The isolation of petunia cDNA for DFRA from a blue-flowered cultivar and the generation of the Gateway entry vector were described by Zhu et al. (2023). The sequence has been deposited in GenBank under the accession number MW929212. Mutagenesis was performed in the entry vector using primers carrying the intended changes (Supplemental Table S4). First, thirty cycles of linear PCR (1 min 94°C, 30 s 55°C, 8 min 72°C) was performed separately with each of the two primers using 500 ng plasmid, 0.4 µM primer, 0.2 mM dNTPs (each) and Phusion high fidelity DNA-polymerase (Thermo-Fischer) in 25 µL. Equal volumes of the amplified single stranded molecules were mixed, heated 5 min at 98°C and allowed to cool to 50°C in ca. ten minutes. After digesting the sample with DpnI to remove unmutated plasmid, it was transformed to E. coli DH5α.

After verification by sequencing, the inserts were transferred from the entry vectors to the destination vectors in an LR reaction according to the Gateway manual (Invitrogen), to pK2GW7 (Karimi et al., 2002) for stable transformation by agrobacterium, and to pEAQ-HT-DEST1 (Peyret and Lomonossoff, 2013) for agroinfiltration. The resulting expression vectors were verified using restriction enzymes and transformed into Agrobacterium tumefaciens C58C1(pGV2260) (Deblaere et al., 1985) using the freeze-thaw method (Wise et al., 2006).

Transient expression of DFR was performed in the top leaves of 6-week-old N. benthamiana plants according to Bashandy et al. (2015) and Zhu et al. (2023). Expression was allowed to continue for three days, after which 100 mg of leaf tissue was harvested in liquid nitrogen and milled to powder with an Oscillating Mill MM400 (Retsch, Germany). Extraction buffer (450 µL; 100 mM Tris-Cl pH 7.5, 0.2% 2-mercaptoethanol and a cocktail of protease inhibitors (Complete Mini EDTA free, Roche)) was added to frozen tissue powder, the thawed homogenate was kept on ice for 15 min and cleared twice in a microcentrifuge (full speed, 4°C).

DFR assay was performed as described by Zhu et al. (2023). Briefly, 10 µL of N. benthamiana extract was used in an assay of volume 100 µL, comprising a final concentration of 79 mM MOPS-KOH, pH 7.0, 1 mM NADPH, 1% DMSO, and 100 µM DHK, dihydroquercetin (DHQ), or dihydromyricetin (DHM). After 10 min at 30°C, the reaction was terminated by adding 20 µL glacial acetic acid. The sample was then extracted with 163 µL and 150 µL ethyl acetate, combined and dried. The dried extracts were dissolved in BuOH-HCl (95 parts n-butanol and 5 parts 37% HCl, supplemented with 1/30 volume of 2% NH₄Fe(SO₄)₂ ·12H₂O in 2 N HCl) and heated for 30 min at 95°C. The absorbance of the butanoylated product was measured at 535, 550, and 560 nm for DHK, DHQ, and DHM, respectively, corrected by subtracting the absorbance at 660 nm and converted to µM substrate used by multiplying with 62.9, 76.6, and 91.7 for DHK, DHQ, and DHM, respectively (Zhu et al., 2023).

(2R,3R)-dihydrokaempferol (DHK), (2R,3R)-dihydroquercetin (DHQ) and (2R,3R)-dihydromyricetin (DHM) and pelargonidin were purchased form TransMIT PlantMetaChem (Giessen, Germany) and other chemicals from Sigma-Aldrich unless specified otherwise.

For the assays shown in Figure 1, two leaves of two plants from each construct were infiltrated and collected three days after infiltration. Samples from the two plants were assayed on separate days, and for each extract, the assay was repeated at 30 min intervals between the assays. The technical repeats were averaged and the four biological replicates were statistically evaluated using ANOVA with post-hoc Tukey HSD Test.

