The biosynthetic pathway of potato solanidanes diverged from that of spirosolanes due to evolution of a dioxygenase

Potato (Solanum tuberosum), a worldwide major food crop, produces the toxic, bitter tasting solanidane glycoalkaloids α-solanine and α-chaconine. Controlling levels of glycoalkaloids is an important focus on potato breeding. Tomato (Solanum lycopersicum) contains a bitter spirosolane glycoalkaloid, α-tomatine. These glycoalkaloids are biosynthesized from cholesterol via a partly common pathway, although the mechanisms giving rise to the structural differences between solanidane and spirosolane remained elusive. Here we identify a 2-oxoglutarate dependent dioxygenase, designated as DPS (Dioxygenase for Potato Solanidane synthesis), that is a key enzyme for solanidane glycoalkaloid biosynthesis in potato. DPS catalyzes the ring-rearrangement from spirosolane to solanidane via C-16 hydroxylation. Evolutionary divergence of spirosolane-metabolizing dioxygenases contributes to the emergence of toxic solanidane glycoalkaloids in potato and the chemical diversity in Solanaceae.

P lants produce a vast array of specialized metabolites, many of which play essential roles in plant adaptation to environment and contribute to human health 1 . Steroidal glycoalkaloids (SGAs), a class of specialized metabolites, are typically found in the plant family Solanaceae. While they serve as chemical protectant against plant pathogens and herbivores 2,3 , some are notorious as antinutritional substance exerting negative effect on quality and marketability of Solanum (Family: Solanaceae) staple food crops, such as potato (Solanum tuberosum), tomato (S. lycopersicum), and eggplant (S. melongena) 4 . Therefore, controlling the SGA level is one of important targets for breeding of Solanaceae species.
SGAs consist of two structural components; the aglycone unit composed of nitrogen containing C 27 steroid derived from cholesterol and oligosaccharide attached to the hydroxy group at C-3 3,5 . Based on the skeletal structure of the aglycone, SGAs can be divided into two general classes, solanidane or spirosolane ( Fig. 1) 3,6 . Minor structural variations of these two ring types such as C-5 saturation/ unsaturation or isomerization at C-22, in combination with various sugar moieties, generate the enormous structural diversity of SGAs 7,8 . In addition, their chemical structures reflect their biological activities, for example, toxicity to animals, anti-cancer properties, and anti-microbial activities 3,9 . Most representatives of solanidane glycoalkaloids are potato toxins, α-solanine 9 and αchaconine 10, that comprise upward of 90% of the total SGAs in cultivated potatoes 10,11 . Additionally, more than 50 different SGAs, including spirosolane glycoalkaloids, have been identified in a variety of wild potato species and cultivars 11,12 . Tomato and eggplant, on the other hands, contain only spirosolane glycoalkaloids 3,13 . In tomato, α-tomatine 5 and dehydrotomatine 6 are predominant in green tissues, while the non-bitter SGA esculeoside A is the major component in the red mature fruits 14,15 . α-Solasonine and α-solamargine are the two major spirosolane glycoalkaloids produced in eggplant 16 .
SGA biosynthesis can be divided into two main parts: aglycone formation and glycosylation. Recent research in potato and tomato identified several SGA biosynthetic genes involved in aglycone formation. Three cytochrome P450 monooxygenases (CYPs) named as PGA2 (GAME7), PGA1 (GAME8), and PGA3 (GAME4) have been found to be involved in hydroxylation of cholesterol 1 at C-22 and C-26 and oxygenation at C-26, respectively 17,18 . A 2-oxoglutarate-dependent dioxygenase (DOX) named as 16DOX (GAME11), and an aminotransferase were reported to be required for the C-16α-hydroxylation and C-26 amination during SGA biosynthesis 18,19 . These enzymes and functions are common to potato and tomato, suggesting that they are involved in the biosynthetic steps common to solanidanes and spirosolanes (Fig. 1). In addition, several uridine diphosphate-dependent glycosyltransferases (UGTs) involved in the glycosylation steps of SGA biosynthesis have been identified in potato and tomato [20][21][22][23][24] . However, the steps and enzymes involved in the metabolic branch point between solanidaneskeleton and spirosolane-skeleton formation remains an unsolved mystery.
In our efforts to elucidate the missing steps in SGAs biosynthesis, we focused on SGA biosynthesis in potato. Here, we find that solanidane glycoalkaloids are biosynthesized from spirosolane glycoalkaloids via ring rearrangement (Fig. 1). We discover DPS, a member of the DOX superfamily, is a key enzyme required for solanidane biosynthesis. DPS in vitro exhibits a novel activity converting the spirosolane-skeleton to a solanidaneskeleton (Fig. 1). Furthermore, studies of the genome organization of DPS homologs in a number of species from the Solanum genus reveal the evolutionary origin of this transformation. These results offer insights into the evolution of chemical diversity of SGAs in Solanum species.
