwacke et al., 2019) web applications for Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), and MapMan ontologies, respectively. All pipelines were run with default settings, except for KAAS, in which several best-annotated plant species were selected: Arabidopsis thaliana (ath), Zea mays (zma), Vitis vinifera (vvi), Glycine max (gmx), and Solanum lycopersicum (sly).

DEA and WGCNA

Illumina reads were trimmed using fastp software (Chen et al., 2018), followed by mapping with STAR (Dobin et al., 2013). Raw counts were computed using FeatureCounts (He et al., 2013) based on the newly annotated gene models. Normalized counts were used for principal component analysis (PCA) (Supplemental Figure 16A). Differential expression analysis (DEA) was performed using the LIMMA R package (Ritchie et al., 2015), and FPKM (fragments per kilobase of transcript per million) normalization was conducted using the DESeq2 package (Love et al., 2014). Genes were considered differentially expressed when the adjusted p value was lower than 0.05. For co-expression analysis, the WGCNA R package was used (Langfelder and Horvath, 2008), with FPKM values for all samples filtered by expression using the filterByExpr function of edgeR (Robinson et al., 2010). Data for resveratrol and oxyresveratrol synthesis were imported to perform module–trait correlations. Additionally, enrichment analyses for MapMan, KEGG, and GO categories across different modules were performed using the gprofiler2 R package (Kolberg et al., 2020).

Phylogenetic analysis

Sequences were aligned using MAFFT (Katoh, 2002). The resulting alignments were used to build maximum-likelihood trees with IQ-TREE software (Nguyen et al., 2015) and visualized using FigTree. The bootstrap consensus tree was inferred from 1,000 replicates.

Protein extraction and identification

Lyophilized cells were first washed according to Lücker et al. (2009), with modifications. All subsequent steps were carried out on ice, and centrifugations were performed at 4 °C. Throughout the procedure, each wash was followed by centrifugation for 15 min at 15,300 × g. Briefly, 80–130 mg of lyophilized cells were washed several times with ethyl acetate:ethanol (1:2, v/v) until the supernatant was colorless. The resulting pellet was washed twice with chilled acetone and twice with 10% (w/v) trichloroacetic acid (TCA) in acetone at −20°C. This was followed by three washes with 10% (v/v) aqueous TCA. Finally, the pellets were washed twice with 80% (v/v) chilled acetone and left to dry at 4 °C.

Total proteins were extracted from the washed tissue according to Hurkman and Tanaka (1986), with modifications. Briefly, all washed tissue was homogenized in 1ml of extraction buffer (pH 7.5) containing 0.7 M sucrose, 0.1 M KCl, 0.5 M Tris, 50 mM EDTA, 1% PVPP, 1% DTT, 1% deoxycholate, and a cocktail of protease inhibitors containing 4-(2-aminoethyl) benzenesulfonyl fluoride (AEBSF), E-64, bestatin, leupeptin, aprotinin, and sodium EDTA (Sigma-Aldrich) and incubated for 30 min with frequent vortexing. An equal volume of Tris-saturated phenol (pH 7.5) was added, and the mixture was incubated for 30 min with vortexing every 5 min. Phase separation was achieved by centrifugation at 15,000 × g for 40 min. The upper phenol phase was recovered, and the aqueous phase was re-extracted as described above with 1 ml of Tris-saturated phenol (pH 7.5). Both phenol phases were pooled and washed twice with an equal volume of washing buffer (pH 7.0) containing 0.7 M sucrose, 0.1 M KCl, 0.5 M Tris, 50 mM EDTA, 1% DTT, and protease inhibitors as described above. The recovered upper phenol phase was precipitated for 24 h with 5 volumes of 0.1 M ammonium acetate in methanol. The resulting precipitate was washed three times with cold 0.1 M ammonium acetate in methanol and twice with chilled 80% (v/v) acetone, allowed to dry, and solubilized in fresh 6 M urea. Protein concentration was determined using the RC DC (reducing agent and detergent compatible) protein assay (Bio-Rad), which is based on a modified Lowry protein assay method (Raghupathi and Diwan, 1994).

