The leaves of E. senticosus with similar growth states, same age and growth potential in Wulingshan National Nature Reserve (Chengde, China) were selected as test materials (n = 100). Total RNA and genomic DNA were extracted by an RNAprep pure plant kit (Beijing Tiangen Biotechnology Co., Ltd., China) and a plant genomic DNA kit (Beijing Tiangen Biotechnology Co., Ltd., China), respectively. Total RNA was then reversely transcribed to cDNA by using RevertAid™ First strand cDNA synthesis Kit (Beijing Tiangen Biotechnology Co., Ltd., China).

According to the manufacturer’s instruction of the DNA Bisulfite Conversion Kit (TIANGEN Biotech), DNA samples were treated with bisulfite with 3 biological repetitions.

The potential function of the promoter of MDD (GenBank: OL467356) was analyzed by PlantCARE online program [10]. Li et al. [11] were used to predict the CpG island in the promoter of MDD. To amplify the promoter of MDD after bisulfite treatment, the primers were then designed using Prime premier 6 [12]. The primers are listed in Table 1, in which the primers for amplifying the upstream and downstream of CpG island were designated with CMDDJHS₂-X₄ and CMDDJHS₄-X₆, respectively. The processed DNA was amplified by PCR with a Methylation-Specific Kit (Beijing Tiangen Biotechnology Co., Ltd., China). The total reaction system was 20 µL with 1.0 µL bisulfite-treated DNA (50 ng/µL), 1.0 µL each prime (15 µm), 1.6 µL dNTPs, 0.4 µL MSP DNA poly and 2.0 µL MSP DNA buffer. The amplification conditions of upstream CpG island were pre-denaturation at 94°C for 5 min, then followed by 40 cycles of Denaturation at 94°C for 20 s, annealing at 54°C for 30 s and extension at 72°C for 15 s, and then supplementary extension at 72°C for 5 m. The reaction for amplifying the downstream CpG island was performed with thermal cycling condition of initial denaturation at 95°C for 5 m, then 40 cycles of 94°C for 20 s, 53°C for 30 s, and 72°C for 40 s, and a final extension at 72°C for 5 m.

After separated by 1.5% agarose gel electrophoresis and purified by an agarose gel DNA Recovery Kit (Beijing Tiangen Biotechnology Co., Ltd., China), PCR products were cloned into PGM-T vector and then transformed into Escherichia coli TOP10 competent cells. From each sample, 10 positive strains were selected for sequencing (Beijing Nosai gene Co., Ltd., China).

BIQ analyzer 2.0 software [13] was used to view and compare the sequences of MDD with or without treatment of bisulfite. Type A, Type B and Type C are grouped according to the DNA methylation ratio of MDD from low to high.

According to the DNA methylation type (Types A, B and C) in the promoter of MDD, mRNA in mature leaves was extracted and grouped. The cDNA library of each sample with 3 biological replicates was constructed and sequenced on the Nova SEQ 6000 sequencing platform (Illumina, USA). The splice sequences in raw data and low-quality reads were removed to obtain high-quality clean data. Each library sequence was then spiced with the Trinity software (http://trinityrnaseq.github.io/). The transcripts with low expression were filtered according to FPKM < 1, TPM < 1, or IsoPct < 5.0%, and then clustered into unigenes using Tgicl (v2.0.6) (https://sourceforge.net/projects/tgicl/). Unigene sequence was compared with the data in Non-redundant (Nr), Swiss-Prot, Gene Ontology (GO), Clusters of Orthologous Groups of proteins (COG), euKaryotic Orthology Groups (KOG), the Kyoto encyclopedia of genes and genomes (KEGG), Pfam database to obtain annotation information.

PlantTFDB [14] and PlantRegMap [15] were used to predict transcription factors that may bind to the promoter of EsMDD gene with reference to the transcription factors of Nicotiana tabacum. The transcriptome sequencing results of E. senticosus were used to screen the differentially expressed transcription factors, which were then mapped by HeatMap function of TBtools [16]. MEGA7 [17] was used for phylogenetic analysis of the transcription factor families and string [18] was used for protein-protein interaction prediction.

According to the transcriptome sequencing results, the known key enzyme gene sequences of the terpenoid synthesis pathway of E. senticosus were compared using BLAST. The key enzyme genes of the terpenoid synthesis pathway in the samples were screened, and qRT-PCR primers were designed (Table 1). According to the qRT-PCR primers of the key enzyme genes of E. senticosus terpenoid synthesis pathway in Table 1, the cDNA of the three selected DNA methylation types of E. senticosus samples was used as the template for qRT-PCR reaction with Abiprism 7900-HT (Applied Biosystems, Foster City, CA 94404, USA). Each sample was set with 3 biological replicates. The total reaction system was 10 µL. 2 × Talent qPCR premix was 5.0 µL (Beijing Tiangen Biotechnology Co., Ltd., China). Forward primer was 0.3 µL (primer concentration 15 µm), and reverse primer was 0.3 µL. E. senticosus cDNA template was 0.5 µL (concentration 50 ng/µL), and 50 × Rox reference dye was 1 µL and RNase free H₂O was 2.9 µL. The reaction conditions were 95°C pre-denaturation for 3 m with denaturation at 95°C for 5 s, annealing at 55°C for 10 s, and supplementary extension at 72°C for 15 s; The reaction was ended after 40 cycles. The gene expression of each sample was calculated by ^2−ΔΔCt method with SDS 2.4 software.