The petunia line M61×W80 (Strazzer et al., 2023) is a mutant for the F3′H and F3′5′H encoding genes (ht1 ht1, hf1 hf1, and hf2 hf2) but heterozygous for a functional DFR-encoding gene (DFRA dfra). Transformation of M61×W80 was performed as described by Conner et al. (2009). In brief, greenhouse-grown petunia leaves were surface sterilized for 1 min in 70% ethanol, washed with sterile water, and incubated for 10 min in one-quarter-strength household bleach. After washing five times with sterile water, the leaves were cut into 5 × 5 mm² squares and infected for 5 min with an Agrobacterium suspension (1/7 diluted overnight culture). After blotting with sterile filter paper, the leaf pieces were placed on MS medium with vitamins (Duchefa Biochemie, Netherlands) supplemented with 20 g/L sucrose, 10 g/L glucose, 1 mg/L folic acid, 1 mg/L 6-benzylaminopurine, and 0.1 mg/L 1-naphthaleneacetic acid, and solidified with 0.8% agar. After 3 d at 23°C in the dark, the leaf pieces were transferred to a selection medium containing, in addition, 100 mg/L kanamycin and 100 mg/L ticarcillin and clavulanic acid mixture (Duchefa Biochemie). Leaf disks were grown at 23°C under fluorescent light and transferred to fresh plates every two weeks until shoots emerged, which were rooted in the same medium but half strength, without hormones and kanamycin, and subsequently transferred to the growth room with fluorescent lights, temperature of 23°C, and daylength of 16 hours. Plants were grown in peat:vermiculite mixture (1:1) and fertilized with commercial fertilizer (Substral, Germany).

Five to fifteen independent transgenic lines were recovered for each mutant construct and analyzed for anthocyanidin content (Table S1). The first flowers were discarded before the analysis started.

For anthocyanidin analysis, petal limbs of each flower were cut, weighed, and crushed in liquid nitrogen. Anthocyanins were extracted using five volumes of 80% methanol and 1% concentrated HCl by sonication for 15 min. After clearing by centrifugation, the extract was mixed 1:1 with 4 N HCl and heated for one hour at 95°C to hydrolyze them to anthocyanidins (the aglycones).

HPLC analysis was performed using an Agilent 1100 Series HPLC system (Agilent, CA, USA) equipped with a Zorbax Rc-C18 column (Agilent). The mobile phases were 0.08% trifluoroacetic acid (TFA) in water and 0.08% TFA in acetonitrile and a linear gradient from 10–90% acetonitrile was run for 20 min at a flow rate of 1 mL/min. The analytes were monitored at 525 nm, and the peak areas were reported using the Agilent software. For estimating the amount of pelargonidin in the samples, purified pelargonidin was dissolved in methanol with 1% concentrated HCl and diluted to 2.5-20 µg/ml for a standard curve (Table S1).

High-resolution mass spectrometry (UPLC-PDA-HDMS) analyses of the petunia samples were performed in the Viikki Metabolomics Unit at the University of Helsinki, Finland. Waters Synapt G2 HDMS mass spectrometer (Waters, Milford MA, USA) was used with the Waters Acquity UPLC system (Waters) via an ESI source with an additional diode array detector (Acquity PDA, Waters). The mass range was set from 90–800 with a 0.1 s scan rate in HDMS, while a wavelength range of 210–499 nm with a scan rate of 40 points/s was used in the PDA detector. Samples were analyzed in positive, sensitivity ion mode (ESI+), with a capillary voltage of 3.0 kV, source temperature of 120°C, desolvation temperature of 360°C, cone gas flow rate of 20 L/h, and desolvation gas flow rate of 800 L/h.

The secondary metabolites were separated on an Acquity BEH C18 UPLC column (1.7 µm, 50 × 2.1 mm, Waters, Ireland) at 40°C. The mobile phase consisted of water (A) and acetonitrile (B) (Chromasolv grade) both containing 0.1% formic acid. A linear gradient of eluents was used as follows: the gradient started at 95% A and was decreased to 10% over 10 min. Then, the ratio was switched back to 95% at 10.1 min and left to equilibrate for 1 min, with a total run time of 11 min. The injection volume was 3 µL and the flow rate of the mobile phase was 0.6 mL/min. The temperature of the sample tray was set at 10°C.

