LjCPR gene sequences were obtained from Miyakojima MG-20 (Li et al., 2020) and the Gifu v1.2 genome database (https://lotus.au.dk/expat/) by using BLAST with A. thaliana CPR1 and CPR2 as the query. Phylogenetic analysis on LjCPRs was conducted using 29 other CPR sequences from different plant species, which were obtained from the web-based resource for Arabidopsis P450, cytochrome b, NADPH-cytochrome P450 reductases, and family 1 glycosyltransferases (www.P450.kvl.dk), NCBI (https://www.ncbi.nlm.nih.gov/), and various plant genome databases. The classification of CPR classes I and II was based on A. thaliana CPR1 and CPR2, as previously reported (Mizutani and Ohta, 1998). Accession numbers of CPR genes and genome databases used in this study are listed in Supplementary Table S1 . Multiple amino acid sequences were aligned using ClustalW and were used for tree construction using the neighbor-joining method with MEGA7. Gene co-expression analysis of L. japonicus was performed using an online transcriptomic database from the gene expression atlas web server https://lotus.au.dk/expat/ using the Gifu v1.2 genome database version. Based on the sequence identity between CPR genes of Miyakojima MG-20 and Gifu ecotype, gene IDs LotjaGi1g1v0345200.1, LotjaGi4g1v0301400.3, and LotjaGi4g1v0301300.1 from Gifu genome database were used for LjCPR1, LjCPR2-1, and LjCPR2-2 genes in this study ( Supplementary Table S2 ).

L. japonicus Gifu B-129 wild-type (WT) and LORE1 insertion lines (Fukai et al., 2012; Urbański et al., 2012) were provided by Miyazaki University, Japan, and Aarhus University, Denmark, through the National BioResource Project (NBRP). Seeds of L. japonicus were surface-sterilized using 2% (v/v) sodium hypochlorite and 0.02% (v/v) Tween 20 for 15 min in a seesaw shaker, rinsed three times with ultrapure water obtained from a Milli-Q Synthesis system (Millipore, Burlington, MA, USA), and placed onto a 0.8% agar plate. The seeds were allowed to germinate at 23°C for 4 days in the dark and 2 days under a 16-h light:8-h dark photoperiod.

β-Amyrin, α-amyrin, lupeol, erythrodiol, uvaol, oleanolic acid, ursolic acid, and asiatic acid were purchased from Extrasynthese (Genay, France). Betulin and MeJA were purchased from Sigma-Aldrich (St. Louis, MO, USA). Betulinic acid was purchased from Tokyo Chemical Industry (Tokyo, Japan). Soyasapogenol B and soyasapogenol A were purchased from Tokiwa Phytochemical (Chiba, Japan). Sophoradiol, 24-hydroxy-β-amyrin, and soyasapogenol E were kindly provided by Dr. Kiyoshi Ohyama (Tokyo Institute of Technology, Japan).

Induction of hairy roots was performed as reported previously (Suzuki et al., 2019), with slight modifications. Agrobacterium rhizogenes ATCC15834 strains were cultured on YEB plates for 2 days and suspended in sterilized water. The roots of 7-day-old WT seedlings were cut off, and A. rhizogenes were infected into the cross-sections of hypocotyls. After co-cultivation for 4 days, the infected seedlings were cultured on cefotaxime-containing hairy root elongation (HRE) solid medium for 2 weeks under a 16-h light:8-h dark photoperiod. After dissection, hairy roots were cultured under dark conditions. The root tip of a randomly chosen healthy WT hairy root clone was subcultured in 5 mL of cefotaxime-containing HRE liquid medium for 2 weeks and then transferred to 5 mL of HRE liquid medium without antibiotics with shaking at 90 rpm for another 2 weeks. Isolated hairy roots were cultured for 2 months at room temperature with subculturing every 3–4 weeks. Finally, hairy roots were cultured in 100 mL of HRE liquid medium at 25°C with shaking at 90 rpm for 4 weeks.