The leaves of E. senticosus with different DNA methylation types were dried and ground, and then shake with 0.5 mL 8% vanillin ethanol solution [19] and 5 mL 72% sulfuric acid solution. The samples were then heated for 10 min at 60°C, and cooled for 15 min in ice bath. The absorbance value of each sample at 534 nm was detected by vanillin-concentrated sulfuric acid coloration method [5]. The total saponin contents were calculated by the regression equation. SPSS 22.0 was used to analyze the correlation between DNA methylation ratio in the promoter of MDD and total saponin content.

According to the transcriptome annotation information, the data of DNA methyltransferase and DNA demethylase genes of E. senticosus were also extracted. The results were then identified by the NCBI-CDD function. MEGA7 and DNA methyltransferase and DNA demethylase in other species were used for phylogenetic analysis.

The string was used to analyze the protein-protein interaction between MDD protein in Arabidopsis and other key enzymes in the terpenoid synthesis pathway.

The results of Li Lab showed that there were 2 CpG islands in the promoter of MDD. The upstream CpG island was in −2265–2148 bp and the downstream CpG island was in −1787–1663 bp. BIQ Analyzer 2.0 software was used to compare the MDD promoter the sequence treated with bisulfite with the untreated DNA sequence. The results showed that there were 10 and 18 DNA methylation sites in the upstream and downstream CpG islands of MDD, respectively (Fig. 1A), The DNA methylation sites in all samples were located at −2267, −2264, −2214, −2176, −2162, −2157, −2100, −1885, −1849, −1769, −1758, −1742, −1715, −1708, −1644, −1519 bp in the upstream of the transcriptional site. The differential methylation sites were located at −2205, −2196, −2130, −1923, −1572, −1430 bp, the methylation sequencing map of −1923 bp site is shown in Fig. S1 for example. There were 3 types of DNA methylation sites, corresponding to DNA methylation ratios, which were designated as Type A, Type B, Type C. There were 18. 19 and 21 DNA methylation sites in samples of Types A, B and C and the methylation rate of cytosine were 0.68%, 0.72% and 0.79%, respectively (Fig. 1B).

The total saponin contents in the 3 groups of samples of Types A, B and C were determined by vanillin-concentrated sulfuric acid method. According to the average absorbance (0.155 ± 0.000471 for Type A, 0.247 ± 0.000816 for Type B, and 0.338 ± 0.000420 for Type C) the total saponin contents of were calculated as 4.65%, 2.32%, and 1.83% for Types A, B and C, respectively. The correlation analysis showed that there was a significantly negative correlation between the DNA methylation ratio of MDD and the saponin content (P < 0.05). The correlation equation was y = −23.532x + 20.112 (R² = 0.7562), where “y” standed for the saponin content.

High-throughput transcriptome sequencing was performed on the 3 groups of E. senticosus samples of Types A, B and C on Illumina Nova Seq 6000 platform. After removing junction and low-quality reads, 20.51–29.33 M clean reads and 6.15–8.80 Gb clean bases were obtained. The GC content of all samples was about 43%–45%, and that of all Q30 bases was over 94% (Table 2), which indicated that the quality of sequencing data was acceptable.

Samples are the sample names; Clean reads are the number of Clean reads; Clean bases are the total number of bases in Clean data; GC is the percentage of GC content in Clean data; ≥Q30 is the percentage of bases whose clean data quality values are greater than or equal to 30. Q30 is the percentage of bases with clean data quality values greater than or equal to 30.

