Unlocking translational control of specialized metabolism in plants through 5′UTR structure

Abstract

Plant-specialized metabolites are essential for plant fitness and human health, with their biosynthesis pathways tightly regulated at multiple levels. However, the translational regulation of their biosynthesis remains poorly understood. Here, we reveal a 5′ untranslated region (5′UTR)–mediated translational mechanism that controls glucosinolate production in Arabidopsis. A forward genetic screen exploring the metabolic interaction between auxin and glucosinolates identified two dominant Arabidopsis alleles, each carrying a single-nucleotide substitution located 13–base pair apart within the 5′UTR of MYB28, a master regulator of aliphatic glucosinolate biosynthesis. These mutations markedly increase MYB28 protein abundance without affecting transcript levels, leading to enhanced glucosinolate production. Mutational profiling of the 5′UTR revealed that alterations in RNA tertiary structure influence translation efficiency, establishing a link between RNA conformation and metabolic output. Our findings uncover a previously uncharacterized layer of posttranscriptional regulation in plant-specialized metabolism and highlight the 5′UTR as a potential target for precision breeding to enhance crop performance and nutritional quality.

Point mutation in the MYB28 5′UTR enhances translation efficiency and drive increased glucosinolate production in plants.

INTRODUCTION

Plants produce millions of specialized metabolites, which are crucial for plant survival, and many of them are related to human health (1). One such example is a class of compounds called glucosinolates (2, 3). Glucosinolates are nitrogen- and sulfur-containing metabolites found in plants belonging to the order of Brassicales (4, 5), which includes economically important Brassica crops—such as cabbage, broccoli, radish, mustard, and kale—that humans consume in their daily diet (6, 7). Glucosinolates are plant defense compounds, as their hydrolyzed products are toxic to herbivores and pathogens. They are also known as health-promoting compounds due to their various health-beneficial properties, including anticancer and anti-inflammatory activities (8, 9). Therefore, enhancing glucosinolate contents increases nutritional values and disease resistance in Brassica crops (10). Hundreds of glucosinolate structures are detected in plants, and they are synthesized from amino acids (11). Arabidopsis thaliana, a model plant species, produces more than 30 glucosinolate structures, which are derived mainly from chain-elongated methionine (Met) or tryptophan (Trp) (12, 13). Met-derived glucosinolates are called aliphatic glucosinolates, and Trp-derived glucosinolates are called indole glucosinolates (14, 15). Glucosinolate biosynthesis, its regulation, and the transport of glucosinolates have been extensively studied (16, 17).

Glucosinolate biosynthesis begins with the conversion of amino acids to corresponding aldoximes, mediated by cytochrome P450 monooxygenases of the CYP79 family (CYP79). In Arabidopsis, CYP79F1 (AT1G16410) and CYP79F2 (AT1G16400) produce aliphatic aldoximes from chain-elongated Met (18), while CYP79B2 (AT4G39950) and CYP79B3 (AT2G22330) function redundantly to convert Trp to indole-3-acetaldoxime (IAOx) (19, 20). Consistently, silencing CYP79F1 and CYP79F2 results in nearly the absence of aliphatic glucosinolates (21), while cyp79b2 cyp79b3 double mutants lack indole glucosinolates (22). Next, two functionally redundant cytochrome P450 enzymes, CYP83A1 (AT4G13770, REF2) and CYP83B1 (AT4G31500, REF5), convert aldoximes into thiohydroximates, with CYP83A1 preferring aliphatic aldoximes, while CYP83B1 has 50 times higher activity toward IAOx than CYP83A1 (23). Following steps in glucosinolate core biosynthesis require a set of enzymes—including glutathione S-transferases (GSTs), γ-glutamyl peptidases (GGPs), C-S lyase (SUR1), UDP-glycosyltransferases (UGTs), and sulfotransferases (SOTs)—to convert intermediate compounds into core glucosinolate structures (12). It has been shown that specific transcription factors regulate the biosynthesis of specific types of glucosinolates (24). In Arabidopsis, MYB28, MYB29, and MYB76 act as transcription activators of aliphatic glucosinolate biosynthesis genes. Among them, MYB28 and MYB29 play predominant roles, as the myb28 myb29 double mutant completely lacks aliphatic glucosinolates, whereas the respective single mutants still produce aliphatic glucosinolates (25). MYB76 appears to function under specific conditions (26). In contrast, MYB34, MYB51, and MYB122 regulate the expression of indole glucosinolate biosynthesis genes (27, 28).

The indole glucosinolate pathway functions beyond the production of defense compounds. This biosynthetic route is intricately linked to auxin biosynthesis, as IAOx serves as an intermediate for indole-3-acetic acid (IAA) in addition to indole glucosinolates (29). IAA, a potent endogenous auxin, plays an essential role in regulating various aspects of plant growth and development (30). While the YUCCA pathway constitutes the primary route for IAA biosynthesis in plants (31), the IAOx-derived IAA pathway represents an alternative route that has been reported in both dicots and monocots (22). The conversion of ¹³C₆-labeled IAOx to ¹³C₆-IAA in Arabidopsis confirms this pathway’s functionality (32). CYP79B2 and CYP79B3 are transcriptionally induced under various biotic and abiotic stress conditions (20, 33, 34). Arabidopsis cyp79b2 cyp79b3 double mutants lacking CYP79B2 and CYP79B3 exhibit growth defects such as shortened hypocotyls and reduced IAA levels under elevated temperature conditions (26°C) (22). Conversely, the overexpression of CYP79B2 and CYP79B3 leads to increased IAA accumulation and characteristic high-auxin phenotypes even under normal growth conditions (22). Consistently, mutants defective in indole glucosinolate biosynthesis downstream of IAOx—such as sur1, ref5, and ugt74b1—display pronounced high-auxin phenotypes including elongated hypocotyls and epinasty leaves, accompanied by elevated IAA levels. This is attributable to the metabolic flux of IAOx being redirected predominantly toward the IAA biosynthesis (35–37). However, the enzymatic steps and intermediates connecting IAOx to IAA remain incompletely characterized, and the molecular basis of the regulatory mechanism that governs IAOx flux between defense-related metabolites and auxin biosynthesis is still largely unresolved.

To investigate the metabolic cross-talk between IAOx and indole glucosinolate biosynthesis, we conducted a forward genetic screen using the Arabidopsis ref5 mutant. ref5 is defective in CYP83B1, a key enzyme that converts IAOx into downstream indole glucosinolate intermediates (38). Through the suppressor screen, we identified two dominant mutants that rescue the high-auxin morphological phenotype in ref5. These mutants carry single-nucleotide substitutions in the 5′ untranslated region (5′ UTR) of MYB28, a positive regulator of aliphatic glucosinolate biosynthesis. The 5′ UTR of mRNA serves as a critical regulatory platform for post-transcriptional regulation, despite not being included in the final protein encoded by the main open reading frame (ORF) (39). This region regulates various facets of mRNA processing, such as transcript stability (40), subcellular localization (41), and, importantly, translational efficiency (42). Yet, how specific 5′ UTR features contribute to metabolic feedback regulation in plants remains largely unknown (43).

Strikingly, the identified 5′ UTR mutations in MYB28 enhance MYB28 protein accumulation without affecting transcript abundance, resulting in elevated glucosinolate levels and restored auxin homeostasis in ref5. This unexpected finding uncover another layer of post-transcriptional regulation in plant specialized metabolism and highlight the 5′UTR’s potential as a target for engineering beneficial plant compounds.

RESULTS

Isolation and genetic characterization of ref5 suppressors

To elucidate the mechanisms underlying the biosynthesis of IAA derived from IAOx, an intermediate of indole glucosinolates (Fig. 1A), we performed an unbiased suppressor screen using an ethylmethanesulfonate (EMS)–mutagenized population of ref5 (35). Approximately 150,000 ref5 seeds were mutagenized and sown across 48 flats, with seeds from each flat pooled for subsequent analysis. The ref5 mutant displays a distinctive epinastic leaf morphology caused by elevated IAA levels (fig. S1). We hypothesized that ref5 suppressor mutants carrying mutations in genes involved in auxin metabolism might exhibit a reduced epinastic leaf morphology compared to ref5. To identify these suppressors, we screened M2 plants grown on soil for altered leaf morphology. From a screen of ~30,000 M2 plants, we isolated two independent suppressor lines (#19-6 and #30-3, originating from flats #19 and #30). These lines were selected because their leaf morphology differed from the characteristic ref5 phenotypes, such as narrow, downward-curled leaves and elongated hypocotyls (Fig. 1B and fig. S1). The restored high-auxin–associated morphological traits (hereafter referred to as the rhax phenotype) in the suppressors correlate with reduced IAA content in these suppressors compared to ref5 (Fig. 1C).

