Conserved trans-regulatory logic and evolving cis-modules drive the diversification of inducible diterpenoid biosynthetic gene clusters across the Oryzoideae Authors: Youming Liu, Shiho Tomiyama, Ikuya Motegi, Naoki Yamamoto, Aiping Zheng, Masaki Mori, Ryoka Kawahara-Miki, Yoshimasa Tsujii, Koji Miyamoto, Hiroyasu Furuumi, Yutaka Sato, Hideaki Nojiri, Kazunori Okada Published: 2026 ABSTRACT AbstractDiterpenoid phytoalexins are major inducible defense metabolites in rice, whose biosynthetic genes are organized into biosynthetic gene clusters (BGCs). While BGC architecture varies across the Oryzoideae, their shared inducibility is known; however, the regulatory logic and evolutionary trajectory governing this response remains poorly understood. Here, we investigate the evolution of clustered gene regulation by examining the interplay between conserved trans-acting factors and progressively diverged cis-regulatory elements. We demonstrate that diterpenoid BGC induction is governed by a stable trans-acting framework built upon rapidly evolving cis-regulatory landscapes. Comparative genomics reveals that the basic helix-loop-helix (bHLH)-type diterpenoid phytoalexin factor (DPF) is deeply conserved across the Oryzoideae and broader grass lineages. By contrast, the promoters of diterpenoid biosynthetic genes exhibit extensive sequence divergence, particularly between AA and non-AA genome lineages of Oryza, which represent distinct genome groups defined by cytogenetic and phylogenetic relationships. Despite this overall divergence, we identified discrete, gene-specific regulatory modules (evolutionary ‘islands’) that recur across species. These modules are enriched in canonical DPF-recognizable N-boxes and newly defined N-box-like elements (NLEs). Functional assays demonstrate that variation in the number and arrangement of these cis-elements quantitatively calibrates DPF-mediated activation, providing a mechanistic basis for species-specific differences in metabolic inducibility. Together, our findings elucidate how BGC regulatory systems diversify through a synergy of ‘trans stability and cis plasticity’, offering a framework for understanding the evolution of inducible specialized metabolism in plants.Keywords: biosynthetic gene clusters, cis-regulatory evolution, diterpenoid phytoalexins, diterpenoid phytoalexin factor (DPF), transcriptional regulation, Oryzoideae Introduction Phytoalexins are low-molecular-weight secondary metabolites in plants that are synthesized and accumulated in response to biotic or abiotic stress and exhibit broad-spectrum antimicrobial activity (Ahuja et al. 2012). The biosynthetic genes for certain secondary metabolites, including terpenoids, are frequently found to be in biosynthetic gene clusters (BGCs), enabling coordinated regulation and evolution of pathway components. In cultivated rice (Oryza sativa), two major diterpenoid phytoalexins, momilactones and phytocassanes, have been identified (Cartwright et al. 1981, Koga et al. 1995, Koga et al. 1997, Horie et al. 2015). Their biosynthetic genes are arranged into BGCs on chromosomes 4 and 2, respectively (Shimura et al. 2007, Q. Wang et al. 2012), and the corresponding metabolites accumulate in response to various environmental and artificial stresses, including ultraviolet irradiation, copper chloride treatment, and pathogen infection (Cartwright et al. 1981, Kodama et al. 1988a, 1988b, Hasegawa et al. 2010). Across wild rice species within the Oryzoideae, both complete and partial diterpenoid phytoalexin BGCs have been identified; however, their gene content, organization, and duplication patterns vary substantially among species (Fig. 1). While complete BGCs are capable of producing the corresponding products, BGCs in a subset of species exhibit structural variation and differences in gene copy number (Miyamoto et al. 2016, Wu et al. 2022, Yan et al. 2022, Priego-Cubero et al. 2025). However, the mechanisms governing the induction of BGC genes in wild rice species, as well as their evolutionary relationships to the regulatory systems characterized in cultivated rice, remain poorly understood. Schematic representation of BGCs on corresponding chromosomes 2 and 4 in cultivated rice (O. sativa) and wild relatives. (a) Momilactone BGC on chromosome 4. (b) Phytocassane BGCs on chromosome 2. Species abbreviations: Os, O. sativa; Or, O. rufipogon; Op, O. punctata; Oo, O. officinalis; Ob, O. brachyantha; Lp, L. perrieri. Details of biosynthetic gene symbols and their corresponding locus IDs for O. sativa, O. rufipogon, O. punctata, O. brachyantha, and L. perrieri were obtained from Miyamoto et al. (2016), whereas the corresponding information for O. officinalis was retrieved from Priego-Cubero et al. (2025). In O. sativa, the regulation of these clusters is mediated by a set of transcription factors (TFs). OsTGAP1, a gene encoding a bZIP-type TF, exhibits root-specific and stress-responsive expression (Okada et al. 2009, Miyamoto et al. 2014, Yoshida et al. 2017), and additional regulators, including several OsWRKY TFs, also contribute to this transcriptional network (Yokotani et al. 2013, Akagi et al. 2014, Fukushima et al. 2016, Wang et al. 2023). The bHLH-type TF OsDPF (O. sativa diterpenoid phytoalexin factor) has been shown to activate diterpenoid phytoalexin BGCs via an N-box (5′-CACGAG-3′)-dependent mechanism, and both loss- and gain-of-function of OsDPF markedly affect diterpenoid accumulation (Yamamura et al. 2015, Ishikawa et al. 2024, Liao et al. 2025). However, whether DPF orthologues in wild rice species play analogous regulatory roles, and how their target promoter architectures have evolved, remain unknown. More broadly, it is widely accepted that the evolution of gene regulation is driven largely by changes in upstream cis-regulatory sequences that reshape downstream expression profiles, whereas trans-acting regulators are often constrained by pleiotropic effects (Wray 2007, Carroll 2008). This cis-trans perspective provides a useful framework for understanding how inducible regulation can be maintained while allowing diversification in clustered specialized metabolism. However, for diterpenoid phytoalexin BGCs in rice relatives, the extent to which conserved TFs and evolving cis-regulatory modules jointly shape the induction of BGC genes has not been systematically examined. In this study, we investigated the regulatory architecture of diterpenoid phytoalexin BGCs across the Oryzoideae from a cis-trans perspective. Specifically, we examined the conservation of DPF orthologues and their genomic