R. delavayi plants were cultivated in the experimental field of National Forest Garden of GuiZhou Province, China. The roots, leaves, petals, pistils, stamens, toruses, scapes, and developing flowers (stages 1–5) were collected. Wild-type Arabidopsis (Arabidopsis thaliana) and T-DNA insertion mutant (UGT78D2) were purchased from the Nottingham Arabidopsis Stock Center (NASC) and maintained in a long day condition (16 h light/8 h dark photoperiod). For RT-PCR and anthocyanin analysis, Arabidopsis seedlings were harvested at 7 days after germination on 1/2 MS medium containing 3% sucrose (anthocyanin induction medium). Nicotiana tabacum plants used for transformation were grown in a glasshouse at 22°C with a 12 h light/12 h dark photoperiod, and full-blooming flowers of T1 transgenic Nicotiana tabacum were sampled for further analysis. All samples mentioned above were frozen immediately in liquid nitrogen, and kept at −80°C until further use.

UDP-glucose, UDP-galactose, UDP-rhamnose, UDP-arabinose, cyanidin, delphinidin, and malvidin were purchased from Sigma-Aldrich (USA), and pelargonidin, petunidin, peonidin, cyanidin 3-O-glucoside, delphinidin 3-O-glucoside, pelargonidin 3-O-glucoside, malvidin 3-O-glucoside, petunidin 3-O-glucoside, and peonidin 3-O-glucoside were obtained from Phytolab (Germany).

The extraction and detection of anthocyanins were conducted as previously described (Sun et al., 2016). Briefly, 0.3 g freeze-dried flowers of R. delavayi were pulverized in liquid nitrogen and extracted with H₂O: MeOH: HCl (75/24/1v/v/v) overnight at 4°C in darkness. After centrifugation, the extracts were filtered (with 0.22 μm microporous membrane) and separated by an HPLC system (Shimadzu) equipped with an ACCHROM XUnion C18 column (250 mm × 4.6 mm, 5 μm). The absorbance of anthocyanin pigments was monitored at 520 nm, and the flow rate was 1 ml/min. Then, the mobile phase A (5% formic acid in H₂O) and B (methanol) worked as follows: 0–10 min, 14–17% B; 10–35 min, 17–23% B; 35–60 min, 23–47% B; 60–67 min, 47–14% B; 67–70 min, 14% B. According to the procedures depicted by Sun et al. (2015) mass spectrometer was selected for qualitative analysis (Sun et al., 2015). The contents of anthocyanin in each sample were calculated via the external standard calibration of cyanidin 3-O-glucoside standards (Fanali et al., 2011). The calibration curves used were linear (R² > 0.99) and the concentration ranges were 5–1,000 μg/ml. Mean values were obtained from three biological replicates per sample.

Total RNA for real-time PCR was extracted from flowers and other vegetative tissues of R. delavayi using RNA pure Plant Kit (CWBIO, China). Synthesis of first strand cDNA from 1 μg total RNA was performed using oligo (dT) and M-MLV reverse transcriptase (Takara, Japan). Then, real-time detection was conducted by using BioRad CFX96 Real-Time PCR System (BIO-RAD, Hercules, CA, USA) and TransStart^® Green qPCR SuperMix (TRANSGEN, China) with primers designed based on sequence information from transcriptome analysis performed previously (Supplementary Table 1). Amplification of RdActin under the identical conditions was carried out to normalize the levels of Rd3GTs as follows: 60 s at 95°C, followed by 40 cycles of 5 s at 95°C and 60 s at 60°C. Each sample was carried out in three biological replicates, and the relative expression levels of target genes were calculated by formula 2^−ΔΔCt. Meanwhile, the specific amplification of each gene was confirmed by melting curve analysis and agarose gel electrophoresis.

