Metabolic Rewiring of the Shikimate Pathway Enables Rose‐Scented Rice

ABSTRACT

Rice fragrance is a key determinant of grain quality, yet the aroma profile of cultivated rice is highly uniform and largely dominated by the popcorn‐like compound 2‐acetyl‐1‐pyrroline. Here we expanded the aromatic repertoire of rice by engineering a rose‐like fragrance through reconstruction of the phenylalanine‐derived 2‐phenylethanol (2‐PE) pathway in the endosperm. To increase phenylalanine precursor supply, we co‐expressed an engineered G211R/G212S variant of the rice 3‐deoxy‐D‐arabino‐heptulosonate 7‐phosphate synthase isoform OsDHS2 (LOC_Os08g37790; mOsDHS2) with heterologous phenylacetaldehyde synthase (PAAS) and phenylacetaldehyde reductase (PAR) in the endosperm. As a result, the engineered grains accumulated up to 2014 μg/kg 2‐PE, conferring a distinct rose‐like aroma. Unexpectedly, metabolic rewiring also reshaped grain composition, resulting in nearly doubled soluble protein content and increased levels of several vitamin B6‐related metabolites, including the active coenzyme forms pyridoxal 5′‐phosphate and pyridoxamine 5′‐phosphate. These findings reveal extensive associated metabolic coordination between aromatic amino acid pathways and central seed metabolism. Our work establishes a strategy for creating designer fragrances in staple crops while simultaneously improving nutritional traits, providing a versatile platform for producing high‐value metabolites in cereal grains.

1. Introduction

Rice ( Oryza sativa L.) feeds more than half of the global population and is a cornerstone of global food security (Bin Rahman and Zhang 2023). Beyond yield and nutritional value, grain quality traits strongly influence consumer preference and market value. Among these, aroma is a defining attribute of premium rice varieties. However, the fragrance of nearly all aromatic rice cultivars is dominated by a single compound, 2‐acetyl‐1‐pyrroline (2‐AP), which produces a characteristic popcorn like scent (Buttery et al. 1983). While this compound underlies the popularity of many elite cultivars, the reliance on a single volatile metabolite has effectively constrained the diversity of fragrance profiles available in rice. Engineering novel aroma traits in staple crops therefore represents an unexplored opportunity to diversify food quality and expand the sensory landscape of agricultural products.

Plant aromas originate largely from specialized volatile metabolites derived from several major metabolic pathways, including terpenoid, fatty acid and aromatic amino acid metabolism (Pichersky et al. 2006; Schwab et al. 2008; Rowe and Shepherd 2016). Among these, volatiles derived from aromatic amino acids contribute many characteristic floral and flavour compounds widely used in the food, fragrance and pharmaceutical industries (Knudsen et al. 2006). One prominent example is 2‐phenylethanol (2‐PE), a rose‐scented volatile that constitutes a key component of many perfumes and flavourings. In plants, 2‐PE is synthesized from phenylalanine via the intermediate phenylacetaldehyde, catalyzed sequentially by phenylacetaldehyde synthase (PAAS) and phenylacetaldehyde reductase (PAR) (Skaliter et al. 2022). Although this pathway has been characterized in several plant species, its introduction into cereal grains to create novel aroma traits has not been explored.

The metabolic precursor phenylalanine is produced through the plastid‐localized shikimate pathway, which provides the central entry point to aromatic amino acid biosynthesis (Lynch and Dudareva 2020). In microorganisms, carbon flux into this pathway is commonly controlled by feedback inhibition of 3‐deoxy‐D‐arabino‐heptulosonate‐7‐phosphate synthase (DAHPS/DHS), the first committed enzyme of the pathway. In plants, however, regulation of DAHPS/DHS is more complex and differs from the canonical microbial model. Earlier biochemical studies showed that plant DAHPS/DHS activity is generally not inhibited by phenylalanine (Phe), although sensitivity to tyrosine (Tyr) or tryptophan (Trp) has been reported in some species and isoforms (Huisman and Kosuge 1974; Graziana and Boudet 1980; Reinink and Borstlap 1982; Suzich et al. 1985; Maeda and Dudareva 2012). Consistent with this, biochemical characterization of Arabidopsis DHS isoforms showed that Tyr and Trp inhibit AtDHS2, but not AtDHS1 or AtDHS3, highlighting isoform‐dependent regulation of the entry step of the plant shikimate pathway (Yokoyama et al. 2021). Moreover, suppressor of tyra2 (sota) mutations in Arabidopsis DHS genes alleviate this regulation and lead to strong accumulation of aromatic amino acids and increased carbon assimilation (Yokoyama et al. 2022). These findings suggest that deregulating shikimate‐pathway flux is a promising strategy to enhance the biosynthesis of high‐value aromatic metabolites.

Metabolic flux through aromatic amino acid pathways is also closely linked to vitamin B6 metabolism, as many aminotransferases and decarboxylases involved in amino acid interconversion require pyridoxal 5′‐phosphate (PLP) as an essential cofactor (Percudani and Peracchi 2003; Vanderschuren et al. 2013). Vitamin B6 exists as six interconvertible vitamers, including pyridoxine, pyridoxal, pyridoxamine and their phosphorylated derivatives, with PLP serving as the biologically active coenzyme form (Farrington et al. 1993; Hellmann and Mooney 2010). Consequently, large perturbations of aromatic amino acid biosynthetic flux may have broader metabolic consequences, potentially affecting vitamin B6 homeostasis and seed nutritional composition.

Rice endosperm has recently emerged as a versatile platform for metabolic engineering and molecular farming. Its advantages include tissue‐specific expression, genetic stability, scalability and the ability to accumulate high levels of recombinant products. Previous studies have demonstrated successful production of therapeutic proteins, vitamins and specialized metabolites in rice grains through targeted metabolic engineering (Tian et al. 2019; Zhu et al. 2022; Niu et al. 2025). However, reconstructing an entire volatile biosynthetic pathway by multigene stacking remains challenging, and the systemic metabolic consequences of such pathway reconstruction in cereal seeds are largely unknown.

Here we report the metabolic engineering of a rose‐like fragrance in rice through reconstruction of the phenylalanine‐derived 2‐phenylethanol biosynthetic pathway in the endosperm. By combining a deregulated mutant of a key shikimate pathway enzyme with heterologous PAAS and PAR enzymes, we redirected aromatic amino acid metabolism toward high‐level 2‐PE production. Unexpectedly, this metabolic rewiring not only generated a strong floral aroma but also substantially altered grain composition, including enhanced protein accumulation and enrichment of vitamin B6 metabolites. Together, these results demonstrate that cereal grains can be engineered to produce designer aroma compounds while simultaneously reshaping seed metabolism, establishing rice as a promising platform for the sustainable biosynthesis of high‐value plant metabolites.