contexts, compared upstream regions of key cluster genes to identify conserved promoter modules and candidate DPF-recognition motifs, and evaluated the regulatory potential of these modules using functional assays in rice protoplasts and plants. By integrating comparative genomics with promoter dissection, this work establishes an experimental framework for elucidating how conserved trans-acting regulators interface with lineage-specific cis-regulatory evolution to control clustered specialized metabolism in grasses. Results Genomic and evolutionary conservation of DPF in the rice subfamily To examine the distribution and evolutionary conservation of DPF-related TFs in plants, we first conducted protein Basic Local Alignment Search Tool (BLASTP) searches using OsDPF as a query against proteomes from 14 representative plant species. Candidate homologues were identified using the thresholds of E-value <1e−10, alignment length > 100 amino acids, and sequence identity >40% (Fig. 2a). Using these criteria, DPF as well as the previously reported DPF-like1 and DPF-like2 proteins from O. sativa were consistently identified (Yamamura et al. 2015). DPF homologues were detected across multiple Oryzoideae species and more broadly within Poaceae. Reciprocal BLASTP analyses, performed using the same thresholds, confirmed OsDPF as the best hit for each candidate protein, supporting their assignment as putative orthologues within grasses. The result also showed that DPF and DPF-like1 cluster within the same clade (Fig. 2a). Notably, DPF-like1 proteins were detected only in AA-genome species, specifically O. sativa (cultivated rice) and O. rufipogon (a wild progenitor of cultivated rice). Given their close sequence similarity to OsDPF and OrDPF, as well as their restricted taxonomic distribution, DPF-like1 likely represents a paralogue derived from tandem duplication of DPF in the AA-genome lineage (Fig. 2b). Evolutionary conservation and genomic context of DPF in grasses. (a) Maximum-likelihood phylogenetic tree of DPF and DPF-like protein sequences from representative species. The phylogenetic tree was constructed with IQ-TREE. Branch support values were assessed using 1000 ultrafast bootstrap replicates, and branch lengths are scaled to amino acid substitutions per site. Proteins identified as reciprocal best BLAST hits to OsDPF are marked with dots preceding the protein IDs. (b) Microsynteny analysis of the chromosomal regions harbouring DPF within the Oryzoideae. Genes are represented as blocks along chromosomes, with orthologous genomic segments connected by lines. The DPF and DPF-like1 locus is highlighted. (c) Pairwise synteny comparisons between O. sativa and grass species outside the Oryzoideae, showing the genomic context of DPF homologues in non-Oryzoideae Poaceae species. We next investigated the genomic context of DPF homologues using microsynteny analysis with MCScanX across Oryzoideae. In nearly all examined species, the DPF homologue was located at a corresponding locus on chromosome 1, indicating strong conservation of the surrounding genomic region and high collinearity throughout diversification (Fig. 2b). An exception was observed in Oryza officinalis, where the DPF homologue was located on chromosome 3 (OoffiChr03g00862). The combination of reciprocal BLASTP analyses and conserved microsynteny supports the identification of DPF homologues as orthologues across the Oryzoideae. In addition, DPF-like1 was consistently located in close proximity to DPF on the chromosome and was restricted to O. sativa and O. rufipogon, further supporting its origin as a lineage-specific paralogue. Beyond Oryzoideae, DPF homologues were also identified in other Poaceae species (Fig. 2c), indicating that the presence of DPF is not restricted to the Oryzoideae. Notably, Echinochloa crus-galli, one of the few non-Oryza species reported to harbour a momilactone BGC (Guo et al. 2017), also contains a DPF homologue. This observation is consistent with the possibility that DPF-mediated transcriptional regulation may be associated with diterpenoid biosynthetic pathways beyond Oryza, a hypothesis further explored in subsequent analyses. Functional activity of DPF homologues from wild rice species To examine the functions of DPF homologues, we examined their temporal expression profiles under CuCl2 treatment. All DPF homologues exhibited CuCl2-induced upregulation, with expression levels generally peaking at 48-h post-treatment, although a few species displayed biphasic expression patterns (Fig. 3a). These results demonstrate that DPF orthologues exhibit inducibility across Oryzoideae. Functional identification of DPF homologues involved in diterpenoid phytoalexin biosynthesis. (a) Time-course expression of DPF homologues in wild rice species following CuCl2 treatment (means ± SE, n = 3). (b) Diterpenoid phytoalexin accumulation in O. sativa protoplasts expressing wild rice DPF homologues. Protoplast samples were collected 2 days after plasmid transformation. Asterisks indicate significant differences from GUS control (P < 0.05, 0.01, 0.001). Os, Or, Op, Oo, Ob, and Lp represent DPF and its homologues. checked all through the text and put corresponding amendments. To evaluate the functions of DPF homologues from wild rice species, their full-length cDNAs were transiently expressed in O. sativa protoplasts. These cDNAs were cloned into the pUbi-Rfa-Tnos vector under the control of the maize ubiquitin promoter. The production of diterpenoid phytoalexins in the transformed protoplasts was then analysed by liquid chromatography–tandem mass spectrometry (LC–MS/MS). As a result, similar to O. sativa DPF, all tested homologues induced the production of diterpenoid phytoalexins (Fig. 3b). To further verify the functional conservation of DPF in a wild rice cellular context, OoDPF was expressed in O. officinalis–derived protoplasts, which also resulted in detectable accumulation of diterpenoid phytoalexins (Supplementary Fig. S1). Collectively, these results demonstrate that DPF homologues from wild rice species not only share a common transcriptional induction pattern but are also functionally conserved as key activators of diterpenoid phytoalexin biosynthesis across the Oryzoideae. Exploration of conserved functional regions in DPF To identify functionally important regions within DPF homologues, conserved sequence segments were identified through comparative analyses of six DPF homologues from the Oryzoideae. A MEME analysis of all examined DPF homologues showed strong statistical significance (E-value ≤1e-30), leading to