The flowers and 45-DAF fruits from A. longiligulare and A. villosum were collected and frozen at −80°C. The fruit was separated into pericarps and seeds. In addition, the mature fruits (approximately 90-DAF) of A. villosum and A. longiligulare were dried to a constant weight at 50°C and then stored in a dryer.

Approximately 0.3 g of fresh or dry materials was ground frozen in liquid nitrogen, vortexed and then extracted with 1.5 mL hexane using an ultrasonic cleaner for 0.5 h followed by incubation at 40°C for 1 h. Samples were then centrifuged at 12,000 rpm for 5 min, after which the resulting supernatants were pipetted into new 2 mL tubes. Next, 1 mL of hexane extract was pipetted into 1.5 mL vials for GC-MS analysis. The extracts were then analyzed using an Agilent 7890B Gas Chromatograph with a 5977B inert Mass Selective Detector (Agilent, United States). Helium was used as the carrier gas (1 mL/min), then separated on an HP-5MS column (30 m × 250 μm × 0.25 μm film thickness). For separation, the GC oven temperature was programmed for an initial temperature of 35°C for 5 min followed by an increase of 12°C/min to 300°C, which was held for 5 min. A NIST17/demo1 Mass Spectral Library was used for metabolite identification. The terpenoid compounds were identified by the mass spectral library, after which the main monoterpenoids investigated in this study, including α-pinene, camphene, myrcene, limonene, linalool, camphor, borneol and bornyl acetate, were further identified using their authentic standards. The contents were quantified based on bornyl acetate standard curves, and there were three biological replicates for each tissue.

The total RNA of each sample (including seeds, pericarps and flowers) was isolated using an OmniPlant RNA kit (CWbiotech, China), following the manufacturer’s protocol. For the RNA extraction of seeds, the following step was added to remove the polysaccharides: after blending by vortexing the lysate and sample, 1/20 volume of isopropyl alcohol and 1/20 volume of high salt solution mixture (1.2 M NaCl + 0.8 M sodium citrate) were added to the lysate and blended upside down. The RNA quality was then verified using an ultraviolet spectrophotometer (TIANGEN, China) and checked using RNase free agarose gel electrophoresis. RNA with an OD260/OD280 of 1.8–2.2 was used for subsequent analyses.

RNA extracted from the 45-DAF seeds was used for the transcriptome sequencing without replicates. The library preparation and transcriptome sequencing, de novo assembly and annotation were conducted by Gene De novo Co., as described previously (He et al., 2018; Wang et al., 2018).

The unigenes of candidate AlTPS genes were screened out from the Pfam annotation (Finn et al., 2016) of the transcriptome data of A. longiligulare using the keyword “terpene synthase,” combining with the local-Blast using the sequences of AvTPS1–AvTPS10 (Wang et al., 2018). The unigenes shorter than 500 bp were filtered out.

The gene-specific primers used for AlTPS2, AvTPS2, and AlTPS3 cloning are listed in Supplementary Table 1. Because AvTPS2 lacks a complete ORF (open reading frame), its 5′-end and 3′-end were amplified using a SMARTer^® RACE Kit (Takara, Japan) according to the user manual. The full-length cDNA was amplified using Prime STAR Max DNA Polymerase (Takara, Japan) following the manufacturer’s instructions. The PCR conditions used were as follows: 98°C for 1 min followed by 30 cycles of 98°C for 10 s, 50–60°C for 15 s, and 72°C for 15 s, and then final extension at 72°C for 5 min. The product was then ligated into pLB cloning vector (Tiangen, China), which was subsequently transformed into Escherichia coli DH5α and sequenced.

ORFs of AlTPS3, AlTPS2, and AvTPS2 excluding the N-terminal transit peptides were amplified, and proper restriction enzyme sites were added at each end by PCR using the primers described in Supplementary Table 1. The ORFs were then ligated into the pET32a expression vector for AvTPS2 and AlTPS3 and the pMAL-c5X vector for AlTPS2 using an In-Fusion Cloning Kit (Takara, Japan). Next, the positive constructs were transformed into competent Rosetta (DE3) cells, and the positive colonies were inoculated into LB media containing 50 μg/mL carbenicillin and 25 μg/mL chloramphenicol and incubated at 37°C until the OD 600 reached 0.4–0.6. The proteins were subsequently induced by incubation in the presence of 0.1 mM isopropyl-β-D-1-thiogalactopyranoside (IPTG) at 16°C for 16 h. Next, the recombined proteins were purified with NI-NTA resin (Qiagen, Hilden, Germany) or MBP-tag columns (Qiagen, Smart-Life Sciences, China), following the relative manual recommendations: for NI-NTA resin column, unbound proteins were washed with 20 mM imidazole phosphate buffer (pH 7.4) and then target proteins (AvTPS2 and AlTPS3) were eluted by 200 mM imidazole phosphate buffer; for MBP-tag columns, unbound protein was washed by eluant A (20 mM Tris-HCl, 200 mM NaCl, 1 mM EDTA and DTT, pH 7.4) and then target protein (AlTPS2) was collected using eluant B (eluant A adding 20 mM maltose). The purified proteins were dialyzed in PD-10 Desalting Columns (GE Healthcare).