2. Results

2.1. An Engineered OsDHS2 Variant Increases Aromatic Amino Acid Accumulation in Nicotiana benthamiana

Because 2‐PE biosynthesis depends on phenylalanine as its immediate precursor, its production is tightly linked to flux through the shikimate pathway (Figure 1A). In microorganisms, DAHPS/DHS, the first committed enzyme of this pathway, is commonly feedback‐inhibited by aromatic amino acids. In plants, however, this regulatory mechanism is more complex and cannot be directly generalized from microbial systems. Earlier biochemical studies showed that plant DAHPS/DHS activity is generally not inhibited by phenylalanine, although sensitivity to tyrosine or tryptophan has been reported in some species or isoforms (Huisman and Kosuge 1974; Graziana and Boudet 1980; Reinink and Borstlap 1982; Suzich et al. 1985; Maeda and Dudareva 2012). More recently, detailed biochemical analysis of Arabidopsis DHS isoforms showed that Tyr and Trp inhibit AtDHS2, but not AtDHS1 or AtDHS3, demonstrating that feedback regulation in plants is isoform dependent rather than uniformly mediated by all three aromatic amino acids (Yokoyama et al. 2021). In addition, point mutations in Arabidopsis DHS enzymes were shown to abolish Tyr/Trp sensitivity and markedly increase aromatic amino acid accumulation in planta (Yokoyama et al. 2022). Consistent with this general strategy, expression of a bacterial feedback‐insensitive DAHPS/DHS in Arabidopsis and tomato also enhanced the accumulation of aromatic amino acids and downstream specialized metabolites (Tzin et al. 2012, 2013). These findings prompted us to engineer a rice DAHPS/DHS variant with reduced sensitivity to plant‐specific feedback regulation in order to increase precursor supply for 2‐PE biosynthesis.

FIGURE 1.

Engineering and functional validation of an engineered G211R/G212S OsDHS2 variant. (A) Schematic of regulation of DHS by aromatic amino acids in the shikimate pathway.ADH, arogenate dehydrogenase; ADT, arogenate dehydratase; CM, chorismate mutase; DHS, 3‐Deoxy‐D‐arabino‐heptulosonate 7‐phosphate synthase; E4P, erythrose 4‐phosphate; PEP, phosphoenolpyruvate; Phe, phenylalanine; PPA‐AT, prephenate aminotransferase; Trp, tryptophan; Tyr, tyrosine. (B) Local sequence alignment of four rice DHS isoforms and AtDHS2 around the engineered G211R/G212S residues. (C‐D) Predicted structural models showing the putative tryptophan‐binding sites in wild‐type OsDHS2 and the G211R/G212S mutant. (E) Molecular docking analysis comparing the predicted binding poses of phenylalanine, tryptophan and tyrosine in wild‐type OsDHS2 and the G211R/G212S mutant (mOsDHS2). (F) Expression of wild‐type OsDHS2 and mOsDHS2 in tobacco leaves driven by the 35S promoter. (G‐I) Quantification of aromatic amino acids (phenylalanine, tryptophan, tyrosine) in transiently transformed tobacco leaves. Data represent mean ± SD. Statistical significance was assessed using an unpaired t‐test (*p < 0.05, **p < 0.01, ***p < 0.001).

Following the nomenclature established for grass DHS isoforms (El‐Azaz et al. 2023), the four rice genes are designated OsDHS1a (LOC_Os07g42960), OsDHS1b (LOC_Os03g27230), OsDHS2 (LOC_Os08g37790) and OsDHSnc (LOC_Os10g41480). OsDHS2 was selected for engineering because it is a conserved, feedback‐regulated rice isoform and contains residues corresponding to positions implicated in deregulated Arabidopsis DHS variants. The G211R/G212S variant is hereafter referred to as mOsDHS2 (Figure 1B). Molecular docking was used to compare the predicted interactions of wild‐type and mutant OsDHS2 with Phe, Tyr and Trp (Figure 1C–E, Table S1–S3). Relative to the wild‐type protein, the mutant showed shifted docking poses and higher predicted binding energies for Tyr and Trp, whereas the predicted interaction with Phe changed only slightly. Although docking alone does not establish inhibitory activity, these results are consistent with the view that the G211R/G212S substitutions mainly affect Tyr/Trp‐related regulatory interactions rather than Phe binding.

To test the metabolic effect of this mutant in planta, wild‐type OsDHS2 and mOsDHS2 were transiently expressed in Nicotiana benthamiana leaves under the CaMV 35S promoter (Figure 1F). Overexpression of wild‐type OsDHS2 had little effect on the levels of downstream aromatic amino acids, consistent with tight endogenous control of shikimate‐pathway flux. By contrast, expression of mOsDHS2 resulted in substantially increased accumulation of phenylalanine, tyrosine and tryptophan, with phenylalanine levels elevated approximately sixfold relative to the control (Figure 1G–I). These findings indicate that the engineered mOsDHS2 variant promotes aromatic amino acid accumulation in the heterologous Nicotiana benthamiana system.

2.2. Reconstruction of a Heterologous 2‐Phenylethanol Pathway Confers a Rose‐Like Fragrance in Rice

In plants, PAAS catalyses the decarboxylation and oxidation of phenylalanine to produce phenylacetaldehyde, which is subsequently reduced to 2‐phenylethanol (2‐PE) by PAR (Skaliter et al. 2022). The activities of PAAS from rose and PAR from tomato have been experimentally validated (Tieman et al. 2006; Sakai et al. 2007). Although putative rice homologues of these two enzymes were identified by sequence similarity, they showed low conservation with the characterized enzymes (Figure 2B,C and Figures S2 and S3). We therefore introduced rose PAAS and tomato PAR into rice to establish the 2‐PE biosynthetic pathway. To test whether deregulated mOsDHS2 could further promote 2‐PE production, we first performed a transient expression assay in Nicotiana benthamiana. Co‐expression of PAAS and PAR with an empty vector or wild‐type OsDHS2 resulted in only low 2‐PE accumulation, whereas co‐expression with mOsDHS2 markedly increased 2‐PE levels in Nicotiana benthamiana leaves (Figure 2E). These results suggest that mOsDHS2 increases aromatic amino acid availability in this heterologous leaf system and likely contributes to increased 2‐PE production when co‐expressed with PAAS and PAR. However, this assay does not establish the individual contribution of mOsDHS2 to 2‐PE production in rice grain.

FIGURE 2.

Generation and characterization of rice lines overexpressing mOsDHS2, PAR and PAAS. (A) Schematic representation of the 2‐phenylethanol (2‐PE) biosynthetic pathway in rice endosperm. (B, C) Phylogenetic analysis of PAAS and PAR genes across different plant species. (D) Schematic of the overexpression construct: PAR, PAAS and mOsDHS2 were driven by the endosperm‐specific promoters PGluc, PGlub1 and PGlub4, respectively. (E) 2‐Phenylethanol content in transiently transformed Nicotiana benthamiana leaves. (F‐H) Relative transcript levels of mOsDHS2, PAR and PAAS in transgenic rice seeds, normalized to OsActin1 (LOC_Os03g50885). (I) HPLC quantification of aromatic amino acids (phenylalanine, tyrosine and tryptophan) in transgenic seeds. (J) GC–MS analysis of 2‐phenylethanol content in seeds. (K) Intensities of five aroma attributes in rice accessions. Data are shown as mean ± SD. Statistical significance was determined using one‐way ANOVA with Duncan's multiple range test. Bars labelled with different letters indicate significant differences at p < 0.05.