the identification of three conserved regions: at the N terminus, the C terminus, and within the bHLH domain (Fig. 4a; Supplementary Fig. S2). Identification and functional analysis of conserved regions in DPF. (a) Schematic representation of conserved regions in DPF identified by MEME analysis. (b) AlphaFold-predicted structure of DPF. The structure is shown in cartoon representation and coloured according to pLDDT confidence scores, with the colour scale representing confidence from lower to higher predicted accuracy. (c) Schematic diagrams of GFP-fused DPF constructs used for functional analysis. Full-length DPF and truncated variants lacking the N-terminal conserved region, the bHLH domain, or the C-terminal conserved region are shown. Black lines indicate deleted regions. FL, full length; ΔN, N-terminal deletion; ΔbHLH, bHLH domain deletion; ΔC, C-terminal deletion. (d) Accumulation of diterpenoid compounds in rice protoplasts expressing full-length or truncated DPF variants. Diterpenoid levels were quantified and compared with those in protoplasts transformed with a GUS control. Data represent means ± SD. Statistical significance was assessed using one-way analysis of variance (one-way ANOVA) followed by Tukey’s post hoc test. Different lower-case letters (a–c) indicate P < 0.05. Error bars represent mean ± SD (n = 3). The protein structure of OsDPF was predicted using AlphaFold 3. Although the predicted structure yielded a low global predicted Template Modeling score (pTM) score (reflecting the overall structural uncertainty), the protein was found to be largely composed of intrinsically disordered regions (~75%) (Fig. 4b). Such a largely disordered architecture is unlikely to be unique to DPF, as extensive intrinsic disorder is commonly observed in TFs, including bHLH proteins, particularly in regions outside the conserved DNA-binding and dimerization domain (Tarczewska and Greb-Markiewicz 2019). Notably, three evolutionarily conserved segments consistently exhibited high predicted Local Distance Difference Test (pLDDT) scores (a metric for local model confidence), suggesting the presence of locally stable and potentially functionally constrained structural elements embedded within a flexible framework. Based on the conserved sequence and structural features identified above, we next evaluated their functional relevance to gene induction and subcellular localization. GFP-fused OsDPF variants lacking each conserved segment were constructed and transiently expressed in rice protoplasts, followed by quantification of momilactone and phytocassane production (Fig. 4c and d). Deletion of any conserved region resulted in reduced accumulation of these phytoalexins. Notably, removal of either the N-terminal conserved region or the bHLH domain completely abolished production, whereas deletion of the C-terminal region caused only a moderate reduction. While the essential role of the bHLH domain in DNA binding is well established, the critical contribution of the conserved N-terminal region suggests an additional regulatory function independent of direct DNA binding. Furthermore, subcellular localization analysis revealed that all truncated OsDPF variants localized predominantly to the nucleus (Supplementary Fig. S3), indicating that the observed functional defects are unlikely to be attributed to altered nuclear targeting. A conserved cis-regulatory module underlying DPF-dependent transactivation in wild rice species Building on the conserved role of DPF as a trans-acting regulator across the Oryzoideae, we next investigated the conservation and divergence of cis-regulatory modules within their promoter regions. For this analysis, we collected promoter sequences of BGC genes from various Oryzoideae species. To increase taxon sampling and strengthen the robustness of our inferences, we expanded our scope beyond the five focal species to include Oryza coarctata, which was recently proposed to harbour a basal momilactone cluster, and Oryza australiensis (EE genome) (Zhao et al. 2023, Morrison et al. 2025). Using BLAST searches with O. sativa BGC cyclase and cytochrome P450 genes as queries, we identified putative phytocassane BGCs in both O. coarctata and O. australiensis (Supplementary Fig. S4). We extracted 3 kb upstream regions relative to the annotated 5′ untranslated region (5′ UTRs) of homologous biosynthetic genes and performed comparative promoter analyses using mVISTA (Frazer et al. 2004). Pairwise mVISTA comparisons revealed high upstream conservation between O. sativa (AA genome) and other AA-genome taxa (e.g. O. rufipogon), but pronounced divergence in non-AA genomes (i.e. other Oryza genome groups such as BB, CC, EE, and FF). Despite this global divergence, we identified discrete, gene-specific conserved promoter blocks (modules) that recur across species (Fig. 5a–d and Supplementary Fig. S5). In total, we delineated multiple conserved modules upstream of diterpenoid phytoalexin biosynthetic genes. Distribution of conserved upstream modules of diterpene biosynthetic cyclase genes. The 3-kb upstream regions of cyclase genes from each Oryzoideae species were extracted and compared with the corresponding regions of homologous genes in cultivated rice (O. sativa) using mVISTA with default parameters. Oc: O. coarctata, Oa: O. australiensis. Regions showing high sequence conservation were identified and displayed as peaks. Genes located within the phytocassane BGC and momilactone BGC are identified in the figure. The presence of cis-regulatory elements within each conserved module is indicated by different symbols: circles representing the N-box (5′-CACGAG-3′), and diamonds representing the NLE (5′-CADGAG-3′). Cis-element inspection revealed that most modules contain binding sites recognized by bHLH TFs, including the DPF-preferred N-box (5′-CACGAG-3′), and a novel N-box variant termed the N-box-like element (NLE, 5′-CADGAG-3′, where D = A, T, or G). Consistent with our findings on trans-acting factors, many genes retained highly conserved modules harbouring N-boxor NLE motifs across multiple genomes. Beyond bHLH-type sites, cis-element scans also identified putative binding sites for other stress- and hormone-responsive TF families within these conserved modules (Supplementary Fig. S6). Together, these data indicate that DPF-dependent transactivation is mediated by a mosaic of conserved promoter modules that persist despite extensive promoter-level divergence. The recurrent enrichment of N-boxes and NLEs across orthologous promoters provides a cis-regulatory substrate for the