MeJA elicitation was conducted to test its effect on triterpenoid biosynthesis in L. japonicus hairy roots. MeJA preparation was performed as reported previously (Akhgari et al., 2019). A 20-mM MeJA stock solution was made by dissolving it in 40% (v/v) ethanol and then was filter sterilized (0.22 µm). The 4-week culture of WT hairy root was cut into similar portions (roughly 200 mg fresh weight each), cultured into 100 mL of HRE liquid medium without antibiotics, and incubated at 25°C with shaking at 90 rpm for another 4 weeks. The 4-week-old hairy root cultures were supplemented with final concentrations of 100 µM of MeJA. For control cultures, equal volumes (500 µL) of 40% ethanol were added to 100 mL of culture medium. The hairy roots were incubated under the same conditions as mentioned above and collected after 0, 3, 6, and 12 h for gene expression analysis and 0, 12, 24, and 24 h for metabolite analysis. The samples were flash-frozen in liquid nitrogen and stored at −80°C until use.

Total RNA was extracted from 100 mg of 4-week-old frozen L. japonicus MeJA-treated and control hairy roots using RNeasy Plant Mini Kit (Qiagen, Germantown, MD, USA). The RNA obtained was purified using the After Tri-Reagent RNA Clean-Up Kit (Favorgen Biotech Corp., Ping Tung, Taiwan) after digesting contaminated genomic DNA with recombinant DNase I (RNase-free) (TaKaRa Bio, Shiga, Japan). First-strand cDNA was synthesized from purified total RNA by PrimeScript RT Master Mix (Perfect Real Time) (TaKaRa Bio). qPCR analysis was performed using The LightCycler^® 96 (Roche, Basel, Switzerland) and FastStart Essential DNA Green Master (Roche). The primers used for qPCR analysis are listed in Supplementary Table S3 . The expression of ubiquitin (UBQ) gene was analyzed as a reference gene.

The loss-of-function mutant lines used in this study were the LORE1 mutant collection with Gifu B-129 genetic background annotated into the Miyakojima MG-20 v3.0 genome assembly. LjCPR1 and LjCPR2-1 were mapped into gene IDs Lj1g3v1548790.1 and Lj4g3v2107220.1 in the Miyakojima MG-20 v3.0 genome version, respectively. Several LORE1 lines mapped to each gene ID were selected from LORE1 mutant library (lotus.au.dk) and screened to obtain homozygous mutants. Two independent lines of Ljcpr1 (30003941 and 30059903) and Ljcpr2-1 (30037476 and 30065390) homozygous insertion mutants were obtained. Each mutant line contains other exonic or intronic LORE1 insertions other than LjCPR genes ( Supplementary Table S4 ). These chosen mutant lines were cultivated in soil, and their progenies were cultivated in a hydroponic system.

For soil-cultured plants, the 7-day-old WT and mutant seedlings were moved to pots with a mixture of soil and vermiculite and cultivated for 3 months. The produced seeds were then collected for hydroponic cultivation. The seed pods from three independent lines for each Ljcpr1 and Ljcpr2-1 homozygous insertion mutant were counted. Then, the pod length of 26 randomly selected pods from each Ljcpr1 and Ljcpr2-1 homozygous insertion mutant line was measured using a digital vernier caliper. The 7-day-old WT and mutant seedlings of the soil-cultured plant progenies were first cultured in 5-mL tubes containing basal nutrient solution. After 2 weeks, the hydroponic plants were scaled up into 50-mL tubes containing basal nutrient solution and cultivated until the flowering stage. The hydroponic medium was renewed weekly.