To introduce a floral fragrance into rice grains while minimizing possible effects on vegetative growth, we reconstructed the phenylalanine‐derived 2‐PE pathway specifically in the endosperm (Figure 2A). The codon‐optimized PAAS and PAR genes, together with mOsDHS2, were driven by the endosperm‐specific promoters P_Glub1, P_Gluc and P_Glub4, respectively (Figure 2D). The construct (Rose Aroma‐Like Pathway: RALP) was introduced into two japonica cultivars, Zhonghua 11 (ZH11) and Yunjingyou 14 (YGY), yielding 24 and 13 independent transgenic lines, respectively, as confirmed by marker‐gene PCR. Among these lines, grains from 14 independent lines emitted a clear rose‐like aroma during seed milling. Three representative ZH11 lines and two YGY lines were then selected for detailed analysis. Quantitative PCR confirmed strong endosperm‐specific expression of mOsDHS2, PAAS and PAR (Figure 2F–H and Figure S5). The RALP transgenic seeds accumulated substantially higher levels of aromatic amino acids, particularly phenylalanine, than the corresponding controls (Figure 2I). GC–MS analysis further detected substantial 2‐PE accumulation in the RALP transgenic grains. In transgenic grains, 2‐PE content reached up to 947 ± 29 μg/kg, whereas it was below the detection limit in control grains (Figure 2J). To examine whether this increase was associated with a perceptible aroma phenotype, we also carried out a sensory evaluation using ten independent evaluators. The evaluators scored five aroma descriptors, including rose aroma, grassy, dairy, nutty and popcorn‐like notes. The transgenic grains received consistently high scores for rose aroma, while the other aroma attributes received much lower scores (Figure 2K), providing preliminary sensory support for the rose‐like aroma detected by volatile analysis. Together, these results show that rice grains carrying the three‐gene cassette accumulated higher levels of aromatic amino acids and 2‐PE and exhibited a perceptible rose‐like fragrance. These observations are consistent with increased precursor availability and 2‐PE pathway activity, but the individual contributions of mOsDHS2, PAAS and PAR cannot be resolved from these rice lines.

2.3. Metabolic Engineering Alters Grain Morphology and Physicochemical Properties

To evaluate potential effects of metabolic pathway engineering on plant growth and yield, we assessed key agronomic traits in transgenic and wild‐type plants. Plant architecture‐related traits, including plant height, tiller number and panicle architecture, were largely unchanged in both ZH11 and YGY backgrounds (Figure 3A–E and Figures S4 and S5). However, transgenic lines produced longer grains accompanied by a reduction in thousand‐grain weight. Further analysis of grain quality traits revealed pronounced differences in endosperm structure. Transgenic seeds displayed substantially increased chalkiness, with some lines reaching nearly 98% chalky grains (Figure 3B,F and Figure S5K). In contrast, total starch content was not significantly altered (Figure 3G), suggesting that the chalky phenotype did not arise from impaired starch biosynthesis.

FIGURE 3.

Agronomic performance and grain quality of transgenic overexpression lines in the ZH11 rice background. (A, B) Representative images of brown rice from wild‐type (WT) and transgenic lines, showing grain length and width. Scale bar: 1 cm. (C, D) Quantification of grain length and grain width in WT and transgenic lines. (E) Thousand‐grain weight (TGW) of WT and transgenic lines. (F) Chalkiness rate in seeds of WT and transgenic lines. (G) Starch content in seeds of WT and transgenic lines. (H) Scanning electron microscopy (SEM) images of grain cross‐sections. (I) Gelatinization profiles of rice flour under standard heating conditions (95°C, 13 min). (J‐O) Rapid Visco Analyser (RVA) parameters of WT and transgenic lines: Peak viscosity (J), trough viscosity (K), breakdown viscosity (L), peak time (M), final viscosity (N) and setback viscosity (O). Data are presented as mean ± SD. Statistical significance was assessed using an unpaired t‐test (*p < 0.05, **p < 0.01, ***p < 0.001).

Scanning electron microscopy revealed pronounced structural changes in starch granule organization. Compared with wild‐type grains, the transgenic endosperm contained starch granules with irregular surfaces and looser packing (Figure 3H). Consistent with these structural differences, several pasting properties associated with grain cooking quality were significantly altered (Figure 3I–O). Together, these observations indicate that metabolic rewiring of aromatic amino acid pathways perturbs endosperm structural organization, leading to altered grain physicochemical properties.

2.4. Untargeted Metabolomics Reveals Large‐Scale Metabolic Rewiring

To investigate the broader metabolic consequences of pathway reconstruction, we performed untargeted metabolomic profiling of mature grains from both genetic backgrounds. Principal component analysis revealed clear separation between transgenic and wild‐type samples, with tight clustering among biological replicates (Figure 4A), indicating high data reproducibility. Among 2212 detected metabolites, 733 showed altered abundance in YGY lines and 654 in ZH11 lines (Figure 4B). A substantial fraction (358 metabolites) overlapped between the two genetic backgrounds (Figure 4C), suggesting that many metabolic changes are directly associated with the engineered pathway. KEGG pathway analysis showed that the differential metabolites were mapped to several metabolic pathways, including phenylalanine, tyrosine, tryptophan and vitamin B6 metabolism‐related pathways (Figure 4D). Consistent with our engineering strategy, metabolites in the three aromatic amino acid pathways accumulated at higher levels in transgenic grains (Figure 4E). Several vitamin B6‐related metabolites were also altered, and targeted biochemical assays showed higher vitamin B6 content in the transgenic lines (Figure 4H,I). Given the central role of aromatic amino acids in nitrogen metabolism, we further measured seed protein content. Strikingly, soluble protein levels were nearly doubled, reaching approximately 200 mg/g in transgenic grains (Figure 4F,G). These findings indicate that engineering the shikimate pathway induces extensive metabolic reprogramming that extends beyond volatile biosynthesis.

FIGURE 4.

Metabolomic profiling and associated compositional changes in transgenic rice grains. (A) Principal component analysis (PCA) score plot showing sample clustering and reproducibility. (B) Volcano plot depicting differentially accumulated metabolites between transgenic and wild‐type samples. (C) Venn diagram indicating shared and unique metabolites across samples. (D) Sankey diagram representing KEGG pathway mapping of differential metabolites. (E) Heatmaps of pathway‐associated metabolites: (I) Phenylalanine, (II) Tyrosine, (III) Tryptophan, (IV) Vitamin B6. (F‐I) Biochemical traits in different genetic backgrounds: Soluble protein in ZH11 (F) and YGY (G); vitamin B6 content in ZH11 (H) and YGY (I). Data are presented as mean ± SD. Statistical significance was assessed using an unpaired t‐test. Statistical significance is indicated as *p < 0.05, **p < 0.01, ***p < 0.001.

2.5. Integrated Multi‐Omics Analysis Reveals Dynamic Rewiring of Aroma Biosynthesis During Grain Filling

To investigate how the engineered rose‐like aroma develops during seed maturation, we quantified 2‐phenylethanol (2‐PE) and its precursor aromatic amino acids at three grain‐filling stages (10, 20 and 30 days after flowering, DAF). GC–MS analysis showed that 2‐PE content peaked at 20 DAF, reaching 2014 ± 43 μg/kg, before declining to 1304 ± 18 μg/kg by 30 DAF (Figure 5G). Notably, 2‐PE levels in transgenic grains under the YGY background were lower than those in the ZH11 background (Figure S6A–G), whereas the compound remained nearly undetectable in control grains throughout development (Figure 5A–F). The engineered pathway also markedly increased 2‐acetyl‐1‐pyrroline (2‐AP) content in mature YGY grains (Figure S6H). Furthermore, phenylalanine, tryptophan and tyrosine levels remained consistently elevated in transgenic lines across all stages (Figure 5H–J), indicating sustained enhancement of flux through the upstream shikimate pathway.

FIGURE 5.