Computational modeling of the structures of wild-type DFRA and its mutants was performed using ColabFold (Mirdita et al., 2022), an online version of AlphaFold2 (Jumper et al., 2021). Relaxed versions of the top-ranked models were used in interaction studies. Additionally, the double mutant LD142VL was prepared by manual amino acid replacement using UCSF ChimeraX (Pettersen et al., 2021), with no significant binding pocket structure changes. Ligand-binding modeling tools, such as DiffDock (Corso et al., 2022) have also been utilized to map ligand-binding sites. Unfortunately, DiffDock could not identify the exact binding sites for primary substrates such as DHM. A comparison of the binding sites of the solved Vitis vinifera structure revealed that the NADPH-binding site was better established than that of the primary substrates. Dihydroflavonol substrates were positioned manually, with substrate-enzyme interactions with nearby residues mimicking those in the existing crystal structure (Petit et al., 2007). Manual ligand-binding interaction prediction and molecular graphic analysis were performed using ChimeraX (Pettersen et al., 2021).

To examine the possibility of altering the substrate preference of the petunia DFR with a minimum number of amino acid changes, we analyzed previous results addressing this. In a classical study addressing the substrate specificity of the petunia DFR, encoded by DFRA, Johnson et al. (2001) constructed hybrids between DFRA and a gerbera gene encoding DFR. The substrate specificity of the hybrid enzymes was assessed by transferring the hybrid genes to RL01, the same petunia laboratory line used by Meyer et al. (1987). The gerbera gene is a cDNA molecule representing the allele GDFR1-1 that encodes an enzyme with a strong preference for DHK (Zhu et al., 2023). Petunia DFR has a preference for DHM (three hydroxyl groups, Figure S1), and accepts DHQ but poorly accepts DHK (Haselmair-Gosch et al., 2018).

Johnson et al. (2001) identified a stretch of 26 amino acids that switched the petunia-type substrate preference to the gerbera type, leading to pelargonin production in transgenic RL01. Compared with five enzymes that allow pelargonidin biosynthesis (DFR from gerbera, rose, snapdragon, carnation, and maize), the petunia DFR is unique at four positions. One is aspartate (D) at position 143 (petunia numbering), where the other enzymes have asparagine (N). This is considered by many the key amino acid for substrate specificity (Johnson et al., 2001; Chu et al., 2014; Berardi et al., 2021).

Johnson et al. (2001) made several mutations in the gerbera DFR by introducing hydrophobic amino acids in place of hydrophilic ones in the 26-amino-acid long stretch and identified a change (N134L, gerbera numbering, corresponding to position 143 in petunia numbering) that further increased the gerbera enzyme preference for DHK based on their in vivo assay.

Based on this information, we made three mutations in the petunia DFRA gene (Figure S2). First, we mutated the putative substrate determining aspartic acid residue to asparagine, D143N (petunia numbering). Second, to mimic the enhanced pattern of DHK use in GDFR1-1 identified by Johnson et al. (2001), we introduced leucine at the same position, D143L. Third, to bring the petunia DFR a step closer to GDFR1-1, which prefers DHK, we mutated additionally the prior amino acid from leucine to valine, LD142VL.

In order to assay the DFR enzymes for their substrate preference, the wild-type and mutated petunia DFRA sequences, along with the maize A1 cDNA encoding DFR, were expressed in N. benthamiana using agroinfiltration (Zhu et al., 2023).