Label-free proteomic analysis

A time-series proteomic experiment was carried out using quadruplicates of whole-cell extracts from mulberry cell suspensions, either untreated (control) or treated with 50 mM MBCD + 0.1 mM MeJA for 24, 48, and 72 h. Trypsin digestion and peptide clean-up were performed as described in Esteve-Sánchez et al. (2020). Thirty micrograms of desalted peptide digests were injected directly onto a reverse-phase Agilent AdvanceBio Peptide Mapping column (2.1 mm × 250 mm, 2.7 μm particle size) mounted on an Agilent 1290 Infinity UHPLC system coupled through an Agilent Jet Stream interface to an Agilent 6550 iFunnel Q-TOF mass spectrometer (Agilent Technologies). Peptides were separated at 50 °C using a 140-min linear gradient of 3%–40% ACN in 0.1% formic acid at a flow rate of 0.400 ml/min. Source parameters were as follows: gas temperature, 250 °C; drying gas, 14 l/min; nebulizer, 35 psi; sheath gas temperature, 250 °C; sheath gas flow, 11 l/min; capillary voltage, 3,500 V; fragmentor voltage, 360 V. Data were acquired in positive-ion mode using Agilent MassHunter Workstation Software LC–MS Data Acquisition B.08.00 (Build 8.00.8058.0). The mass spectrometer was operated in high-sensitivity mode, and MS and tandem mass spectrometry (MS/MS) data were acquired in Auto MS/MS mode, in which the 20 most intense precursor ions (charge states 2–5) within the 300–1700 m/z range and above a threshold of 1,000 counts were selected for MS/MS analysis. MS/MS spectra (50–1700 m/z) were collected with the quadrupole set to “narrow” resolution and were acquired until 25,000 total counts were reached or for a maximum accumulation time of 333 ms.

Each MS/MS spectrum was preprocessed using the extraction tool of Spectrum Mill Proteomics Workbench (Agilent) to obtain a peak list and improve spectral quality by merging MS/MS spectra with the same precursor (Δm/z < 1.4 Da and chromatographic Δt < 15 s). The reduced dataset was searched against the M. alba reannotated protein database (Mal_HE_154X_1) and contaminant proteins in identity mode using the MS/MS search tool of Spectrum Mill Proteomics Workbench with the following settings: trypsin, up to 2 missed cleavages, carbamidomethylation of Cys as a fixed modification, oxidation of Met as a variable modification, and mass tolerances of 20 ppm for precursor ions and 50 ppm for product ions. Peptide hits were filtered with a score ≥ 6 and percent scored peak intensity (%SPI) ≥ 60.

LC–MS raw files were imported into Progenesis QI for Proteomics (Nonlinear Dynamics) v4.0 label-free analysis software. Quantification was performed based on MS1 intensity. The data file yielding the most features (peaks) was used as a reference to align the retention times of all other chromatographic runs and to normalize MS feature signal intensities (peak areas). Correction for experimental variation was performed by calculating the robust distribution of all ratios (log[ratio]). MS features were filtered to include only those with charge states from two to five. After defining the experimental design as “between subjects,” samples were clustered according to experimental groups (C24, C48, C72, MBCD–MeJA24, MBCD–MeJA48, and MBCD–MeJA72), and average intensity ratios of matched features across experimental groups, as well as p values from one-way ANOVA were automatically calculated. To identify the proteins from which the detected features originated, the filtered Spectrum Mill peptide hit files were imported into Progenesis QIp. Peptide assignment conflicts were resolved in favor of the highest-scoring assignments (or left unresolved in cases of equal scores and identical sequences), and the inferred protein list was filtered with a score ≥ 15. Protein abundance was automatically calculated using the Hi-3 method as described by Silva et al. (2006), as implemented in Progenesis QI for Proteomics. Differential protein abundance across experimental groups was assessed using the advanced statistical tools implemented in Progenesis QIp, including ANOVA, hierarchical clustering, and power analysis. Proteins with ANOVA significance p ≤ 0.05 (410 proteins) were used for PCA (Supplemental Figure 16B).

