Regular Articles
Metabolic Discrimination of Different Rhodiola Species Using 1H-NMR and GEP Combinational Chemometrics
2019 年 67 巻 2 号 p. 81-87
詳細
2019 年 67 巻 2 号 p. 81-87
Rhodiola is widely consumed in traditional folk medicine and nutraceuticals. To establish a procedure for the hydrogen (1H)-NMR spectroscopic fingerprinting of secondary metabolites from three different Rhodiola species, the variation among three Rhodiola species were studied using 1H-NMR metabolomics combined with multivariate data analysis. Gene expression programming (GEP) was used to generate a formula to distinguish Rhodiola crenulata from two other Rhodiola species. Finally, HPLC was used to demonstrate the results. Same metabolites were compared by quantitative 1H-NMR (qNMR). Three Rhodiola species were clearly discriminated by 1H-NMR fingerprinting involved 22 nuclear magnetic signals of chemical constituents. y = d166 × 2 + C1 + d56 + d236 − d128 × C2 can be used to distinguish R. crenulata from two other Rhodiola species by GEP. The gallic acid concentration in R. crenulata was significantly higher than in the other. Rhodiola species as was the level of salidroside. R. crenulata also exhibited substantially higher levels of α-glucose. The fatty acid level in Rhodiola kirilowii was lower than the other species. These findings demonstrated that 1H-NMR fingerprinting combined with principal component analysis (PCA), partial least squares discriminant analysis (PLS-DA), hierarchical cluster analysis (HCA) and GEP can be used to distinguish different Rhodiola species and these methods were applicable and effective approaches for metabolic analysis, species differentiation, and quality assessment. In addition, gallic acid, salidroside, α-D-glucose, glycine, alanine, caffeic acid and tyrosol and are the discriminators.
Rhodiola is a genus of perennial plants in the family Crassulaceae that is widely used in Chinese and Tibetan medicine, which grows in high-altitude and cold regions of the Northern Hemisphere.1) Rhodiola has been primarily used to treat qi deficiency, blood stasis, heart pain, stroke hemiplegia, fatigue asthma and to increase physical endurance, work productivity and longevity as well as to enhance energy levels.2,3)However, scarce and expensive resources, along with confusion over the correct plant species and the production of counterfeit goods has been a big challenge. Recently, it has been reported that different Rhodiola species vary in secondary metabolite production, which leads to inconsistent medicinal value and health care efficacy.4–7) Thus, the content differences of secondary metabolites could be employed to distinguish R. crenulate, R. kirilowii, and Rhodiola fastigiate, which has been commonly used in Tibetan medicine.8)
Plant metabolomics has recently attracted considerable attention because the methods focus on the holistic characteristics of plants and are an important development improving modern medicinal plant research.9,10) 1H-NMR analytical strategies used in metabolomics is considered the most promising analytical tool because the method allows the simultaneous detection of primary and secondary metabolites in a single run and produces a non-biased abundant metabolic profile.11–15) In addition, this method analyses various classes of chemical constituents qualitatively and quantitatively with simple sample preparation steps and good reproducibility.16) Thus, 1H-NMR-based metabolomics combined with multivariate data analysis (e.g., principal component analysis (PCA), partial least squares discriminant analysis (PLS-DA) and hierarchical cluster analysis (HCA)) has recently demonstrated suitability for use in metabolomics studies for the discrimination, authentication, and quality assessment of food, crops and herbal medicines.17)
Gene expression programming (GEP) is an evolutionary algorithm that is used to develop computer programs on the basis of a search and optimization technique using.18,19) GEP is also a promising tool to predict biological effects and physical chemistry outcomes. One advantage of this GEP over the combination method is that it can be represented as mathematical expressions.20) GeneXproTools is a powerful soft computing software package utilised to perform classification and clustering based on GEP.21,22) In this study, GEP was used in 1H-NMR metabolomics by GeneXproTools.