Clean reads generated after filtering the transcriptome sequencing data were assembled using the Trinity program. A total of 374,281 transcripts were generated, of which the longest transcript was 17,973 bp, and the shortest was 197 bp. The average length of the transcripts was 303 bp, and the length of the N50 transcript was 243 bp. Further filtering of the low-expressed transcripts, there were 364,816 unigenes with an average length of 284 bp and an N50 of 240 bp (Table 3). The unigenes with lengths between 200–300 bp accounted for the largest proportion of the total (345,555, 94.72%), and unigenes with lengths between 200–1,000 bp, 1,000–2,000 bp and 2,000–3,000 bp accounted for 97.57% (355,956), 1.51% (5,497) and 0.62% (2,258) of the total, respectively. In addition, the length of 1,105 (0.30%) unigenes were longer than 3,000 bp (Table 3)

Types are the type listed; Unigenes were unigenes obtained after assembly and filtration; Transcripts are transcripts obtained after assembly; Total sequence number is the total number of sequences; Total sequence length is the total length of all sequences; Maximum length is the maximum length of the assembled sequences; Minimum length is the minimum length of the assembled sequences; Average length is the average length of all sequences obtained from the assembly; GC% is the sum of the number of GC bases as a percentage of the total number of bases; N50: the length of the assembled transcripts sorted from longest to shortest, and the length of the transcripts accumulated to 21%–50% of the total length.

The assembled unigenes sequences were compared with the data in KEGG, NR, Swiss-Prot, GO, COG, KOG and Trembl databases by BLAST program. Amino acid sequences of unigenes were compared with the data from Pfam database using HMMER software to obtain the annotation information. The results are shown in Fig. 2A. A total of 192,820 unigenes were annotated by at least 1 of the above 7 databases, accounting for 52.85% of the total unigenes. In the NR database, 177,470 (48.65%) unigenes were annotated, and the highest matching rate was obtained, and the COG was the least annotated with 64,532 (17.69%). Comparing with the NR database, it was found that the single gene sequence of E. senticosus showed the highest match rate (4.69%) with that of Daucus carota, and the lowest match rate with Ramularia collo-cygni (1.22%) (Fig. 2B).

A total of 364,817 unigenes were divided into 3 groups and 54 classes in the GO database according to their functions. The cellular component group mainly includes cells (cell, 38,186), membranes (membrane, 27,338) and organelles (organelle, 27,408); the molecular function mainly involves the catalytic activity (50,432), binding (42,708) and transporter activity (3,699). The biological process mainly involves metabolic process (50,154), cellular process (42,115) and localization (14,885) (Fig. 2C). Only 13.17% of Unigene annotation information in KOG was annotated to General function prediction, and only 0.09% annotated to cell motility (Fig. 2D).

The transcriptome sequencing data were used to screen differentially expressed genes (DEGs) among samples of Types A, B and C with FDR < 0.01 and | log2 FC | ≥ 1 as the screening criteria. A total of 4,618 DEGs were detected in the Type A vs. Type C, with 2,203 up-regulated genes and 2,415 down-regulated genes; a total of 345 DEGs were detected in the Type B vs. Type C, with 116 up-regulated genes and 229 down-regulated genes; a total of 4,657 DEGs were detected in the Type A vs. Type B, including 2,228 up-regulated genes and 2,429 down-regulated genes. Among these, 719 DEGs were differentially expressed in Type A and Type B only, 683 DEGs in Type A vs. Type C only, and 25 DEGs in Type B vs. Type C only (Fig. 3A). A total of 91 DEGs were differentially expressed in the all 3 types (Fig. 3B). These findings suggested that there was a significant difference in transcript expression among samples of Types A, B and C.

In plants, MVK (Mevalonate kinase), MDD, FPS (Farnesyl pyrophosphate synthase), SS, SE, and bAS (β-amyrin synthase) are key enzymes which involved in the terpenoid synthesis. Eight differentially expressed genes were screened according to the KO numbers of the above 6 genes in 3 groups. The further judgment showed that the sequences of FPS and SS1 and SS2 in the transcriptome sequencing data were the same with those obtained by cloning in our previous study [5]. All seven genes, except bAS, were highly expressed in the Type A, but low expressed in Type C. These expression patterns showed a significant negative correlation with the DNA methylation ratio of E. senticosus MDD gene promoter (P < 0.05). In contrast, among the bAS gene family, bAS2 was significantly negatively correlated with the DNA methylation ratio of MDD (P < 0.05) and positively correlated with the saponin content (P < 0.05), but bAS1 was not (Fig. 3C).

To verify the reliability of the transcriptome data, eight differentially expressed genes involved in terpenoid synthesis were validated by qRT-PCR. The results showed that the trends of the relative expression of differentially expressed genes were all consistent with the transcriptome sequencing data, which indicating the reliability of transcriptome sequencing results (Fig. 3D).

The results of predicted cis-acting elements in the promoter of MDD are shown in Table 4. In addition to TATA-box and CAAT-box, the largest number of elements were related to light response, such as TCT-motif and Box-4. Other cis-acting elements were responded to drought, gibberellin, methyl jasmonate, and abscisic acid (Fig. 4).

PlantTFDB was used to predict the transcription factor binding sites in the promoter region of E. senticosus MDD with Nicotiana tabacum as the reference. The results showed that the promoter region of MDD could bind to 34 of 324 transcription factors. Among them, 15 can bind to DNA methylation sites in CpG island (Fig. 4). Nineteen of 120 differentially expressed transcription factors binding to MDD promoter were found among samples with different DNA methylation types of MDD. Sixty-two were positively correlated with the DNA methylation ratio, while 58 were negatively related with it (Fig. 5A).