Fig. 1. Isolation and genetic characterization of ref5 suppressors.

(A) Schematic of IAOx-mediated metabolic network in A. thaliana and the major IAA biosynthesis pathway. REF5 deficiency causes IAOx accumulation, leading to elevated IAA levels. (B) Representative 3-week-old plants of WT, ref5, #19-6, and #30-3. (C) IAA content in aerial tissues of 2-week-old WT, ref5, #19-6, and #30-3 plants (n = 3). Data are presented as means ± SD. The means were compared by one-way analysis of variance (ANOVA), and statistically significant differences (P < 0.05) were identified by Tukey’s test and indicated by letters to represent differences among groups. (D) Genetic segregation analysis of F1 and F2 populations derived from crosses between #19-6 × ref5, #30-3 × ref5, and #19-6 × #30-3, which exhibited rhax-like and ref5-like phenotypes. (E) Bulked segregant analysis (BSA) for identifying the #19-6 mutation on chromosome 5. The scatter plot compares SNP occurrence ratios between a rhax-looking pool (x axis) and a ref5-looking pool (y axis). The vertical dashed lines (0.6 to 0.7) and the horizontal dashed line (0) represent the thresholds used to select potential mutations. Red dots denote SNPs unique to the rhax-looking pool within the 0.6 to 0.7 threshold. (F) Summary of identified SNPs pinpointed by BSA that met the expected SNP occurrence ratios [corresponding to the red dots in (E)]. The table includes coexpression analysis showing the percentage of overlap between the top 50 coexpressed genes for each candidate (from the ATTED-II database) and genes that were differentially expressed in #19-6 (versus ref5). The bar graphs visualize this percentage overlap. (G) Heatmap of transcript expression levels [z-score transformed transcript per million (TPM)] for 12 BSA-identified candidate genes in WT, ref5, and #19-6. AT5G59100 is not expressed in our sample.

To understand the genetic characteristics of these newly identified mutations, #19-6 and #30-3 were crossed with ref5. All F₁ progeny exhibited the rhax phenotype, and F₂ populations segregated in a 3:1 ratio (rhax-like to ref5-like) (Fig. 1D), indicating that the suppressor phenotypes are caused by single dominant mutations. As #19-6 and #30-3 are dominant mutants with similar phenotypes, we performed an allelism test by crossing them and analyzing their progeny. F₁ plants from the cross between #19-6 and #30-3 uniformly exhibited the rhax phenotype, and their F₂ progeny showed no segregation, with all individuals displaying the rhax phenotype (Fig. 1D). This suggests that the two suppressors carry either closely linked mutations or, more likely, allelic variants.

Identification of mutations responsible for rhax phenotype

To identify the mutations responsible for the restored morphological phenotype observed in #19-6, bulk segregation analysis (BSA) was performed. Genomic DNA was extracted from two pooled F2 populations: one showing the rhax phenotype and the other showing the ref5 phenotype (Fig. 1D). From the BSA sequences, a total of 18,766 sequence variants were identified. Given that EMS primarily induces G-to-A or C-to-T base changes (44), a focused analysis identified 4860 mutations as EMS-induced single-nucleotide polymorphisms (SNPs). For a dominant mutation causing the rhax-like phenotype, the SNP occurrence ratio at the target site was expected to be ~0.66 in the rhax-looking pool and 0 in the ref5-looking pool. Of the 4860 EMS-induced SNPs, only 13 SNPs met these stringent criteria, exhibiting a zero SNP occurrence ratio in the ref5-looking pool and a ratio between 0.6 and 0.7 in the rhax-looking pool (Fig. 1E and fig. S2A). We found that 12 of the 13 candidate SNPs were located close together in chromosome 5 (Fig. 1, E and F).

As the mutation responsible for rhax phenotype follows a dominant trait, we hypothesized that this mutation activates the gene function. To determine the impact of the mutation on transcript changes of identified genes, RNA sequencing (RNA-seq) was carried out using wild-type (WT), ref5, and #19-6 plants. The analysis revealed substantial transcriptional alterations in #19-6 compared to ref5, with 993 genes found to be up-regulated and 885 genes down-regulated (fig. S2B). However, despite these widespread changes, none of the 12 candidate genes showed an alteration in their transcript levels in #19-6 compared to ref5 or WT (Fig. 1G). As the candidate gene itself exhibited no change in its transcript levels, we investigated the expression levels of genes coexpressed with each of the 12 candidate genes. First, we identified the top 50 coexpressed genes for each candidate gene using the ATTED-II database. Next, we compared the transcript levels of the top 50 coexpression genes in the #19-6 mutant to ref5. This analysis revealed that 60% of the top 50 coexpressed genes of AT5G61420 were significantly up-regulated in #19-6 compared to ref5 (Fig. 1F and fig. S2B). This suggests that the SNP in AT5G61420 likely activates its function, leading to the altered expression of its coexpressed genes.

Mutations in MYB28 5′UTR increase aliphatic glucosinolates

The gene AT5G61420 encodes MYB28. The 19-6 mutant carries a C-to-T SNP in the 5′UTR of MYB28, located 152–base pair (bp) upstream of the ATG translation start site of MYB28 isoform 2 (Fig. 2A). Given the earlier genetic evidence suggesting that 30-3 is likely an allele of 19-6, the 5′UTR region of MYB28 in 30-3 was subsequently sequenced. This analysis revealed an additional C-to-T SNP positioned 165-bp upstream of the ATG and 13-bp upstream of the 19-6 mutation site (Fig. 2A). Based on these findings, we renamed dominant ref5 suppressors 19-6 and 30-3 as ref5 myb28-1D and ref5 myb28-2D, respectively. This confirmed that both ref5 myb28-1D and ref5 myb28-2D mutants had different but closely located mutations within the MYB28 5′UTR. The identification of mutations in the 5′UTR, rather than coding regions, indicates a regulatory role of 5′UTR at the translational level, which represents a less common but gradually recognized mode of gene regulation (45, 46).

Fig. 2. The C-to-T mutations in the MYB28 5′UTR enhance aliphatic glucosinolate production in both WT and ref5 backgrounds.

(A) Schematic of MYB28 isoforms, MYB28.1 and MYB28.2, showing the location of the C-to-T mutations in ref5 myb28-1D (152-bp upstream of ATG) and ref5 myb28-2D (165-bp upstream of ATG), marked by red boxes. In the diagrams, blue boxes represent UTRs, orange boxes represent exons, and connecting lines represent introns. (B) Table of 30 genes that are both up-regulated in ref5 myb28-1D and coexpressed with MYB28. The table displays log₂ fold changes of gene expression with corresponding adjusted P values (in parentheses) from the RNA-seq analysis. (C) Quantification of three major aliphatic glucosinolates (3MSOP, 4MSOB, and 8MSOO) in 3-week-old aerial tissues of WT, ref5, ref5 myb28-1D, and ref5 myb28-2D (n = 3). Data are presented as means ± SD. (D) Representative photo of 3-week-old WT, myb28-1D, and myb28-2D plants. Control plants shown in Figs. 2D, 6B, and 7D, and fig. S12B are the same as they were grown from the same batch of plants. (E) 3MSOP, 4MSOB, and 8MSOO aliphatic glucosinolate levels in aerial tissues of 3-week-old WT, myb28-1D, and myb28-2D plants. Data are presented as means ± SD (n = 3). Data in the graphs were analyzed using one-way ANOVA, and statistically significant differences (P < 0.05) were determined by Tukey’s test. Different letters indicate significant differences among groups.

MYB28 is a master transcription activator of aliphatic glucosinolate biosynthesis genes (47). Our RNA-seq analysis indicated that most of the aliphatic glucosinolate biosynthesis genes were significantly up-regulated in ref5 myb28-1D compared to WT or ref5 (Fig. 2B). Consistently, the levels of three major aliphatic glucosinolates—3MSOP (3-methylsulfinylpropyl glucosinolate), 4MSOB (4-methylsulfinylbutyl glucosinolate), and 8MSOO (8-methylsulfinyloctyl glucosinolate)—were significantly increased in ref5 myb28-1D compared to WT and ref5 plants, with ref5 myb28-2D exhibiting comparable increases (Fig. 2C and fig. S3).