Integrated transcriptomic and metabolomic analysis during grain filling. (A‐F) Chromatographic profiles of 2‐phenylethanol (2‐PE) in ZH11 wild‐type and RALP transgenic lines across grain‐filling stages. (G) 2‐Phenylethanol content in ZH11 and RALP grains at different grain‐filling stages. (H‐J) Fold changes of aromatic amino acids during grain filling: Phenylalanine (H), Tryptophan (I) and Tyrosine (J). (K) Dynamic changes in the number of differential metabolites and differentially expressed genes (DEGs) across grain‐filling stages. (L) Venn diagram of differential metabolites across stages and KEGG enrichment analysis of metabolites consistently altered throughout development. (M‐N) Heatmaps showing 17 core metabolites (M) and 31 key genes (N) in the phenylalanine and vitamin B6 metabolic pathways. (O‐P) Biochemical measurements: Starch content (O) and vitamin B6 content (P) in grains at different filling stages. A, B and C represent 10, 20 and 30 days after flowering, respectively. Data are presented as mean ± SD. Statistical significance was assessed using an unpaired t‐test. Statistical significance is indicated as *p < 0.05, **p < 0.01, ***p < 0.001.

To place these changes in a systems context, we performed integrated metabolomic and transcriptomic profiling across grain filling (Figure S7). Principal component analysis revealed clear genotype‐driven separation at all stages (Figure S9). Metabolomic trajectories showed progressive divergence from the wild type over time, reflected by a steadily increasing number of differential metabolites (Figure 5K and Figure S8). By contrast, the number of differentially expressed genes (DEGs) followed an opposite trend, with the highest number detected at 10 DAF and a gradual decline toward 30 DAF. This divergence between transcriptomic and metabolomic dynamics suggests a temporal decoupling between regulatory reprogramming and metabolic output.

We propose that early‐stage transcriptional reprogramming is required to accommodate the abrupt increase in carbon flux through the engineered shikimate pathway, enabling coordination of amino acid biosynthesis and downstream pathways. As grain development proceeds, the system transitions toward a new metabolic steady state, requiring fewer transcriptional adjustments, while metabolites continue to accumulate as endosperm storage compounds. This model is consistent with the observed progressive expansion of the metabolite pool despite reduced transcriptional changes at later stages (Figure 5K).

Focusing on metabolites consistently altered across all stages, we identified 17 core compounds enriched in phenylalanine and vitamin B6 metabolism (Figure 5L–N). Notably, most genes associated with these pathways were consistently upregulated in transgenic grains, closely mirroring metabolite accumulation patterns (Figure 5N). Gene‐ and metabolite‐set enrichment analyses further confirmed coordinated activation of the phenylalanine metabolic network across all developmental stages (Figure S8I–K and Figure S9I–K). Consistent with this coordinated regulation, vitamin B6 levels were significantly elevated in transgenic grains throughout grain filling, whereas total starch content remained largely unchanged (Figure 5O,P), indicating selective metabolic reallocation rather than global disruption of carbon storage.

To resolve the regulatory architecture underlying this metabolic rewiring, we performed weighted gene co‐expression network analysis (WGCNA), identifying 17 transcriptional modules (Figure S10). Correlation analysis between module eigengenes and key metabolites revealed several modules, including Me11, Me12, Me13 and Me15, that were strongly associated with phenylalanine‐derived intermediates, including L‐phenylalanine and phenylethylamine (Figure 6A). Functional enrichment indicated that these modules are linked to amino acid metabolism and cofactor biosynthesis (Figure 6B–E), supporting their roles as transcriptional hubs coordinating flux through the engineered pathway. Integration of transcriptomic and metabolomic datasets further revealed coordinated changes in gene expression and metabolite accumulation associated with phenylalanine and vitamin B6 metabolism (Figure 6F,G). Moreover, the vitamin B6 metabolism‐related genes PDX1, PDX3 and SOS4 were upregulated in the transgenic lines, with particularly pronounced increases in PDX3 and SOS4 expression. Quantitative PCR validation of representative genes confirmed sustained upregulation across developmental stages (Figure 7A–H), supporting the robustness of the multi‐omics analysis.

FIGURE 6.

Integrated transcriptomic and metabolomic analysis in rice grains. (A) Correlation heatmap between weighted gene co‐expression modules and specific metabolites. (B‐E) Bubble plots showing KEGG enrichment profiles for genes from the Me11 (B), Me13 (C), Me12 (D) and Me15 (E) modules. (F, G) Integrated pathway maps combining transcriptomic and metabolomic data: Shikimate and phenylalanine metabolism (F) and vitamin B6 metabolism (G). Circles represent metabolites measured in the genetic‐background comparison experiment, whereas boxes represent metabolites measured in the grain‐filling time‐course experiment, together with the corresponding enzyme encoding transcripts. Heatmaps display significant genotype‐dependent differences (p < 0.05). Colour scales: Red–blue for gene expression (red = upregulated, blue = downregulated) and purple–green for metabolite accumulation (purple = increased, green = decreased). All heatmap data were log2‐transformed and normalized. The six boxes correspond to transgenic and wild‐type seeds at progressive grain‐filling stages: RALP‐A, RALP‐B, RALP‐C, ZH11‐A, ZH11‐B and ZH11‐C (left to right). The four circles denote different genetic backgrounds: RALP/ZH11, RALP/YGY, ZH11 and YGY (left to right). DAHP, 3‐deoxy‐D‐arabino‐heptulosonate 7‐phosphate; DHQ, 3‐dehydroquinate; E4P, erythrose 4‐phosphate; EPSP, 5‐enolpyruvylshikimate‐3‐phosphate; G3P, glyceraldehyde 3‐phosphate; GL, glucosidase; Gln, glutamine; Glu, glutamate; GT, glycosyl transferase; PAA, phenylacetaldehyde; P‐ase, phosphatase; PEA, phenylethylamine; PEP, phosphoenolpyruvate; PL, pyridoxal; PLP, pyridoxal 5′‐phosphate; PLR, PL reductase; PM, pyridoxamine; PMP, pyridoxamine 5′‐phosphate; PN, pyridoxine; PN‐Glu, pyridoxine 5′‐β‐D‐glucoside; PNP, pyridoxine 5′‐phosphate; R5P, ribose 5‐phosphate; S3P, shikimate 3‐phosphate; T‐ase, transaminase.

FIGURE 7.

Validation of core genes and schematic summary of the engineered pathway. (A‐H) Relative expression levels of eight core genes in transgenic (RALP) and wild‐type (ZH11) grains, determined by qRT‐PCR. A, B and C represent 10, 20 and 30 days after flowering, respectively. Data are presented as mean ± SD. (I) Schematic overview of the engineered 2‐phenylethanol (2‐PE) biosynthetic pathway and its associated metabolic changes in rice endosperm. The engineered mOsDHS2 step, the shikimate pathway and phenylalanine biosynthesis are depicted in the plastid, whereas the conversion of phenylalanine to 2‐PE by PAAS and PAR is shown in the cytosol. The engineered grains exhibited increased 2‐PE accumulation, rose‐like aroma, vitamin B6 levels and soluble protein content.

Collectively, these results are consistent with progressive and system‐wide metabolic changes during grain filling in rice lines carrying the engineered three‐gene cassette. In addition to the accumulation of a floral volatile, these changes suggest an association between aromatic amino acid metabolism and vitamin B6 homeostasis. At the phenotypic level, enhanced nitrogen allocation may support increased protein accumulation, whereas altered carbon partitioning impacts starch organization and grain weight. Together, these findings establish rice endosperm as a programmable metabolic platform for the biosynthesis of high‐value aromatic compounds, while highlighting the dynamic and stage‐dependent nature of metabolic regulation in developing seeds.