Selection of candidate DOXs responsible for solanidane biosynthesis. The DOX superfamily is one of the largest enzyme superfamily in plant kingdom and associated with the oxygenation reactions of various specialized metabolites 27 . Recently, involvement of DOXs in SGA biosynthesis and modification was reported 14,15,18,19 . In our previous study, St16DOX, which showed the highest fragments per kilobase of exon per million mapped fragments (FPKM) value in tuber sprout among the 256 DOX transcripts in RNA-Seq Gene Expression Data obtained from Spud DB (http://solanaceae.plantbiology.msu.edu/), was identified to encode the steroid 16α-hydroxylase 19 . In this study, we focused on the gene Sotub01g007130, which showed the second highest FPKM value in the tuber sprouts among the potato DOXs (Supplementary Table 1). Quantitative reverse transcription PCR revealed that Sotub01g007130 shared a similar expression pattern with previously identified SGA biosynthetic genes 19 ( Supplementary Fig. 5), and we henceforth named Sotub01g007130 as DPS (dioxygenase for potato solanidane synthesis). Many SGA biosynthetic genes were reported to be clustered on chromosomes 7 and 12 in potato and tomato 18 , while DPS is positioned on chromosome 1. We found that DPS exists in a gene cluster (spanning about 160 kbp) with seven DOXs designated as St7070, St7080, St7090, St7100, St7110, St7120, and St7150, respectively. St7090, St7100, St7120, and St7150 appeared to encode truncated proteins lacking canonical motifs conserved among the DOX superfamily 27 . Among these clustered DOXs, St7070, St7100, and St7110 also showed high FPKM values in potato tuber sprouts (Supplementary Table 1). We successfully amplified the full-length open-reading frame (ORF) sequences of DPS and St7070 using the cDNAs derived from tuber sprouts of potato. DPS shares 85.8% amino acid sequence identity to St7070 and a shorter C-terminus of 36amino acid residues than St7070. Fig. 2 Bioconversion of the solanidane-skeleton to the spirosolaneskeleton in potato hairy roots. a LC-MS detection of conversion from administered α-solamarine to α-solanine in 16DOX-disrupted potato hairy roots (16DOXko HR). Total ion current (TIC) chromatogram obtained in positive ionization mode with a full-scan range of 350-1250m/z are shown. Production of α-solanine was confirmed by comparison with the authentic standard of α-solanine in terms of retention time and mass spectrum provided in Supplementary Fig. 1. b Inhibition of the bioconversion of αsolamarine to α-solanine by uniconazole-P or prohexadione. After administering α-solamarine for 3 days in the presence of uniconazole-P or prohexadione, the SGAs accumulated in 16DOXko HR were extracted and quantified using LC-MS. Bars represent the ratio of α-solamarine to αsolanine accumulation in hairy roots. Error bars represent ± standard deviation of mean (n = 5).
Characterization of DPS-silenced transgenic potato plants. To investigate the contribution of DPS to solanidane biosynthesis, we generated DPS-silenced transgenic potato plants (i.e., DPSi lines) by using RNA-interference (RNAi) vector containing a stem-loop partial fragment of DPS cDNA ( Supplementary Fig. 6). DPSi plants showed~80% reduction in DPS transcript levels relative to non-transformed (NT) plants (Fig. 3a), and no off-target silencing effect on St7070 was observed (Fig. 3b). The levels of solanidane glycoalkaloids (α-solanine 9 and α-chaconine 10) in the leaves of in vitro-grown DPSi plants were reduced by ∼90% compared with those of the NT plants (Fig. 3c), suggesting that DPS is involved in solanidane biosynthesis in potato. Analysis of endogenous metabolites accumulated in DPSi plants detected two new peaks, of which the retention time (Rt) and mass spectrum matched with those of the authentic compounds, α-solamarine 7 and β-solamarine 8 ( Fig. 3d and Supplementary Fig. 7). The amounts of α-solamarine 7 and β-solamarine 8 accumulated in DPSi plants were comparable to those of α-solanine and α-chaconine in NT potato plants, respectively (Fig. 3e). These results implicate DPS as a key gene in the conversion from spirosolanes to solanidanes in potato.
Characterization of DPS function. Recombinant His-tagged protein of DPS was prepared with a bacterial expression system in Escherichia coli and purified by cobalt-affinity chromatography ( Supplementary Fig. 8). The purified DPS was assayed with several spirosolanes as substrates, and the reaction products were analyzed using liquid chromatography-mass spectrometry (LC-MS). DPS metabolized α-solamarine 7 to form an unknown product, of which the retention time and the mass fragment pattern was not identical to α-solanine 9 (Fig. 4a, b). The product gave a parental ion mass at m/z 882.5 [M + H] + which is two mass smaller than that of the substrate α-solamarine: m/z 884.5 [M + H] + ( Fig. 4b and Supplementary Fig. 9). Negative control reactions using either denatured DPS, lacking cofactors, or containing ethylenediaminetetraacetic acid (EDTA) as an Fe chelator failed to yield any products ( Supplementary Fig. 10). The other spirosolane glycoalkaloids (β-solamarine 8 and α-tomatine 6) and the spirosolane aglycones ((22S,25S)-spirosol-5-en-3β-ol 3 and tomatidine 4) were also accepted as substrates, and the mass decrease of 2m/z was detected for each product . On the other hand, DPS is not active with solanidanes as substrates ( Supplementary Fig. 16).