The ref5 myb28-1D and ref5 myb28-2D mutants are suppressors of ref5, which has a defect in REF5 functioning in Trp-derived indole glucosinolate biosynthesis. To determine whether the elevated aliphatic glucosinolate levels in ref5 myb28-1D and ref5 myb28-2D depend on the ref5 mutation, we introduced the phenotype-recovered myb28-1D and myb28-2D alleles into the WT background by crossing these suppressors with WT. The resulting myb28-1D and myb28-2D single mutants carry a functional REF5 but retain the myb28-1D and myb28-2D homozygous mutations. As shown in Fig. 2D, myb28-1D and myb28-2D single mutants are morphologically indistinguishable from WT. However, aliphatic glucosinolate levels in myb28-1D and myb28-2D were significantly higher than WT (Fig. 2E and fig. S3), suggesting that the SNPs in the 5′UTR of MYB28 are sufficient to enhance aliphatic glucosinolate production independently of the ref5 mutation. While analyzing the F2 segregating lines, we found that both heterozygous and homozygous myb28-1D mutants exhibited increased aliphatic glucosinolate levels compared to WT, indicating that the chemical phenotype is also dominant (fig. S4). Together, these findings reveal that the myb28-1D and myb28-2D mutations in the 5′UTR of MYB28 increase MYB28 function and aliphatic glucosinolate production without changing MYB28 transcript levels, likely by enhancing MYB28 protein translation.

A single C-to-T substitution in the MYB28 5′UTR enhances MYB28 protein accumulation

MYB28 is predicted to have two isoforms, MYB28.1 and MYB28.2, and the identified 5′UTR mutations can affect both isoforms if they are functional (Fig. 2A). To identify functional isoform(s), we performed complementation assays using the myb28 myb29 double mutant, which lacks aliphatic glucosinolates. Constructs containing the coding sequences (CDSs) of either MYB28.1 or MYB28.2, each fused to a hemagglutinin (HA) tag driven by the 35S promoter, were introduced into the myb28 myb29 double mutant. Seven independent T1 transgenic lines were obtained for each isoform. All seven transgenic lines expressing MYB28.2 restored aliphatic glucosinolate levels (3MSOP, 4MSOB, and 8MSOO) in mature siliques to WT levels, whereas none of the transgenic lines expressing MYB28.1 complemented the deficiency (fig. S5A). Reverse transcription polymerase chain reaction (RT-PCR) confirmed comparable transcript levels of both isoforms in the respective lines (fig. S5B), indicating that the phenotypic differences among these transgenic lines are not due to transcriptional variation. This suggests that MYB28.2 is the primary driver of aliphatic glucosinolate biosynthesis regulation, and the functional implications of 5′UTR mutations specifically regulate MYB28.2. In the rest of this study, MYB28 refers specifically to MYB28.2.

To test whether the C-to-T substitution in the MYB28 5′UTR enhances MYB28 protein levels, we generated transgenic myb28 myb29 plants expressing HA-tagged MYB28 under its native promoter. Two constructs were created: one containing the WT “C” allele (p2Kb^C::MYB28:HA) and the other harboring the mutant “T” allele (p2Kb^T::MYB28:HA) at position −152 bp in the 5′UTR, corresponding to the myb28-1D mutation position. Each construct consisted of a 2-kb promoter region (p2Kb) containing 5′UTR, driving the expression of MYB28 genomic DNA (gDNA), followed by a 3′ HA tag, a STOP codon, and the 3′UTR. These constructs were designed to compare the functional consequences of the C and T variants on MYB28 regulation and activity (Fig. 3A).

Fig. 3. A single C-to-T substitution in the MYB28 5′UTR enhances translation, resulting in increased MYB28 protein levels and elevated GFP reporter expression.

(A) Schematic of the MYB28 genomic locus (top) and expression constructs driven by the 2-kb native promoter (bottom), containing either the native (p2Kb^C::MYB28:HA) or modified (p2Kb^T::MYB28:HA) 5′UTR. (B) RT-PCR analysis of MYB28:HA transcript levels in 5-week-old rosettes of T1 transgenic lines expressing MYB28:HA with either the native (p2Kb^C) or modified (p2Kb^T) 5′UTR. ACTIN2 was used as an internal control. (C) Western blot analysis of MYB28-HA protein levels in 5-week-old rosettes of corresponding T1 transgenic lines. The expected protein size of MYB28-HA is 42.2 kDa. Ponceau staining confirms equal loading. Numbers below the blots indicate relative MYB28-HA band intensity normalized to Ponceau. (D) Levels of 3MSOP, 4MSOB, and 8MSOO aliphatic glucosinolate in mature siliques of 5-week-old T1 transgenic myb28 myb29 plants expressing either the native (p2Kb^C::MYB28:HA) or modified (p2Kb^T::MYB28:HA) 5′UTR constructs. n.d., not detected. (E) Schematic representation of the MYB28 5′UTR with GFP driven by the 35S promoter, showing expression under the control of either the native (p35S::5′UTR^C:GFP) or modified (p35S::5′UTR^T:GFP) 5′UTR. The yellow line in the 5′UTR indicates the site of mutation (152-bp upstream of the translation start codon). (F) RT-PCR analysis of GFP transcript levels in 3-week-old rosettes from T1 transgenic lines expressing GFP under the control of either the native (5′UTR^C) or modified (5′UTR^T) 5′UTR. ACTIN2 was used as an internal control. (G) Western blot analysis of GFP protein levels in corresponding T1 transgenic lines. The expected protein size of GFP is 27.0 kDa. Ponceau staining confirms equal loading. Numbers shown below the blots represent the relative GFP band intensity normalized to Ponceau.

Four independent transgenic lines were isolated from each construct pool, which showed comparable transcript levels of MYB28:HA among transgenic lines (Fig. 3B and fig. S6). However, MYB28-HA protein levels were significantly higher in the four transgenic lines expressing the mutant T allele (p2Kb^T::MYB28:HA) compared to those expressing the WT C allele (p2Kb^C::MYB28:HA) (Fig. 3C). When aliphatic glucosinolate levels were measured in mature siliques of T1 transgenic lines, the construct carrying the mutant T allele (p2Kb^T::MYB28:HA) substantially increased 3MSOP, 4MSOB, and 8MSOO levels compared to the construct carrying the WT C allele (p2Kb^C::MYB28:HA) (Fig. 3D). These results suggest that the single C-to-T mutation in the 5′UTR alone is sufficient to increase MYB28 protein levels without altering transcript levels, strongly supporting the notion that the myb28-1D mutation plays a role in the MYB28 regulation at the level of translation. To further determine the translational regulation mediated only by the MYB28 5′UTR or requiring other components, green fluorescent protein (GFP) reporter lines were constructed. In these lines, GFP expression was driven by the constitutive 35S promoter carrying either the native (p35S::5′UTR^C:GFP) or the mutated (p35S::5′UTR^T:GFP) MYB28 5′UTR (Fig. 3E). The mutation site was specifically introduced at the myb28-1D mutation position within the 5′UTR (Fig. 3E). Four independent lines expressing each construct were isolated (Fig. 3F and fig. S6). Although their GFP transcript levels showed variation among the lines, GFP protein levels were unequivocally higher in the lines expressing GFP carrying the mutated 5′UTR (p35S::5′UTR^T:GFP) compared to those with the native 5′UTR (p35S::5′UTR^C:GFP) (Fig. 3, F and G), suggesting that the 5′UTR alone can affect translation efficiency.

Mechanistic analysis of 5′UTR-mediated translational regulation

Several studies have shown the importance of the 5′UTR in gene function (39, 48). One mechanism involves upstream open reading frames (uORFs), which attenuate translation by sequestering the ribosomes and delaying main ORF initiation (49). We found that MYB28 5′UTR contains a predicted uORF (50), where both myb28-1D and myb28-2D mutations are located. However, neither mutation disrupts the CDS of the uORF as both result in the same uORF product size despite causing S31F and L27F amino acid substitutions of the uORF-encoded peptide, respectively. These findings suggest that enhanced translation in myb28-1D and myb28-2D mutants may not be attributable to the recognition of the uORF for translation.

We further generated the myb28^uORF-Δ3bp mutant using the CRISPR-Cas9 system. This mutant carries a 3-bp deletion (Δ3bp) mutation located 30-bp downstream of the uORF translation start site (Fig. 4A), resulting in a uORF translation product shortened by one amino acid. The myb28^uORF-Δ3bp mutant developed normally and was phenotypically indistinguishable from WT but had significantly higher levels of aliphatic glucosinolates (3MSOP, 4MSOB, and 8MSOO) in 3-week-old whole aerial tissues compared to WT (Fig. 4, B and C).

Fig. 4. A 3-bp deletion within MYB28 uORF in the WT background leads to increased aliphatic glucosinolate levels.