Genomic DNA was extracted and purified from the leaves using FavorPrep™ Plant Genomic DNA Extraction Mini Kit (Favorgen Biotech Corp.) to screen for homozygous Ljcpr1 and Ljcpr2-1 mutants using PCR with CPR-specific primers and a LORE1-specific primer ( Supplementary Table S5 ; Supplementary Figure S1 ). PCR was performed using KOD FX Neo following the manufacturer’s instructions (Toyobo, Osaka, Japan) using the same DNA concentration for all samples. Triterpenoids were extracted from the soil- and hydroponic-cultured roots of homozygous and heterozygous mutants and were analyzed as described below.

The multiplex guide RNA (gRNA)-expressing CRISPR-Cas9 vector, pMgP237-2A-GFP (Hashimoto et al., 2018), was used for genome editing of L. japonicus hairy root. The target sequences of the gRNAs ( Supplementary Figure S2 ) were selected from LjCPR1 gene using the web-based tool CRISPRdirect (https://crispr.dbcls.jp/) (Naito et al., 2015). Two gRNA target sequences were simultaneously transferred into the pMgP237-2A-GFP vector as described previously (Nakayasu et al., 2018), generating the T1/T2-pMgP237 vector. Three sets of different gRNA designs (Nos. 2, 4, and 5) targeting LjCPR1 gene were constructed using the primers listed in Supplementary Table S6 . A. rhizogenes ATCC15834 was transformed with the pMgP237 empty vector or the T1/T2-pMgP237 vector.

The induction of hairy roots was described above. Crude genomic DNA extraction and PCR were performed using KOD FX Neo following the manufacturer’s instructions (Toyobo). Mutagenesis was confirmed using PCR with specific primers for each gRNA design ( Supplementary Table S6 ; Supplementary Figure S3 ) and analyzed using a heteroduplex mobility assay (HMA) and the MCE-202 MultiNA microchip electrophoresis system (Shimadzu, Kyoto, Japan) following the manufacturer’s instructions. The target sequences amplified from putative mutants were cloned into the pJET1.2/blunt vector (CloneJET PCR Cloning Kit; Thermo Fisher Scientific, Waltham, MA, USA). Insertion and deletion mutations were confirmed by sequencing several randomly selected clones.

Multiple sequences of nucleotides and amino acids were aligned using MEGA11 software (Tamura et al., 2021). The protein structure of LjCPR1 from wild-type and L1-4.2 mutant lines was modeled using the SWISS-MODEL (Arnold et al., 2006). The LjCPR1 models were constructed using two crystal structure templates, A0A0R4J338.1.A (NADPH-cytochrome P450 reductase of Glycine max) and 5GXU.1.A (NADPH-cytochrome P450 reductase 2 of A. thaliana), with 87.7% and 72.8% identity, respectively. These two templates were used due to the different properties exhibited by the crystal structures. The crystal structure of A0A0R4J338.1.A presents the complete protein structure of G. max CPR, including the transmembrane region. Meanwhile, 5GXU.1.A is a truncated A. thaliana CPR crystal structure without the transmembrane region and includes the position of the covalently bonded flavin mononucleotide (FMN) ligand on the CPR protein. The protein models were then visualized using PyMOL (DeLano, 2020) to compare the conformations of the wild-type and mutant proteins.

Triterpenoid extraction was performed as reported previously (Suzuki et al., 2019), with slight modifications. Plants at the flowering stage and hairy roots were lyophilized and powdered using a multi-bead shocker (Yasui Kikai, Osaka, Japan). Powdered tissues (20.00 ± 0.3 mg) were extracted three times with 1 mL of methanol using a sonication-assisted method. Completely dry extracts were resuspended in 2 mL of MeOH in 4 M of HCl (1:1). The extracts were incubated at 80°C for 1 h to remove the sugar moieties of triterpenoid saponins. The hydrolyzed products were extracted three times with hexane:EtOAc (1:1) and dried completely. The obtained pellet was resuspended in 500 μL of MeOH:chloroform (1:1). A portion of the solution was dried in a gas chromatography–mass spectrometry (GC-MS) vial. Additionally, 100 μL of the solution was evaporated and trimethylsilylated using a mixture of 50 μL of N,N-dimethylformamide (Kishida Chemical Co., Ltd., Osaka, Japan) and 50 μL of BSTFA : TMCS (99:1) (TCI) at 80°C for 30 min. For semi-quantitative analysis, an asiatic acid authentic standard was applied to the plant tissue powder before extraction.

