Metabolomics approach reveals annual metabolic variation in roots of Cyathula officinalis Kuan based on gas chromatography–mass spectrum

The experimental plants were sowed among four successive years (2011–2014) in Baoxing country, Ya’an city, Sichuan province, China. In March, 2015, they were simultaneously collected. In total, 32 batches of authentic roots of Cyathula officinalis Kuan were identified by prof. Meng-liang Tian at College of Agronomy, Sichuan Agricultural University, where we deposited the voucher specimens of C. officinalis. All the plant materials were grew in a same farm by QiXiang farmer professional cooperative with same cultivated techniques before this study. We sorted these roots into 4 groups in terms of their growth years. They were labeled group A, group B, group C, and group D, which denoted that they had been grown for 1, 2, 3 and 4 years until sampling. Each group was composed of 8 biological replicates and each replicate was named by a unique sample identifier, which combined with group label and random number, like A1 or D8 (details see in Additional file 1). After washing the roots with pure water, all samples were immediately frozen in liquid N₂ and stored at −80 °C until processing.

Samples for HPLC detection were prepared based on Ref. [8] with little modification. 1000 mg dried root powder was accurately weighted and then extract by methyl alcohol (20 ml) in ultrasonic for 30 min. Each sample group contained 8 replicates and each extraction repeated 3 times. Before HPLC analysis, extracting solutions were filtrated by 0.45 μm nylon membrane filter and diluted with equal amount of water.

Samples for GC–MS detection were prepared as follow: 100 mg fresh root tissue of each sample was accurately weighed and mixed with 0.4 ml methanol-chloroform (v/v; 3:1) and vortex for 10 s. 20 μl ribitol (0.2 mg/ml, stock in dH₂O) was added in mixtures as internal standard. Mixtures were homogenized in ball mill (JXFSTPRP-24, Shanghai jinxin industrial development Co., Ltd) for 5 min at 55 Hz, subsequently centrifuged at 12,000 rpm for 15 min at 4 °C. The supernatant (approximately 0.4 ml) was transferred to a new GC/MS glass vial. The extracts were dried in a vacuum concentrator without heating for about 1.5 h. 80 μl methoxymethyl amine salt (dissolved in pyridine, final concentration of 20 mg/ml) was added into dried metabolites, afterwards incubated at 80 °C for 20 min in an oven after mixing and sealing. After that, 100 μl BSTFA (containing 1% TCMS, v/v) was added into each sample and incubated at 70 °C for an hour. When sample cooled to room temperature, 10 μl FAMEs (Standard mixture of fatty acid methyl esters, C8–C16:1 mg/ml; C18–C30:0.5 mg/ml in chloroform) was added to it, and finally mixed well for GC–MS detection.

The cyasterone assaying used below parameters: Chromatographic column: Agilent Zorbax Eclipse XDB-C18 (5 μm particles, 4.6 mm × 150 mm). Flow rate: 0.8 ml/min. Temperature: 35 °C. Determine wavelength: 243 nm. Mobile phase: water and acetonitrile. Isocratic elution: 0–20 min, 18% water.

Gas chromatography–mass spectrum analysis was performed using an Agilent 7890 gas chromatograph system coupled with a Pegasus HT time-of-flight mass spectrometer. The system utilized a DB-5M Scapillary column coated with 5% diphenyl cross-linked with 95% dimethylpolysiloxane (30 m × 250 μm inner diameter, 0.25 μm film thickness; J&W Scientific, Folsom, CA, USA). A 1 μl aliquot of the analyte was injected in splitless mode. Helium was used as the carrier gas, the front inlet purge flow was 3 ml/min, and the gas flow rate through the column was 20 ml/min. The initial temperature was kept at 50 °C for 1 min, then raised to 330 °C at a rate of 10 °C/min, then kept for 5 min at 330 °C. The injection, transfer line, and ion source temperatures were 280, 280, and 250 °C, respectively. The energy was −70 eV in electron impact mode. The mass spectrometry data were acquired in full-scan mode with the m/z range of 30–600 at a rate of 20 spectra per second after a solvent delay of 366 s. Before statistical analysis, the GC–MS method was validated by internal standard compound (Ribitol), whose standard deviation of retention time was 0.014 (n = 32).