(A) Schematic diagram of the MYB28 5′UTR showing the 3-bp deletion within the uORF (myb28^uORF-Δ3bp). Below are sequencing chromatograms confirming the deletion. The green box represents the 117 bp uORF of MYB28. Asterisks (*) and black arrows indicate the myb28-1D and myb28-2D mutation positions. (B) Representative photo of 3-week-old myb28^uORF-Δ3bp and WT plants. (C) Levels of 3MSOP, 4MSOB, and 8MSOO aliphatic glucosinolate in 3-week-old aerial tissues of myb28^uORF-Δ3bp compared to WT. Data are presented as means ± SD (n = 3). Student’s t test was used to calculate statistical differences. **P < 0.01 and *P < 0.05. (D) Tertiary structure of the WT (native) MYB28 5′UTR as determined by dimethyl sulfate probing and predicted structures for the C228U, C241U, and Δ3bp variants. The mutation sites are highlighted by red arrows. The start codon AUG is highlighted in red font. Blue lines indicate predicted long-range base pair interactions (pseudoknots) between distant nucleotides. High-resolution RNA structure images are presented in fig. S9.

Given that all three mutants—myb28-1D, myb28-2D, and myb28^uORF-Δ3bp—can produce the same size of WT uORF encoded peptide or a 1–amino acid shorten one if translated, we hypothesized that the translational regulation mediated by the MYB28 5′UTR may depend on the integrity of its 5′UTR region, and the mutations in the MYB28 5′UTR may modulate translation efficiency through RNA structure. By using the public database FOLDAtlas of experimental dimethyl sulfate probing-based global RNA structures in WT Arabidopsis (51), we found that WT MYB28 5′UTR has a pseudoknot structure that normally represses the translation (Fig. 4D). In support of this pseudoknot structure, we generated two mutant 5′UTR versions, A303G that maintains the pseudoknot pairings and A303G/U353C that weakens the pseudoknot pairings by disrupting one base pair. By transfecting protoplasts generated from WT Arabidopsis, we showed that the A303G mutant UTR had poor efficiency in driving the translation of the GFP reporter, while the A303G/U353C mutant UTR had significantly higher translation efficiency (fig. S7). According to RNA structure prediction by the RNAStructure program (52), the C241U (myb28-1D) and C228U (myb28-2D) mutations alter the overall structure of MYB28 5′UTR and break the pseudoknot, which is in favor of translation. Similarly, the Δ3bp mutation also breaks the pseudoknot structure (Fig. 4D).

To further corroborate the link between the 5′UTR structure and the translation efficiency, we designed two additional mutants (C254G/U256A and A352U) for translation tests in WT Arabidopsis protoplasts (fig. S8A). C254G/U256A also breaks the pseudoknot, while the U256A introduces a premature stop codon that leaves out the last three residues (Lys-Ser-Asp) of the uORF-encoded peptide. On the other hand, the A352U mutation maintains the pseudoknot and adopts a long double-stranded structure that is unfavorable for ribosome activity. The GFP reporter gene was driven by different 5′UTRs, including the translational leader from tobacco etch virus in the original pRTL2 vector (53), as well as WT and mutated MYB28 5′UTRs. By normalizing the protein levels to the mRNA amounts, we found that C241U had the strongest translation efficiency, consistent with the myb28-1D phenotype and its impact on translation shown in fig. S3. The C254G/U256A mutant also increases translation efficiency compared to WT (fig. S8B). Both variants have a relatively less structured 5′UTR RNA in favor of translation (figs. S8A and S9). However, the A352U mutant, having a long double-stranded structure with the pseudoknot, showed significantly decreased translation efficiency compared to the WT (fig. S8B), suggesting a strong correlation between the MYB28 5′UTR RNA structure and the translation efficiency. To determine whether the uORF-encoded peptide is required for translation repression, we generated an additional 5′UTR variant containing a synonymous mutation (U227G). This mutation has a T-to-G mutation at position 227 of the 5′UTR while preserving the encoded valine if the uORF is translated (Fig. 5A). Although the amino acid sequence remains unchanged, the U227G substitution alters the inhibitory 5′UTR structure, making an overall open structure lacking the pseudoknot or long double-strand helices present in the native 5′UTR (Fig. 5B and fig. S10). In Arabidopsis protoplast transfection assay, the U227G 5′UTR increased GFP translation by approximately sevenfold, indicating that translation repression can occur independently of the uORF-encoded peptide, likely through structure features of the 5′UTR.

Fig. 5. A synonymous mutation altering the 5′UTR structure enhances translation efficiency.

(A) A synonymous mutation, U227G, keeps the amino acid valine in the uORF CDS. (B) However, the U227G mutation alters the inhibitory 5′UTR structure to an overall open structure, with no pseudoknot or long double-strand helices. The AUG start codon was highlighted in red font. The mutation G is highlighted in red font and by a red arrow. A high-resolution RNA structure image is presented in fig. S10. (C) This mutation enhances the GFP translation up to about sevenfold as compared with the function of the WT 5′UTR sequence. Both constructs were driven by the same CaMV 35S promoter and were expressed in Arabidopsis protoplasts. Data are presented as means ± SD (n = 3). The t test was used to calculate statistical differences. **P < 0.01 and *P < 0.05.

Metabolic interactions among auxin, indole glucosinolate, and aliphatic glucosinolate biosynthesis pathways

ref5 myb28-1D and ref5 myb28-2D mutants were originally isolated due to the restored high auxin phenotypes in ref5 (Fig. 1, A and B). It is well known that MYB28 activates the expression of aliphatic glucosinolate biosynthesis genes, but not indole glucosinolate biosynthesis genes (12, 16). To determine whether the activation of MYB28 is indeed responsible for the restored high auxin phenotypes in ref5 myb28 mutants, we overexpressed MYB28.1 or MYB28.2 in ref5 to test if the activation of MYB28 can restore high auxin phenotype of ref5 phenocopying ref5 myb28-1D. We isolated four independent lines from each construct as they showed comparable expression levels (fig. S11, A and B). All ref5 plants overexpressing MYB28.2 (#11 to #14) showed a restored high auxin phenotype similar to ref5 myb28-1D, while MYB28.1 overexpression lines (#1 to #4) were indistinguishable from ref5 (fig. S11A), which is consistent with functional complementation of myb28 mby29 by overexpression of MYB28.2 but not by MYB28.1 (fig. S5A). This result suggests that the activation of MYB28 can abolish the high auxin phenotype in ref5. We then disrupted MYB28 function in the ref5 myb28-1D background using a CRISPR-Cas9 construct targeting an exon of MYB28. Three MYB28 CRISPR-mediated mutations were successfully generated: CRISPR-1, which has a −148-bp deletion; CRISPR-2, which has one base pair deleted; and CRISPR-3, which has one base pair inserted. Both 1-bp insertion and deletion mutations resulted in an early translation termination site, effectively creating null alleles of MYB28 (Fig. 6A). All three myb28 mutations reversed the rhax phenotype of ref5 myb28-1D to a high-auxin morphological phenotype similar to the original ref5 mutant (Fig. 6B). Consistently, IAA levels in these mutants were higher than in ref5 myb28-1D or WT but comparable to those observed in ref5 (Fig. 6C). ref5 myb28 double mutants that were generated using the CRISPR system showed increased IAA contents compared to myb28 and ref5, whereas the myb28 single mutant is indistinguishable from the WT (fig. S12, A and B). Together, our data suggests that the single-nucleotide change in the 5′UTR of MYB28 in ref5 myb28-1D is responsible for its recovered high-auxin phenotype and that MYB28 activation affects the auxin homeostasis, particularly in the ref5 background (fig. S12, B and C).

Fig. 6. Knockout of MYB28 reverses the suppression of ref5 phenotype by the myb28-1D mutation.

(A) Schematics of MYB28 CRISPR-mediated knockout lines in ref5 myb28-1D background. Highlighted boxes indicate 20-bp target site (yellow), PAM sequence (green), and the induced mutations (red). All three mutations result in a premature stop codon. (B) Phenotypes of 3-week-old plants including WT, ref5, ref5 myb28-1D, and the three CRISPR-mediated knockout lines (CRISPR-1/ref5 myb28-1D, CRISPR-2/ref5 myb28-1D, and CRISPR-3/ref5 myb28-1D). Control plants shown in Figs. 2D, 6B, and 7D, and fig. S12B are the same as they were grown from the same batch of plants. (C) Free IAA levels in aerial tissues of 2-week-old plants from the indicated genotypes. Data are presented as means ± SD (n = 3). Different letters indicate statistically significant differences as determined by one-way ANOVA with Tukey’s HSD test (P < 0.05).