Tobacco leaves were extracted three days after infiltration and assayed for DFR activity using the BuOH-HCl method originally described by Porter et al. (1986) and refined by Zhu et al. (2023). Wild-type petunia DFR prefers DHM and exhibits low activity with DHK. While the D134N mutation draws substrate preference towards less-hydroxylated dihydroflavonols, it does not induce a dramatic change in DHK acceptance. In contrast, the D134L mutation inverts the substrate preference to favor DHK and rejects DHM. The LD142VL mutation further improved DHK use, making the enzyme DHK specific. Maize DFR A1 showed a preference for DHK, but accepted well also DHQ and DHM (Figure 1).

For stable expression in transgenic petunia and to observe possibly increased pelargonidin production in planta, the DFR-encoding genes were expressed from a vector where the transgene is driven by a 35S promoter. The host for the transgenes was the petunia line M61×W80, which is a cross between two inbred petunia laboratory lines (Strazzer et al., 2023). The line harbored mutations in genes encoding F3′H and F3′5′H but a wild-type DFRA allele. The transgenic lines (and the non-transgenic control) were analyzed for the pattern of anthocyanidins in their corolla limbs by extracting anthocyanins, hydrolyzing them into their aglycones, and separating the aglycones using HPLC.

The flowers of non-transgenic M61×W80 appeared pink because of the accumulation of a small amount of anthocyanins derived from peonidin and malvidin, and a minute amount from pelargonidin (Figures 2A–C). To compensate for the variability in the anthocyanin content of flowers produced and analyzed at different times, the ratio of the peak areas of pelargonidin to those of peonidin-malvidin (which was not chromatographically resolved in our HPLC) was calculated (rPg). Many transgenic lines did not express the transgene (or escaped kanamycin selection during transformation), and lines with rPg less than the highest value observed in the nontransgenic M61×W80 control (0.41) cannot be distinguished from non-expressors or escapes (see Supplemental Tables S1–S3).

Transgenic lines expressing wild-type DFRA increased the pelargonidin content of the petals by a small amount, which is in line with the fact that the enzyme accepts DHK at a low rate. All the mutant DFRA enzymes led to elevated pelargonidin levels. The best mutant was LD142VL; however, it still fell short of the maize A1 DFR in producing pelargonidin (Figure 2; Tables S1–S3). Another approach to evaluate transgenic lines is to score the highest pelargonidin producer for each DFRA mutant, similar to how a breeder selects the best-performing line for production or further crossing. For the D143N mutant, the amount was modest, with 18 µg/g fresh weight of the petal limb, but the LD142VL mutant led to 146 µg/g, which was more than half of the best transformant carrying the maize A1 gene at 251 µg/g. The transgenic petunia cultivar African Sunset, carrying a 35S-A1 construct and bred for its intense orange color, is shown as a control and accumulates pelargonidin up to 329 µg/g fresh weight (Figure 2; Tables S2, S3).

With the intention of gaining insight into how the mutations we made change the interaction between the enzyme and its substrates, we modelled the active site of the wild type and mutated petunia DFR. The only DFR this far crystallized for 3D structure determination is the grapevine (Vitis vinifera) VvDFR, which accepts DHQ as a substrate (Petit et al., 2007). We used ColabFold (Mirdita et al., 2022), a web-based version of Alphafold2 (Jumper et al., 2021), to predict the 3D structure of the petunia enzyme. As shown in Figure 3A, the DHM in the model is positioned analogous to the DHQ in the grapevine crystal structure (Petit et al., 2007). In these computational models, the interactions between the B-ring hydroxyls of the substrate and the enzyme occur between aspartate (D143) and glutamine (Q237). Both D at position 143 (Figure 3A) and N (Figure 3B) at the corresponding position in VvDFR can interact with the substrate B-ring hydroxyls through hydrogen bonding, giving no obvious clue as to why the mutation D143N in DFRA would increase activity with DHK and DHQ. However, replacing the polar amino acid with a nonpolar one in D143L (Figure 3C) abolished this interaction, considerably dropping the enzyme activity with DHQ and DHM, whereas the activity with the less polar DHK increased (hypothetically due to an increased hydrophobic environment and 3′,5′-hydroxyl repulsion). Further replacement of the L at position 142 with V (mutation LD142VL) makes nearly no change in the active site and the modelling does not help in explaining the increased activity with all substrates.