In a recent study, 10 batches of raw R. crenulata material of different origins were successfully distinguished by 1H-NMR.23) However, this method has not been applied to large sample populations, and metabolite differences among Rhodiola species are unclear. Furthermore, an inadequate number of metabolites were recognized from the 1H-NMR spectra, and some metabolites of great significance to quality control evaluations were not quantified. Therefore, this study aims to distinguish three species of Rhodiola native to China and to analyse and compare the potential discriminating metabolites using 1H-NMR-based metabolomics.
A total of 41 Rhodiola samples (Supplementary Table 1) representing three species were collected from the major producing areas of China in Sichuan, Gansu, Qinghai, Yunnan, and Tibet in July and August 2016. The samples were identified by Professor Yi Zhang, and voucher specimens were deposited in the College of Ethnic Medicine, Chengdu University of Traditional Chinese Medicine, China.
Solvents and ChemicalsMethanol-d4 (CD3OD, 99.8%), deuterium oxide (D2O, 99.9%), chloroform-d (CDCl3, 99.8%), and dimethyl sulfoxide-d6 (DMSO-d6, 99.9%) were purchased from Cambridge Isotope Laboratories (Miami, FL, U.S.A.). 3-(Trimethylsilyl) propionic-2,2,3,3-d4 acid sodium salt (TMSP, 99%) was purchased from Sigma-Aldrich (St. Louis, MO, U.S.A.). Monopotassium phosphate (KH2PO4, 99.5%) was obtained from Sinopharm Chemical Reagent Co., Ltd. (Shanghai, China). Standard compounds were purchased from Must Biological Technology Co., Ltd. (Chengdu, China). HPLC grade acetonitrile was purchased from Fisher (Pittsburg, PA, U.S.A.). Analytical grade methanol was purchased from Qiaosun (Guangzhou, China). Deionised water was get from a Hi-tech water system (Shanghai, China). Detailed information is shown in Supplementary Table 2, and the structures of all standard compounds were unambiguously identified based on their 1H- and 13C-NMR spectral data, LC-MS data, and published literature.23–25)
Sample Collection and PreparationAll medicinal materials were oven-dried at 50°C until a constant weight was achieved to completely eliminate moisture. Each powdered sample (200 mg) was vortexed in CD3OD (1.0 mL) and 0.3 mL pH 6.0 buffer containing KH2PO4 in D2O (containing 0.04% TMSP as the internal chemical shift standard), and the sample was extracted by ultrasonication for 30 min at room temperature. After extraction, the sample was centrifuged at 16000 × g for 5 min and subsequently filtered through a 0.45 µm membrane filter. Exactly 0.6 mL of filtrate was transferred into a standard 5 mm NMR tube for 1H-NMR analysis.
1H-NMR Analysis1H-NMR spectra were recorded on an Agilent 400 spectrometer (Agilent Technologies, U.S.A.), operating at a frequency of 400.12 MHz. A presaturation sequence was applied to suppress the residual water signal. For each sample, 64 transients were collected with a spectral width of 6393.9 Hz, a relaxation delay of 2.0 s, and an acquisition time of 2.0 s and the pulse width of 8 µs. A 0.3 Hz line broadening function was applied to all spectra prior to Fourier transformation.
Data ProcessingFourier transformation and phase and baseline corrections were applied to the data. Calibration of the data was performed by shifting the TMSP signal to 0.0 ppm using MestReNova 11.0.4 software (Mestrelabs Research SL, Santiago de Compostela, Spain). The NMR spectra were reduced to integrate buckets of equal widths of 0.04 ppm each in the range of δ 10.5–0.5, excluding residual methanol (δ 3.26–3.34) and water (δ 4.74–4.90) signal, using MestReNova software. All of the integrated values were normalized to TMSP signal intensity. The resulting datasets were then imported into SIMCA-P version 13 (Umetrics, Umeå, Sweden), and mean-centred pre-processing was applied for multivariate statistical analysis.