3. Discussion

3.1. Engineering Cereal Endosperm as a Platform for Floral Fragrance Biosynthesis

2‐Phenylethanol (2‐PE) is a widely used floral fragrance compound whose natural abundance in plants is extremely low, making its production heavily reliant on inefficient extraction from floral biomass or microbial fermentation systems (Bugorskiĭ and Zaprometov 1978; Watanabe et al. 2002; Tieman et al. 2006; Sakai et al. 2007; Hazelwood et al. 2008). For example, extraction from rose petals requires several tons of biomass to obtain small quantities of 2‐PE, whereas tomato fruits contain only about 1.0 × 10⁻³ ng/g fresh weight (Li et al. 2023), highlighting the limitations of plant‐derived production systems. Here, we show that cereal endosperm can be reprogrammed to synthesize high levels of 2‐PE through pathway reconstruction and flux redirection. By combining a deregulated variant of a key shikimate pathway enzyme (mOsDHS2) with heterologous expression of PAAS and PAR, we established a complete biosynthetic route from central carbon metabolism to a volatile aromatic product. Previous studies have shown that deregulation of DAHPS/DHS can enhance aromatic amino acid biosynthesis and downstream specialized metabolism in plants (Tzin et al. 2012, 2013; Yokoyama et al. 2022), while PAAS and PAR catalyse the conversion of phenylalanine to 2‐PE in multiple plant species (Kaminaga et al. 2006; Chen et al. 2011). This multi‐layer engineering strategy substantially increased phenylalanine availability and enabled its efficient conversion into 2‐PE, resulting in orders‐of‐magnitude accumulation of a non‐native floral volatile in rice grains. Compared with previous efforts that mainly modulated endogenous aroma compounds such as 2‐acetyl‐1‐pyrroline (Buttery et al. 1983), our work expands the accessible chemical space of cereal grains and provides a framework for introducing entirely new scent profiles. More broadly, these findings position the crop endosperm as a scalable and sustainable platform for the biosynthesis of high‐value aromatic metabolites (Tian et al. 2019; Zhu et al. 2022). However, although the functional effect of mOsDHS2 in promoting precursor supply is clear, the precise mechanism underlying its deregulated behaviour remains to be resolved.

Because the rice lines analysed here carried a single three‐gene cassette containing mOsDHS2, PAAS and PAR, the individual contribution of each enzyme to 2‐PE production in rice grain cannot be resolved from the present experiments. The Nicotiana benthamiana assays showed that mOsDHS2, unlike wild‐type OsDHS2, increased aromatic amino acid accumulation and strongly enhanced 2‐PE production when co‐expressed with PAAS and PAR. However, these heterologous leaf assays do not establish that mOsDHS2 is necessary for 2‐PE production in rice endosperm, nor do they exclude the possibility that increased phenylalanine availability alone could promote 2‐PE formation through endogenous or introduced conversion capacity. Consistent with this possibility, deregulated Arabidopsis DHS variants substantially broaden aromatic amino acid and related keto‐acid pools (Yokoyama et al. 2022). Future studies using rice lines expressing mOsDHS2 alone, PAAS/PAR without mOsDHS2, and each downstream enzyme separately, together with isotope‐based flux analysis, will be required to quantify the contribution of each module.

Recent work further showed that rice OsDHS1b (LOC_Os03g27230) is naturally partially deregulated and can increase aromatic amino acid and shikimate‐pathway intermediate levels when expressed in Nicotiana benthamiana, whereas OsDHS2 (LOC_Os08g37790) is more strongly regulated (El‐Azaz et al. 2025). Thus, endosperm‐specific overexpression of native OsDHS1b, without introducing engineered amino‐acid substitutions, represents a promising alternative strategy for increasing precursor supply in rice. Direct comparison of OsDHS1b and mOsDHS2 in otherwise identical rice constructs will be needed to evaluate their relative effects on 2‐PE yield and grain quality.

Our molecular docking analysis suggested that the G211R/G212S substitutions alter the predicted interactions of OsDHS2 with aromatic amino acids, particularly Tyr and Trp, but these simulations do not by themselves define the mode of inhibition. Previous structural and mechanistic studies of DAHP synthases, including type II enzymes, support feedback regulation through allosteric mechanisms rather than simple competition with the catalytic substrates E4P or PEP (Shumilin et al. 2002; Blackmore et al. 2013; Jiao et al. 2020; Balachandran et al. 2022). This caution is especially relevant in plants, where DAHPS/DHS regulation differs from the canonical microbial model. Recent work in Arabidopsis showed that phenylalanine does not inhibit plant DHS activity, whereas tyrosine and tryptophan inhibit only specific isoforms, highlighting the isoform dependence of feedback regulation at the entry step of the plant shikimate pathway (Yokoyama et al. 2021). We therefore regard the docking results as supportive, hypothesis‐generating evidence rather than mechanistically definitive, while the main evidence for functional deregulation comes from the strong in planta accumulation of aromatic amino acids observed after transient expression of mOsDHS2. Future biochemical characterization using purified‐enzyme assays will be required to define the regulatory mechanism of mOsDHS2 more directly.

3.2. Dynamic Decoupling of Transcriptional and Metabolic Responses During Pathway Rewiring

A central insight from this study is the temporally distinct dynamics of transcriptomic and metabolomic responses during grain filling. While metabolic divergence from the wild type progressively increased over time, the number of differentially expressed genes showed the opposite trend, peaking at early stages and declining thereafter. This apparent decoupling suggests a two‐phase response to metabolic engineering. We propose a model in which early developmental stages are characterized by extensive transcriptional reprogramming to accommodate the abrupt increase in carbon flux through the engineered shikimate pathway. Such large‐scale transcriptional adjustment has been observed in response to metabolic perturbations and flux imbalances in plants (Tan et al. 2025; Wang et al. 2025). As development proceeds, the system transitions toward a new steady state in which fewer transcriptional adjustments are required, while metabolites continue to accumulate due to sustained flux and storage‐associated processes in the endosperm. This ‘early transcriptional burst‐late metabolic accumulation’ model provides a conceptual framework for understanding how engineered metabolic pathways are integrated into developmental programmes. This model further suggests that transcriptional plasticity and metabolic output can be temporally uncoupled, a phenomenon increasingly recognized in systems biology but rarely documented in crop metabolic engineering contexts (Chen et al. 2011).

3.3. System‐Wide Metabolic Coordination and Unintended Network Propagation

Beyond the targeted production of 2‐PE, our data reveal extensive propagation of metabolic changes across interconnected pathways. In particular, the coordinated activation of phenylalanine metabolism and vitamin B6‐related metabolites points to broad metabolic reprogramming associated with the engineered pathway. Vitamin B6, in its active form PLP, serves as an essential cofactor for numerous enzymes involved in amino acid metabolism (Percudani and Peracchi 2003; Mooney et al. 2009). The observed increase in PLP and PMP levels in transgenic grains, together with elevated aromatic amino acid metabolism, is consistent with coordinated adjustments in cofactor‐related metabolism during endosperm remodelling. However, the present data do not establish a direct mechanistic relationship among aromatic amino acid flux, vitamin B6 homeostasis and protein accumulation. Instead, these changes are more cautiously interpreted as associated features of system‐wide metabolic reprogramming in the engineered seeds. This result highlights that metabolic engineering of primary pathways can have broad downstream consequences, even when the precise causal relationships among the affected pathways remain to be resolved.