To identify the nature of the DPS reaction, we performed highresolution mass spectrometry (HRMS) (Supplementary Fig. 17) and nuclear magnetic resonance (NMR) (Supplementary Data 1) analyses of the DPS enzymatic product obtained by reacting with α-tomatine 6, which is commercially available in enough quantities for structural analysis. The structure of the DPS reaction product was determined as shown in Supplementary  Fig. 18 (Supplementary Fig. 19, Supplementary Tables 2, 3). The solanidane-skeleton scaffold is formed, and an iminium ion structure is constructed between C-16, the nitrogen atom, and C-22. Thus, DPS seems to catalyze hydroxylation at C-16α of the spirosolane-skeleton to form a hemiacetal that then undergoes condensation with the secondary amine of the piperidine F ring to form the tertiary amine of solanidane, and hence, this enzymatic transformation by DPS is quite unusual (Fig. 4c). To gain insight into the reaction mechanism of dehydration and E/F ring rearrangement by DPS, we investigated the origin of the oxygen atom of the oxy ion in the DPS enzymatic reaction product by 18 O 2 -labeling experiments. DOXs typically catalyze oxygenation/hydroxylation reactions, and DOX-dependent reactions in the presence of 18 O 2 result in the incorporation of 18 O atom into the product. Accordingly, when the enzyme assay with Sl23DOX, which catalyzes C-23 hydroxylation of α-solamarine, was conducted in the presence of 18 O 2 , the reaction product gave 2 mass higher than the product assayed with 16 O 2 , indicating 18 O incorporation into the 23-hydroxylated product by Sl23DOX ( Supplementary Fig. 20a-c). In contrast, the DPS enzymatic assay in the presence of 18 O 2 did not yield any 18 O-labeled product (Supplementary Fig. 20d-f). This result indicates that the oxy ion at C-16α in the DPS-reaction product is not derived from molecular oxygen and also suggests that the hydroxy moiety introduced by the DPS-dependent oxygenation leaves as water to form the solanidane. Based on these findings, we propose the reaction mechanism shown in Fig. 4d. First, DPS catalyzes hydroxylation at C-16α of spirosolane-skeleton to form a 16hemiacetal intermediate. Second, the lone pair on the nitrogen atom of the intermediate migrates to between C-22 and the nitrogen atom. Concomitantly, the carbon-oxygen bond at C-22 cleaves, and the hydroxy group at C-16α introduced by DPS leaves as water to afford a 16-oxo species. Third, the nitrogen atom attacks the C-16 carbonyl carbon to form an imine-oxy zwitterion, the DPS reaction product.
Kinetic analyses revealed that the k cat /K m value of DPS for αsolamarine 7 was 1123-fold greater than that for (22S,25S)spirosol-5-en-3β-ol 3, the aglycone of α-solamarine 7 (Supplementary Fig. 21 and Supplementary Table 4). The strong substrate preference of DPS for the glycosides suggests that conversion of spirosolane to solanidane occurs after the glycosylation step at the C-3 hydroxy group (Fig. 1). This is consistent with the observation that α-tomatine 6 was converted to its solanidane glycoalkaloid demissine 11 with retention of the lycotetraose at the C-3 hydroxy group ( Supplementary Fig. 4). Recombinant St7070 assays with the above substrates gave the same reaction products as DPS, but the substrate preference of St7070 was significantly different from that of DPS. St7070 showed lower activity for α-solamarine 7 than DPS ( Fig. 4a) and also showed only 4.5-fold higher preference for αsolamarine 7 than for its aglycone (22S,25S)-spirosol-5-en-3β-ol 3 (Supplementary Fig 21 and Supplementary Table 4).
Thus, DPS was found to catalyze the conversion of spirosolanes to form the corresponding solanidanes in vitro, but the resultant DPS-reaction products were not the solanidane end-products completed in planta. To resolve this inconsistency, the DPSreaction products were fed to 16DOXko HR to examine their conversion to the corresponding end-products in vivo. The DPSreaction product with α-solamarine 7 was converted into αsolanine 9 in 16DOXko HR ( Supplementary Fig. 22). Similarly, administration of the DPS-reaction product with β-solamarine 8 to 16DOXko HR restored the accumulation of α-chaconine 10 ( Supplementary Fig. 22). These results indicate that the DPSreaction products serve as intermediates of solanidane glycoalkaloid biosynthesis in potato.

Evolution of SGA biosynthetic pathways in Solanum species.