conserved, inducible activation of diterpenoid phytoalexin biosynthesis observed across the Oryzoideae. Functional activity of conserved modules and cis-elements mediating DPF-dependent transactivation To further elucidate the relationship between DPF and conserved promoter modules involved in diterpenoid metabolism, we examined the upstream regions of two diterpenoid cyclase genes: CPS2, which is responsible for phytocassane biosynthesis, and KSL4, which participates in momilactone biosynthesis. The DPF-recognition sites of CPS2 have been well characterized (Fig. 6a) (Yamamura et al. 2015, Liao et al. 2025). In contrast, KSL4 is characterized by the presence of a canonical N-box only in AA-genome species, suggesting that the presence or absence of this motif may influence the DPF responsiveness of KSL4 homologues (Fig. 7a). Characterization of the promoter module upstream of the CPS2 genes. (a) The left panel shows an mVISTA plot of the upstream sequences of the CPS2 lineage, in which grey shading indicates conserved region, defined as region 1. The right panel presents the multiple sequence alignment of this conserved region. Regions containing the NLE and N-box cis-elements are indicated by underlines. (b) Promoter activity assays of OoCPS2 and LpCPS2 upstream regions. The positions of the NLE and N-box cis-elements within the module are indicated by different symbols. Relative LUC/RLUC ratios were used for comparison. (c) Functional analysis of the OsCPS2 promoter mutants targeting the NLE and N-box cis-elements. The full-length OsDPF expression construct was used as the effector, and the ΔbHLH OsDPF construct was used as the negative-control effector. Statistical significance was determined by one-way ANOVA followed by Tukey’s post hoc test. Different lower-case letters (a–e) indicate P < 0.05. Error bars represent mean ± SD (n = 3). Characterization of the promoter module upstream of the KSL4 genes. (a) The left panel shows an mVISTA plot of the upstream sequences of the KSL4 lineage, in which shading indicates conserved regions, defined as region 1, region 2, and region 3. (b) The multiple sequence alignments of conserved region 1, region 2, and region 3. Regions containing the NLE and N-box cis-elements are indicated by underlines. (c–e) Promoter activity assays of OpKSL4, OoKSL4-a, and OoKSL4-b upstream regions. The positions of the NLE and N-box cis-elements within the module are indicated by different symbols. Circles for the N-box (5′-CACGAG-3′), and diamonds for the NLE (5′-CADGAG-3′). Relative LUC/RLUC ratios were used for comparison. The full-length OsDPF expression construct was used as the effector, and the ΔbHLH OsDPF construct was used as the negative-control effector. Statistical significance was determined by one-way ANOVA followed by Tukey’s post hoc test. Different lower-case letters (a–d) indicate P < 0.05. Error bars represent mean ± SD (n = 3). For CPS2, tandem N-box cis-elements were identified within the conserved promoter module. In several wild rice species, one of these N-boxes was replaced by an NLE (5′-CADGAG-3′). Using O. officinalis and Leersia perrieri as representatives, we cloned the upstream regions of their CPS2 homologues and tested their promoter activity in O. sativa protoplast reporter assays. In both cases, DPF strongly activated these promoters. However, mutation of either the NLE or the N-box almost completely abolished activation, indicating that DPF-mediated transcription in wild rice strictly requires these elements (Fig. 6b). We further performed transient expression assays using the native OoDPF effector together with the OoCPS2 promoter reporter in protoplasts of its native host, O. officinalis. These assays confirmed that both robust promoter activation and the requirement for either an N-box or an NLE are maintained in wild rice (Supplementary Fig. S7). To assess the functional redundancy of the tandem N-boxes and compare the transcriptional strength of the N-box and NLE, we generated a series of site-directed mutants. Replacement of one N-box with an NLE reduced reporter activity, and replacement of both N-boxes further diminished activation. In contrast, mutation of either cis-element to a non-functional sequence (5′-CCCGGG-3′) completely abolished promoter activation. This indicates that both sites within the module are functionally essential and that the N-box exhibits a higher apparent affinity for OsDPF than the NLE (Fig. 6c). For KSL4 homologues, multiple conserved sequences were identified in upstream modules; however, the positions of putative DPF-recognition sites were not conserved. Based on sequence alignments (Figs. 5 and 7a), a canonical N-box was identified only in AA-genome species. In the BB-genome species Oryza punctata, only an NLE was identified in the OpKSL4 promoter. In the CC-genome species O. officinalis, an NLE was identified in OoKSL4-a, whereas OoKSL4-b lacked both an N-box and an NLE. These sequence fragments were defined as regions 1–3 (Fig. 7b). These results indicate that the upstream regions of KSL4 homologues have undergone more extensive divergence than those of CPS2, suggesting a distinct evolutionary trajectory for KSL4 regulation. In OpKSL4, reporter assays revealed that two tandem NLEs within the conserved module conferred DPF-dependent activation, and their deletion markedly reduced promoter activity (Fig. 7c). Similarly, in OoKSL4-a, DPF recognized the NLE to activate downstream expression (Fig. 7d). Although these functionally relevant NLEs are embedded within conserved modules, their presence is not universal, suggesting lineage-specific turnover of NLE-containing cis-elements associated with DPF-mediated regulation and reflecting diversification within an otherwise conserved regulatory framework. Conversely, OoKSL4-b, which lacks both an N-box and an NLE, was non-responsive to DPF (Fig. 7e). Notably, OoKSL4-b is likely responsible for producing novel diterpenoid compounds rather than momilactone (Priego-Cubero et al. 2025, Hasegawa et al. 2026). When a 57-bp DPF-recognition module from the OsCPS2 promoter (containing two tandem N-boxes) was inserted upstream of OoKSL4-b, the modified promoter acquired DPF-responsive activity (Fig. 7e, Supplementary Fig. S8). This demonstrates that functional promoter modules are portable and can be transplanted to heterologous loci to incorporate them into the DPF regulon. Analysis of genome-edited plants lacking conserved modules in the promoter region of OsCPS2 and OsKSL4 We next examined the in planta relevance of these modules using CRISPR-Cas9-mediated genome