3.4. Trade‐Offs in Carbon‐Nitrogen Allocation and Grain Quality

While pathway reconstruction enabled efficient production of a high‐value metabolite, it also imposed measurable trade‐offs at the physiological level. Transgenic grains exhibited increased protein accumulation together with elevated aromatic amino acid levels, alongside reduced grain weight and increased chalkiness. These phenotypes likely reflect altered carbon‐nitrogen allocation, in which enhanced nitrogen assimilation and amino acid biosynthesis compete with starch accumulation for shared carbon resources (Deng et al. 2019; Zhang et al. 2023). Grain chalkiness has long been associated with disrupted starch granule organization and imbalanced storage compound deposition during endosperm development (Kaneko et al. 2016; Shi et al. 2024; Li et al. 2025; Chen et al. 2026). Previous studies have also reported higher protein content in chalky grains, supporting a link between nitrogen allocation and grain appearance (Areum and Jin 2009; Peng et al. 2024). Our results extend these observations by showing that engineered metabolic flux can directly shift carbon partitioning, thereby affecting both grain quality and yield‐related traits.

3.5. Design Principles for Next‐Generation Metabolic Engineering in Crops

The findings of this study suggest several generalizable strategies for improving metabolic engineering outcomes in cereal systems. Temporal control of pathway activity, for example, through stage‐specific promoters, may help mitigate trade‐offs by restricting flux redirection during critical phases of starch accumulation (Zang et al. 2026). In addition, subcellular compartmentalization of pathway enzymes has been shown to enhance metabolic efficiency and reduce toxicity of intermediates (Alam et al. 2017). Manipulation of amino acid transport processes may further enable more precise partitioning of metabolic intermediates between competing pathways. More broadly, our work highlights the importance of integrating pathway design with developmental and systems‐level considerations. Future efforts to engineer crop metabolism will benefit from approaches that coordinate flux optimization, cofactor balance and resource allocation, thereby enabling the production of high‐value compounds without compromising agronomic performance.

While the engineered lines exhibited increased grain chalkiness, this phenotype does not necessarily preclude their practical utilization, but rather points to alternative application scenarios. Notably, glutinous rice varieties, which are widely used for products such as rice cakes and fermented beverages, also display high chalkiness due to their elevated amylopectin content, yet retain substantial economic value in specialized markets. In this context, the engineered aromatic rice developed here represents a multifunctional germplasm with diverse application potential. The introduction of a rose‐like fragrance provides a novel sensory attribute that can be blended with premium rice to enhance product differentiation and market value. In addition, the high accumulation of 2‐phenylethanol suggests the feasibility of using rice grains as a renewable source for natural fragrance extraction, offering an alternative to traditional plant‐based production systems. Importantly, the substantial increase in grain protein content, combined with a relatively balanced amino acid composition, highlights its potential as a high‐quality feedstock. Together, these features suggest that, rather than representing a limitation, the altered grain properties reflect a reprogramming of end‐use functionality, expanding the utility of rice from a staple food crop toward a versatile platform for food, feed and bioproduct applications.

4. Conclusion

In summary, this study establishes rice endosperm as a programmable metabolic platform for the biosynthesis of non‐native aromatic compounds. By combining pathway reconstruction with systems‐level analysis, we reveal that metabolic engineering induces dynamic, stage‐dependent rewiring of gene expression and metabolite accumulation, alongside broader network‐associated metabolic coordination and physiological trade‐offs. These findings provide both a practical framework for crop‐based bioproduction and new insights into the integration of engineered pathways within plant metabolic networks.

5. Materials and Methods

5.1. Vector Construction

To produce 2‐PE, we assembled a three‐gene construct stepwise. The mOsDHS2 mutant (G211R/G212S) was generated from wild‐type ZH11 OsDHS2 cDNA by overlap PCR with mutagenic primers (Table S4). Separately, we acquired synthetic, rice‐codon‐optimized PAAS (Rosa hybrida, GenBank: DQ192639.1) and PAR ( Solanum lycopersicum , GenBank: EF613490.1) genes. For expression, mOsDHS2, PAAS and PAR were placed under the control of the endosperm‐active promoters P_Glub4, P_Glub1 and P_Gluc, respectively, each followed by a T_Nos terminator. Finally, we inserted this three‐cassette assembly into pRHVc‐GFP, creating the transformation vector pRHVc‐mOsDHS2‐PAAS‐PAR.

5.2. Plant Transformation

The recombinant binary vector was introduced into Agrobacterium tumefaciens strain EHA105 via the freeze–thaw method. For plant transformation, we used embryogenic calli induced from mature seeds of the rice cultivars Zhonghua 11 (ZH11) and Yunjingyou 14 (YGY). Calli were immersed in the Agrobacterium suspension (OD₆₀₀ = 0.8–1.0) for 5 min, then co‐cultured on solid N6 medium in the dark at 25°C for 2 days. After washing, calli were transferred to selection medium (N6 medium with 50 mg/L hygromycin B and 250 mg/L carbenicillin) and subcultured every 2 weeks for approximately 4 weeks. Hygromycin‐resistant calli were moved to MS regeneration medium containing 1 mg/L NAA, 2 mg/L 6‐BA and 30 g/L sucrose (with hygromycin and carbenicillin) to regenerate plantlets. Once developed, plantlets were acclimatized in soil and grown to maturity in the greenhouse. We extracted genomic DNA from young leaves of PCR‐positive plants (primers in Table S4) to confirm transgene integration. Homozygous lines from the T₂ or T₃ generation were used for all further experiments.

5.3. Plant Materials and Growth Conditions

All T₂ or T₃ generation materials were planted in the field at the experimental farm of South China Agricultural University, Guangzhou, China. Throughout the entire growth cycle, water and fertilizer were administered in accordance with conventional agronomic protocols.

5.4. Transient Expression and Amino Acid Assay in Nicotiana benthamiana

Nicotiana benthamiana seeds were surface‐sterilized and sown in seedling trays filled with a peat: vermiculite mixture (3:1, v/v). Plants were grown in a growth chamber set to a 16 h light/8 h dark photoperiod with a light intensity of 150 μmol m⁻² s⁻¹, day/night temperatures of 25°C/22°C, and a relative humidity of 60%–70%. When seedlings reached the 4‐ to 5‐leaf stage, they were transplanted into 10‐cm‐diameter pots and kept under the same conditions for an additional 4 weeks before being used for transient transformation assays.

The recombinant plasmid was introduced into Agrobacterium tumefaciens strain GV3101 by the freeze–thaw method. A single transformed colony was picked and inoculated into 5 mL of YEB liquid medium, then grown overnight at 28°C with shaking at 200 rpm. The next day, this culture was diluted 1:100 into fresh YEB medium and shaken until the mid‐log phase (OD₆₀₀ = 0.6–0.8). Bacterial cells were pelleted by centrifugation at 4000 × g for 10 min at 4°C, resuspended in infiltration buffer (10 mM MES, pH 5.6, 10 mM MgCl₂, 200 μM acetosyringone) and adjusted to a final OD₆₀₀ of 0.5. The resuspended cells were kept in the dark at room temperature for 2–3 h before infiltration. The bacterial suspension was infiltrated into the abaxial side of fully expanded mature leaves (3–4 leaves per plant) using a needleless syringe. Infiltrated plants were returned to the growth chamber. At 48–72 h post‐infiltration, leaf discs were punched from the infiltrated zones. A portion of the discs was used immediately for analyses, and the remainder was flash‐frozen in liquid nitrogen and stored at −80°C.