The enzymatic reaction conditions and methods for the analysis of AlTPS2, AvTPS2, and AlTPS3 were conducted as described by Wang et al. (2018). The enzyme reacted with the substrate GPP for 1 h at 30°C, after which samples were dephosphorized by treatment with 1.0 μL alkaline phosphatase (Thermo Fisher, United States) for another hour at 37°C with 200 μL hexane overlaid. The hexane phase was then extracted and used for GC-MS analysis. Enzymatic products were detected based on GC-MS as previously described (Wang et al., 2018). The NIST17 Mass Spectral Library was used for metabolite identification. Additionally, the standards of linalool, camphene, limonene and borneol were utilized for further identification. Each enzyme assay was performed with at least three replicates.

The primers for site-directed mutation were listed in Supplementary Table 1. The mutant of AlTPS3-A496G and AvBPPS-G495A were performed using a site-directed mutagenesis kit (TIANGEN, Beijing, China). PCR mixture in a volume of 50 μL consisted of 1 U of FastAlteration DNA Polymerase, 5 × FastAlteration Buffer, 10 μM primers, and 50 ng of recombinant plasmid pET-32a -AlTPS3. The PCR mixture was denatured at 95°C for 2 min, followed by 30 cycles of 20 s at 94°C, 10 s at 62°C, and 2.5 min at 68°C, and a final 5-min extension at 72°C, the PCR product was incubated at 37°C with DpnI restriction enzyme for 1 h to remove the template plasmid. Then, the PCR product was transformed into competent cells of E. coli, and the mutant site was confirmed by sequencing. The subsequent prokaryotic expression, protein purification were performed following the methods mentioned above. Enzyme assay was performed to compare the products of wild type and mutant using the methods described above with the overnight incubation with GPP instead of 1 h.

The secondary structures of the proteins were predicted through SwissModel Workspace,¹ using 1N24.1.A [crystal structure of (+) -bornyl diphosphate synthase] protein as a template for homology modeling. The reliability of protein models was evaluated by Procheck² and the energy was minimized by SPDBV. The secondary structures of GPP and BPP were downloaded from ZINC.³ The molecular docking was performed using AutoDock vina with grid points and spacing set as 50 × 50 × 50 and 0.375 Å (Trott and Olson, 2010). The resulting complexes were visualized with PyMOL.

Quantitative real-time PCR (qRT-PCR) was performed to quantify the gene transcriptional expression. The primers (qRT-TPS2F/R and qRT-TPS3F/R) used for qRT-PCR are shown in Supplementary Table 1. qRT-PCR was conducted using 2 × M5 HiPer SYBR Premix EsTaq (Mei5 Biotechnology, China) on a CFX 96 Real-Time PCR Detection System (BioRad, United States) following the manufacturers’ instructions. The thermal cycling conditions were 95°C for 10 s, followed by 40 cycles of 95°C for 5 s and 58°C for 30 s. During the reaction, the target gene transcript levels were monitored using a reference gene and calculated using the 2^–△△Ct method. All experiments were performed with three biological replicates and three technical replicates.