According to the methods mentioned above, cDNAs synthesized from flower of R. delavayi were used as PCR templates. Specific primers obtained from the assembled transcriptomic information (Supplementary Table 1) were used to clone the full-length coding sequence of Rd3GT1 and Rd3GT6. After PCR amplification, the products were sub-cloned into the T/A cloning vector pMD18-T (Takara, Japan), and transformed into Escherichia coli JM109 competent cells. After positive screening, the correct recombinant clone was verified by sequencing.

Alignment of amino acid sequences was carried out by using the DNAMAN 6.0 software. The multiple sequence alignment was performed through using Clustal Omega. Based on this alignment, a phylogenetic tree was drawn using MEGA5.1 with neighbor-joining method and 1,000 bootstrap replicates.

For ectopic expression of Rd3GTs in Arabidopsis and Nicotiana tabacum, their full-length CDS were amplified from the pMD18-T vector and inserted into the binary vector pBI121 digested with Xba I and BamH I, generating the pBI121-Rd3GTs overexpression constructs. Then, the resulting constructs were confirmed by sequencing and introduced into Agrobacterium tumefaciens strain GV3101 for Arabidopsis and Nicotiana tabacum transformation. Subsequently, inflorescences of Arabidopsis mutant (UGT78D2) were transformed via floral dip method, as described by Clough (Clough and Bent, 1998). The harvested seeds were selected on 1/2 MS medium supplemented with 50 mg/l kanamycin to set T1 seeds. After growing the seeds on anthocyanin induction media (1/2 MS medium containing 3% sucrose) for 7 days, the T2 transgenic seedlings were collected and used for molecular and anthocyanin analysis. At the same time, Nicotiana tabacum transformation was also performed according to the method previously depicted by Sparkes (Sparkes et al., 2006). Through resistant selection, T1 transgenic Nicotiana tabacum seedlings were obtained and cultured in green house, and after flowering, their full-blooming flowers were used for sample collection and later on analysis. For confirming the expressions of Rd3GTs, RT-PCR analysis was conducted both in Arabidopsis and Nicotiana tabacum. Furthermore, total anthocyanin concentration in T2 transgenic Arabidopsis seedlings and T1 transgenic Nicotiana tabacum flowers was determined using the method described above with three biological replicates.

The full-length CDS of Rd3GT1 and Rd3GT6 was introduced into the EcoR I/EcoR I and BamH I/Hind III site of pET32a (+) expression vector, respectively, and transformed into E. coli strain BL21 (DE3). On the second day, transformed E. coli expressing the recombinant Rd3GTs protein was cultured in liquid Luria-Bertani (LB) medium at 37°C till the value of OD₆₀₀ of 0.6 was reached; after that, the cultures were induced several times for 36 and 48 h at 15°C through adding 0.2 mM of isopropyl-β-d-thiogalactopyranoside (IPTG). The cells were then collected by centrifugation (5,000 rpm, 10 min, 4°C) and suspended in phosphate-buffered saline without protein inhibitor (PBS, pH 7.4). After sonication in ice, the cell debris was removed and filtered with a 0.45 μm filter (Millipore). Next, the supernatant which contained soluble recombinant protein was purified via Ni-NTA pre-packed column (TransGen, China) and its purity was checked by sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE). The concentration of recombinant Rd3GTs protein was measured by NanoDrop 1,000 Spectrophotometer (Thermo scientific, Waltham, MA, USA).

In vitro enzymatic activity of Rd3GT1 and Rd3GT6 was tested by using cyanidin and UDP-Glu as substrates. The assay reaction was carried out at 30°C for 5 min in a final volume of 200 μl containing 20–30 μl purified Rd3GT1 and Rd3GT6 protein, 100 mM potassium phosphate buffer (pH 8.0), 10 mM UDP-glucose, and 100 μM cyanidin. After quenching by adding 5% HCl (50 μl), the reaction mixture was centrifuged (12,000 rpm, 5 min, 4°C) and analyzed by HPLC using the method described above in “Detection of Anthocyanins”; then, the corresponding glucosylation product was confirmed by comparing it with the standard. Meanwhile, the protein extracted from BL21 (DE3) cells that express empty pET-32a (+) vectors was always used as a negative control. For determination, the substrate specificity of recombinant Rd3GT1 and Rd3GT6, delphinidin, pelargonidin, petunidin, peonidin, and malvidin was selected as acceptors using UDP-glucose as a donor. In addition, 10 mM UDP-galactose, UDP-rhamnose, and UDP-arabinose were also used to examine the specificity of sugar donors.