editing in cultivated rice. Lines were generated to delete promoter segments encompassing the DPF-recognizing motifs in OsCPS2 and OsKSL4 (Fig. 8a). For OsKSL4, a 961-bp promoter segment (−1967 to −1006) was deleted (OsKSL4pΔ). For OsCPS2, a 146-bp promoter segment (−482 to −336) containing the tandem N-boxes was removed (OsCPS2pΔ). Both homozygous mutant of OsKSL4pΔ and OsCPS2pΔ genome-edited mutant lines were identified in the T1 generation (Supplementary Fig. S9), then homozygous T2 plants were used for gene expression and metabolite quantification. Effect of N-box and NLE deletion in the promoter regions of OsKSL4 and OsCPS2 on phytoalexin production. (a) 3-kb promoter regions of OsCPS2 and OsKSL4. Upstream editing sites targeted by two distinct gRNAs in separate CRISPR-Cas9 vectors for each gene in cultivated rice (O. sativa) are indicated. (b) Expression levels of OsCPS2 and OsKSL4 in WT and genome-edited rice leaves after CuCl2 treatment over time. Left: OsCPS2 expression in WT and OsCPS2pΔ genome-edited rice. Right: OsKSL4 expression in WT and OsKSL4pΔ genome-edited rice. Data are shown as mean ± SE, n = 3. (c) Accumulation of momilactones and phytocassanes following CuCl2 time-course treatment. Filled bars indicate accumulation in WT, and open bars indicate accumulation in the respective genome-edited rice plants. Data are shown as mean ± SE, n = 3. Asterisks indicate statistically significant differences between genome-edited and WT plants based on a one-tailed Student’s t-test: P < 0.05 (*), P < 0.01 (**), P < 0.001 (***). For each genome-edited genotype, one homozygous T2 line was analysed. When treated with CuCl2, transcription of OsCPS2 and OsKSL4 was significantly attenuated in the deletion mutants compared with wild-type (WT) plants (Fig. 8b). Consistently, the accumulation of momilactones and phytocassanes was markedly reduced in both mutants (Fig. 8c). These results indicate that the loss of these promoter modules diminishes transcriptional activation under stress. However, residual inducibility remained detectable, implying that additional regulatory elements or TFs also contribute to their activation in planta. Discussion In rice and its Oryzoideae relatives, the biosynthetic genes responsible for momilactone and phytocassane are organized in the genome as BGCs. The formation of gene clusters is thought to facilitate the coordinated expression of functionally related genes through their physical linkage, thereby promoting both transcriptional co-regulation and joint evolutionary retention (Osbourn 2010, Nützmann et al. 2016, Cawood and Ton 2025). At the scale of the Oryzoideae subfamily, these two diterpenoid BGCs have been stably maintained for a long evolutionary period (with the momilactone BGC likely originating around the divergence of Oryza coarctata, ~14.5 Mya, and the phytocassane BGC around the divergence of Zizania latifolia, ~26–30 Mya) (Yan et al. 2022). Such long-term persistence strongly suggests that maintenance of these BGCs requires not only conservation of enzymatic functions, but also a stable regulatory environment capable of sustaining their expression over deep evolutionary time. In cultivated rice, the TF OsDPF shows a strong co-expression pattern with the biosynthetic genes within both diterpenoid BGCs, and perturbation of OsDPF expression, either by knockout or overexpression, leads to pronounced changes in the accumulation of the corresponding metabolites (Yamamura et al. 2015). This regulatory behaviour indicates that OsDPF is not a redundant accessory factor, but rather functions as a central transcriptional activator within this regulatory module. Consistent with this notion, collinearity and orthologue identification analyses revealed that DPF homologues are widely conserved throughout the Poaceae, whereas no clear homologues were detected outside this family. This distribution pattern suggests that DPF likely originated early in Poaceae evolution and was subsequently recruited into the regulatory control of newly assembled or progressively refined diterpenoid gene clusters. This hypothesis is further supported by cross-species functional assays. Even in the more distantly related species L. perrieri (~14–16 Mya divergence from cultivated rice), the DPF homologue retained the ability to activate diterpenoid biosynthetic genes. Taken together with the phylogenetic distances involved, these results strongly imply that DPF has been subject to strong purifying selection over tens of millions of years, preserving its core transcriptional activation function. Such strong evolutionary constraints are consistent with the general properties of TFs occupying highly connected positions in regulatory networks, where functional mutations are more likely to induce widespread downstream perturbations. Similar evolutionary constraints have been documented for other plant TFs. A primary example is MYC2, a bHLH-type master regulator of jasmonate signalling that performs conserved roles in defence and growth regulation across angiosperms (Kazan and Manners 2013). Both the protein sequence and core function of MYC2 are highly constrained, as functional mutations often impose significant fitness costs due to growth–defence trade-offs. DPF likely belongs to an analogous class of functionally constrained trans-acting factors; its evolutionary stability provides a robust regulatory scaffold for complex metabolic pathways. When TF function is under strong evolutionary constraint, regulatory network flexibility is predominantly achieved through modification in cis-regulatory elements (King and Wilson 1975). In this study, systematic identification and functional characterization of putative DPF recognition sites revealed a hallmark of regulatory evolution within diterpenoid BGCs: cis-regulatory elements do not evolve in a strictly binary fashion (presence vs. absence) but instead undergo gradual fine-tuning while preserving basal transcriptional activation. For instance, sequence analysis of the OsCPS2 promoter identified a short conserved region containing two canonical N-boxes. Within the Oryzoideae, one of these N-boxes has frequently diverged into an NLE. Reporter assays demonstrated that DPF-mediated activation of CPS2 is dependent on this conserved region and can be driven by either the N-box or the NLE. Notably, OsDPF from cultivated rice exhibited higher transactivation via the canonical N-box than the NLE, suggesting that subtle cis-element variation is translated into quantitative differences in transcriptional output. Consistent with this interpretation, a