Chroma TOF 4.3X software of LECO Corporation and LECO-Fiehn Rtx5 database were used for raw peaks exacting, the data baselines filtering and calibration of the baseline, peak alignment, deconvolution analysis, peak identification and integration of the peak area [22]. The RI (retention time index) method was used in the peak identification, and the RI tolerance was 5000. Metabolite data were normalized by dividing each peak area value by the area of internal standard (Ribitol). After that, the data were log10 transformed, mean-centered and divided by the standard deviation of each variable before performing statistical analysis. All the statistical analyses, such as ANOVA, PCA, PLS-DA, HCA, were performed by using MetaboAnalyst 3.0 [23]. The Minimum Standards of Reporting Checklist contains details of the experimental design, and statistics, and resources used in this study (Additional file 2).

As the only certificated quality marker by Chinese Pharmacopoeia (edition 2015) [8], cyasterone was detected in four groups by HPLC. The results showed that the content of cyasterone in each sample could meet the minimum requirement (≥0.030%) of Chinese Pharmacopoeia. However, the 4-years growth ages plants have the highest content of cyasterone (0.087%), followed by 3-years group (0.076%), 2-year group (0.065%) and 1-year group (0.039%) (details see Additional file 3). These results were also partial proved by previous studies [15]. Therefore, considered the content of marker compound, herbs with 1–4 growth years ages were all qualified and the 4th year was the best harvest year.

To obtain an overview of annual metabolic changes in roots of C. officinalis, we carried on the GC–MS approach to all samples. The representative GC–MS chromatograms showed obvious variation in different growth ages (Fig. 1a–d). In total, 752 chromatographic peaks were detected and then numbered 1–752 in sequence of retention time. Peak 442 was contributed by ribitol as reference substance. In order to remove the systematic noise, we just extracted those peaks that were successfully discovered (peak area value >0) at least 6 times in either groups (n = 8). In result, 341 peaks (not include peak 442) were retained. Among them, 166 peaks were given identified chemical name (details see Additional files 1, 4). It should be noted that 4 chemicals were identified twice. They were 3-hydroxypropionic acid (peak 99 and 107), aspartic acid (peak 295 and 349), xylitol (peak 431 and 432), and diglycerol (peak 446 and 457). This may result from the insufficient precision of methods. To the end, the final dataset was composed of all the 166 peaks of 32 samples and their peak area values, which was used to the following ANOVA, PCA, HCA, PLS-DA etc.

We firstly carried on univariate analysis method, ANOVA, to have an overview of all 166 metabolites data, attempted to simply find potentially important metabolites. In result, 63 metabolites showed potentially significant difference about their content between 4 groups using the standard P < 0.05 (detail see Additional file 5). Compared with the 1st growth year (group A), 15 metabolites changed significantly when herbs grew for two years (group B), while 40 metabolites changed in 3rd year (group C) and 32 metabolites in 4th year (group D). This trend was as well illustrated in Fig. 2a, which clearly showed us most metabolic variations happened in 3rd year after sawing. Similar trend was also revealed in Fig. 2b by comparing the number of metabolites changed significantly between previous year and following year. Therefore, we speculated that the 3rd year, when the most large-scale of metabolites changed, more likely a turning point for C. officinalis metabolism.

Top 10 typical metabolites were listed Additional file 5, using the standard of P < 0.0001. In Fig. 3 their changes in different growth years were illustrated by box plots, which showed that those metabolites more or less prone to accumulate in some certain years. For instance, xylitol and citramalic acid were hardly detected in the first 3 growth years until grew for 4 years. While mannose and oxalic acid accumulated rich in fist year but decreased during the 2–4 years. Gluconic acid, 3-hydroxybutyric acid, glutaric acid and cellobiose were easily detected in 3–4 years while little existed in 1–2 years. However, this trend was inverted for pipecolinic acid and ribonic acid.