Protein–RNA correlation analysis

Spearman’s rank correlation coefficient (ρ) for each mRNA–protein pair was used to correlate transcriptome and proteome levels. Transcriptomic and proteomic data were evaluated by means of WGCNA (Langfelder and Horvath, 2008). FPKM values and normalized protein abundances were separately transformed into Z scores, and WGCNA was performed using a signed network with a soft power of 6 for DEG–DAP mRNA–protein pairs. Modules were defined by dynamic tree cutting with a minimum size of 10 and a deep split of 4.

Cloning of C2′H cDNA and construction of the binary vector

First-strand cDNA was synthesized from total RNA using a cDNA synthesis kit (NZY First-Strand cDNA Synthesis Kit, NZYTech) according to the manufacturer’s instructions. The C2′H1 and C2′H4 coding regions (Supplemental Data 1) were PCR-amplified from a pool of elicited cell RNAs collected at different time points (primers are listed in Supplemental Table 7). The amplification reactions consisted of a polymerase activation step of 2 min at 95 °C; 40 cycles of denaturation (20 s at 95 °C), annealing (10 s at a selected temperature), and extension (20 s at 70 °C); followed by a final extension step of 10 min at 70 °C; and a hold at 4°C. Annealing temperatures were set based on the lowest melting temperature of the primers used. Amplified DNA fragments were cloned into pENTR/D-TOPO plasmids (Invitrogen) as recommended by the manufacturer, and inserts were sequenced using the LightRun Tube service (Eurofins, Germany), with Sanger sequencing. Correct inserts were transferred into the Gateway-compatible vector pB2GW7 under the control of the CaMV 35S promoter using an LR Clonase reaction (Invitrogen, Thermo Fisher Scientific) performed according to the manufacturer’s instructions. This construct contained a spectinomycin-selectable marker reporter gene. The C2′H5 gene was obtained by commercial synthesis (Gene-Script; Piscataway, NJ, USA) and was also transferred into the Gateway-compatible vectors pB2GW7 and pKAN-ALLIGATOR2. The resulting binary vectors (Supplemental Figure 17), together with pJCV52–VviSTS42 and pBINY53–VviSTS48, were transferred into chemically competent Rhizobium radiobacter (Agrobacterium tumefaciens) strain C58-C1 (pG2260) using the freeze–thaw method and selected on LB plates supplemented with spectinomycin and rifampicin. Colonies were verified by PCR.

Transient expression in N. benthamiana

pB2GW7-C2′Hs and pBINY53–VviSTS48 were infiltrated into N. benthamiana leaves. Agrobacterium C58-C1 harboring the pCH32-35S:p19 vector, which expresses the silencing suppressor p19 of tomato bushy stunt virus, was also used. All Agrobacterium strains were cultured in liquid medium to the late exponential phase, and cells were harvested by centrifugation at 3,000 × g for 15 min at room temperature (24 °C). Cell pellets were resuspended in agroinfiltration buffer (10 mM morpholinoethanesulfonic acid [MES]-KOH, 10 mM MgCl₂, and 150 mM acetosyringone; pH 5.6) and incubated for 3 h at room temperature (24 °C). Cells were mixed in different combinations at a 1:1 ratio, to a final total Agrobacterium optical density at 600 nm (OD₆₀₀) of 1, and then injected into young, fully expanded leaves of 4-week-old N. benthamiana plants (Schöb et al., 1997). Additionally, control samples were obtained by injecting resveratrol (70 μM in 10 mM MES) 4 days after agroinfiltration and oxyresveratrol (70 μM in 10 mM MES) into non-agroinfiltrated leaves. Leaf samples were collected 3 and 6 days after agroinfiltration and stored at −80 °C.

Transient expression in grapevine cells

Agrobacterium tumefaciens strains harboring the indicated constructs were used to transiently transform Vitis cell suspensions. Transient transformation experiments were performed as described in Martínez-Márquez et al. (2023). Strains harboring the binary plant vector pKAN-ALLIGATOR-C2′H1, used alone or mixed with a strain containing pJCV52–VviSTS42 (Hidalgo et al., 2017), were co-cultured in a 1:1 ratio in Vitis cell suspensions. Six days after Agrobacterium infection, stilbene content was analyzed.