With the establishment of DPS-associated metabolic pathway converting spirosolanes to solanidanes, we next linked this discovery to the evolution of SGA metabolism in the Solanum genus. We conducted a Basic Local Alignment Search Tool (BLAST) search of DPS against the Tomato ITAG protein database (https://solgenomics.net/). This analysis identified a single distinct homolog Solyc01g006585 in the tomato genome, which shows 85% amino acid sequence identity to DPS. Recombinant Solyc01g006585 catalyzed the same reaction as DPS on the spirosolanes and was inactive with the solanidanes (Supplementary Figs. [23][24][25][26][27][28][29]. Very low or no expression of Solyc01g006585 observed in RNA-seq data from the Tomato Functional Genomics Database (http://ted.bti.cornell.edu) (Supplementary Table 5) explains the lack of solanidane glycoalkaloids and accumulation of spirosolane glycoalkaloids (i.e., α-tomatine etc.) in tomato. Solyc01g006585 is embedded in the gene cluster containing three DOX genes in tomato chromosome 1 (Fig. 5a). Genes flanking this cluster showed considerable homology with genes flanking the DOX clusters harboring DPS, suggesting a common origin of these regions. Additionally, the eggplant DPS homolog, SMEL_001g151230 is present on the chromosome 1. SMEL_001g151230 protein, which shares 69% amino acid sequence identity to DPS, also had the same catalytic activity as DPS ( Supplementary Fig. 30), while the expression of SMEL_001g151230 is quite low in aerial parts of eggplant 28 (Supplementary Table 6). These results indicate that the evolution of these orthologous DPS genes from duplication and functional divergence of the DOX family may have occurred before the speciation of Solanum species. Consequently, it is most likely that functional expression of the DPS gene or its orthologs is a prerequisite for the endogenous accumulation of toxic solanidane glycoalkaloids in Solanum species.
Based on the classification of the plant DOXs 27 , all the clustered DOXs in potato, tomato, and eggplant are classified into the clade DOXC20. DOXC20 includes the α-tomatine 23hydroxylase (Sl23DOX/GAME31) involved in conversion of bitter α-tomatine 6 to non-bitter esculeoside A during tomato fruit ripening 14 . Sl23DOX/GAME31 shares 55% amino acid sequence identity to DPS, and the homologs of Sl23DOX are clustered in tomato chromosome 1 and 2 (Fig. 5a). In the phylogenetic analysis, DPS forms a distinct branch (the DPS branch) which is clearly separated from the 23DOX branch that includes Sl23DOX/GAME31 (Fig. 5b). Among the eight DOX  (Fig. 5b). Solyc01g006585 and SMEL_001g151230 are placed in the DPS branch, while the other DOXs clustered in tomato genome are in the 23DOX branch (Fig. 5b). These results indicate that functional diversification between Sl23DOX/GAME31 and DPS may have arisen from a common ancestral DOXC20 before the speciation of Solanum species.

Discussion
In this study, we revealed a key reaction step diverging the biosynthetic pathways of two general classes of SGAs, solanidanes and spirosolanes. By the feeding experiments, we found that solanidane glycoalkaloids are biosynthesized from spirosolanes in potato. Gene silencing experiment and biochemical characterization revealed that potato DPS, belonging to the DOX superfamily, is involved in the metabolic conversion of spirosolanes to solanidanes. Thus, DPS is a key enzyme in an evolutionary origin of the biosynthetic pathway of solanidane glycoalkaloids branched from spirosolanes in potato (Fig. 1).
The strong preference of DPS for glycosides observed in the in vitro assays (Supplementary Table 4) let us hypothesizes that glycosylation at the C-3 hydroxy group occurs before formation of the solanidine backbone. Previously, SGT1 and SGT2 were reported to catalyze the initial transgalactosylation and transglucosylation of solanidine 12 at the C-3 hydroxy group, respectively. These two glycosyltransferases accept spirosolane aglycones as well, and the activities were comparable to the activity for solanidine 21,22 . Additionally, 16DOX-silencing which disrupts 16α-hydroxylation of 22,26-dihydroxycholesterol resulted in the accumulation of glycosides of 22,26-dihydroxycholesterol 2 19 . These previous findings support our hypothesis that conversion of spirosolane to solanidane occurs after the glycosylation step at the C-3 hydroxy group (Fig. 1).
Analysis of the catalytic activity of DPS clearly demonstrated that DPS is not sufficient to complete the conversion from spirosolanes to the solanidane end-products accumulated in planta. The structure of the DPS-reaction product allowed us to propose a pathway for completing the solanidane-skeleton formation, which requires two imine-reduction reactions following to DPS reaction ( Supplementary Fig. 31). Feeding of the DPSreaction products to 16DOXko HR indicated their conversion to α-solanine and α-chaconine, suggesting the presence of putative enzyme(s) involved in the reduction of the imine intermediates in vivo (Supplementary Fig. 22). Analysis of transgenic tomato hairy roots constitutively expressing the DPS gene detected a new peak, of which the retention time (Rt. 18.7) and mass spectrum matched with those of the reaction product by DPS using αtomatine as a substrate, but the production of demissine, which is the solanidane end-product derived from α-tomatine, was not detected (Supplementary Fig. 32). These results suggest that, in tomato, putative reductase(s) catalyzing the reduction of the imine intermediate is absent or not expressed. Thus, it is likely that the reductase(s), together with DPS, will be an important enzyme for solanidane glycoalkaloid biosynthesis in potato.