GC-MS analyses were performed as reported previously (Suzuki et al., 2019) on a 5977A MSD mass spectrometer (Agilent Technologies, Santa Clara, CA, USA) connected to a 7890B gas chromatograph (Agilent Technologies) with an HP-5MS UI (30 m × 0.25 mm, 0.25-μm film thickness; Agilent Technologies) capillary column for qualitative analysis. The injection temperature was set at 250°C. The column temperature program was as follows: 80°C for 1 min, increase to 300°C at a rate of 20°C/min, and hold for 28 min. The carrier gas was helium at a flow rate of 1.0 mL/min. The ion source temperature was 230°C, and the quadrupole temperature was 150°C. The derivatized sample (1 µL) was injected in splitless injection mode. Peaks were identified by comparing their Rt and mass spectra with those of authentic standards ( Supplementary Figure S4 ). Samples were analyzed in selected ion monitoring (SIM) mode for relative quantification by extracting the mass chromatogram in respective extracted ion chromatogram (EIC) for each metabolite as listed in Supplementary Table S7 .

The significance of differences was determined using a one-way single-factor analysis of variance (ANOVA). The significance of the means was separated using Tukey’s test. p-Values less than 0.05 (p < 0.05) were considered significant in this study. All the statistical analyses were performed with JASP (JASP Team 2020).

The two ecotypes of L. japonicus most commonly used for research are Miyakojima MG-20 and Gifu B-129 (Hashiguchi et al., 2011). The most recent Miyakojima MG-20 transcripts submitted by Li et al. (2020) showed that L. japonicus has one copy of CPR class I gene and two copies of CPR class II genes ( Supplementary Figure S5-A ). This latest genome version of Miyakojima MG-20 was obtained using the Illumina HiSeq 2500 platform and PacBio sequencing system (Li et al., 2020). Interestingly, based on a high-quality L. japonicus Gifu v1.2 genome database constructed using 100× PacBio read coverage and RNA-seq analysis (Kamal et al., 2020), seven CPR genes were annotated. One CPR gene belonged to CPR class I, while six CPR genes belonged to CPR class II ( Supplementary Figure S5-B ). In both ecotype genomes, CPR class I gene was located on chromosome 1, while several copies of CPR class II genes were located close to each other on the same locus of chromosome 4 ( Supplementary Figure S5 ).

Based on phylogenetic tree analysis, LjCPR class I genes from Miyakojima MG-20 and Gifu B-129 were in the same clade, while the LjCPR class II genes from Miyakojima MG-20 and Gifu B-129 branched into two different clades, namely, LjCPR2-1 and LjCPR2-2 ( Supplementary Figure S6 ). Two copies of the CPR class II in Miyakojima MG-20 genome belonged to two different LjCPR2 clades. However, from the six LjCPR class II genes found in the Gifu genome, three genes belonged to the LjCPR2-1 clade, and three genes belonged to the LjCPR2-2 clade. A multiple sequence alignment analysis and amino acid and nucleotide sequence identity matrix were constructed to evaluate homology among LjCPR genes ( Supplementary Figure S7 ; Supplementary Tables S2A, B ). LjCPRs showed that LjCPR class I from both ecotypes had 100% nucleotide similarity ( Supplementary Tables S2A, B ). All LjCPR2-2 genes from Gifu showed 100% nucleotide similarity, but they are not identical to Miyakojima MG-20 LjCPR2-2, with 99.1% amino acid similarity ( Supplementary Tables S2A, B ). In contrast, one isoform of LjCPR2-1 from Gifu (LotjaGi4g1v0301400.3) showed a sequence identical to that of LjCPR2-1 from Miyakojima ( Supplementary Table S2B ).