Multi-sequence alignment was conducted using the DNAMAN software. A phylogenetic tree was constructed by the neighbor-joining method using the MEGA-X software, after which the constructed phylogenetic tree was imported into the iTOL⁴ for modification. Plastid transport peptide was analyzed using the ChloroP 1.1 Server.⁵

Seeds are the main medicinal part of Fructus Amomi, which are abundant in volatile terpenoids, therefore, the volatile terpenoids in 45-DAF seeds of A. longiligulare and A. villosum were analyzed by GC-MS. In total, 21 monoterpenoids and 16 sesquiterpenoids were detected in A. longiligulare, while 20 monoterpenoids and 21 sesquiterpenoids were detected in A. villosum, and both the total contents of monoterpenoids and sesquiterpenoids of A. villosum are higher than those of A. longiligulare (Supplementary Table 2). There were nine monoterpenoids only detected in A. longiligulare or A. villosum, for example linalool was only detected in A. longiligulare, while γ-terpineol was only detected in A. villosum. Monoterpenoids are the majority terpenoids in the seeds, and the percentages of main monoterpenoids, including bornyl acetate, bornel, camphor, D-limonene, myrcene, camphene and α-pinene, among the total volatile terpenoids are 50 and 56% in A. longiligulare and A. villosum, respectively (Supplementary Figures 1A,B). Specifically, the percentage of bornyl acetate of A. villosum is higher than that of A. longiligulare, however, the percentage of camphor of A. longiligulare is higher than that of A. villosum (Supplementary Figures 1C,D). Additionally, considering the mature and dried fruits are used as traditional medicine, the main monoterpenoids of mature and dried seeds of A. longiligulare and A. villosum were analyzed as well, and compared with the data of developing seeds. In mature seeds, bornyl acetate and camphor are the most abundant monoterpenoid in A. villosum and A. longiligulare, respectively (Supplementary Table 3). The contents of bornyl acetate and borneol of A. villosum are higher than those of A. longiligulare, but the content of camphor of A. longiligulare is higher than that of A. villosum. Camphor, borneol and bornyl acetate are produced from the same precursor BPP (Figure 2B); the total contents of these BPP-related terpenoids are higher in A. villosum than in A. longiligulare in spite of 45-DAF seeds or mature seeds (Table 1).

In order to mine the BPPS gene and to explore the other genes involved in the terpenoid biosynthesis of A. longiligulare, the 45-DAF seeds were used for transcriptome sequencing. High-throughput sequencing and de novo assembly yielded 62,433 unigenes, and 36,746 unigenes (58.86%) were totally annotated using the major public databases Nr, KOG, Swiss-Prot and KEGG (Supplementary Table 4 and Supplementary Figure 2). The transcriptome data were submitted to the NCBI and assigned the SRA accession number SRR10769494. Based on the KEGG annotation, 1,527 unigenes were mapped to the biosynthesis of secondary metabolites, and 88 unigenes were annotated to terpenoid backbone biosynthesis. The KEGG annotation data of the 45-DAF seeds transcriptome of A. villosum (Wang et al., 2018) were used to compare the genes involved in the upstream pathway of terpenoid biosynthesis with A. longiligulare. The results revealed that all genes involved in the MVA and MEP pathway were annotated in both A. longiligulare and A. villosum transcriptome data, but more unigenes were annotated in total in A. villosum (62) than in A. longiligulare (58) (Supplementary Table 5). The identities of relative genes of A. villosum and A. longiligulare were all higher than 80%.

Based on the Pfam annotation of the transcriptome data of A. longiligulare and the local-Blast result via AvTPSs (Wang et al., 2018), six AlTPS unigenes with longer length were screened out (Supplementary Table 6). Of these genes, three (AlTPS1-AlTPS3) were predicted to encode the monoterpenoid synthase gene, and AlTPS3 presented the highest RPKM expression value. According to the nucleotide sequence alignment, AlTPS3 had the highest (95%) identity with AvBPPS, which has been identified as bornyl diphosphate synthase from A. villosum (Wang et al., 2018). AlTPS2 shared 99% identity with AvTPS2, which had been previously screened from the transcriptome of A. villosum but yet cloned (Wang et al., 2018). AlTPS2, AvTPS2 and AlTPS3 were cloned in this work. The complete ORFs of both AlTPS2 and AvTPS2 were 1,764 bp, encoding 587 amino acids with a predicted 34-amino-acid plastid transit peptide on the N-terminal (Figure 2A). The complete ORF of AlTPS3 was 1,791 bp, and it encoded 597 amino acids with a predicted 47-amino-acid transit peptide on its N-terminal, however, AvBPPS encoded 596 amino acids and its predicted transit peptide was shorter than AlTPS3 (Figure 2A). The gene and deduced amino acid sequences of AlTPS2, AvTPS2, and AlTPS3 have been submitted to GenBank under the accession numbers MN829548, MN829551, and MN829549, respectively.