We identified the presence of DPS isoform designated as St7070 in potato genome, and biochemical characterization revealed that St7070 catalyzed the conversion of the spirosolaneskeleton to the solanidane-skeleton. Gene silencing of DPS resulted in a significant reduction (~90%) in solanidane glycoalkaloid levels, but no off-target silencing effect on St7070 was observed (Fig. 3a-c). These results indicate that DPS is the major isoform involved in solanidane glycoalkaloid biosynthesis in potato, although there may be a minor contribution from St7070. We also identified functional DPS orthologs in the nonsolanidane producing species, tomato and eggplant, and however, these orthologs were almost not expressed in these species. It has been reported that the transcription factor JRE4/GAME9 comprehensively regulate SGA biosynthetic genes in tomato 10,29,30 . Overexpression of the transcription factor JRE4/ GAME9 in potato led to significant increase of DPS transcription levels 10 , while the expression of Solyc01g006585 was not induced by JRE4/GAME9 overexpression in tomato 10,29,30 . These observations indicate that transcriptional regulation of DPS or its orthologs is a crucial factor diverging between spirosolane and solanidane glycoalkaloids producers. There are two possible evolutionary scenarios; (i) gain of the DPS expression in solanidane-glycoalkaloids producing species such as domesticated potatoes and wild potatoes or (ii) loss of the DPS expression in non-solanidane glycoalkaloids producers such as tomato and eggplant. Further genetic studies are required to clarify the evolutionary events that underpin the differential expression of DPS and its orthologs in Solanum species.
Several studies reported that tandem gene duplication and subsequent changes in gene function drive the generation of new biosynthetic pathways or enhancement of existing metabolic flux [31][32][33][34] . Therefore, it is noteworthy that the genes belonging to the DPS branch are highly duplicated in tandem on potato chromosome 1 (Fig. 5a). The fact that there is only one DPS ortholog in each tomato and eggplant genome indicates that the clustered genes including DPS and St7070 in potato have arisen from a single-copy ancestral gene via gene duplication after the speciation of Solanum species. DPS encodes a characteristic shorter polypeptide missing about 40 amino acid residues at the C-terminus as compared to typical DOXC members 14,19,27 ( Supplementary Fig. 33). Nucleic acid sequence alignment suggests that DPS was modified to become shorter via a single nucleotide-insertion ( Supplementary Fig. 34). In addition, DPS and St7070 exhibited markedly differences in substrate specificity and catalytic efficiency, with DPS having very strong substrate preference for glycosides compared to St7070 (Supplementary Table 4). Given the major contribution of DPS to solanidane glycoalkaloids biosynthesis, we presume that such improved catalytic property of DPS, in addition to its elevated gene expression, provide an evolutionary benefit by providing the metabolic pathways from spirosolane to solanidane glycoalkaloids in cultivated potatoes. Several wild potato species contain both spirosolanes and solanidanes 12,25 , whereas domesticated potato cultivars produce almost only solanidane glycoalkaloids. To clarify the underlying mechanism of the different glycoalkaloid profiles between wild and cultivated potatoes, further experiments are needed, such as analyzing differences in gene expression, coding sequences, and catalytic properties for DPS and its homologous genes between wild and cultivated potatoes.
The SGA biosynthetic pathway is widely conserved among Solanum species 18,28 , and evolutionary divergence of spirosolanemetabolizing DOXs contributes to the chemical diversity of SGAs in the Solanum genus 15 . Recent studies identified that Sl23DOX/ GAME31 in tomato catalyzes C-23 hydroxylation of αtomatine 14,15 (Fig. 5c), and this reaction is responsible for the first step of detoxification metabolism from toxic spirosolane glycoalkaloid α-tomatine to non-toxic esculeoside A. Phylogenetic analysis indicates that functional differentiation between Sl23DOX/GAME31 and DPS in the DOXC20 subfamily likely has arisen from a common ancestral enzyme (Fig. 5). In tomato, spirosolane-metabolizing Sl23DOX/GAME31 contributes to the domestication of cultivated tomatoes by the reduction of toxicity and bitterness in mature fruits. In contrast, metabolism of spirosolane glycoalkaloids to solanidane glycoalkaloids in potato may represent evolutionary advantage to produce SGAs more toxic and effective in protection against their enemies.
In conclusion, we have revealed the genetic and enzymatic origin of solanidane glycoalkaloid biosynthesis in potato. This provides insight into the evolution of chemical diversity of SGAs in Solanum species and its evolution's footprints on Solanum genomes. Thus, our findings further elucidate understanding of how plants have evolved and developed their specialized metabolic pathways, demonstrating that evolutionary divergence of the DOX family is one of the key driving forces for phytochemical diversity.
Feeding experiments. St16DOX-disrupted potato hairy roots were established in our laboratory 26  B increased linearly to 100% over 4 min, held at solvent B 100% for 5 min; solvent B then returned immediately to 10%, followed by a 5 min re-equilibration period. In the α-tomatine 6 feeding experiment, the mobile phases were water with 0.1% (v/v) formic acid (A), acetonitrile (B), and methanol (C), with a linear gradient elution as follows: from solvent A 90%/solvent B 5%/solvent C 5% to solvent A 45%/solvent B 27.5%/solvent C 27.5% over 30 min; from solvent A 45%/solvent B 27.5%/solvent C 27.5% to solvent A 25%/solvent B 37.5%/solvent C 37.5% over 5 min; from solvent A 25%/solvent B 37.5%/solvent C 37.5% to solvent A 0%/solvent B 50%/solvent C 50%, held at solvent A 0%/solvent B 50%/solvent C 50%; solvent B and solvent C then returned immediately to 5%; followed by a 5 min reequilibration period. Conversion of α-tomatine 6 to corresponding solanidane was confirmed by comparison with demissine 11 extracted from leaves of the wild potato S. acaule 38 .
