Seeds of P. angulata were harvested from the Langxi county, Anhui province of China in August 2018. The identity of the plant material was confirmed by Dr. Xiaodong Li at the Wuhan Botanical Garden, Chinese Academy of Science. P. angulata plants were grown in a growth chamber at 22°C under long-day (16 h of light/8 h of darkness) conditions. Transgenic plants were grown at an open field, located at the campus of Shanghai University, China.

Using Ws24ISO (GenBank accession: AXG64144.1) as a query, a BLASTP analysis of our recently constructed P. angulata calyx-derived transcriptome (unpublished data) revealed the candidate Pa24ISO. Deduced amino acid sequences of Pa24ISO were aligned with the previously reported sequences of SSR1 and SSR2 (see their accession numbers in Supplementary Table 2) using ClustalW software embedded in MEGA7. A maximum likelihood tree was inferred in MEGA7 using bootstrap method with1000 replications. Arabidopsis DWF1 was used to root the tree.

The PDS (phytoene desaturase) gene, associated with the photobleaching phenotype, was used as a reporter gene to test the TRV-based VIGS system for P. angulata, The cDNA sequence of PaPDS, as shown in Supplementary Figure 1, was obtained from the P. angulata calyx transcriptome, and a 382 bp-fragment of PaPDS was amplified from the P. angulata calyx cDNA. For silencing the Pa24ISO gene expression, a 297 bp-region, which is specific to Pa24ISO, but is not conserved in the putative PaSSR1 and PaSSR2, was selected (see their sequences in Supplementary Figures 2–4). All the primers used in this study are shown in Supplementary Table 1 in the Supporting Information. The TRV system constitutes the pTRV1 and pTRV2 vectors, which represent two RNA strands of the TRV virus system. The amplified DNA fragments of targeted genes were cloned into the pTRV2 vector via restriction cloning sites, yielding the pTRV2-PaPDS and pTRV2-Pa24ISO plasmids.

For down-regulating the expression of PaPDS gene, the pTRV2 (empty vector), pTRV1, and pTRV2-PaPDS were separately transformed into the Agrobacterium GV3101. The transgenic Agrobacterium strains were inoculated into LB medium supplemented with kanamycin at 50 mg/ml and rifampicin at 50 mg/ml at 28°C, and used for infection when OD₆₀₀ value of the Agrobacterium culture reaching 1.5. For infiltration of the P. angulata leaf, the TRV1-contained Agrobacterium culture was mixed at a 1:1 ratio with the cultures containing pTRV2-PaPDS, as well as the empty vector pTRV2 as a control. To improve the infection efficiency, 0.01% Silwet L-77 was added into the Agrobacterium mixer. Four to six-leaf-stage plants were selected for the infection. The infected plants were kept in darkness for 48 h, and then grown at an open field at the campus of the Shanghai University, China. The phenotype was then evaluated after 6–8 weeks. The same procedure was applied to generate the Pa24ISO-silenced plants.

The Real-time PCR (qRT-PCR) was performed on Bio-Rad CFX96™ Real-Time PCR instrument (Bio-Rad, Inc., United States) using TransStart Green qPCR SuperMix (Transgen). Primers used in this article are listed in Supplementary Table 1. The PCR program consisted of an initial step of 94°C for 30 s; 40 cycles of 94°C for 5 s and 60°C for 30 s; and then a dissociation stage of 95°C for 10 s, 65°C for 5 s and 95°C for 5 s. Each gene was assessed at least three biological replicates. The relative expression levels of the genes were calculated by the 2^-ΔΔCt method.

Every 1 g of powdered P. angulata samples was extracted with 1.7 ml of 90% methanol by sonication for 30 min, and this procedure was repeated three times. To normalize variations between the different extractions, isofraxidin at a final concentration of 1.5 μg/ml was added as an internal standard. The solvent extracts were collected by centrifugation at 12,000 g for 20 min, and passed through a membrane filter (0.22 mm) prior to the Ultra Performance Liquid Chromatography–Tandem Mass Spectrometry (UPLC–MS) analysis.

One microliter of the extracts was injected for the UPLC-MS analysis. LC-MS analysis was performed using a Q-Exactive Focus mass spectrometer, coupled with a VanquishTM UPLC system (Thermo Fisher Scientific, Bremen, Germany) and a HESI source (Thermo Fisher Scientific). The column (100 mm × 2.1 mm, 1.8 μm) was used to separate the sample, the column temperature was 40°C, and the flow rate was 0.35 ml/min. The mobile phases contain 0.1% formic acid (solvent A) and acetonitrile (solvent B), and the solvent gradient is set as follows: 15% solution B (0–1.0 min), 20–25% solution B (2.0–13.0 min), 25–40% solution B (13.0–25.0 min), 40–90% solution B (25–26 min), 90% solution B (26.0–29.0 min), and 15% solution B (29.0–33.0 min). The MS detection was performed with a full MS mode. The parameters of the mass spectrometers were as follows: spray voltage, 3.2 kV; source capillary temperature, 320°C; sheath gas flow rate (nitrogen), 25 ml/min; Aux gas flow rate (nitrogen), 8 ml/min; Aux gas heater temperature, 30°C, Scan range 150.0–700.0 m/z.