Transcriptomic data were retrieved from the database to analyze the gene expression level of different CPR classes in L. japonicus. Unfortunately, the transcriptomic database was unavailable for the L. japonicus Miyakojima MG-20 genome (Li et al., 2020). Therefore, in this study, only the expression level of LjCPR genes from the Gifu v1.2 transcriptomic database was analyzed. LjCPR1 was mapped to probe ID LotjaGi1g1v0345200. LjCPR2-1 was mapped to probe ID LotjaGi4g1v0301400. The identical sequence of Gifu LjCPR2-2s was mapped to the LotjaGi4g1v0301300 probe, labeled as LjCPR2-2 in this study. CPR class I and II genes of L. japonicus showed distinct expression patterns ( Figure 1 ). As reported in our study before, based on Miyakojima MG-20 v3.0 transcriptomic database, CPR class I was found to be constitutively expressed with lower and more stable expression levels. In contrast, CPR class II was generally found to have higher expression levels than CPR class I, which varied depending on tissues and treatments (Istiandari et al., 2021). Similarly, in the L. japonicus Gifu v1.2 transcriptomic database, both LjCPR2-1 and LjCPR2-2 showed higher inducible expression than LjCPR1 ( Figure 1 ). However, LjCPR2-2 expression was even lower than LjCPR1 expression in some of the samples but seemed to be tissue-specific. Interestingly, LjCPR2-2 was predominantly expressed in immature flowers and showed the highest expression level among the three CPR genes. Nevertheless, LjCPR2-1 expressions were generally higher in almost all samples than those of LjCPR2-2 and LjCPR1, implying LjCPR2-1 is the dominant LjCPR class II in L. japonicus.

Triterpenoid biosynthetic pathway in L. japonicus is shown in Figure 2 . The major pentacyclic triterpene backbones, β-amyrin, α-amyrin, and lupeol, are converted from 2,3-oxidosqualene by bAS, aAS, and LUS, respectively ( Figure 2 ). The carboxylation of β-amyrin, α-amyrin, and lupeol at the C-28 position by CYP716A51 leads to oleanolic acid, ursolic acid, and betulinic acid production (Suzuki et al., 2019). CYP93E1 catalyzes hydroxylation at the C-24 position of β-amyrin to produce 24-hydroxy β-amyrin (Shibuya et al., 2006; Seki et al., 2008). CYP72A61 converts β-amyrin into sophoradiol by adding a hydroxyl group at the C-22 position (Fukushima et al., 2011). 24-Hydroxy β-amyrin is further oxidized by CYP72A61 that catalyzes hydroxylation at the C-22 position, producing the major soyasaponin aglycone soyasapogenol B (Fukushima et al., 2011).

MeJA (100 μM) was added to 1-month-old L. japonicus hairy roots that were sampled at different time points to elucidate the effect of phytohormone elicitation on the L. japonicus triterpenoid biosynthetic pathway ( Figure 2 ). L. japonicus accession Gifu B-129 was used in this study. The qRT-PCR result of extracted RNA from these treated hairy roots showed differences in the regulation of some triterpenoid biosynthetic genes ( Figure 3 ). While LjCPR1 gene was quickly downregulated by MeJA addition, LjCPR2-1 and LjCPR2-2 expressions were significantly upregulated up to four times when compared to the mock sample 3 h after MeJA addition. Interestingly, similar to the LjCPR1 expression pattern, CYP716A51 and LUS genes also showed downregulation by MeJA addition even 12 h after the treatment. However, very high and quick upregulation was observed in bAS, CYP93E1, and CYP72A61 expression levels. The expression of bAS was upregulated more than 20 times when compared to the control 12 h after the treatment, while CYP72A61 and CYP93E1 upregulation was the highest 6 h after the treatment at approximately five times higher than those of the control. Another triterpene OSC gene, aAS, was not detected in all samples. To observe the effect of MeJA on primary metabolites such as phytosterols, CAS and LAS expressions were also analyzed. CAS and LAS are cycloartenol and lanosterol synthase, respectively, which represent the branch-off entry of phytosterol biosynthesis after 2,3-oxidosqualene cyclization ( Figure 2 ). Both CAS and LAS expressions were instantly downregulated 3 h after MeJA addition, but then the expressions increased after 6 h and returned to levels similar to those of the control 12 h after the treatment ( Figure 3 ).