comparative analysis of N-box occurrence within the promoters of diterpenoid biosynthetic genes highlights a distinct evolutionary trend. The frequency of N-box motifs was negligible (nearly 0%) in more distantly related species such as Oryza brachyantha, which is incapable of synthesizing momilactone or phytocassane and harbours only fragmentary BGCs. In contrast, N-box occurrence increased to over 50% in cultivated rice, ~60% in O. australiensis, and reached 100% in L. perrieri (Supplementary Fig. S10). These findings suggest an association between functional BGCs capable of end-product synthesis and the presence of N-box-mediated regulatory potential, although the observed pattern may reflect lineage-specific gain and/or loss of these motifs during diversification. Thus, rather than indicating a single conserved regulatory configuration, this pattern suggests that DPF recognition sites have undergone dynamic cis-regulatory changes across Oryzoideae lineages. A related yet more flexible regulatory scenario was observed for KSL4. In several wild rice species, canonical N-box motifs were absent from the KSL4 upstream region; nevertheless, DPF maintained its ability to activate expression by recognizing NLEs. This finding highlights a degree of binding plasticity in DPF that allows the regulatory network to remain functional despite substantial cis-regulatory divergence. Such cis-regulatory divergence may also influence metabolic output by altering which KSL4 homologues are transcriptionally incorporated into the DPF-responsive regulon. Because KSL4 catalyzes the formation of diterpene hydrocarbon skeletons from syn-CDP, differences in the DPF responsiveness of KSL4 homologues may affect not only inducibility but also the structural diversity of diterpene products. Consistent with this possibility, OoKSL4-b, corresponding to OoKSL4-1 in Hasegawa et al. 2026, lacks DPF-responsive N-box/NLE motifs in our analysis and was reported not to function as the major 9β-pimara-7,15-diene-producing KSL4 homologue, unlike OoKSL4-a/OoKSL4-2. While promoter sequences have diverged far more extensively than their corresponding coding regions, comparative analyses strikingly revealed that the DPF–BGC regulatory relationship remains deeply conserved across species. Even in cases like CPS2, where the majority of the 3 kb upstream region shows negligible sequence similarity, all species retained a highly compact transcriptional activation ‘island’ of only a few dozen base pairs. This region was not only evolutionarily conserved but also functionally portable: when fused to the promoter of a gene (OoKSL4-b) not normally regulated by DPF, it was sufficient to confer DPF-dependent activation. These observations indicate that cis-regulatory information can be preserved in a minimalist, modular form while the surrounding sequences remain free to diverge, thereby balancing regulatory stability with evolutionary flexibility. Taken together, this study utilizes the Oryzoideae subfamily as a model to elucidate the regulatory logic underlying the long-term maintenance of complex metabolic pathways. Highly conserved trans-acting factors serve as stable regulatory anchors, while cis-regulatory elements undergo constant variation, retention, and redeployment to fine-tune gene expression and metabolic output. Within the context of BGCs, this synergy between trans stability and cis plasticity minimizes the systemic risks associated with TF evolution while enabling the adaptive diversification of specialized metabolism. Materials and Methods Identification of DPF homologues and phylogenetic analysis The OsDPF protein sequence was used as a query to identify homologous sequences in representative 14 monocot and dicot genomes (O. sativa, O. rufipogin, O. punctata, O. officinalis, O. brachyantha, L. perrieri, Z. latifolia, E. crus-galli, Zea mays, Hordeum vulgare, Sorghum bicolor, Phyllostachys edulis, Ananas comosus, Arabidopsis thaliana) by BLASTP searches. Candidate sequences were collected based on sequence similarity, and reciprocal BLASTP analyses against the O. sativa proteome were performed to identify putative reciprocal best hits. Protein sequences were aligned using MAFFT v7 with the L-INS-i algorithm (--localpair --maxiterate 1000), which incorporates local pairwise alignment information and iterative refinement (Katoh et al. 2002). Poorly aligned regions were trimmed prior to phylogenetic analysis. Maximum-likelihood phylogenetic trees were constructed using IQ-TREE v2.4.0 (Minh et al. 2020). The best-fit amino acid substitution model was selected using ModelFinder according to the Bayesian Information Criterion. Branch support was evaluated using 1000 ultrafast bootstrap replicates and Shimodaira-Hasegawa-like approximate likelihood ratio tests (1000 replicates). Branch lengths are presented as the number of amino acid substitutions per site. Microsynteny analysis of the DPF locus To assess the conservation of genomic context surrounding DPF, microsynteny analyses were performed across Poaceae species. For each identified DPF orthologue, genomic regions spanning the DPF gene and its flanking genes were extracted from genome assemblies and corresponding gene annotation files. Pairwise and multi-species synteny was examined using MCScanX pipelines, and results were visualized to highlight conserved collinear blocks (Y. Wang et al. 2012). Identification of conserved sequence regions in DPF proteins Conserved sequence motifs in DPF proteins were identified using MEME (v5.5) (Bailey et al. 2015). Amino acid sequences of OsDPF and its homologues were analysed in protein mode with the zero or one occurrence per sequence model. The maximum number of motifs was set to three, with motif widths ranging from 6 to 150 amino acids. Motif discovery was performed using the classic objective function with a zero-order Markov background model. Other parameters were set to default values. Plant materials and growth conditions Seeds of O. brachyantha (accession W1711), O. rufipogon (accession W1943), O. punctata (accession W1514), O. officinalis (accession W0002), and plantlet of L. perrieri (W1529) were obtained from the National Institute of Genetics (Shizuoka, Japan). These plants were cultivated on soil in a biotron at 28°C with a 14-h light and 10-h dark condition until they vigorously grew. Cultivated rice Oryza sativa L. ‘Nipponbare’ and O. officinalis were used as the protoplast material for transient assays. The cultivated rice seeds were sterilized with 75% ethanol for 1 min and then rinsed with tap water. Subsequently, the