C2′H protein purification and activity assay

C2′H5 was sub-cloned into pEZY19 from the corresponding pTOPO-D_C2′H5 entry clone through a Gateway LR reaction (Invitrogen). For small-scale protein purification, Escherichia coli BL21 (DE3) cells were transformed with pEZY19_10HIS-C2′H5 and grown in 100 ml of LB medium supplemented with 100 μg/ml carbenicillin at 30 °C, until an optical density at 600 nm of 0.6–0.8 was reached (≈2 h). Isopropylβ-D-thiogalactopyranoside (IPTG) was added to a final concentration of 1 mM, and cells were harvested after 18 h of incubation at 16 °C. Protein purification was performed using the HisCube Ni-INDIGO His-Tag Protein Purification MINI Kit (Cube Biotech) according to the manufacturer’s instructions. Briefly, the bacterial pellet was resuspended in 10 ml of INDIGO lysis buffer supplemented with 1 mg/ml hen egg white lysozyme and one tablet of cOmplete Protease Inhibitor Cocktail (Roche). Lysis was performed at 4 °C for 1 h in a rotisserie shaker (Thermo Fisher Scientific). The clarified lysate obtained after centrifugation was purified by Ni-affinity chromatography. Elution fractions were quantified using the Bradford protein assay (Bio-Rad) and analyzed by SDS–PAGE.

Confirmation of C2′H5 purification was performed by LC–MS/MS analysis of the resulting tryptic peptides. Protein identification was carried out through database searches of peptide ion fragment spectra, following the method described in Sellés-Marchart et al. (2008), using the same search parameters outlined above (see the section “Label-free proteomic analysis”).

C2′H5 activity was assayed in a final reaction volume of 100 μl containing 100 mM Bis–Tris (pH 6.5), 110 μM p-coumaroyl-CoA, 5 mM α-ketoglutarate, 5 mM sodium ascorbate, 0.5 mM FeSO₄, 1 mg/ml BSA, and approximately 1 μg C2′H5, in a black 96-well plate. Fluorescence (excitation 325 nm/emission 455 nm [Ex/Em]) was measured using a VARIOSKAN fluorometer (Thermo Fisher Scientific) every hour over a 6 h period. Reactions were stopped by adding 20 μl of 3 M NaOH, which resulted in complete hydrolysis of CoA thioesters after incubation for 10 min at 37 °C, and were then neutralized with 20 μl of glacial acetic acid. A UMB calibration curve was generated using different concentrations ranging from 0 to 1.25 μM. Fluorescence measurements were performed as described above, and the calibration curve was used to calculate UMB concentrations in the VARIOSKAN fluorometer assay.

For downstream confirmations, enzymatic reactions were centrifuged at 15,000 × g for 10 min, and 10 μl of the supernatant was analyzed by HPLC using an Agilent 1260 Infinity Binary LC system coupled to either a diode array detector (DAD) or a fluorescence detector (Agilent, Santa Clara, CA) on a Gemini-NX 5 μm C18, 110 Å (250 × 4.6 mm) column (Phenomenex, Torrance, CA). Elution was performed in isocratic mode with 25% (v/v) aqueous MeOH containing 0.1% (v/v) formic acid at a flow rate of 1 ml/min at 40°C. Hydroxycinnamic acids were detected at 324 nm by DAD, and UMB was detected both at 324 nm by DAD and at 324/455 nm (Ex/Em) by fluorescence detection.

For structural confirmation of the reaction products, 5 μL of the supernatant was analyzed by UHPLC–ESI (electrospray ionisation)–MS/MS using an Agilent 1290 Infinity UHPLC system coupled to an Agilent 6490 triple quadrupole mass spectrometer with a JetStream ion source, equipped with a Zorbax Extended C18 RRHT (50 × 2.1 mm, 1.8 μm) column. Elution was performed in isocratic mode with 25% (v/v) aqueous MeOH containing 0.1% (v/v) formic acid at a flow rate of 0.2 ml/min at 30 °C. The mass spectrometer was operated with a fragmentor voltage of 380 V and a cell acceleration voltage of 5 V, in product ion scan mode, triggering fragmentation in Q2 at collision energies of 0, 20, 25, and 30 eV for the following m/z values of the target compounds: 163 for p-coumaric acid (negative mode), 179 for 2,4-dihydroxycinnamic acid (negative mode), and 163 for UMB (positive mode).