Correlation analysis was carried out via calculating pairwise Pearson correlations between each Rd3GT gene and anthocyanin. p < 0.05 was taken as statistically significant by Student's t-test.

To uncover the detailed biochemical basis of the red color of R. delavayi flower, its anthocyanin profiles were identified and quantified during flower development by using HPLC (Figure 1A). The results showed that a total of eight peaks (A1–A8) were identified in flowers (Figure 1B). These eight peaks were then confirmed as delphinidin 3-O-galacoside, cyanidin 3-O-galacoside, delphinidin rhamnoside, cyanidin 3-O-glucoside, cyanidin 3-O-arabinoside, and cyanidin derivatives according to the MS analysis (Table 1). But pelargonidin glycosides, one kind of the basic anthocyanin, were not detected in R. delavayi flower. To correlate anthocyanin accumulation with gene expression, anthocyanin contents in five developmental stages of R. delavayi flower were determined. During flower development, anthocyanin accumulations declined gradually from stage 1 to stage 3 and raised at stage 4 with the minimum levels at stage 5 (Figure 1C). Interestingly, among different anthocyanins, anthocyanidin 3-O-glycoside was used most of the times, which accounted for 93.68–96.31% of the total anthocyanins (Figure 1D).

On the basis of transcriptome data of different tissues of R. delavayi, a total of six potential 3GT genes were identified through blastn alignment with reference genes from proximal species and Arabidopsis. To confirm the key 3GT genes related to the biosynthesis of anthocyanins in R. delavayi flowers, we performed the expression analysis of these genes during flower development. As shown in Figure 2, transcript abundance of Rd3GT1 (R² = 0.8785) decreased from stage 1 to stage 2, followed by a raised transcript level to stage 5, and thus exhibited a negative correlation to the accumulation patterns of anthocyanin in R. delavayi flowers. For Rd3GT6, its transcripts (R² = 0.9526) declined from stage 1 to stage 3, and significantly upregulated at stage 4 with minimum transcript levels noted at stages 5, which perfectly correlated with the anthocyanin accumulation profiles in R. delavayi flowers. But for Rd3GT7 (R² = 0.0297), Rd3GT9 (R² = 0.0765), Rd3GT11 (R² = 0.1002), and Rd3GT12 (R² = 0.0003), there was a weak or nearly no correlation between their expressions and anthocyanin accumulations. Therefore, Rd3GT1 and Rd3GT6 were preliminarily identified as the key 3GT genes involved in the formation of R. delavayi flower color and selected for further functional characterization.

Based on sequence information of transcriptome data, the coding region sequences of Rd3GT1 and Rd3GT6 were successfully cloned from flower of R. delavayi. The ORFs of Rd3GT1 and Rd3GT6 were 1,395 and 1,365 bp long encoding 464 and 454 amino acids residues, respectively. Multiple sequence alignment of Rd3GT1 and Rd3GT6 with UGT78G1 (flavonoid 3-O-glucosyltransferase from Medicago truncatula, A6XNC6) and UGT78D2 (flavonoid 3-O-glucosyltransferase from Arabidopsis thaliana, NM_121711.5) revealed that both Rd3GTs carried the conserved 44-residue C terminal PSPG signature motif, and the last residue within this motif was histidine (Figure 3A). Next, a neighbor-joining phylogenetic tree of plant flavonoid UGTs was constructed, and three major clusters, which exhibited activities specific toward 3-OH, 5-OH, and 7-OH glycosylation, were recognized. Rd3GT1 and Rd3GT6 were grouped with 3-OH cluster implying that these two glucosyltransferases might belong to flavonoid 3-O-glycosyltransferase and glycosylate at the 3-OH of flavonoid substrates (Figure 3B).