To understand how the activation of MYB28 affects IAA metabolism, we analyzed IAOx, IAA, and indole glucosinolate content in ref5 myb28 mutants. In addition to higher levels of aliphatic glucosinolates (Fig. 2C), we found increased levels of indole-3-ylmethyl glucosinolate (I3M), a major indole glucosinolate, in ref5 myb28 mutants compared to ref5 (Fig. 7A), while they exhibited reduced IAOx and IAA levels compared to ref5, consistent with the suppression of the high-auxin phenotype. This indicates a metabolic rebalancing between indole glucosinolate biosynthesis and auxin homeostasis by the activation of MYB28. Given that MYB28 is a transcription activator, we examined the expression levels of glucosinolate biosynthesis genes (Fig. 7B). As expected, Most of the aliphatic glucosinolate biosynthesis genes are up-regulated in ref5 myb28-1D compared to WT and ref5 (Fig. 7B and fig. S13). However, the expression of CYP79B2 and REF5, the major indole glucosinolate biosynthesis genes, was not increased, although CYP79B3 expression is slightly up-regulated in ref5 myb28-1D (Fig. 7B). Notably, REF2 is substantially up-regulated in ref5 myb28-1D compared to ref5. It was shown that REF2 can take IAOx as a substrate despite its low activity compared to REF5 (29, 54). So, it is likely that the transcriptional activation of REF2 and downstream common glucosinolate biosynthesis genes, such as SUR1 and UGT74B1, increases indole glucosinolate production in ref5 myb28-1D, which reduces the IAOx flux toward IAA.

Fig. 7. The dominant myb28 mutations suppress the high-IAA phenotype of ref5 and increase indole glucosinolate I3Maccumulation, indicating a metabolic interaction between indole and aliphatic glucosinolate pathways.

(A) Comparison of IAOx, IAA, and I3M levels in ref5 myb28-1D and ref5 myb28-2D mutants relative to WT and ref5 controls. IAOx and IAA levels were measured in 2-week-old aerial parts; I3M glucosinolate level was measured in 3-week-old aerial tissue. Data are presented as means ± SD (n = 3). (B) Schematic representation of the indole glucosinolate and aliphatic glucosinolate biosynthesis pathways. Heatmaps show log₂ fold changes in individual gene expression from RNA-seq analysis, with a blue to red gradient (range: −2 to 4) indicating the magnitude of change. A bar graph represents REF2 transcript levels in WT, ref5, and ref5 myb28-1D. Data are presented as means ± SD (n = 3). (C) Comparison of IAOx and IAA levels in 2-week-old aerial parts of ref2, ref5 ref2, and ref5 myb28-1D ref2 mutants relative to WT, ref5, and ref5 myb28-1D (n = 3). Data are presented as means ± SD. (D) Morphological phenotypes of 3-week-old WT, ref5 ref2, and ref5 myb28-1D ref2 plants. Control plants shown in Figs. 2D, 6B, and 7D, and fig. S12B are the same as they were grown from the same batch of plants. Graph datasets were compared using one-way ANOVA, and statistically significant differences (P < 0.05) were identified by Tukey’s test. Different letters indicate significant differences among groups. FC, fold change.

To test this, we disrupted REF2 in both the ref5 myb28-1D and ref5 backgrounds. IAOx and IAA levels were substantially increased in ref5 myb28-1D ref2 and ref5 ref2 mutants compared to their respective controls, with the levels of IAOx and IAA being comparable in the two genotypes (Fig. 7C). Both the ref5 myb28-1D ref2 and ref5 ref2 mutants exhibited severe growth defects, likely due to auxin over accumulation (Fig. 7D). The fact that auxin levels and high-auxin phenotype in ref5 ref2 were not relieved by MYB28 activation in ref5 myb28-1D ref2 suggests that decreased auxin content in ref5 myb28-1D requires functional REF2. The restored high-auxin phenotype in ref5 myb28-1D is primarily due to increased IAOx flux toward indole glucosinolates from IAOx-derived IAA production. Together, these results suggest that the broad substrate specificity of REF2 plays a crucial role in maintaining the metabolic balance between auxin homeostasis and indole glucosinolate production under IAOx surplus conditions, in addition to aliphatic glucosinolate biosynthesis.

DISCUSSION

Plant-specialized metabolites make crucial contributions to plant fitness, mediating interactions with the environment, conferring resistance to biotic and abiotic stresses, and providing numerous benefits to human health and nutrition (55, 56). The biosynthesis of these metabolites is typically regulated through tightly coordinated transcriptional programs in response to internal or external stimuli, and decades of research have elucidated the mechanisms of transcriptional regulation underlying this metabolic flexibility (57–60). In this study, we identify a previously unknown layer of posttranscriptional regulation in plant-specialized metabolism, mediated by subtle sequence variation in the 5′UTR of a biosynthetic regulator gene. Through a forward genetic screen in the high-auxin ref5 background, we found two dominant myb28 alleles harboring single-nucleotide substitutions in the 5′UTR of MYB28, a key regulator of aliphatic glucosinolate biosynthesis (47). Detailed genetic analyses—including allelic dominance tests, native promoter complementation, and GFP reporter assays—all support that these mutations enhance translational efficiency without altering transcript abundance. Therefore, our data marked the direct influence of translational regulation on specialized metabolism in plants.

The 5′UTR influences translation efficiency, mRNA stability, and spatial-temporal expression patterns of genes in plants (61). Structural and sequence elements within the 5′UTR—such as uORFs, RNA structures, and cis-regulatory motifs—are known to fine-tune translation in response to developmental or environmental signals (61). These elements act through diverse mechanisms. For example, some 5′UTRs function as riboswitches that directly bind small metabolites, triggering conformational changes that repress translation, as observed in pathways involving thiamine or S-adenosylmethionine biosynthesis (62, 63). Other 5′UTRs encode catalytic structures, such as hammerhead ribozymes, which self-cleave mRNAs to modulate transcript stability (64). uORFs are another widely conserved class of regulatory elements, often acting as translational attenuators by interfering with ribosome scanning or reinitiation, leading to reduced translation of the downstream main CDS (65–69). In the case of MYB28, its 5′UTR contains a predicted uORF, and the identified myb28-1D (C241U) and myb28-2D (C228U) mutations both reside within this region. This suggested that these mutations may disrupt uORF-mediated repression, thereby enhancing MYB28 translation. Neither mutations disrupt the coding region of the uORF nor alter the predicted uORF-encoded peptide length despite introducing amino acid substitutions (S31F in myb28-1D and L27F in myb28-2D). Akin to the myb28-1D (C241U) and myb28-2D (C228U) mutations that constitutively activate translation without stimuli, a CRISPR-Cas9–generated myb28^uORF-Δ3bp mutant, in which three nucleotides were deleted immediately downstream of the uORF start codon, also exhibited elevated levels of aliphatic glucosinolates (Fig. 4, A to C). Given that uORF translation attenuates the translation of downstream main CDSs (65–67) and that the dominant myb28 and Δ3bp mutations, which constitutively activate the translation of MYB28, should not disrupt the CDS of the predicted uORF despite a single amino acid substitution or deletion, it is most parsimonious to reason that uORF-mediated attenuation, although possibly involved in translational repression of MYB28 in WT plants, is unlikely to contribute to the constitutive translational activation of MYB28 in myb28-1D, myb28-2D, and myb28^uORF-Δ3bp. Moreover, the synonymous mutation U227G residing in the uORF still significantly enhances the translation efficiency (Fig. 5), suggesting that these mutations somehow bypass the repressive regulation of the uORF.

Considering that the mutations in this study (myb28-1D, myb28-2D, Δ3bp, C254G/U256A, U227G, and A352U) spread throughout the 5′UTR, it is less likely that they affect specific RNA modifications to modulate translation efficiency. This led us to consider alternative modes of regulation, particularly the role of RNA structure. Experimental data deposited in the FOLDAtlas in vivo structural dataset (51) indicated that the WT MYB28 5′UTR forms a pseudoknot structure, a tertiary organization known to impede ribosome scanning and initiation (70, 71). Directly modulating the base pairs within the pseudoknot (A303G maintaining the pseudoknot and A303G/U353C weakening the pseudoknot) changed translation efficiency as predicted (Fig. 7). The computational modeling also indicated that the myb28-1D, myb28-2D, and myb28^Δ3bp mutations all disrupt this pseudoknot, correlating with enhanced translational efficiency (Fig. 4C). This hypothesis was further validated experimentally through systematic mutagenesis in Arabidopsis protoplasts, and mutants from the genetic screening (C228U and C241U), CRISPR-Cas activity (Δ3bp), and our designing (C254G/U256A and U227G) that disrupted the pseudoknot all significantly increased translation, whereas a designed mutation predicted to stabilize the pseudoknot (A352U) reduced translation efficiency (fig. S8A). On the basis of the functional mutagenesis analysis, we conclude that the translation promoting mutations all alter the repressive WT 5′UTR structure and result in constitutively activated translation.