Real-time quantitative RT-PCR analysis. Total RNA was extracted using the RNeasy Plant Mini Kit (Qiagen) and the RNase-Free DNase Set (Qiagen). Total RNAs of potato (S. tuberosum cv Sassy) were prepared from the leaves, stems, flowers, roots, stolons, tuber peels, and tuber sprouts. The extracted total RNAs of potato were used to synthesize first-strand cDNAs, using the ReverTra Ace ® qPCR RT Master Mix with gDNA remover. Quantitative reverse transcriptionalpolymerase chain reaction (RT-PCR) was performed with a LightCyclerNano (Roche), using FastStart Essential DNA Green Master (Roche) with the following primer sets: 1 and 2 for DPS, 3 and 4 for St7070, and 5 and 6 for EF1α, which is a housekeeping gene used for real-time RT-PCR normalization 39 (Supplementary  Table 7). Cycling was carried out as follows: 95°C for 10 min, 45 cycles at 95°C for 10 s, 60°C for 10 s, amplification at 72°C for 15 s, and holding at 95°C for 30 s. This was followed by ramping up from 60 to 95°C at 0.1°C s −1 to perform a melting curve analysis. Three biological replicates were analyzed in duplicate. The values obtained for the EF1α gene were used as internal references in potato, and the gene expression levels were normalized against these values. Data acquisition and analysis were performed using LightCyclerNano software (Roche).
Cloning of DPS and St7070 cDNAs. The potato cDNA template was prepared from mRNA isolated from sprouts of S. tuberosum cv Sassy, as described above. The cDNA fragments that contained the open-reading frames of DPS and St7070 were amplified by RT-PCR, using cDNA from tuber sprouts as a template, with primers 7 and 8 for DPS and primers 9 and 10 for St7070 (Supplementary Table 7). These primers were designed from the potato unigene sequences Sotub01g007130 and Sotub01g007070, respectively, using the Potato ITAG protein database (http:// solanaceae.plantbiology.msu.edu/index.shtml). The PCR products were cloned into the pCR ® 4 Blunt-TOPO ® vector (Thermo Fisher Scientific).
Generation of DPSi transgenic potato plants. The DPS-RNAi binary vector carried the inverted repeat of the partial DPS fragment interposing the third intron of the Arabidopsis (Arabidopsis thaliana) At4g14210 gene 40 under the control of the cauliflower mosaic virus 35S (CaMV35s) promoter in the T-DNA region. pRI201-AS (TaKaRa) was digested with KpnI and SmaI to disrupt multi-cloning site 2. The protruding end of the digested site was blunted with T4 DNA polymerase (TaKaRa) and then the blunted ends were ligated. The resulting plasmid was named pRI201-ΔMCS2 in this work. The third intron of At4g14210 was PCR amplified from pKT258 19 with primers 11 and 12, which contained restriction sites (Supplementary Table 7). The amplified fragments were cloned into the XbaI and NdeI sites of pRI201-ΔMCS2, and the resulting plasmid was named pRI201-ΔMCS2_PDS3. A 262 bp fragment of the DPS coding region was PCR-amplified using primers 13 and 14 (Supplementary Table 7). The fragment was inserted into the XbaI/BamHI and NdeI/SalI sites of pRI201-ΔMCS2_PDS3 to construct a DPSi vector. Potatoes (S. tuberosum cv Sassy) were then transformed using A. tumefaciens EHA105 cells harboring the DPSi vector. Briefly, internodal stem sections were excised from in vitro-grown potato shoots and cut into~5 mm sections. Explants were co-cultivated with agrobacterium for 48 h at 20°C on MS medium containing 3% (w/v) sucrose, trans-zeatin (2 μg ml −1 ), indole-3-acetic acid (50 ng ml −1 ), kanamycin (50 μg ml −1 ), and acetosyringone (100 μM) under dark conditions. After the co-cultivation period, the stem pieces were transferred to shoot induction medium composed of agarose-solidified MS medium containing 3% (w/ v) sucrose, trans-zeatin (2 μg ml −1 ), indole-3-acetic acid (50 ng ml −1 ), kanamycin (50 μg ml −1 ), and meropenem hydrate (50 μg ml −1 ), and cultured at 20°C under 16-h/8-h light/dark cycles. Stem pieces were transferred to fresh shoot induction medium every 2 weeks, until plantlets emerged. Well-developed plantlets were excised and transferred to MS medium containing 3% w/v sucrose, kanamycin (50 μg ml −1 ), and meropenem hydrate (50 μg ml −1 ). After 2-3 weeks, well-rooted shoots were selected and transformants were individually screened by genomic PCR using primer sets 15 and 16 (Supplementary Table 7), which targeted the kanamycin-resistance gene in the T-DNA region that was integrated into the genome.