Based on GC-MS analysis, the change in expression levels of triterpenoid biosynthetic genes due to MeJA treatment affected the triterpenoid production in L. japonicus hairy roots ( Figures 2 , 4 ). The triterpenoids analyzed in this study were annotated with numbers corresponding to the chromatogram peaks of MeJA-treated hairy root extracts shown in Supplementary Figure S8 . Consistent with the significant upregulation of bAS and CYP93E1 expressions ( Figure 3 ), MeJA addition resulted in a significant increase in β-amyrin and 24-OH β-amyrin levels ( Figure 4 ). The β-amyrin level was increased 10 times 12 h after the treatment and continued to increase until it reached 20 times higher than that of the control even 48 h after the treatment. Similar to β-amyrin level, the 24-OH β-amyrin level also increased rapidly after 12 h to 10 times higher than that of the control and reached a maximum of 24 h after the treatment at 15 times higher than that of the control. Sophoradiol production level also increased 24 h after the treatment. In addition, soyasapogenols showed gradually increasing production levels even 48 h after the treatment. However, no change in oleanolic acid and betulinic acid levels, and even a significant reduction of the ursolic acid level, was observed upon MeJA treatment. However, the lupeol level showed a significant increase after 24 h up to four times that of the control, similar to α-amyrin production. All triterpenoids’ peaks were annotated, and the chromatogram area was measured by comparing the retention time and mass spectrum with their authentic standards. The analyzed triterpenoids were the major triterpenoid constituents in L. japonicus (Suzuki et al., 2019), which served as the representative for triterpenoids in this study. The levels of other minor triterpenoids were too low to be detected by GC-MS. Campesterol, β-sitosterol, and stigmasterol, as the three major sterols in plants, were analyzed to observe the effect of MeJA on the phytosterol pathway. Due to the lack of standard compounds, the phytosterol peak was annotated based on similarity with a mass spectrum from the NIST library ( Supplementary Figure S9 ). All phytosterols showed only a small increase after 24 h and returned to levels similar to those of the control after 48 h ( Figure 4 ).

The triterpenoid profiles of LORE1 insertion mutant lines of L. japonicus accession Gifu B-129 (Fukai et al., 2012; Urbański et al., 2012) were analyzed to investigate the effect of loss-of-function of either LjCPR1 or LjCPR2 genes. Based on a previous study, LUS and CYP716As genes that are responsible for lupeol and betulinic acid production, respectively, were upregulated in secondary aerenchyma of hydroponic-cultured L. japonicus (Suzuki et al., 2022) and soybean plants under flooded conditions (Takahashi et al., 2022). Therefore, to investigate if plant culture condition affects LjCPRs’ involvement with specific CYPs, the Ljcpr1 and Ljcpr2-1 loss-of-function mutant lines were both cultured in soil and a hydroponic system. The seeds of two independent homozygous mutant lines, 30003941 (L1-A) and 30059903 (L1-B), which contain a single LORE1 insertion into the first and fourth exons of LjCPR1, were obtained ( Figure 5A ). Other seeds of two independent homozygous mutant lines 30037476 (L2-1A) and 30065390 (L2-1B), which contain a non-single insertion LORE1 insertion into the first and fourth exons of LjCPR2-1, were obtained ( Figure 5B ).