The 5′UTR-mediated translational activation resulted in a substantial increase in MYB28 protein levels, causing overaccumulation of aliphatic glucosinolates. Although MYB28 is well known to regulate aliphatic glucosinolate biosynthesis and sulfur metabolism (72), elevated MYB28 activity also induces the expression of REF2, thereby diverting the shared intermediate IAOx away from IAA production toward indole glucosinolate biosynthesis. This metabolic rerouting restores auxin homeostasis and mitigates the characteristic growth defects of ref5 (Figs. 1, A and B and 6A). A previous study also showed that the overexpression of REF2 rescues both reduced indole glucosinolate content and severe dwarfism of cyp83b1 (ref5 knockout line) (29).

Our genetic study further shows that while ref5 myb28-1D restores high-auxin phenotypes of ref5, the activation of MYB28 fails to reduce auxin content in the absence of functional REF2. Both ref5 myb28-1D ref2 and ref5 ref2 overproduce IAA and display similarly severe growth defects, likely due to toxicity from IAA hyperaccumulation (Fig. 7D). These results indicate a pivotal role for REF2 in rebalancing metabolic pathways upon IAOx surplus conditions, attributable to its broad substrate specificity. REF2 can convert a wide range of aldoximes, including IAOx, to thiohydroximates, although it shows stronger activity toward aliphatic aldoximes. Some enzymes often show varying degrees of substrate promiscuity, being recognized as a key driver of metabolic diversification by enabling flexible substrate use (73). Our findings suggest that this promiscuity, particularly at metabolic branch points, may be shaped by selective pressures to regulate metabolic flux. The interplay between enzyme promiscuity and transcriptional control of enzyme-encoding genes may maintain metabolic homeostasis, thereby differentially influencing defense (glucosinolates) and growth (auxin).

Under our experiment conditions, only the MYB28.2 isoform complemented the glucosinolate production of the myb28 myb29 double mutant (fig. S5). MYB28.2 has both R2 and R3 DNA binding domains, while MYB28.1 carries a partial R3 domain without the R2 domain (fig. S14). This indicates that the MYB28.1 isoform is either dispensable or only functions in specific tissue or cell types or under specific conditions, suggesting different roles for MYB28 isoforms in regulating plant metabolic pathways.

Collectively, our results support a model in which MYB28 translation is regulated by a structural switch in the 5′UTR and mutations that destabilize the inhibitory pseudoknot bypass the translational repression, leading to increased MYB28 protein accumulation without stimuli. This mechanism expands our current understanding of 5′UTR functions in plants, demonstrating that they not only serve as passive ribosomal barriers but can also actively maintain repressive RNA structures that fine-tune metabolic pathway regulators (51, 52, 74). Given that the 5′UTR-mediated translation regulation offers a direct route to controlling enzyme abundance and activating biosynthetic processes, bypassing transcriptional regulation (46), it may serve as an efficient mechanism for reallocating the shared precursors under rapidly changing environmental conditions. Beyond that, 5′UTR may be a potential target for precision breeding of defense and health-promoting metabolites in crops, although we have not tested it here. In Arabidopsis, mutations disrupting a pseudoknot structure within the MYB28 5′UTR elevated aliphatic glucosinolate production up to 10-fold (Fig. 2, C and E), achieving quantitative enhancement without transcriptional overexpression. Crucially, these single-nucleotide substitutions increase translation and glucosinolate production in a WT background without affecting overall plant growth and morphology (Fig. 2, D and E).

The MYB28 5′UTR sequences are highly conserved across diverse Brassicaceae species (figs. S15 and S16). Notably, the nucleotide positions corresponding myb28-1D and myb28-2D are conserved in all species examined. Although the functional roles of the MYB28 5′UTRs in other Brassica crops remain to be determined, the strong sequence conservation and the substantial increase in glucosinolate accumulation caused by a single-base substitution in Arabidopsis suggest that the 5′UTR of MYB28 orthologs could be promising targets for precise metabolic enhancement.

In conclusion, this study demonstrates that 5′UTR-mediated translational regulation plays an essential role in coordinating MYB28 expression with the broader metabolic network in Arabidopsis. The dynamic reallocation of metabolic flux between growth (auxin) and defense (glucosinolate) pathways, driven by MYB28 translational control, showcases how non-CDSs contribute to balancing competing physiological demands in plants. The discovery of translational activation mediated by single base pair substitutions in the 5′UTR provides a transformative mechanism for precision breeding to acquire desirable traits in crops.

MATERIALS AND METHODS

Plant materials and growth conditions

A. thaliana ecotype Columbia-0 (Col-0) was used as the WT control. Arabidopsis mutant lines ref5 (ref5-1), ref2 (ref2-1), and ref5 ref2 (ref5-1 ref2-1) were genotyped following previously established methods (35). Briefly, the genotyping of the ref5 mutation used primers P45 and P46 (table S1) to amplify a 337-bp region in exon 2 of REF5, followed by digestion with the Xho I enzyme before separation of the PCR products using a 4% agarose gel via gel electrophoresis. Genotyping of ref2 used primers P43 and P44 (table S1) to amplify a 188-bp region of exon 1 of REF2, which was then digested with the Hae III enzyme. The PCR products were separated using 2% agarose gels via gel electrophoresis. Arabidopsis single-mutant lines myb28 (SALK_136312) and myb29 (SM_3_34316)—containing T-DNA insertion and transposon element, respectively—were available from the Arabidopsis Biological Resource Center. The double knockout mutant myb28 myb29 was previously generated by crossing myb28 and myb29 (75) after confirmation of lines through genotyping. Briefly, genotyping of the myb28 and myb29 mutations used primers P981 with P982 (table S1) and P983 with P984 (table S1), respectively, following the established protocol (76). Plants were cultivated under controlled environmental conditions as previously described (77), with a growth chamber maintained at 22° ± 1°C under a 16-hour light/8-hour dark photoperiod. White fluorescent light was provided at an intensity of 140 μmol m⁻² s⁻¹.

For seedlings grown on Murashige and Skoog (MS) medium, seeds were surface-sterilized in 20% (v/v) bleach supplemented with 0.005% Triton X-100 (Sigma-Aldrich, St. Louis, MO, USA) for 10 min, followed by five rinses with sterile distilled water. After sterilization, seeds were stratified at 4°C in the dark for 3 days to synchronize germination. Stratified seeds were then sown on one-half MS agar plates containing 1× MS basal salts, 2% (w/v) sucrose, and 0.7% (w/v) agar, with or without antibiotic supplements.

For soil-grown plants, seeds were cold-stratified at 4°C for 3 days before being sown directly onto moistened potting soil (Berger Germination, SKU: 07634-BM2). Plants were subsequently grown under the same environmental conditions described above.

Glucosinolate extraction

Glucosinolates were extracted from Arabidopsis plants using 50% (v/v) methanol with a tissue concentration of 100 mg/ml, followed by sonication for 15 min (78). The extracts were centrifuged at 10,000g for 10 min, and the supernatant was collected for further analysis. For desulfoglucosinolate quantification, samples were extracted in 50% methanol containing 125 μM sinigrin (internal standard). A 100-μl aliquot of the extract was mixed with 200 μl of QAE Sephadex solution (Sigma-Aldrich) and incubated for 5 min at room temperature. The Sephadex beads were then washed twice with 50% methanol and twice with Milli-Q water. Subsequently, 100 μl of Milli-Q water containing sulfatase (Sigma-Aldrich) was added, and samples were incubated at 37°C for 6 hours to facilitate desulfation. Chromatographic separation was performed using an UltiMate 3000 HPLC system (Thermo Fisher Scientific) equipped with a cooled autosampler (10°C) and a diode array detector. Desulfoglucosinolates were separated on an Acclaim 120 C18 column (150 mm by 4.6 mm, 5 μm; Thermo Fisher Scientific) with a mobile phase consisting of solvent A (Milli-Q water) and solvent B (100% acetonitrile) and a linear gradient program of solvent B 2 to 12% over 10 min, 12 to 15% over 15 min, 15 to 95% over 30 s, and 95% for over 2 min. The flow rate was set at 0.75 ml min⁻¹, and the column temperature was maintained at 40°C. The content of desulfo-glucosinolates was quantified using the peak area at 220 nm and response factors (79). Glucosinolate peaks were validated with authentic standards: 3MSOP (PhytoLab), 4MSOB (PhytoLab), 8MSOO (ChemFaces), and I3M (PhytoLab).

Purification and quantification of IAA and IAOx

IAA and IAOx were purified from Arabidopsis plants using established methods (77), with samples resuspended in Milli-Q water for analysis. Quantification was performed via ultrahigh-performance liquid chromatography tandem mass spectrometry (UHPLC-MS/MS) using a Vanquish Horizon UHPLC system (Thermo Fisher Scientific) equipped with an Eclipse Plus C18 column (2.1 mm by 50 mm, 1.8 μm; Agilent) and a TSQ Altis Triple Quadrupole MS/MS detector (Thermo Fisher Scientific).