We created mutations that changed the amino acids in a critical position of the petunia dihydroflavonol 4-reductase that interact with the B-ring of the dihydroflavonol substrate according to the 3D structure of the grapevine DFR. Based on sequence comparisons, it has been suggested that the aspartic acid at position 143 of the petunia enzyme is an important determinant of the poor use of DHK. However, replacing aspartic acid with asparagine, which is found at the corresponding position in enzymes that accept DHK, had little effect. However, a hydrophobic leucine at this position almost completely abolished the use of DHQ and DHM while improving the use of DHK. By additionally changing the preceding leucine to valine, the petunia DFR was converted from DHK avoider to DHK favorer.

We further showed that when expressed from a 35S promoter in transgenic petunia, the mutant enzymes led to pelargonidin biosynthesis in the M61×W80 background, which accumulates DHK because of mutations in the hydroxylases responsible for DHQ and DHM biosynthesis (F3′H and F3′5′H, respectively). The intensity of the orange color of our transgenic lines was not comparable to that of the commercial cultivars that carried the maize DFR-encoding gene A1; however, they did compare well with the M61×W80 control transformants with 35S-A1 (Figure 2).

Similar to the lines produced with A1 in Cologne (Meyer et al., 1987; Oud et al., 1995), crossing the laboratory lines carrying the DFRA mutants with commercial varieties could result in a significant improvement in color intensity. Even better, targeted mutagenesis using gene editing of the DFR-encoding gene residing in the petunia genome may yield stable expression of the altered phenotype in one step. The D143N mutation is a single-base mutation that is not considered a genetic modification in many countries. However, as we demonstrated, the change in substrate preference is subtle. To obtain the best results, several nucleotides must be modified. D143L was achieved with two nucleotide changes and LD142VL with three changes. Techniques for this type of targeted mutagenesis include oligonucleotide mutagenesis in protoplasts (Beetham et al., 1999) and CRISPR/Cas-based prime editing (Anzalone et al., 2019). According to the current legislation, these multiple changes would cause the plants to be classified as genetically modified organisms. In the EU, even single nucleotide changes with gene editing techniques are considered genetically modified and are strictly regulated. Very recently, the European Commission’s legal proposal for new genomic techniques includes that simple edits, including up to 20 nucleotide substitutions, would be considered to be equivalent to conventional plants (Anonymous, 2023). Also, the rapid deregulation of the originally illegal orange petunia in the US (Anonymous, 2021) could indicate a more positive turn in public attitudes toward new tools in the breeder’s toolbox, at least for breeding non-food crop plants.

In addition to human consumers (Anonymous, 2016), orange and scarlet red flowers attract bird pollinators (Rodríguez-Gironés and Santamaría, 2004). However, pelargonidin-based anthocyanins, which are typical of orange and scarlet red, do not provide the protective functions of these compounds. Plants that harbor pelargonins in their flowers still synthesize (stress-induced) cyanidin derivatives in their vegetative parts. Examples can be found in gerbera (Deng et al., 2014) or in the pelargonium itself (unpublished observations). Assays of antioxidant capacity have shown that pelargonidin is a weaker antioxidant than cyanidin or delphinidin, and vicinal hydroxyl groups seem to be important for this function (Rice-Evans et al., 1996; Wang et al., 1997). Strong activity of the hydroxylases F3′H and F3′5′H does not allow substantial formation of pelargonidin, but there are plants, like petunia, that have evolved to additionally control this through the substrate preference of the DFR enzyme. Besides DFRA, enzymes that avoid reduction of DHK have been identified in many species, including Cymbidium orchids, strawberries, and arabidopsis (Johnson et al., 1999; Leonard et al., 2008; Miosic et al., 2014).