Among the 28 DNA methylation sites, 5 sites did not have bindable transcription factors (−2162, −1923, −1811, −1742, −1658 bp). Among the remaining 23 sites, only 4 sites (−2205, −2196, −1572 and −1430 bp sites) have differences in methylation status across three types, which was likely to bind to four gene families, including ERF (Ethylene-responsive factor), LBD (Lateral Organ Boundaries Domain), C2H2 (Cys2/His2-type zinc finger protein) and GRAS. Among them, 18 members of ERF gene family mainly bind to −2205 bp and −2196 bp sites, and a small number of them bind to −1430 bp sites. There were 9 members of LBD family, mainly binding to −2205 bp and −2196 bp sites, and a small number of them binding to −1572 bp site. In addition, 1 member of the C2H2 gene family is combined with −2196 bp and −1572 bp sites, and 1 member of the GRAS gene family is combined with −1430 bp site.

The cluster analysis showed that the above 4 differentially expressed transcription factors were significantly different across different DNA methylation types of MDD (P < 0.05), which were mainly divided into 2 expression patterns. The first one was high expression in Type A and low expression in Type C. The second one was low expression in Type A and high expression in Type C. ERF and GRAS mainly showed in the second expression pattern, while LBD and C2H2 mainly showed in the first expression pattern. But there are also some exceptions. For example, EsERF1 (TRINITY_DN14850_c2_g1), which belongs to ERF family, appeared in the first expression pattern; EsLBD1-1 (TRINITY_DN10977_c0_g1) and EsLBD1-2 (TRINITY _ DN10977_c0_g2), which were LBD family members, showed in the second pattern and the first pattern, respectively (Fig. 5B).

After that, the second most differentially expressed transcription factors, ERF and LBD, were also analyzed. Phylogenetic analysis was performed using the N. tabacum ERF gene family in PlantTFDB with differentially expressed ERFs in the promoter region of the E. senticosus MDD. After the comparison with ClustaW, the NJ method is used 1500 times to construct the phylogenetic tree, and the results are shown in Fig. 5C. The ERFs of E. senticosus were divided into several different subfamilies, in which EsERF1 was a member of a different subfamily from the other E. senticosus ERF members. EsLBD1-1 and EsLBD1-2, which belong to the LBD family and are different transcripts of the same gene, were identified to have a binding site at −1572 bp after comparing with the N. tabacum LBD. Protein-protein interaction analysis was performed on the LBD members bound to this site, and the results were shown in Fig. 5D. The members at this site interacted with bHLH (Basic helix-loop-helix, XP_009609302.1), ERF (XP_009588287.1) and KNOX (Knotted1-like homebox, XP_009613677.1, XP_009598343.1, XP_009617782.1).

DNA methylation is affected by many factors, among which DNA methyltransferase and DNA demethylase have a direct effect on it. Due to the lack of key information of DNA methyltransferase in N. tabacum, the DNA methyltransferase gene of A. thaliana and the DNA demethylase of Lonicera japonica are used as the identification standard. The comparison method is ClustaW, and the test is carried out 1500 times by using the NJ method to construct the phylogenetic tree. The results are shown in Fig. 6A. A total of 1 DNA methyltransferase gene, named EsMT0 (TRINITY_DN4200_c0_g1), and 1 DNA demethylase gene, named EsDM0 (TRINITY_DN722_c1_g1), were identified in E. senticosus. Clustering analysis of the 2 genes showed that EsMT0 clustered with the DNA methyltransferase of A. thaliana, while EsDM0 clustered with the DNA demethylase of L. japonica.

Analysis of the differential expression of these 2 genes in the 3 groups of samples revealed that the DNA methyltransferase gene EsMT0 was significantly correlated with the methylation ratio (P < 0.05), and it was significantly low expressed in Type A and significantly high expressed in Type C. Its expression was significantly and positively correlated with the methylation ratio; while EsDM0 did not show a significant correlation with DNA methylation ratio (Fig. 6B). Meanwhile, another 22 methyltransferases and 20 demethylases were found in the proteome of E. senticosus (Table S1, Table S2), but the genes corresponding to these enzymes were not differentially expressed in the transcriptome.

The protein-protein interaction analysis of MDD in A. thaliana was performed using String program, and the results were shown in Fig. 6C. MDD interacts with multiple genes in the terpenoid synthesis pathway, including HMGS, HDR, MVK in the upstream, IPP1 and its homologous genes IPP2, FPS1, FPS2 and SS1 in the downstream. AT1G31910 and AT1G26640 are unclassified proteins.