To have a further visually insight on every metabolite content change in different growth years. We performed a heat-map combined with HCA, using extracted dataset composed by 63 metabolites filtered by ANOVA (P < 0.05). As illustrated in Fig. 4, 32 samples trended to separate into 3 classes. Samples with same growth age had trend to cluster in same class. For instance, samples of group C all clustered in class III and group D all clustered in class II. Seven samples from group A and 6 samples from group B were together clumped in class I and further separated into two subclasses. Unusually, the rest 3 samples of group A or B, A2, B1, B5, were together put into class III. This means that those 3 samples were more similar to group C. The fact that only group D was clustered in an independent class, class II, indicated that its metabolic profile was stabilized when herbs of 4 growth age. Therefore, compared with the traditional cultivation, more proper harvest time for C. officinalis may be the 4th year after sowing because of the micro-change in metabolic profile between samples.

Showed in Fig. 4, 4 metabolite classes were found in the heat-map, each of which revealed the content distribution in different sample classes or growth years. Metabolites in Class (1), like glycine (peak 217), 2-hydroxy-3-isopropylbutanedioic acid (peak 379), valine (peak 157), were least abundant in 3 growth years than any other growth ages. However, metabolites in class (4), like 2-deoxyerythritol (peak 199), acetol (peak 430), citramalic acid (peak 322), were most abundant in 4 growth years. Most of metabolites in Class (2), like ribonic acid (peak 424), pipecolinic acid (peak 253), N-carbamylglutamate (peak 537), had high content in the 1–2 growth years compared with the next 3 or 4 growth years, while metabolites in class (3), like gluconic acid (peak 618), 3-hydroxypropionic (peak 99), glutaric acid (peak 277), cellobiose (peak 674), lactic acid (peak 47), trended to accumulate in 3–4 growth years when compared with 1–2 growth years.

As one of common used unsupervised methods of multivariate analysis, we performed PCA on the 166 metabolites dataset. Considering only the first two principal components, PC1 and PC2, explained 27.8% variance. An obvious separation between the earliest (group A and group B) and latest (group C and group D) growth years in Fig. 5a was discovered, which exhibited the notable metabolic difference between them. The considerable overlap between the group A and group B indicated their metabolic profiles were similar. This fact as well had been told by Fig. 4, where group A and group B were clustered in two subclasses although belonged to a same upper class. Compared with earliest growth years, the latest groups, group C or group D, were better separated due to samples of group D gathered more closely in this PCA score plot. This result proved again that group D had less internal difference relatively.

Principal component analysis analysis demonstrated the presence of discriminating factors which allowed the separation between earliest growth years and latest years. However, this kind of unsupervised method did not allow observing well separation of every two groups, such as group A/group B or group C/group D. To verify whether our current metabolites dataset provided enough information could detail the significant difference of any two groups, we decided to use a supervised method as PLS-DA. In result, this method dose exhibited the ability to discriminate each group in three dimensional score plot with three principle components (Fig. 5b), accounting for 32.1% variance. As showed in Fig. 5a, b, the main factor that drived the separation between each group was PC1. In order to summarize the importance of metabolites for constructing PC1, we listed the top 15 metabolites with high variable influence on projection (VIP) score (Fig. 6a). They were gluconic acid, xylitol, glutaric acid, pipecolinic acid, ribonic acid, mannose, oxalic acid, digalacturonic acid, lactic acid, 2-deoxyerythritol, acetol, 3-hydroxybutyric acid, citramalic acid, N-carbamylglutamate, and cellobiose. These results were correlated with main loadings of PLS regression (Fig. 6b). In order to verify whether those 15 metabolites have ability to discriminate different growth years as potential chemical markers, we re-performed PLS-DA based on 15 metabolites. Luckily, the new score plot (Fig. 7) who produced by this simple dataset was very similar with the score plot produced by 166 metabolites (Fig. 5b). That means, using the 15 metabolites to evaluate the quality of herbs is feasible.