Characterization of DPSi transgenic potato plants. Total RNA was then prepared from the leaves of four independent lines of the following in vitro-cultured plants: #13, #17, and #21. Quantitative RT-PCR analysis of DPS and St7070 were performed using primers 1/2 and 3/4, respectively (Supplementary Table 7), as described above. The SGAs that accumulated in the DPSi transgenic plants were extracted as described previously 19 with minor modification: Briefly, 50 mg of fresh plant material was frozen and homogenized with a mixer mill at 4°C. The homogenate was then extracted with 300 μl MeOH. After centrifugation, the supernatant was collected. The extraction was repeated three times and the extract, dried in vacuo, was re-dissolved in 400 μl MeOH. A 5 μl sample of solution was diluted with 295 μl MeOH and filtered through a 0.22-μm polytetrafluoroethylene membrane filter. An aliquot (2 μl) was then analyzed by UPLC-MS. LC-MS analysis was performed as described above, under "Feeding experiments", with minor modifications. Briefly, the mobile phases were water with 0.1% (v/v) formic acid (A) and acetonitrile (B), using the following gradient conditions: solvent B ramped linearly from 10% to 32.5% over 30 min; solvent B increased linearly to 100% over 1 min, held at 100% solvent B for 4 min; solvent B then returned immediately to 10%; followed by a 5 min re-equilibration period. The mass spectra were obtained in positive electrospray ionization mode, with a capillary voltage of 3 kV and a sample cone voltage of 60 V. Mass spectrometry scan mode with a mass range of m/z 350-1250 was used. Quantification of SGAs was carried out using selected ion monitoring (SIM). α-Solanine 9, α-chaconine 10, α-solamarine 7, and β-solamarine 8 were monitered at m/z 868, 852, 884, and 868, respectively, and were quantified using the calibration curves of the authentic samples.
Expression of the recombinant proteins in E. coli. The coding sequences of DPSi and St7070 were digested from the pCR ® 4 Blunt-TOPO ® vector (Thermo Fisher Scientific) with NdeI and SalI. Since Solyc01g006580 could not be amplified from cDNA in any tissue in tomato, the coding sequence of Solyc01g006580 with NdeI and SalI sites was synthesized (Eurofins Genomics Inc.) The DNA fragments were then ligated into the NdeI-SalI sites of pCold ProS2 (TaKaRa). E. coli strain BL21 (DE3) (Clontech) transformed with the constructed plasmid was grown at 37°C in LB medium containing ampicillin (50 μg ml −1 ) until the OD 600 reached 0.5. Recombinant protein production was induced by adding 0.1 mM isopropyl β-D-1thiogalactopyranoside and the culture was grown for 24 h at 15°C. The culture was then centrifuged at 7720 × g for 10 min at 4°C, and the cell pellets were resuspended in 5 ml of cold sonication buffer containing 50 mM Bis-Tris-HCl (pH7.2), 150 mM NaCl, 10% (v/v) glycerol, and 5 mM dithiothreitol. The solution was then sonicated three times for 30 s each on ice using a Bandelin Sonopuls HD 2070 ultrasonic homogenizer type MS73 (Sigma-Aldrich), at a sound intensity of 200 W cm −2 and centrifuged at 20,630xg for 10 min at 4°C. The recombinant His-tagged protein contained in the soluble fraction was purified using a His SpinTrap TALON column (GE Healthcare), according to the manufacturer's instructions. After two column washes, the adsorbed proteins were eluted twice in 200 μl of an elution buffer containing 50 mM sodium phosphate (pH 7.4), 30 mM NaCl, and 150 mM imidazole, and the elution was mixed. The purified recombinant proteins were used for further analyses.
18 O 2 -treated assay. A freshly prepared solution (1 ml) containing 100 mM Bis-Tris-HCl (pH 7.2), 5 mM 2-oxoglutrate, 10 mM sodium ascorbate, 0.2 mM FeSO 4 , 25 μM α-solamarine, was introduced into 10 ml round bottomed flask with a stir bar. Then, the flask was capped by a three-way cock which was connected to a vacuum pump line and a balloon filled with 18 O 2 , respectively. The reaction mixture was degassed in vacuo, and then 18 O 2 was incorporated into the reaction mixture by purging once and stirring with a stirrer bar. An aliquot of the reaction mixture was quickly transferred to 1.5 ml microtube and mixed with the purified enzyme. The reaction was carried out at 30°C for 30 min. The reaction was then stopped by incubation for 2 min at 90°C. After centrifugation, 20 μl of supernatant was diluted with 180 μl methanol and filtered through 0.2-μm nylon membrane filters (Acrodisc, Waters). An aliquot (2 μl) was analyzed using LC-MS, as described in "Feeding experiments" above.
Structural determination of the DPS enzymatic reaction product. The DPS enzymatic reaction with α-tomatine 6 was performed using 50 ml of reaction mixture as described above, with a minor modification, to determine the structure of the product of the enzymatic reaction catalyzed by recombinant DPS: the reaction was conducted using the crude enzyme overnight. The reaction mixture was extracted with water-saturated butanol after the pH was adjusted to 3.0 by the addition of hydrochloric acid. The butanol layer was corrected and, before the dried residue was dissolved in methanol, DPS product was concentrated in vacuo. The DPS reaction product dissolved in methanol was subjected to preparative scale high performance liquid chromatography (HPLC). The sample was injected into an ODS column (COSMOSIL 5C 18 -PAQ, 20 × 250 mm, Nacalai) at a column temperature of 40°C, and the flow rate was set to 4 mL min −1 . Separation was performed with isocratic mobile phases of 22% acetonitrile in H 2 O containing 0.1% trifluoroacetic acid (v/v), and the elute was detected at 203 nm. Then, the separated DPS reaction product was concentrated.