The deduced amino acid sequences of six candidate AlTPSs and AvTPS2 (Supplementary Table 7), and other functional TPS identified from subfamilies a–g (Supplementary Table 8) were used to construct phylogenetic trees (Figure 2B). AlTPS2, AvTPS2, and AlTPS3 were clustered into the TPS-b subfamily, which is composed of monoterpenoid synthases. AlTPS1 was clustered into the TPS-f subfamily, while AlTPS4 and AlTPS5 were clustered into the TPS-c subfamily, and AlTPS6 was clustered into the TPS-e subfamily. The terpenoid synthases in the TPS-c and TPS-e subfamilies were found to be diterpenoid synthases. Moreover, AlTPS2 and AvTPS2 were clustered into one branch, while AlTPS3 and AvBPPS were clustered into another branch. Alignment of AlTPS2, AvTPS2, AlTPS3, and AvBPPS showed that they all contained the conserved mono-TPS motif, RRX₈W, RXR, DDXXD, and NSE/DTE (Figure 2A). Comparison of the amino acid sequences revealed that the identity of AlTPS2 and AvTPS2 was 98%. The four conserved motifs of AlTPS2 and AvTPS2 were completely consistent, and their other amino acid regions were highly conserved as well (Supplementary Figure 3). However, the amino acid identity of AlTPS3 and AvBPPS was only 83%, although they were clustered into a close branch (Figure 2B). Therefore, we speculated that the functions of AvTPS2 and AlTPS2 might be the same, while the function of AlTPS3 might be very similar to AvBPPS.

The transit peptide of monoterpenoid synthases often reduces soluble protein expression; therefore, the transit peptides of AlTPS2 and AvTPS2 were truncated. The remaining coding regions of AlTPS2 and AvTPS2 were then sub-cloned into expression vectors. The recombinant proteins of AlTPS2 and AvTPS2 were induced and purified using MBP-tag columns and NI-NTA resin, respectively. SDS-PAGE showed that the recombinant proteins of AlTPS2 (fused with MBP-tagged) and AvTPS2 (fused with His-tagged) were about 108 and 89.0 kDa, respectively, as predicted (Supplementary Figure 4). To verify the function of AlTPS2 and AvTPS2, the AlTPS2 and AvTPS2 proteins were incubated with GPP or FPP, after which the reaction products were detected using GC-MS. When the substrate was GPP, AlTPS2, and AvTPS2 produced the monoterpene product linalool (Figure 3A) based on the comparison of retention time and mass spectra with those of the linalool standard (Figure 3B). AlTPS2 and AvTPS2 could not catalyze FPP to form products (data not shown). The optimum pH of AlTPS2 and AvTPS2 was pH 8 and pH 5, respectively (Supplementary Figures 5A,C), and both showed much higher activity in the presence of Mg²⁺ than Mn²⁺ (Supplementary Figures 5B,D).

To investigate the correlation of the expression levels of AvTPS2 and AlTPS2 with metabolite accumulation, qRT-PCR of AvTPS2 and AlTPS2 was performed. As the organ accumulating linalool, flowers were also used for this analysis besides pericarps and seeds. Both AlTPS2 and AvTPS2 were expressed in flowers and pericarps higher than in seeds (Figure 4A). The expression level of AlTPS2 in the flowers of A. longiligulare was much higher than that of AvTPS2 in the flowers of A. villosum, however, the linalool content in the flowers of A. villosum was much higher than that of A. longiligulare, and so as the linalool content in the pericarps of A. villosum (Figure 4B). The AlTPS2 or AvTPS2 expression in different tissues was not consistent with the linalool accumulation, therefore we speculate that in addition to AvTPS2 or AlTPS2 there are other terpenoid synthases responsible for linalool biosynthesis in A. villosum or A. longiligulare.

The transit peptide of AlTPS3 was truncated, and the remaining coding region was sub-cloned into the pET32a expression vector. The AlTPS3 recombinant protein was approximately 69.36 kDa (Supplementary Figures 4C,D). GC-MS analysis of the in vitro enzymatic assay revealed that the AlTPS3 recombinant protein catalyzed GPP to produce camphene as the major product and limonene as the main by-product, which is similar to AvBPPS (Supplementary Figure 6). In addition, minor amounts of tricyclene, α-pinene, β-terpineol, and terpinolene were detected in the AlTPS3 enzymatic product. After dephosphorization, borneol, the dephosphorylated product of borneol diphosphate (BPP), was detected in both AlTPS3 and AvBPPS enzymatic products, as well as limonene and camphene (Figure 5A and Supplementary Figure 7). Other trace by-products, such as β-myrcene and tricyclene, were also detected but not marked in Figure 5A.