5.5. RNA Extraction and Real‐Time PCR

To investigate the expression of mOsDHS2, PAR and PAAS, total RNA was extracted from 0.1 g of frozen, powdered rice seeds at the mid‐filling stage (20 days after flowering) using TRIzol reagent (Invitrogen, Carlsbad, CA, USA), strictly following the manufacturer's instructions. The RNA was reverse transcribed using HiScript Reverse Transcriptase (Vazyme, Nanjing, China). Real‐time PCR was conducted using Fast SYBR Green Master Mix (Thermo Fisher Scientific). The thermal cycling conditions were as follows: initial denaturation at 95°C for 30 s, followed by 40 cycles of 95°C for 10 s and 60°C for 30 s. A melt‐curve analysis was performed at the end of each run by increasing the temperature from 60°C to 95°C in 0.5°C increments to confirm primer specificity and the absence of primer dimers. qRT‐PCR data were standardized using OsActin1 as an internal reference. All assays were conducted in triplicate. Relative expression levels were calculated using the comparative 2^−ΔΔCt method. The primers used for qRT‐PCR are listed in Table S5.

5.6. Measurement of Amino Acid Content

Approximately 0.1 g of powdered seed sample was extracted with 1 mL of 0.1 M HCl containing 10 ppm acetanilide as the internal standard. After sonication for 30 min and centrifugation at 9000 × g for 10 min, the supernatant was subjected to LC–MS/MS analysis. Chromatographic separation was performed on an ACQUITY UPLC BEH Amide column (2.1 × 100 mm, 1.7 μm; Waters, Milford, MA, USA) using an ACQUITY UPLC H‐Class system coupled to a Xevo TQ‐S triple quadrupole mass spectrometer equipped with a ZSpray electrospray ionization source (Waters, Milford, MA, USA). The column temperature was maintained at 40°C, with an injection volume of 2 μL and a flow rate of 0.30 mL min⁻¹. The mobile phase consisted of 0.01% ammonium acetate in water (A) and 0.1% formic acid in acetonitrile (B). The gradient programme was 20%–70% A (0–3.0 min), 70%–20% A (3.0–4.0 min) and 20% A (4.0–7.0 min).

Mass spectrometric detection was performed in positive electrospray ionization mode using multiple reaction monitoring (MRM) for targeted quantification. The monitored transitions were m/z 207.3 → 166.4 for tryptophan, m/z 182.0 → 91.0 and 182.0 → 119.0 for tyrosine, and m/z 166.1 → 103.2 and 166.1 → 120.2 for phenylalanine. Authentic standards of phenylalanine, tyrosine and tryptophan were used to generate external calibration curves for quantification. Three independent biological replicates were analysed.

5.7. Pasting Properties

The pasting properties of rice flour were determined using a Rapid Viscosity Analyser (RVA) (Shanghai Baosheng Industrial Development Co. Ltd., China) according to the method described by Lu and Seetharaman (2014). Briefly, 3.00 g of rice flour (adjusted to 10% moisture basis) was dispersed in 25 mL of distilled water in the RVA sample canister and stirred manually for 10 cycles to ensure homogenization. The slurry was then subjected to a 13 min standard pasting programme. All measurements were performed in triplicate, and average values were reported.

5.8. Metabolite Profiling by UHPLC–MS

Metabolites were extracted from 100 mg of each sample and analysed using a Vanquish UHPLC system coupled to an Orbitrap Q Exactive HF mass spectrometer (Thermo Fisher, Germany). Raw data were processed with Compound Discoverer 3.1 for peak alignment, picking and quantification. Metabolites were identified by matching against the mzCloud, mzVault and MassList databases. Principal component analysis (PCA) and partial least squares discriminant analysis (PLS‐DA) were performed in metaX to visualize group differences. Differential metabolites were defined as those with p < 0.05 (t‐test), and |FC| ≥ 2, and were visualized using ggplot2 and pheatmap in R. Integrated omics analysis of transcriptomic and metabolomic data was conducted in Python 3.5.0 and R 3.4.3 to identify shared KEGG pathways.

5.9. RNA Sequencing and Differential Expression Analysis

RNA sequencing was performed by Novogene Bioinformatics Technology Co. Ltd. (Beijing, China). Briefly, mRNA was enriched from total RNA using oligo(dT)‐attached magnetic beads and fragmented into short fragments. First‐strand cDNA was synthesized using random hexamer primers, followed by second‐strand cDNA synthesis. Sequencing libraries were constructed through end repair, adenylation of 3′ ends, adapter ligation, fragment size selection, PCR amplification and purification. Library quality and concentration were evaluated using a Qubit fluorometer, real‐time quantitative PCR and an Agilent Bioanalyzer. Qualified libraries were pooled at equimolar concentrations and subjected to high‐throughput sequencing on an Illumina platform. Sequencing was carried out using sequencing‐by‐synthesis (SBS) technology.

Raw sequencing reads were processed using Trimmomatic to remove adapter contamination, reads containing more than 10% ambiguous nucleotides (N), and low‐quality reads in which more than 50% of bases had a Phred quality score ≤ 5. Quality metrics including Q20, Q30 and GC content were calculated for each sample. All downstream analyses were performed using clean reads. Clean reads were aligned to the rice reference genome (IRGSP‐1.0) using HISAT2 (v2.2.1). Gene expression levels were quantified using StringTie, and gene‐level count matrices were generated using the prepDE.py3 script for downstream analyses. Differential expression analysis was performed using the DESeq2 package in R. Genes with read counts ≥ 10 in at least half of the samples were retained for analysis. Three biological replicates were included for each treatment group. Differentially expressed genes (DEGs) were identified using the thresholds of adjusted p‐value (padj) < 0.05 and |log₂Fold Change| > 1. The resulting DEGs were subsequently used for functional annotation, enrichment analysis and integrative transcriptome–metabolome analyses.

5.10. Integrative Transcriptome–Metabolome Analysis and WGCNA

To investigate the relationship between transcriptional regulation and metabolite accumulation, weighted gene co‐expression network analysis (WGCNA) was performed using transcriptomic and metabolomic datasets. Raw RNA‐seq count matrices were imported into the DESeq2 package for preprocessing. Genes with read counts ≥ 10 in at least half of the samples were retained. Variance stabilizing transformation (VST) was subsequently applied to normalize gene expression data, and genes within the lowest 25% of expression variance were removed prior to network construction.

Differential metabolites identified from the phenylalanine and vitamin B6 metabolic pathways were used as trait data for integrative analysis. Metabolite abundance data were normalized by total abundance and log₂‐transformed before being matched with transcriptomic samples.

Weighted gene co‐expression networks were constructed using the WGCNA package (v1.72) in R. The soft‐thresholding power was determined using the pickSoftThreshold function by evaluating scale‐free topology across powers ranging from 1 to 20. A soft‐threshold power of 19 was selected when the scale‐free topology fit index (R ²) exceeded 0.8 and the mean connectivity was approximately 50. An unsigned network was constructed with a minimum module size of 30 genes and a module merging threshold (mergeCutHeight) of 0.25.