While both Ljcpr1 mutants are single insertion homozygous mutants ( Supplementary Figure S1 ), PCR genotyping results showed that there were heterozygous insertions on other genes other than LjCPR2-1 in these mutant lines ( Supplementary Figure S1 ). Therefore, the effect of other gene mutations on both Ljcpr2-1 mutant lines should not be ruled out. Other seeds of two independent mutant lines, which contain a LORE1 insertion on the LjCPR2-2 exon, were also screened; however, no homozygous mutant was obtained. Therefore, only Ljcpr1 and Ljcpr2-1 loss-of-function mutants were analyzed in this study. Ljcpr2-1 mutant lines might serve as a representative mutant line for LjCPR class II since they were more dominantly expressed than LjCPR2-2 ( Figure 1 ). Some WT allele-specific primer sets (F and R) and insertion allele-specific primer sets (F and P2) were designed for LjCPR1 and LjCPR2-1 genes and other genes that may have LORE1 insertions in their exons ( Supplementary Figure S1 ; Supplementary Table S2 ) according to a previous report (Urbański et al., 2012).

No significant difference in the physiology of the Ljcpr mutants could be observed in the plant leaves, stems, or roots. However, a notable change was observed in the seed pods ( Figures 5C–E ). The collected seed pods from soil-cultured mutants were then measured and counted to observe the effect of Ljcpr loss on the seed physiology ( Figures 5C, D ). The number of pods and pod length of two homozygous Ljcpr1 mutant lines showed a more significant reduction compared to Ljcpr2-1 mutant seed pods ( Figure 5D ). The physiological change of the seed pod can be observed in Figure 5E .

The triterpenoid sapogenin profile of each soil-cultured mutant plant was analyzed using GC-MS, and the result showed different profiles for Ljcpr1 and Ljcpr2-1 loss-of-function mutant lines ( Figure 6 ), similar to that of the hydroponic-cultured mutants ( Supplementary Figure S10 ). A significant difference was shown both in soil-cultured and hydroponic-cultured mutant plants. β-Amyrin, oleanolic acid, lupeol, 24-OH β-amyrin, and sophoradiol levels were significantly decreased in Ljcpr2-1 mutant lines but showed little or no change in Ljcpr1 mutants ( Figure 6 and Supplementary Figure S10 ). Interestingly, both Ljcpr1 and Ljcpr2-1 mutant lines showed significant decreases in betulinic acid and ursolic acid and no change in soyasapogenol E and A ( Figure 6 and Supplementary Figure S10 ). Both mutants showed lower levels of soyasapogenol B. However, the decrease in soyasapogenol B was more significant in hydroponic-cultured ( Supplementary Figure S10 ) than in the soil-cultured mutant roots ( Figure 6 ). As the three major phytosterols, the campesterol, β-sitosterol, and stigmasterol levels were analyzed and annotated based on mass spectra from the NIST library to investigate the effect of Ljcpr loss-of-function on the primary metabolisms ( Supplementary Figure S9 ). However, due to a shift in retention time, campesterol could not be detected in hydroponic-cultured samples. Campesterol, β-sitosterol, and stigmasterol were shown to be significantly reduced in both Ljcpr1 and Ljcpr2-1 loss-of-function soil-cultured mutants ( Figure 6 ).

To directly confirm the involvement of different LjCPR classes on the triterpenoid biosynthetic pathway, we used the CRISPR-Cas9 system. We previously described a CRISPR-Cas9 vector, pMgP237-2A-GFP, which has been used to generate L. japonicus hairy root knockout mutants (Suzuki et al., 2019). Since LjCPR2-1 and LjCPR2-2 genes have very similar sequence identities, it was very difficult to obtain single or complete null-mutant of double-knockout Ljcpr2s due to a lower probability of removing all the intact sequences. Thus, in this study, only Ljcpr1 knockout mutants were successfully obtained and analyzed. Two target sequences were simultaneously integrated into the vector to generate double tgRNA-pMgP237. Transgenic hairy roots were induced by A. rhizogenes ATCC15834 harboring double tgRNA-pMgP237 or the empty vector as a control. A total of nine target sequences on LjCPR1 ( Supplementary Figure S2 ) were selected using CRISPRdirect software (Naito et al., 2015).