Kinetic analysis of C2′H isoform 5

Reaction medium composition and UMB product detection were performed as described earlier, with the following modifications: the p-coumaroyl-CoA concentration varied from 6.25 to 100 μM, the reaction was allowed to proceed for 15 min before stopping, and only the intensity of the positive-mode m/z = 163 fragments generated at collision energies of 20 and 25 eV was monitored. The area under the peak (AUP) was integrated and converted into UMB concentration using a calibration curve generated with the authentic compound (AUP = 10.43 × 10⁶ μM UMB). Reciprocal reaction rate and substrate concentration pairs were plotted using a Lineweaver–Burk representation to determine kinetic parameters, which were then used to generate the Michaelis–Menten curve.

Web apps

The genomic database MorusalbaDB was created based on EasyGDB (Fernandez-Pozo and Bombarely, 2022), including tools for downloading annotation and ontology files, BLAST searches, genome browsing, sequence extraction, and gene searches. A Web Apollo platform (Lee et al., 2013) was also deployed to manually curate the gene models identified through the annotation pipeline. The MulViz data visualization platform is hosted on a Shiny server to deploy each R-based web application.

Data and code availability

The mass spectrometry proteomics data have been deposited in the ProteomeXchange Consortium via the Proteomics Identification Database (PRIDE) partner repository under the dataset identifier PXD048037. Long-read and short-read RNA sequencing data are available in NCBI GEO under the identifier GSE261571. The new M. alba structural and functional genome annotations can be accessed at MorusalbaDB (http://plantaeviz.tomsbiolab.com:8000/easy_gdb/) and the PlantaeViz downloads section (https://plantaeviz.tomsbiolab.com/). In addition, the structural annotation, protein, and transcript FASTA files are provided as Supplemental Data 2, 3, and 4.

Funding

This work was supported by grants PID2021-128865NB-I00 (Valinet) and PID2024-163039NB-I00 (Metacell), and by the Ramón y Cajal program (grant RYC-2017-23 645), all awarded to J.T.M.; and by the PROMETEO scholarship (PROMETEO/2021/056-01) granted to A.S. by the Generalitat Valenciana (GVA). C.Z. is supported by the China Scholarship Council (CSC; no. 201906300087). Additionally, this work was supported by the Spanish Ministry of Science and Innovation (MCIN/AEI/10.13039/501100011033); ERDF, A way of making Europe; and the European Union NextGenerationEU/PRTR through grants PID2020-113438RB-I00 and TED2021-129617B-I00 awarded to R.B.-M.; by the Valencian Conselleria d’Innovació, Universitats, Ciencia y Societat Digital grant CIAICO/2021/167 awarded to R.B.-M. and S.S.-M.; by a “María Zambrano” grant awarded to J.M.-C from the Spanish Ministry of Universities, NextGenerationEU/PRTR; and by the University of Alicante.

Acknowledgments

Special thanks to Pere Mestre and Philippe Hugueney (INRAE Colmar) for providing the 35S:STS48 construct used for agroinfiltration. Computational analyses were performed on the HPC cluster Garnatxa at the Institute for Integrative Systems Biology (I2SysBio). No conflict of interest declared.

Author contributions

Conceptualization, A.S., D.M., R.B.-M., and J.T.M.; performed experiments, G.A.P., P.R., A.M.-M., M.J.M.-E., J.E., S.S.-M., C.Z., E.L.M., A.S.-H., and J.M.-C.; performed bioinformatics and statistical analyses, A.S., G.A.P., R.A.-U., D.N.-P., and R.A.C.; supervision, R.B.-M. and J.T.M.; writing—original draft, review, and editing, A.S., G.A.P., A.M.-M., P.R., D.M., R.B.-M., and J.T.M. All authors provided critical feedback and approved the final version of the manuscript.

Published: February 4, 2026