The mass spectrometer operated in positive ionization mode (ESI+) with an ion spray voltage of 4800 V; ion transfer tube and vaporizer temperatures of 325° and 350°C, respectively; and sheath, aux, and sweep gas settings of 50, 9, and 1 (arbitrary units). Q1 and Q3 resolutions were set to 0.7 full width at half maximum with a collision-induced dissociation gas pressure of 1.5 mtorr and a scan cycle time of 0.8 s. For multiple reaction monitoring, transitions were optimized via direct infusion of authentic standards (IAA and [¹³C₆]-IAA from Cambridge Isotope Laboratories; IAOx was synthesized as in (77), with solvents consisting of 0.1% formic acid in Milli-Q water (A) and 100% acetonitrile (B), with a gradient elution (0 to 95% B over 4 min) at 0.4 ml min⁻¹, yielding the following parent→product ion transitions: IAA [mass/charge ratio (m/z) 175.983→130.071, collision energy of 18 V), [¹³C₆]-IAA (m/z 182.091→136, 18 V), and IAOx (m/z 175.087→158, 16 V).

Generation of transgenic lines

To generate MYB28 overexpression vectors, the CDSs for the MYB28.1 and MYB28.2 isoforms were amplified from Arabidopsis WT cDNA via PCR. The reactions used Q5 High-Fidelity DNA Polymerase (New England Biolabs, Ipswich, MA, USA) with the primers P845 and P835 (table S1) for MYB28.1 and P834 and P835 (table S1) for MYB28.2. The resulting PCR products were cloned into the Gateway entry vector pCC1155 using the Gateway BP Clonase II Enzyme mix (Thermo Fisher Scientific, Waltham, MA, USA). These entry clones were then individually recombined with the destination vector pCC995 using the Gateway LR Clonase II Enzyme mix (Thermo Fisher Scientific, Waltham, MA, USA) to yield the final constructs: p35S::MYB28.1:HA and p35S::MYB28.2:HA.

To generate MYB28 complementation lines, we made two constructs. The core of each construct consisted of the MYB28.2 genomic region with a C-terminal HA-tag, followed by the MYB28.2 3′UTR. This cassette was driven by a 2-kb MYB28.2 promoter region and the MYB28.2 5′UTR region, which differed between constructs by a single nucleotide, either C or T at 152-bp upstream from the start codon (within the 5′UTR). We performed PCR using Q5 High-Fidelity DNA Polymerase (New England Biolabs, Ipswich, MA, USA) using gDNA from either Arabidopsis WT or ref5 myb28-1D as a template to amplify the DNA fragments, including the 2-kb MYB28.2 promoter region (using primers P854 and P677 in table S1), two versions of the MYB28.2 gene, including the 5′UTR (primers P650 and P855 in table S1) and the 3′UTR regions (primers P856 and P857 in table S1). These three insert DNAs were assembled into the binary vector pCHF3 using NEBuilder HiFi DNA Assembly (New England Biolabs, Ipswich, MA, USA) after the linearization of the vector with Eco RI and Pst I. This process yielded the constructs p2Kb^C::MYB28gDNA:HA:3′UTR and p2Kb^T::MYB28gDNA:HA:3′UTR.

To generate GFP reporter lines, two constructs were designed, each containing an enhanced GFP (eGFP) sequence driven by the 35S promoter and the 5′UTR of MYB28.2, differing by a single nucleotide (C or T) at the 152-bp position upstream of the start codon. The two 5′UTR variants were amplified via PCR using Q5 High-Fidelity DNA Polymerase (New England Biolabs, Ipswich, MA, USA) from Arabidopsis WT (C variant) and ref5 myb28-1D (T variant) gDNA with primers P860 and P861 (table S1). The eGFP fragment was amplified from pYL322d1/N-eGFP (Addgene, plasmid #183151) with primers P862 and P863 (table S1). These amplified DNA fragments were then assembled into the binary vector pCHF3 under the 35S promoter using NEBuilder HiFi DNA Assembly Master Mix (New England Biolabs, Ipswich, MA, USA) after the linearization of the vector with Sac I and Pst I. This yielded the final reporter constructs p35S::5′UTR^C:GFP and p35S::5′UTR^T:GFP.

All constructs were verified by sequencing and introduced into Agrobacterium tumefaciens (GV3101). The overexpression vectors (p35S::MYB28.1:HA and p35S::MYB28.2:HA) were transformed into the ref5 and the myb28 myb29 double mutant background. The complementation vectors (p2Kb^C::MYB28gDNA:HA:3′UTR and p2Kb^T::MYB28gDNA:HA:3′UTR) were transformed into the myb28 myb29 double mutant, while the GFP reporter vectors (p35S::5′UTR^C:GFP and p35S::5′UTR^T:GFP) were transformed into Arabidopsis WT plants. Transformant lines containing either p35S::MYB28.1:HA or p35S::MYB28.2:HA overexpression vectors are referred to as p35S::MYB28.1:HA or p35S::MYB28.2:HA, and transformant lines containing either p2Kb^C::MYB28gDNA:HA:3′UTR or p2Kb^T::MYB28gDNA:HA:3′UTR complementation vectors are referred to as p2Kb^C::MYB28:HA and p2Kb^T::MYB28:HA. GFP reporter lines transformed with vectors p35S::5′UTR^C:GFP and p35S::5′UTR^T:GFP are referred to as p35S::5′UTR^C:GFP and p35S::5′UTR^T:GFP.

All transformations were performed individually via the Agrobacterium-mediated floral dip method (80). Briefly, open flowers were dipped into a solution containing 5% sucrose, 0.01% (v/v) Silwet L-77 (PhytoTech, S7777), and Agrobacterium harboring the specific construct. For selection, more than two dozen T1 plants transformed with the overexpression vectors were screened using 0.2% Basta (Rely 280, BASF, Florham Park, NJ, USA). The complementation and GFP reporter lines were screened for approximately two dozen T1 transformants on MS medium supplemented with 0.5% sucrose, 0.22% MS basal salts, 0.025% MES hydrate, 0.8% agar, and kanamycin (50 mg/liter).

Generation of MYB28 CRISPR lines

To generate the myb28^uORF-Δ3bp line, gRNAs targeting two loci within the uORF of MYB28.2 5′UTR were designed using the CHOPCHOP software (https://chopchop.cbu.uib.no/) (81). Specifically, sequences 5′-AGCAGTGTGGGTAAGATCAA-3′ and 5′-GGAAATTGTTTAATCAGATA-3′ were selected for gRNA-1 and gRNA-2, respectively. A DNA fragment containing a Bsa I restriction site, gRNA-1 sequence, a gRNA scaffold, U6p sequence, gRNA-2 sequence, and another Bsa I restriction site was synthesized (Twist Biosciences, South San Francisco, CA, USA). The DNA fragment was cloned into the binary vector pAGM55273 (Addgene, #153211) via Bsa I digestion and ligated using T4 DNA ligase (New England Biolabs). The resulting recombinant vector was verified by sequencing and transformed into A. tumefaciens (GV3101). WT Arabidopsis plants were then transformed via the floral dip method (80), where open flowers were dipped in a solution containing 5% sucrose, 0.01% Silwet L-77 (PhytoTech, S7777), and Agrobacterium carrying the construct. Transformant plants were selected on MS medium supplemented with 0.5% sucrose, 0.22% MS basal salts, 0.025% MES hydrate, 0.8% agar, and kanamycin (50 mg/liter).

To generate MYB28 CRISPR lines, a gRNA against a target sequence 5′-GGCTTCTAGTTCCAACCCTA-3′, designed by CHOPCHOP, located within the second and third exons of MYB28.1 and MYB28.2, was used. The binary vector pHSE401 (Addgene, plasmid #62201) was digested with Bsa I, and the MYB28 target sequence was then inserted into a linearized pHSE401 vector using the annealed product of oligonucleotides P640 and P641 with T4 ligase (New England Biolabs), resulting in the U6-26P::MYB28gRNA construct. The U6-26P::MYB28gRNA construct was confirmed by sequencing and introduced into Agrobacterium (GV3101). The ref5 and ref5 myb28-1D mutants were transformed via floral dipping (80). Transformant plants were selected on the same kanamycin-containing MS medium as described above.