Transcript levels of Rd3GT1 and Rd3GT6 were also determined in different R. delavayi organs by real-time qRT-PCR. As shown in Figure 4, Rd3GT1 and Rd3GT6 transcripts were detected in all organs tested, and their expressions were tissue specific. Accordingly, both Rd3GT1 and Rd3GT6 were most highly expressed in leaves where anthocyanins were barely detected (Supplementary Figure 4), and exhibited relatively lower expressions in pistil. Nevertheless, expression analysis during flower development displayed that the mRNA levels of Rd3GT1 and Rd3GT6 were dependent on flower development, and significantly correlated with the accumulation of anthocyanin. Taken together, this integrative expression analysis suggests that Rd3GT1 and Rd3GT6 may participate in not only the anthocyanin but also other flavonoid glycoside biosynthesis in R. delavayi.

To investigate the function of Rd3GT1 and Rd3GT6 in vivo, these two genes were overexpressed in the Arabidopsis mutant (UGT78D2), which failed to generate anthocyanin pigments in their cotyledon and hypocotyls. After resistance selection, seeds of the wild-type, UGT78D2 mutant, and T2 transgenic plants were germinated and inductively cultured on 1/2 MS medium with 3% sucrose. As present in Figure 5A, the seedlings of UGT78D2 mutant exhibited purple coloration while overexpressing Rd3GT1 and Rd3GT6, although the mutant transformed with empty vector was still green (Supplementary Figure 5). Meanwhile, RT-PCR analysis was conducted for further confirming the expression of Rd3GTs in transgenic lines, and amplicons of expected size, which were absent in wild type and mutant were observed in transgenic plants (Figure 5C). In addition, extracted anthocyanin metabolites from the abovementioned seedlings were also analyzed by HPLC and HPLC-MS to determine the contents and kinds of individual anthocyanin (Supplementary Table 2). The results revealed that both Rd3GT1 and Rd3GT6 could restore the anthocyanin peaks (peak 1, 3, and 4) that lacked UGT78D2 mutant (Figure 5D), and showed higher anthocyanin contents than that in wild-type as well as UGT78D2 mutant (Figure 5B). Overall, these results confirmed that both Rd3GT1 and Rd3GT6 encode functional UF3GT protein that participated in the biosynthesis of anthocyanin in Arabidopsis.

To further check the potential effect of Rd3GT1and Rd3GT6 on plant flowers, a total of 24 transgenic tobacco plants overexpressing Rd3GT1 and Rd3GT6 were generated. Then, two independent transgenic lines were analyzed for each gene, and RT-PCR analysis was used to confirm the overexpression of Rd3GT1 and Rd3GT6 (Figure 6C). As shown in Figure 6A, flower color of transgenic plants was darker than the control. Next, for examining this color change in detail, anthocyanin was extracted from the corollas of transgenic lines and quantified using HPLC (Figure 6D). The results showed that anthocyanin levels in flowers of transgenic tobacco were all higher than those in wild type (Figure 6B). Simultaneously, the kinds of anthocyanin in transgenic flower were unexpectedly increased from 2 to 5 (Figure 6D), and these anthocyanins were subsequently identified as cyanidin 3-O-rutinoside 5-O-glucoside, cyanidin 3-O-arabinoside, cyanidin 3-O-rutinoside, cyanidin 3-O-xyloside, and cyanidin 3-O-(6-O-malonyl-beta-D-glucoside), respectively (Table 2). Thus, combining all these results, it becomes clear that the enzymes encoded by Rd3GT1 and Rd3GT6 can effectively increase the contents and kinds of anthocyanin in transgenic tobacco flower.