sterilized seeds were incubated in sterile soil with an 8-h light and 16-h dark photoperiod at 28°C for 2 weeks. For O. officinalis materials, seeds were sterilized using the same procedure described above, and seedlings grown for one month after germination were used for protoplast preparation. For genome-edited plants, OsKSL4pΔ and OsCPS2pΔ were grown under the same conditions as cultivated rice. Rice transformation Rice transformation was performed according to the method of Toki et al. (2006). Dehulled seeds of cultivated rice were surface-sterilized with 70% (v/v) ethanol for 1 min, followed by two rounds of surface sterilization with a two-fold diluted sodium hypochlorite solution (available chlorine ≥5.0%) for 15 min each. After sterilization, seeds were thoroughly rinsed with a large volume of Milli-Q water until the smell of hypochlorite was completely removed. Inside a clean bench, excess moisture was completely removed from the seeds on sterile filter paper, and the seeds were placed on callus induction medium [N6 Basal Salt Mixture (1×), MS Vitamins (1×), 3% (w/v) sucrose, 2 mg/L 2,4-D, 0.03% (w/v) N-Z-Amine® A, 0.3% (w/v) proline, and 0.4% (w/v) Gelrite] with the embryo in contact with the medium. After approximately one week of incubation, the induced calli were collected. For transformation, Agrobacterium tumefaciens carrying the CRISPR/Cas9 plasmid pZH-gYSA-MMCas9 containing distinct guide RNA units was used. The bacterium was streaked on preculture medium containing rifampicin, gentamicin, and spectinomycin, and incubated at 28°C for 3 days. A single colony was picked under sterile conditions and suspended in 40 mL of Agrobacterium suspension medium (AAM). The induced calli were added to the bacterial suspension and gently mixed by inversion for 1.5 min. After removal of excess medium, the calli were air-dried on sterile filter paper and transferred to co-cultivation plates lined with filter paper moistened with 1 mL of AAM medium. After 3 days of co-cultivation in the dark, the calli were washed to remove attached Agrobacterium. Calli were first washed eight times with Milli-Q water, followed by two washes with an aqueous meropenem solution (final concentration 5 mg/mL), and finally washed once with Milli-Q water. After excess moisture was removed on filter paper, the calli were placed on selection medium containing hygromycin and meropenem. Approximately two weeks later, calli showing vigorous growth on the selection medium were selected, and deletion mutations in the upstream region were confirmed by PCR. PCR-positive calli were transferred to regeneration medium for root induction. After rooting, the regenerated plantlets were transferred to hormone-free medium for further growth, and subsequently transplanted into soil for continued cultivation. CuCl2 treatment For the expression analysis of DPF homologues in wild rice species, ~3–5 expanded leaves were collected from each wild rice plant and mixed to reduce bias due to differences in the responses of independent leaves. For stress treatment, 1 cm2 detached leaf blades from the mature plants were soaked in 0.5 mM CuCl2 solution at 28°C under light for 6–72 h. Mock (untreated) leaf blades were used as controls. These samples were immediately frozen in liquid nitrogen and stored at −80°C until use for RNA extraction and metabolite analysis. For genome-edited plants, 1-month-old leaves of WT, OsKSL4pΔ, and OsCPS2pΔ rice plants were subjected to the same treatment procedure for RNA extraction and metabolite analysis. Expression analysis Leaf blades from the wild rice plants were collected after 6, 12, 24, 48, and 72 h of CuCl2 treatment. Total RNA was extracted by Sepasol®-RNA I Super G (Nacalai tesque, Kyoto, Japan). cDNA was reverse transcribed using the PrimeScript™ RT reagent kit (Perfect Real Time) (TaKaRa, Shiga, Japan). The cDNA mixture was used for qRT-PCR using GoTaq® qPCR and RT-qPCR Systems (Promega, WI, USA) on a 7300 Real-Time PCR System (Applied Biosystems). The 2−△△CT method was used to calculate relative expression levels (Livak and Schmittgen 2001). The WT and transgenic rice plants OsKSL4pΔ and OsCPS2pΔ were collected after CuCl2 treatment to investigate the expression levels of OsCPS2 and OsKSL4 using the same methods. Plasmid construction for transient assays The homologues of five wild rice DPF from wild rice relatives were reverse transcription (RT)-PCR amplified and subsequently cloned into the cloning vector pENTR/D-TOPO (Invitrogen, MA, USA) using primer sets shown in Supplementary Table S1. The fragments were then recombined into the transient expression vector pUbi-Rfa-Tnos, which has the maize ubiquitin-1 promoter (pUbi) and the nopaline synthase terminator (Tnos) using the attL x attR recombination (LR) reaction (Gateway®). pUbi-β-glucuronidase (GUS)-Tnos (Chujo et al. 2014), in which the expression of GUS was controlled by the polyubiquitin promoter, served as a negative control. For the reporter gene assay, pUbi-Renilla luciferase (RLUC)-Tnos (Ogawa et al. 2017) expressing RLUC was used as the internal control. pGL4.1-CPS2p-0.4 kb and pGL3-KSL4p-2.0 kb, in which the promoter regions OsCPS2, and OsKSL4 in O. sativa were fused to the firefly luciferase gene individually, were used as reporter vectors. For O. sativa and wild rice CPS2 and KSL4, upstream regions with different lengths and their respective N-box or NLE mutagenesis were amplified and inserted into pGL4.10 using In-Fusion® Cloning, TaKaRa. For CRISPR/Cas9 construction, the guide RNA cloning vectors pU6gRNA and pZH_gYSA_MMCas9 were used (Mikami et al. 2015). Then, 20 bp sequences from −332 to −352 and −463 to −483 of the OsCPS2 promoter region and from −1003 to −1023 and −1962 to −1982 of the OsKSL4 promoter region (including Protospacer Adjacent Motif (PAM)) were selected as the target sites of Cas9 by using the guide RNA design tool CRISPR-P 2.0 (Liu et al. 2017). The DNA fragments were cloned into pU6gRNA using the restriction enzyme BbsI. These DNA fragments were subcloned into pZH_gYSA_MMCas9. The CRISPR/Cas9 plasmids were transformed into the A. tumefaciens strain EHA105 to transform calli generated from rice embryos. The genomic regions containing the Cas9 target sites were PCR amplified to confirm the mutations. Protoplast isolation and transient assay The method of protoplast isolation was primarily based on the protocol described by Zhang et al. (2011). In brief, the sheath portion of 2-week rice seedlings was dissected into strips, submerged in an enzymatic digestion buffer [1% Cellulase Onozuka™ R-10 (Yakult