To optimize the expression of Pa24ISO in yeast, the full-length sequences of Pa24ISO were manually synthesized following the S. cerevisiae codon usage preference (see the optimized Pa24ISO cDNA sequence in Supplementary Figure 5), and inserted into a yeast expression vector pESC-HIS3 at BamHI/SalI sites, yielding the construct pESC-HIS3-Pa24ISO. The plasmid pESC-HIS3-Pa24ISO was transferred into a 24-methylenecholesterol-producing yeast strain YS11, which was recently developed by our group (Yang et al., 2021). As a negative control, the empty vector pESC-HIS3 was also transformed. The transgenic yeast colonies were selected on selection medium omitting histidine. Expression of the transformed gene was induced by 2% galactose.

To extract sterols from the transformed yeast cultures, yeast cells were pelleted from 2 ml of the yeast culture, broken with acid-washed glass beads, and extracted with 500 μl of 20% potassium hydrate in methanol (w/v) at 90°C for 2 h. When the saponified sample was cooled down, it was extracted with 2 ml of hexane for 30 min, and then was repeatedly extracted two times with another 1 ml hexane. The pooled hexane extracts were then dried in a vacuum freeze drier, and derivatized in 50 μl of N,O-bis(trimethylsilyl)-trifluoroacetamide (BSTFA) at 60°C for at least 1 h for sterol analysis by GC–MS.

Phytosterol was extracted as described previously (Knoch et al., 2018) with small modifications. One gram of powered sample was extracted with 3 ml chloroform/methanol (2:1, v/v) at 75°C for 1.5 h. After removing the solvent, the dried residue was saponified in 1 ml 10% (w/v) KOH in MeOH for 1 h at 80°C. When the samples were cooled to room temperature, 500 μl of H₂O and 500 μl of hexane were added, and extracted with shaking for 30 min. After centrifugation, the upper clear hexane extract was collected. The lower aqueous phase was then extracted twice with an additional 500 μl of hexane. The pooled hexane extracts were evaporated to dryness, and treated with 50 μl of the silylating reagent BSTFA [bis(trimethylsilyl)trifluoroacetamide]. The trimethylsilyl esters were dissolved in 80 μl of hexane, and a 1-μl sample was used for GC–MS analysis. For quantifying the sterol content, 40 μg of cholesterol (YuanYe, Shanghai, China) was added to each sample as an internal standard. GC–MS analysis was performed using a Shimadzu 2010 Plus GC machine coupled to a QP2020 Ultra mass spectrometer (Shimadzu Corporation, Canby, OR, United States). Samples were separated using a RTX-5 MS column (30 m × 0.25 mm × 0.25 mm) with helium being used as a carrier gas at a flow rate of 1.0 ml/min. The injection temperature was 280°C. The GC oven temperature program was as follows: 90°C for 1 min, then increased by 30°C/min to 300°C and hold at 300°C for 15 min. The mass of sterol compounds was detected in a select ion-scan mode with electronic ionization (70 eV), and monitored at multiple ions of m/z 343, 382, and 472 for campesterol, m/z 341, 365, and 470 for 24-methylenecholesterol, m/z 365, 386, and 470 for 24-methyldesmosterol, and m/z 329, 368, and 458 for the internal standard cholesterol.

Firstly we have shown that physalin B accumulates to higher levels in leaf, compared to other tissues of P. angulata (Figure 2A). This result is different from a previous report (Huang et al., 2014), in which calyx was identified to be the physalin-accumulating site with the highest amounts. This discrepancy is probably due to the difference of the P. angulata cultivars utilized between this and that studies. The identity of physalin B was determined by comparing its MS/MS fragmented pattern (Supplementary Figure 7) with that published in a previous literature (Huang, 2014). A Ws24ISO gene was previously cloned from Withania somnifera (Knoch et al., 2018). In this work, using Ws24ISO as a query, we performed a BLASTP analysis against the transcriptome derived from P. angulata calyx that was recently constructed by our group. One unigene (ID number: PB.4414.9), hereafter refer to as Pa24ISO, was found to show the highest amino acid identity (96.0%) with Ws24ISO. A phylogenetic analysis shows that Pa24ISO closely clustered with Ws24ISO in the tree (Supplementary Figure 7), indicating that it is a potential 24ISO candidate. The expression of Pa24ISO was profiled in different organs with real-time RT-PCR. Pa24ISO transcripts were detected in the highest levels in the leaf of P. angulata, but in much less levels in other tissues (Figure 2B). The expression pattern of Pa24ISO matched well with the accumulation profile of physalin B across these tissues (Figure 2A).