The molecular formula of the DPS enzymatic reaction product was determined to be C 50 H 81 NO 21 Fig. 17). Its structure was determined by two-dimensional NMR measurements, based on correlation spectroscopy (COSY), nuclear Overhauser effect spectroscopy (NOESY), heteronuclear single-quantum correlation spectroscopy (HSQC), and heteronuclear multiple-bond correlation spectroscopy (HMBC) techniques (Supplementary Data 1). The NMR measurements in this study were carried out in CD 3 SOCD 3 and CD 3 OD. The 13 C and 1 H assignments are summarized in Supplementary Tables 3 and 4, and COSY and HMBC correlations are shown in Supplementary Fig. 19. Samples were dissolved in NMR solvent immediately prior to measurement. In CD 3 OD, signals corresponding to H-23, observed at 2.84 ppm (br ddd, J = 21.3, 8.9, 6.8 Hz) and 3.12 ppm (1H, br ddd, J = 21.3, 4.1, 4.1) at the beginning, disappeared within 21 h. This observation indicated H-D exchange in CD 3 OD. Accordingly, C-23 was observed as a severely broadened signal in CD 3 OD by C-D coupling. In both solutions, C-16 signals were observed at 110.95 ppm (in CD 3 SOCD 3 ) and 117.85 ppm (in CD 3 OD), indicating their acetal like nature. Since C-16 of α-tomatine (the substrate of the DPS reaction) bears two carbon, one oxygen, and one hydrogen atom, the chemical shift in C-16 of the DPS reaction product indicates that oxidation occurred on C-16. In contrast, the chemical shift of C-22, within the hemiaminal structure of α-tomatine, observed at 192.21 ppm (in CD 3 SOCD 3 ) and 196.80 ppm (in CD 3 OD) in the DPS product at first suggested that C-22 exists as a carbonyl carbon of ketone or aldehyde. However, the HMBC correlation between H-26 and C-22 indicated that C-22 exists as an imine carbon instead of a carbonyl carbon. Since chemical shifts of typical imine carbons are observed around 140-170 ppm 42 , the chemical shift of C-22 of the DPS product appeared at a considerably lower field density. The observation indicated the presence of an iminium ion structure in the DPS product, and the reported 13 C chemical shift of tri-or tetra-substituted carbon atoms of iminium ions (over 180 ppm, Supplementary Fig. 35) 43 is comparable to the C-22 chemical shift of the DPS reaction product. In particular, the chemical shift of the tetra-substituted iminium ion in Supplementary Fig. 35 (195.1 ppm) is in good accordance with those observed in the DPS product (192.21 or 196.80 ppm). Furthermore, the NOE correlation between H-26 b and H-15β indicated the proximity of these hydrogen atoms. Additionally, the above-mentioned 13 C chemical shift of C-16 indicated that not only one oxygen atom but also one nitrogen atom binds to C-16. Accordingly, the structure of the DPS product was determined to be a zwitterion species, as shown in Supplementary Fig. 18, in which an iminium ion structure is constructed between C-16, the nitrogen atom, and C-22. The previously mentioned H-D exchange of H-23 in CD 3 OD, which further confirmed the iminium ionic form of the DPS product, is illustrated in Supplementary Fig. 36. The cationic nature of the nitrogen atom would increase the acidity of H-23 and accelerate the H-D exchange reaction.
Overexpression of DPS in tomato hairy roots. A binary vector for DPS overexpression was constructed from the binary vector pBINPLUS 46 : the coding region of DPS fused with the 5′-UTR of the A. thaliana alcohol dehydrogenase (AtADH) gene was inserted under the control of cauliflower mosaic virus 35S (CaMV35S) promoter and the heat shock protein terminator in the T-DNA region. The constructed binary vector was electroporated into A. rhizogenes ATCC15834. To generate transgenic hairy roots, tomato hypocotyls from 7-day-old seedlings were infected with A. rhizogenes strain ATCC15834 harboring a binary vector as described previously 29 . Briefly, bottom end of a hypocotyl segments (1.5 cm in length) were touched to a bacterial colony and then the segments were stood on the on solidified B5 medium containing 2% (w/v) sucrose with the contacted end up. Hairy roots emerging from infected sites were excised and subcultured twice every week on solidified B5 medium containing 2% (w/v) sucrose, 300 mg l −1 cefotaxime for disinfection, and 50 mg l −1 kanamycin for drug resistance selection. The selected lines were maintained by subculturing every week in 100 ml glass flasks filled with 20 ml of liquid B5 medium supplemented with 2% (w/v) sucrose with shaking at 100 rpm. in the dark. The accumulated SGAs in DPS overexpressed tomato hairy roots were extracted as described above in "Feeding experiment" and then analyzed using LC-MS by the same condition as α-tomatine 6 feeding experiment in "Feeding experiment".
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Sequence data for DPS (LC547753), St7070 (LC547754), and Solyc01g006585 (LC547755) are deposited in DDBJ. All data is available in the main text or the supplementary materials. Data supporting the findings of this study are available within the article and the supplementary materials. Source data are provided with this paper.