Pharmaceutical, Tokyo, Japan), 0.5% Macerozyme™ R-10 (Yakult Pharmaceutical), 0.1% bovine serum albumin (Sigma-Aldrich, MO, USA), 0.6 M mannitol, 10 mM 2-(N-morpholino) ethanesulphonic acid (MES) at pH 5.7, and 10 mM CaCl2], and then incubated for 4 h in the dark with gentle shaking. After enzymatic digestion, the residue was filtered, and protoplasts were isolated using W5 buffer (154 mM NaCl, 125 mM CaCl2, 5 mM KCl, and 2 mM MES at pH 5.7). In the transient phytoalexin measurement, polyethylene glycol (PEG)-mediated transformation was performed for plasmid introduction. For each sample, 10 μg of plasmid, 100 μL of protoplast solution (2 × 105 cells), 110 μL of 40% PEG 4000 solution (Sigma-Aldrich), 0.2 M mannitol, and 0.1 M CaCl2 were mixed and incubated in the dark for 15 min. Following transformation, the samples were incubated in W5 buffer and stored in the dark at 25°C for 2 days. For the reporter gene assay, 5 μg of effector, 5 μg of reporter, and 1 μg of the internal control were transformed into 2.0 × 105 cells using the PEG-mediated method described above. Following transformation, the samples were incubated in W5 buffer and stored overnight in the dark at 25°C. Subsequently, the samples were collected, and the signal intensities of firefly luciferase (LUC) and RLUC were quantified using the Dual-Luciferase® Reporter Assay System (Promega) on the GloMax® 20/20 Luminometer (Promega). Protoplast nuclei staining Protoplast nuclei staining was used, followed by Meadows and Potrykus 1981. Samples were collected by centrifuge, and the protoplast pellet was re-suspended in staining buffer [0.16 M CaCl2, 0.006% Triton X-100, 20 μg/mL Hoechst 33342 (Nacalai tesque, Kyoto, Japan), pH at 7.5] for 10 min, then samples were observed using a Keyence BZ-X700 microscope. Diterpenoid phytoalexin measurement For phytoalexin measurements in genome-edited rice leaves, CuCl2-treated leaf blades were submerged in 2 mL of 80% methanol at 4°C for 24 h, and 5 μL of the extract was subjected to LC–MS/MS analysis. For transient phytoalexin measurements, phytoalexin extraction was carried out using 100% methanol at 4°C for 6 h. The supernatant (5 μL) from the extracted samples was used for phytoalexin measurement (Miyamoto et al. 2016). Supplementary Material Acknowledgments We are grateful to Lena Steinbach and Daniel Gotman (Karlsruhe Institute of Technology) for their assistance with molecular cloning. We are grateful to Dr Matthew R. Shenton (Institute of Crop Science, NARO) for providing the O. officinalis genome sequence data before its publication. We also thank Dr Masaki Endo (Institute of Agrobiological Sciences, NARO, NARO) for providing the genome-editing vectors pU6gRNA and pZH_gYSA_MMCas9. The wild rice accessions used in this study were distributed from the National Institute of Genetics supported by the National Bioresource Project, MEXT, Japan. We would like to thank Editage (www.editage.jp) for English language editing. Contributor Information Youming Liu, Agro-Biotechnology Research Center, The University of Tokyo, 1-1-1 Yayoi, Bunkyo-ku, Tokyo 113-8657, Japan. Shiho Tomiyama, Agro-Biotechnology Research Center, The University of Tokyo, 1-1-1 Yayoi, Bunkyo-ku, Tokyo 113-8657, Japan. Ikuya Motegi, Agro-Biotechnology Research Center, The University of Tokyo, 1-1-1 Yayoi, Bunkyo-ku, Tokyo 113-8657, Japan. Naoki Yamamoto, College of Agronomy, Sichuan Agricultural University, No. 211 Huimin Road, Wenjiang District, Chengdu 611130, China; Key Laboratory of Southwest China Wildlife Resources Conservation (Ministry of Education), College of Life Science, China West Normal University, No. 1 Shida Road, Shunqing District, Nanchong 637000, China. Aiping Zheng, College of Agronomy, Sichuan Agricultural University, No. 211 Huimin Road, Wenjiang District, Chengdu 611130, China. Masaki Mori, Institute of Agrobiological Sciences, NARO, 2-1-2 Kannondai, Tsukuba, Ibaraki 305-8634, Japan. Ryoka Kawahara-Miki, Faculty of Applied Bioscience, Tokyo University of Agriculture, 1-1-1 Sakuragaoka, Setagaya-ku, Tokyo 156-8502, Japan. Yoshimasa Tsujii, Faculty of Applied Bioscience, Tokyo University of Agriculture, 1-1-1 Sakuragaoka, Setagaya-ku, Tokyo 156-8502, Japan. Koji Miyamoto, Department of Integrated Science and Engineering, Faculty of Science and Engineering, Teikyo University, 1-1 Toyosatodai, Utsunomiya, Tochigi 320-8551, Japan; Advanced Instrumental Analysis Center, Teikyo University, 1-1 Toyosatodai, Utsunomiya, Tochigi 320-8551, Japan. Hiroyasu Furuumi, National Institute of Genetics, 1111 Yata, Mishima, Shizuoka 411-8540, Japan. Yutaka Sato, National Institute of Genetics, 1111 Yata, Mishima, Shizuoka 411-8540, Japan. Hideaki Nojiri, Agro-Biotechnology Research Center, The University of Tokyo, 1-1-1 Yayoi, Bunkyo-ku, Tokyo 113-8657, Japan; Collaborative Research Institute for Innovative Microbiology, The University of Tokyo, 1-1-1 Yayoi, Bunkyo-ku, Tokyo 113-8657, Japan. Kazunori Okada, Agro-Biotechnology Research Center, The University of Tokyo, 1-1-1 Yayoi, Bunkyo-ku, Tokyo 113-8657, Japan. Author Contributions K.O. and H.N. conceived and supervised the study. Y.S. and H.F. prepared wild rice materials. Y.L., Y.T., M.K., A.Z., and N.Y. conducted genomic data in silico analysis. Y.L., I.M., S.T., R.K.M., and M.M. performed molecular cloning, genome editing, and functional analyses. Y.L. drafted the manuscript, and K.O. revised the manuscript with contributions from all authors. All the authors have read and approved the final version of the manuscript. Conflict of Interest The authors declare no conflicts of interest directly relevant to the content of this article. Funding This work was supported in part by JSPS KAKENHI grant no. 20H02922, 24K01690, Japan-Austria Research Cooperative Program between JSPS and FWF (grant number JPJSBP120202002), and NIG-JOINT (15A2021) to K.O. This work was supported by the Science and Technology Department of Sichuan Province (2020-YFH0161) and the Science and Technology Department of Chengdu City (2021-GH02-00085-HZ). Data Availability The data underlying this article are available in the GenBank Nucleotide Database at https://www.ncbi.nlm.nih.gov/genbank/ and can be accessed with OrDPF, PZ118706; OpDPF, PZ118707; OoDPF, PZ118708; ObDPF, PZ118709; LpDPF, PZ118710. References Associated Data Supplementary Materials Data Availability Statement The data underlying this article are available in the GenBank Nucleotide Database at https://www.ncbi.nlm.nih.gov/genbank/ and can be accessed with OrDPF, PZ118706; OpDPF, PZ118707; OoDPF, PZ118708; ObDPF, PZ118709; LpDPF, PZ118710.