To test the enzymatic function of Pa24ISO, the Pa24ISO gene was codon-optimized by manually chemical synthesis for preferential expression in Saccharomyces cerevisiae, and inserted into a yeast expression vector pESC-HIS3. As a positive control, the known gene Ws24ISO was also synthesized, and expressed in the same way as Pa24ISO. Pa24ISO or Ws24ISO was expressed in a 24-methylenecholesterol-producing yeast strain YS11, which was previously made by Yang et al. (2021). The yeast extracts were subsequently prepared, and analyzed by GC–MS for the presence of 24-methyldesmosterol. The negative control was prepared by transferring the empty vector pESC-HIS3 into the YS11 strain.

Cells expressing Pa24ISO showed a clear GC–MS product peak, which was missing from the empty vector control. This new GC–MS peak showed the same retention times (Figure 3A) and the same mass spectra with the 24-methyldemosterol product synthesized by Ws24ISO (Figure 3B). Thus, it is confirmed that the protein encoded by Pa24ISO is able to convert 24-methylenecholesterol to 24-methyldesmosterol (see the proposed reaction in Figure 3C).

To test functions of genes in vivo in P. angulata, it is necessary to establish a virus-induced gene silencing (VIGS) system for P. angulata, as there have been no reports of VIGS in this plant. For this purpose, the tobacco rattle virus (TRV; Burch-Smith et al., 2006) vector based VIGS system was employed, using phytoene desaturase (PDS; Cunningham and Gantt, 1998) as a reporter gene. The sequence of P. angulata PDS (PaPDS) was retrieved from the calyx transcriptome, which is available in our laboratory, and a 382 bp-fragment of PaPDS was selected as a target sequence. Six-to-eight week-old seedlings were infected with the Agrobacterium slurry containing the pTRV2-PaPDS vector or the empty pTRV2 vector as a control. At around 4 weeks post infiltration, a greenish–whitish variegated phenotype was observed in the upper leaves. With the growing of the infected seedlings, some of the newly emerging leaves exhibited a completely white phenotype (Figure 4A). Interestingly, at around 10–12 weeks post infiltration, a complete white phenomenon was also observed in the newly developed flowers and calyx of some of the infected plants (Figure 4A). We harvested the seeds that were generated from the plants showing completely white flowers or calyx, and grew them in soils to see whether the photobleaching phenomenon can be inherited in their offspring. However, they grew as wild type plants, and no photobleaching phenomenon was observed in their newly grown leaves.

To check whether the white phenotype correlated with the decreased PaPDS transcripts by VIGS, the PaPDS expression was monitored by real-time PCR. As shown in Figure 4B, the transcript levels of PaPDS were largely decreased in the photobleached leaf, flower and calyx, respectively, by the PaPDS-VIGS, compared with their corresponding empty vector controls. This result suggests that the TRV-based VIGS system can be used to suppress expression of genes of interest in P. angulata.

After successfully establishing the VIGS system for P. angulata, we embarked experiments to silence the expression of Pa24ISO in vivo. Seven-week old plants were separately infiltrated with each of the Agrobacterium strains, which carried pTRV2-Pa24ISO, pTRV2-PaPDS, or pTRV2 empty vector, respectively. The pTRV2-PaPDS construct was utilized, as an indicator group, to monitor the silencing efficiency under different time lengths post the inoculation. Based on the visible phenomenon by pTRV2-PaPDS, the highest silencing efficiency was achieved at 8 weeks post the infection. Thus, after 8 weeks, the newly developed leaves from the infected plants transformed with the pTRV2-Pa24ISO, as well as the empty vector pTRV2, were harvested. To evaluate the silencing efficiency of Pa24ISO, quantification of the transcript level of Pa24ISO was performed with real-time PCR. In comparison with the control plants, which were infiltrated with the empty vector, the transcript level of Pa24ISO was down-regulated by 46% in the Pa24ISO-silenced plants (Figure 5A). It should be noted that the silencing of Pa24ISO also led to a decrease in the expression of a putative PaSSR1, however, it was not statistically significant (Supplementary Figure 8). Consistent with the reduced levels of Pa24ISO transcript, the Pa24ISO-silenced plants showed a sharp decline (54.3%) in the content of 24-methyldesmosterol, which was accompanied with increased accumulation of 24-methylenecholesterol and campesterol (Figure 5B and Supplementary Figure 8), when compared to the empty vector control samples. This result demonstrates that Pa24ISO is indeed the branch enzyme diverting the common substrate 24-methylenecholesterol toward biosynthesis of 24-methyldesmosterol-derived sterols, in competition with the campesterol-mediated carbon flux, as depicted in Figure 1B.

To investigate the effect of Pa24ISO silencing on biosynthesis of physalin, we quantified the content of physalin B in the newly developed leaves of the infected plants, through LC–MS/MS anlaysis. Physalin B appeared to be the major physalin compound in the P. angulata cultivar used in this study. The Pa24ISO-silenced plants displayed a 47% reduction in physalin B (Figure 5B), as compared with control leaves, indicating that Pa24ISO play a pivotal role in biosynthesis of physalin in P. angulata.