Module–trait relationships were assessed by calculating Pearson correlation coefficients between module eigengenes (MEs) and metabolite abundance levels. Multiple testing correction was performed using the Benjamini–Hochberg (BH) method. Modules with |r| > 0.5 and FDR < 0.05 were considered significantly associated with the target metabolites and selected as key modules.

Genes from significant modules were subsequently subjected to Kyoto Encyclopaedia of Genes and Genomes (KEGG) enrichment analysis to identify biological pathways associated with phenylalanine metabolism and vitamin B6 metabolism. Hub genes within key modules were further identified based on module membership (MM) and gene significance (GS), using thresholds of |MM| > 0.8 and |GS| > 0.5.

5.11. Measurement of Soluble Protein Content

Approximately 0.1 g of powdered grain sample was homogenized in 1 mL of phosphate buffer (pH 7.0) and centrifuged at 10000 × g for 10 min. The supernatant (900 μL) was transferred to a new 2.0 mL tube. The pellet was re‐extracted by homogenization in another 1 mL of phosphate buffer and centrifugation under the same conditions. The supernatants from both extraction steps were combined (~1.8 mL) to obtain the crude protein extract. All assays were conducted in triplicate. A 50 μL aliquot of the extract was mixed with 150 μL of distilled water and 800 μL of Bradford reagent (Coomassie Brilliant Blue G‐250) in a 1.5 mL tube. After incubation at room temperature for 2 min, 200 μL of the mixture was transferred to a microplate well. The absorbance was measured at 595 nm using a microplate reader. The protein concentration was determined from a standard curve (Bradford 1976; He et al. 2018; Hasanuzzaman et al. 2026).

5.12. Measurement of Vitamin B6 Content

Approximately 0.1 g of the powdered sample was accurately weighed. All assays were conducted in triplicate. The concentration of vitamin B6 (VB6) was determined using a VB6 enzymatic assay kit from Solarbio, according to the manufacturer's instructions. The absorbance was measured with a microplate reader, and the concentration was calculated based on the standard curve provided with the kit.

5.13. Measurement of Starch Content

Seeds of each rice line were ground to a fine powder, and total starch content was assayed using a commercial kit (BC0700, Solarbio, China) based on the anthrone–sulphuric acid method. Sample processing was performed strictly according to the manufacturer's instructions, including removal of interfering soluble sugars, acid hydrolysis of starch to glucose and colour development with anthrone reagent. The resulting chromogen was measured at 620 nm using a Shimadzu UV‐1800 UV–visible spectrophotometer. The calculation formula provided with the kit was used to convert the glucose‐equivalent absorbance readings into starch content.

5.14. Measurement of 2‐PE

Grains were frozen with liquid nitrogen and ground to a fine powder in a mortar. Approximately 2.0 g of powdered sample was transferred into a 20 mL crimp‐top vial, mixed with 10 mL of chromatography‐grade dichloromethane, sealed with a rubber septum and cap, and thoroughly shaken. The vial was placed in an ultrasonic bath, with water level above the vial body, and extracted at 40 kHz for 4 h. After extraction, the vial was cooled to room temperature and anhydrous sodium sulphite was added to remove residual water. The supernatant was collected using a 2 mL sterile syringe, filtered through a 0.22 μm organic syringe filter (13 mm) and transferred into a 1.5 mL headspace vial, ready for GC–MS.

GC–MS analysis was performed according to Tieman et al. (2006) and Qi et al. (2015), with minor modifications, using a Shimadzu GCMS‐QP2010 system (Japan). Separation was conducted on an SH‐Rxi‐5Sil MS column (30 m × 0.25 mm × 0.25 μm), with a maximum column temperature of 220°C. High‐purity helium (99.999%) was used as the carrier gas. Samples were analysed under constant‐pressure mode with splitless injection, and the injection volume was 2 μL. The mass spectrometer was operated in electron ionization mode at 70 eV, with an ion source temperature of 200°C, interface temperature of 250°C and quadrupole temperature of 150°C. Data were acquired in full‐scan mode over an m/z range of 35–160.

2‐PE was quantified using external standard calibration. A series of 2‐PE standards at different concentrations were analysed under the same conditions. The standard curve was generated by plotting peak area (y) against concentration (x), yielding the equation y = 1.47 × 10¹¹ x + 1.20 × 10⁹. Sample 2‐PE concentrations were calculated from the standard curve based on their peak areas and expressed as μg/kg fresh weight (FW). All chemical analysis results were presented as the mean of measurements from three independent biological replicates.

5.15. Statistical Analysis

Data were shown as mean ± SD from at least three biological replicates, with details provided in the figure legends. Statistical analyses were performed using SPSS (v21.0) and GraphPad Prism 10.1.2. For two‐group comparisons, two‐tailed Student's t‐tests were used. Multi‐group comparisons were analysed by one‐way ANOVA followed by Duncan's multiple range test, whereas comparisons between transgenic lines and their corresponding wild‐type controls were performed using an unpaired Student’s t‐test. Differences were considered statistically significant at p < 0.05.

Author Contributions

J.W. and J.L. supervised the project and J.W. and M.X. designed the experiments. X.C., W.Y., H.Y., Y.O., K.C., C.L., B.Z. and M.X. performed the experiments. X.C., W.Y., H.Y., Y.Q., M.X., W.W. and J.W. analysed the data. X.C. drafted the manuscript, and all authors contributed to editing and approving the final version.

Funding

This work was supported by Guangdong Basic Research Center of Excellence for Precise Breeding of Future Crops, FCBRCE‐202506. Guangdong Provincial ‘Pearl River Talent Program’ Innovation and Entrepreneurship Team Project, 2023ZT10N019. Natural Science Foundation of Guangdong Province, 2025A1515012347. Laboratory of Lingnan Modern Agriculture Project, NT2021001, NG2021004.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Sequence comparison of AtDHS2 and the four rice DHS isoforms.

Figure S2: Sequence comparison of PAAS proteins across species.

Figure S3: Sequence comparison of PAR proteins across species.

Figure S4: Agronomic traits of transgenic overexpression lines in the ZH11 background. (A, B) Plant height and tiller number per plant. (C) Seed setting rate. (D) Panicle length. (E, F) Number of primary and secondary branches per panicle. (G) Pasting temperature of rice flour. (H) GC–MS analysis of 2‐phenylethanol representative peak profiles in seeds. Data are shown as mean ± SD. Statistical significance was assessed using an unpaired t‐test (*p < 0.05, **p < 0.01).

Figure S5: Agronomic performance of transgenic lines in the YGY background. (A–C) Relative transcript levels of mOsDHS2, PAR and PAAS in transgenic seeds, normalized to OsActin1 (LOC_Os03g50885). (D, E) Representative images of brown rice and grains; scale bar: 1 cm. (F, G) Grain length and width. (H, I) Plant height and tiller number per plant. (J) Thousand‐grain weight (TGW). (K) Chalkiness rate. Data are shown as mean ± SD. Statistical significance was assessed using an unpaired t‐test (*p < 0.05, **p < 0.01, ***p < 0.001).

Figure S6: Quantification of 2‐phenylethanol (2‐PE) and 2‐acetyl‐1‐pyrroline (2‐AP) in YGY transgenic lines. (A–F) Chromatographic profiles of 2‐PE during grain filling in wild‐type and RALP transgenic lines. (G) 2‐PE content in grains measured by GC–MS. (H) 2‐AP content in grains measured by GC–MS. Data are shown as mean ± SD. Statistical significance was assessed using an unpaired t‐test (*p < 0.05, **p < 0.01, ***p < 0.001).

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.