The coding regions of Rd3GT1and Rd3GT6 were cloned into pET-32a (+) vector, and introduced into an E. coli BL21 (DE3) strain. The recombinant soluble proteins of Rd3GTs were purified by Ni-NTA pre-packed column and analyzed by SDS-PAGE (Supplementary Figure 1). The function of Rd3GT1 and Rd3GT6 was characterized using cyanidin as the substrate and UDP-glucose as the sugar donor. HPLC analysis identified that both Rd3GTs could catalyze the transfer of glucose to the 3-OH of cyanidin via comparing it with the reference standard. The recombinant protein extracted from E. coli BL21 expressing the empty vector could not glucosylate cyanidin (Figure 7A). Then, the reaction conditions for Rd3GTs were optimized, and both Rd3GTs exhibited their maximum activity at pH 8.0 and 30°C, which has been applied in the analysis of most plant UFGTs (Masayuki et al., 2000; Kim et al., 2006).

To explore the substrate specificity of Rd3GTs, six anthocyanidin compounds were tested using UDP-glucose (UDP-Glu) as the sugar donor. Rd3GT1 showed the highest conversion rates for delphinidin, whereas pelargonidin, peonidin, and petunidin could not be accepted by it. However, Rd3GT6 was able to glucosylate five anthocyanidins except for malvinidin, and cyanidin exhibited the highest glucosylation rate, which was higher than that of Rd3GT1 toward delphinidin (Supplementary Table 3). Furthermore, HPLC analysis of Rd3GT6 reaction products showed that additional products (P2 and P4) were generated in the reaction of cyanidin and delphinidin (Figures 7A,B). The retention time of P1 and P3 was similar to the authentic cyanidin 3-O-glucoside and delphinidin 3-O-glucoside. Evidently, they were cyanidin 3-O-glucoside and delphinidin 3-O-glucoside, respectively. Then, HPLC-ESI-MS results demonstrated that the molecular weights of P1 and P2 were indistinguishable, indicating that only one UDP-glucose had been transferred to P2. It was reported that hypsochromic shift of UV spectra between Rd3GT6 products and substrates could be used to confirm the regiospecificity of Rd3GT6. Glycosylation at both 3-OH and 4'-OH would produce a hypsochromic shift, while at 7-OH, it does not display any effect (Thomas et al., 1997; Kramer et al., 2003). Therefore, the results suggested that P2 was likely to be cyanidin 4'-O-glucoside, and P4 was likely to be delphinidin 4'-O-glucoside based on hypsochromic shift (Supplementary Table 4).

To further explore the sugar donor specificity of Rd3GTs, we tested three other donors, that is, UDP-galactose (UDP-Gal), UDP-rhamnose (UDP-Rha), and UDP-arabinose (UDP-Ara). Surprisingly, Rd3GT6 exhibited high sugar donor promiscuity when catalyzing cyanidin, whereas Rd3GT1 did not, because it could not transfer UDP-Rha and UDP-Ara (Supplementary Tables 5, 6). In the presence of UDP-Gal, both Rd3GT1 and Rd3GT6 were able to efficiently glycosylate all the anthocyanidin substrates (Figure 8A). And with regard to UDP-Rha, Rd3GT6 could catalyze the glycosylation of cyanidin as well as delphinidin and showed relatively higher glycosylation rates toward delphinidin (Supplementary Table 6). For the catalytic reaction of UDP-Rha, Rd3GT6 produced four products (P1–P4), and all these products were identified as mono-O-glycosides through HPLC-ESI-MS analysis; thus, P1, P3 and P2, P4 were tentatively characterized as 3-O-rhamnoside and 4'-O-rhamnoside by the hypsochromic shift (Figure 8B; Supplementary Table 7). As shown in Supplementary Figure 3, only cyanidin could accept UDP-Ara at conversion rates of 5.96%.