Comparison of the product percentages of AlTPS3 and AvBPPS revealed that camphene was the major product (62.5%) of AlTPS3, while borneol was the second most abundant product (22.1%) (Figure 5B); however, the major product of AvBPPS was borneol (57.6%), while minor products consisted of camphene (21.0%), limonene and β-myrcene (Figure 5B). AlTPS3 produced an approximately 3:1 ratio of camphene and borneol, however, AvBPPS produced the reverse ratio of camphene and borneol (Supplementary Figure 8). The optimum pH of AlTPS3 and AvBPPS was pH 7 and pH 6, respectively (Supplementary Figures 9A,C). The activity of AlTPS3 showed higher dependence of Mg²⁺ than Mn²⁺ (Supplementary Figures 9B,D).

According to the sequence alignment of AlTPS3 and AvBPPS, the sequences of other three conserved motifs were complete identical between AlTPS3 and AvBPPS, there was only one different amino acid, A496 for AlTPS3 and G495 for AvBPPS, in the DTE motif (Figure 3A). In addition, according to the sequence alignment of AlTPS3 and other known BPPSs, including AvBPPS, SoBPPS, LaBPPS, and LdBPPS, the glycine in DTE motif of BPPSs with BPP as main product is highly conserved only the amino acid of AlTPS3 in this site is alanine (Figure 6A). To identify the function of this different amino acid residue, the site-directed mutant of AlTPS3-A496G and AvBPPS-G495A were constructed, and the overnight reaction with GPP was performed. The comparison of mutant enzyme with AlTPS wild-type enzyme revealed that site-directed mutation of A496G in the conserved DTE motif changed the major product from camphene to BPP (borneol as final product), and no other product was detected expect for camphene (41%) and borneol (59%) (Figure 6B). Compared with AvBPPS wild-type, AvBPPS-G495A increased the percentage of camphene and decreased the percentage of BPP (borneol), which is the opposite result of AvTPS3-A496G, although BPP is still the main product of AvBPPS-G495A (Supplementary Table 9). The results suggest that the glycine in DTE motif affects product BPP selectivity of enzyme.

In order to explain how A496G affects the product selectivity of AvTPS3, molecular docking was conducted based on the virtual three-dimensional structures of AlTPS3-WT and AlTPS3-A496G, which were predicted by homology modeling using SoBPPS crystal structure (Whittington et al., 2002). The evaluation results of Procheck show that the protein model is reliable (Supplementary Figure 10). Superimposition of AlTPS3-A496G on AlTPS3-WT in their catalytic pockets reveals that their substrate-binding and BPP-binding pockets are similar, although AlTPS3-A496G has a few more active residues than AlTPS3-WT (Y425 and S450 for the GPP-binding pocket and D494 for BPP-binding pocket) (Figures 7A,B). However, the camphene-binding pocket of AlTPS3-WT (colored in white) is obviously bigger than that of AlTPS3-A496G (colored in blue), and AlTPS3-WT has four more active residues than AlTPS3-A496G (Figure 7C). Compared with AlTPS3-WT, the smaller camphene-binding pocket of AlTPS3-A496G may hinder the production of camphene. Additionally, the binding energy of AlTPS3-A496G with BPP is lower than that of AlTPS3-WT (−6.0 vs. −4.7 Kcal/mol) (Supplementary Table 10), which may facilitate the BPP production. Taken together, we speculate that A496G mutation narrows the camphene-binding pocket and decreases the BPP-binding energy, thus increases the BPP selectivity of products.

To compare the expression of AlTPS3 and AvBPPS, 45-DAF pericarps and seeds of A. villosum and A. longiligulare were used to perform qRT-PCR. The results revealed that both AvBPPS and AlTPS3 were highly expressed in seeds, but hardly expressed in pericarps (Figure 8A). The expression level of AvBPPS in seeds of A. villosum was nearly three times of that of AlTPS3 in seeds of A. longiligulare. The high expression level of AvBPPS was correlated with the high content of corresponding terpenoids, including BPP-related terpenoids (borneol, camphor and bornyl acetate) and camphene in seeds, and the expression level of AlTPS3 matched the contents of corresponding monoterpenoids as well (Figures 8B,C).