Putative Ljcpr1 hairy root mutant lines were selected using PCR and electrophoresis (HMA). Extra bands were observed in nine of the putative mutant hairy root lines but not in the control hairy root lines ( Supplementary Figure S3 ), suggesting that mutations occurred in LjCPR1 gene and produced heteroduplex PCR fragments. Genomic DNA fragments around the target sites were cloned and sequenced to confirm mutations. Mutated alleles were not found in the control lines, EV-1 ( Supplementary Figure S3 ). No WT sequences were detected in all obtained mutants ( Supplementary Figure S3 ). Four mutant hairy root lines with longer nucleotide deletions were chosen for further analysis ( Figure 7 ). These mutant hairy root lines were generated by three gRNAs (gRNA 4A, 4B, and 5B) that successfully cut the target nucleotides on the FAD and FMN/FAD hinge domain of LjCPR1 gene ( Figure 7A ), which resulted in three frameshift mutant lines (L1-4.1, L1-5.1, and L1-5.2) and one non-frameshift mutant line (L1-4.2) ( Figure 7B ).

Additionally, we analyzed the amino acid sequences of all LjCPR1 mutant proteins from the mutant hairy root lines. All frameshift mutations in the hairy root mutant lines L1-4.1, L1-5.1, and L1-5.2 caused premature termination, resulting in a lower number of amino acids (approximately 200) when compared to those in the WT; however, the deletion of 72-bp allele in the non-frameshift mutant L1-4.2 did not cause early termination of the protein ( Supplementary File 1 ). The non-frameshifting mutation resulted in the absence of amino acids 222nd to 246th in the WT LjCPR1. Structural analysis of the mutant protein indicated that the deletion of 72-bp nucleotides encoded one strand of α-helix near the FMN-binding domain ( Supplementary Figure S11A ). Nevertheless, comparing the WT and mutant LjCPR1 protein structure models using two different templates revealed that the loss of 24 amino acids in the LjCPR1 mutant line L1-4.2 did not alter the overall conformation of the protein ( Supplementary Figures S11A, B ). The amino acid path in the mutant, where one α-helix was missing, was directly connected to amino acid chains and continued into a β-sheet structure with the same configuration as the WT LjCPR1 ( Supplementary Figures S11A, B ).

The triterpenoid profiles of control and Ljcpr1 mutant hairy roots were then analyzed using GC-MS ( Figure 8 ). All the frameshift mutants from different target regions showed significantly lower betulinic acid and ursolic acid content when compared to the control. Frameshift mutant lines L1-5.1 and L1-5.2 showed similar levels of lupeol, α-amyrin, and soyasapogenols compared to the control, while L1-4.1 showed significantly lower levels of those triterpenes compared to the control. All frameshift mutant lines showed similar levels of β-amyrin, 24-OH β-amyrin, and sophoradiol compared to the control. Interestingly, the non-frameshift L1-4.2 mutant line showed significantly higher oleanolic acid and betulinic acid than the control. In the phytosterol biosynthesis pathway, the knockout of Ljcpr1 genes showed significantly lower β-sitosterol and stigmasterol amounts than the control. A similar experiment was repeated in another target region of LjCPR1 gene to generate other Ljcpr1-KO hairy root mutant lines ( Supplementary Figure S12 ). Similarly, the GC-MS result on these Ljcpr1-KO hairy root mutants showed significantly lower betulinic and ursolic acid levels and no effect on β-amyrin, 24-OH β-amyrin, sophoradiol, and soyasapogenol contents ( Supplementary Figure S12 ). The total ion chromatogram (TIC) ( Supplementary Figure S13 ) showed a significant decrease in the betulinic acid level of Ljcpr1 knockout mutant hairy roots.