Generating myb28-1D and myb28-2D

To introduce the dominant myb28 mutations into a WT background, we crossed both ref5 myb28-1D and ref5 myb28-2D mutants separately with Arabidopsis WT plants. The F1 progeny from these crosses were allowed to self-pollinate, and we screened the F2 generation for plants carrying the myb28-1D and myb28-2D mutations. For mutation verification, we amplified a 619-bp genomic region containing the 5′UTR and first exon of MYB28.2 using gene-specific primers, P662 and P663 (table S1). PCR products were purified and sent to Genewiz (Plainfield, NJ, USA) for Sanger sequencing. Sequence analysis confirmed the presence of the myb28-1D and myb28-2D mutations in the selected lines.

Generating ref5 myb28-1D ref2

To generate ref5 myb28-1D ref2, ref5 myb28-1D was crossed with ref2. The F1 population was allowed to self, and F2 individuals were selected for genotyping for the ref5, myb28-1D, and ref2 mutations. Genotyping of ref5 and ref2 mutations occurred as described above. Genotyping of the myb28-1D mutation used primers P662 and P663 (table S1) to amplify the 5′UTR region of MYB28.2 and then digested with the Bgl II enzyme before separation of PCR products using 3% agarose gel via gel electrophoresis.

Genomic DNA preparation and bulk segregant analysis

Two groups of plants were collected from F2 segregation populations of ref5 crossed with #19-6 (ref5 myb28-1D). A total of 73 ref5-looking plants and 235 rhax-looking plants were gathered into two separate tubes and immediately frozen in liquid nitrogen. The frozen samples were ground, and gDNA was extracted using CTAB method. Briefly, the extracted DNA was incubated with ribonuclease A (10 mg/ml), then mixed with an equal volume of chloroform:isoamyl alcohol solution (24:1), and further incubated with 0.7 vol of isopropanol before being washed with 70% ethanol, centrifuged, dried, and resuspended in tris-EDTA buffer. The sequencing of the BSA samples was performed at Novogene Corporation (Sacramento, CA, USA) using 150-bp paired-end reads.

The low-quality reads of the BSA sequences were trimmed and filtered using AdapterRemoval with default options. The SIMPLE pipeline (82) was used to obtain mutation sites and the mutation ratio at each site. Arabidopsis TAIR10 was used as the reference genome, and the filtered BSA sequences were mapped to the Arabidopsis genome sequence using the SIMPLE pipeline. The mutation was considered dominant for phenotype. The expected ratio of mutation versus reference type was 2:1 for the rhax-looking pool and 0:1 for the ref5-looking pool. The ratio of the number of mapped reads to the reference and the mutant genome type was used to isolate candidate mutation sites. If the proportion of the mutation is between 0.6 and 0.7 in the rhax-looking pool and 0 in the ref5-looking pool, then the mutation sites were considered candidates closely linked to the target gene.

RNA preparation, RNA-seq, and DEG analysis

Total RNA was extracted from the whole aerial parts of 2-week-old WT, ref5, and ref5 myb28-1D in triplicate using the TRIzol method according to the manufacturer’s protocol (15596018, Thermo Fisher Scientific). Briefly, frozen samples were ground and centrifuged with 1 ml of TRIzol. The supernatant was transferred and mixed with 200 μl of chloroform before centrifuging and transferring the upper aqueous layer. Then, RNA was precipitated with 500 μl of isopropanol, followed by centrifugation to pellet, washing with 75% ethanol, drying, and resuspending in diethyl pyrocarbonate–treated water.

Raw RNA-seq reads were initially subjected to quality control to remove low-quality bases using AdapterRemoval (83), with default options. The quality-filtered reads were then aligned to the Arabidopsis transcriptome sequences (TAIR10) using the RSEM pipeline (84), with Bowtie2. Through this pipeline, both the number of mapped reads and transcript per million values were obtained. To identify differentially expressed genes (DEGs), the number of mapped reads was analyzed using DESeq2. In this analysis, we defined the DEGs as those meeting a significance criterion of an adjusted P value < 0.05.

Reverse transcription polymerase chain reaction

Total RNA was extracted from the rosette leaves of 5-week-old WT, myb28 myb29 double mutant, and MYB28 complementation lines (p2Kb^C::MYB28:HA and p2Kb^T::MYB28:HA), as well as from 3-week-old WT and GFP reporter lines (p35S::5′UTR^C:GFP and p35S::5′UTR^T:GFP), using a previously described RNA preparation method. Total RNA was treated with deoxyribonuclease I (Ambion TURBO DNA-free, #AM1907) before cDNA synthesis with oligo(dT) primers. For MYB28 complementation analysis, PCR was run for 25 cycles with 53°C annealing for MYB28:HA (P751/P989) (table S1) and 55°C for ACTIN2 control (P889/P890) (table S1). The GFP reporter analysis used 30 cycles with 55°C annealing for both GFP (P895/P896) (table S1) and ACTIN2 control. All reactions began with 95°C for 3 min, followed by cycling steps (95°C for 30 s, appropriate annealing temperature for 30 s, 72°C for 1 min), and ended with 72°C for 5 min final extension. PCR products were separated on 1% agarose gels for electrophoresis.

Western blot analysis

Western blot analysis was performed to detect MYB28-HA and GFP fusion proteins in transgenic Arabidopsis lines. Total protein was extracted from rosette leaves of 5-week-old WT, myb28 myb29 double mutant, and complementation lines (p2Kb^C::MYB28:HA and p2Kb^T::MYB28:HA), as well as 3-week-old WT and GFP reporter lines (p35S::5′UTR^C:GFP and p35S::5′UTR^T:GFP). Tissues were homogenized in a harsh lysis buffer (5% SDS, 15% glycerol, 0.175 M tris-HCl, and 0.3 M dithiothreitol) at a 2:1 buffer-to-tissue ratio. After centrifugation, supernatant protein concentrations were determined using the Bradford assay.

Proteins were separated by 10% SDS–polyacrylamide gel electrophoresis and subsequently transferred to nitrocellulose membranes (0.45-μm pore size) using standard wet transfer methods. Following transfer, membranes were blocked with 3% skim milk in TBST (tris-buffered saline with 0.1% Tween 20) for 1 hour at room temperature to prevent nonspecific antibody binding. The blocked membranes were then incubated overnight at 4°C with primary antibodies diluted in blocking buffer (anti-HA at 1:1000 or anti-GFP at 1:1000 for GFP fusion proteins). After thorough washing, membranes were probed for 1 hour at room temperature with species-matched horseradish peroxidase–conjugated secondary antibodies (either anti-rabbit or anti-mouse immunoglobulin G, diluted 1:8000 in blocking buffer). Protein detection was finalized using Pierce ECL Western Blotting Substrate (Thermo Fisher Scientific, #32106), with chemiluminescent signals captured by a ChemiDoc MP imaging system (Bio-Rad). Subsequently, the membranes were stained with a solution containing 0.1% Ponceau S (Sigma-Aldrich) in 5% acetic acid, gently rinsed with distilled water, and lastly imaged to verify uniform loading. Band intensities were measured using ImageJ (v1.53k). Rectangular regions of interest of fixed size were drawn around each target band and corresponding loading control (Ponceau S–stained total protein). Integrated density values were recorded, and the relative band intensity was calculated as: Relative expression = Integrated density (target band)/integrated density (Ponceau S).

Protoplast transfection assay

To generate constructs for protoplast transfection, we first cloned the WT MYB28.2 5′UTR using primers MYB28.2-UTR-F and MYB28.2-UTR-R (table S1). The PCR fragment was inserted into the pRTL2-GFP construct (85) between Xho I and Nco I sites (New England Biolabs). The pRTL2-MYB28.2-GFP plasmid served as a template for site-directed mutagenesis to generate pRTL2-MYB28.2^C241T-GFP, pRTL2-MYB28.2^A352T-GFP, pRTL2-MYB28.2^T227G-GFP, and pRTL2-MYB28.2^C254G/T256A-GFP (table S1). pRTL2-GFP served as a control. All constructs were verified by Sanger sequencing at Eton Bio (Research Triangle Park, NC, USA).

Protoplasts from Arabidopsis leaves were generated according to our published protocol (86). Two days posttransfection, total RNA and protein were simultaneously purified for RT-qPCR using a commercial kit (Bio-Rad, Hercules, CA, USA) and Western blotting analysis, respectively. Proteins were detected using anti-GFP polyclonal antibody (Genscript, Piscataway, NJ, USA), and Western blot analysis was performed as described above.

Statistical analysis

Statistical analyses were conducted using the R language. Data were evaluated for significance via one-way analysis of variance (ANOVA) with Tukey’s post hoc test for multiple comparisons or a two-sample t test for pairwise comparisons. In Tukey’s test, differences among groups were indicated by lowercase letters (shared letters denote non-significant differences), while statistically significant differences from t test were represented by asterisks: ****P < 0.0001, ***P < 0.001, **P < 0.01, and *P < 0.05.

Supplementary Materials

This PDF file includes:

Figs. S1 to S16

Table S1