Statistical Analysis and Metabolite Resonance Signal RecognitionPCA was used to provide an intrinsic overview of the dataset and reveal possible groups and outliers.26) PLS-DA and HCA were also performed to maximize the separation between groups.27) Here, PCA, PLS-DA, and HCA were conducted with mean-centred pre-processing using SIMCA-P software. NMR signals were assigned and recognized with the aid of published literature and the Spectral Database for Organic Compounds.28) Important resonance signals consistent with specific compound structures were recognized by careful comparison with the 1H-NMR spectra of the standard compounds and by adding relevant standard compounds directly to the pH-adjusted NMR samples.29)
GEP AnalysisThe integration results of R. crenulata as one group and R. kirilowii and R. fastigiata as the second group were imported into the GeneXproTools (version 5.0, gepsoft) software. In the first group, 12 R. crenulata samples were used as the entire database and training database, with 3 samples as a test database. In the second group, 20 integrated samples were used as the entire database and training database, with 5 samples as the test database. The run category was selected as classification. A formula to distinguish R. crenulata from the other two species of Rhodiola was identified. Moreover, the key integration area distinguishing R. crenulata from other Rhodiola species was determined using by this formula. The operational characters and weights used are shown in Table 1.
| Weight | Operational character |
|---|---|
| 4 | + |
| 4 | − |
| 4 | * |
| 1 | / |
| 1 | √ |
| 1 | ex |
| 1 | ln |
| 1 | x2 |
| 1 | x3 |
| 1 | x4 |
| 1 | ∛ |
To compare the chemical composition of different Rhodiola species, a known internal standard (in this case TMSP) was used to determine the concentrations of targeted metabolites. Because the signal intensities in the 1H-NMR spectra are absolutely proportional to the molar concentrations of the metabolites, the same metabolites present in different species were compared by qNMR.30,31)
HPLC AnalysisChromatographic separation was performed on an Agilent 1200 Series LC system (Agilent Technologies), consisting of an online degasser, a quaternary pump, a diode array detector, an autosampler, and a column thermostat. The samples were separated on an Agilent ZORBAX Eclipse XDB-C18 (250 × 4.6 mm, 5 µm) at a column temperature of 25°C and flow rate of 1.0 mL/min. The mobile phase was acetonitrile (A) and 0.2% (v/v) phosphoric acid (B) with a gradient elution program of 7% A at 0–8 min, 7–10% A at 8–10 min, 10–12% A at 10–35 min. The detection wavelength was set at 275 nm. The injection volume for all samples was set to 10 µL. The dried powders of Rhodiola samples (0.5 g) were accurately weighed into a clean conical flask and extracted with 50 mL of methanol by ultrasonication for 30 min at room temperature. The sample solutions were filtered through a 0.45 µm membrane filter before HPLC analysis.
Five extraction solvents, including CD3OD–D2O (1 : 0.3, v/v), CD3OD, CDCl3, D2O, and DMSO-d6, were applied and optimized. By contrast the 1H-NMR spectra of the same Rhodiola sample via different extraction methods (Supplementary Fig. 1), more fatty acids and sugars could be respectively extracted by CDCl3 and D2O. Nevertheless, signals intensity at δ 7.70–5.20 corresponding to secondary metabolites in other solvents were obviously higher than that in CDCl3 and D2O. Primary and secondary metabolites could be synchronously extracted by DMSO-d6, but most signals in the 1H-NMR spectra were seriously overlapped. whole metabolites CD3OD and CD3OD–D2O had similar metabolic product, and both could extract. However, sugars and amino acids were more abundant in CD3OD–D2O compared with those in CD3OD. Eventually, CD3OD–D2O was selected as the extraction solvent, because organic acid, glycosides, carbohydrates, and amino acids present could be simultaneously extract by it in the Rhodiola in a single run.
1H-NMR Spectra Analysis and Metabolite Resonance Signals RecognitionRepresentative 1H-NMR spectra of the methanol-water sample extracts from different Rhodiola species is shown in Fig. 1A. As seen from these spectra, the metabolic profiles of Rhodiola species with different botanical origins exhibit a similar composition, with fatty acids, organic acids, and glucose as common metabolites. However, metabolite intensities varied greatly; for instance, the ethyl gallate signal intensity in R. crenulata was substantially higher than that in other species, whereas R. fastigiata showed the highest reducing sugars level among the analyzed Rhodiola species. Thus, some Rhodiola species can be visually discriminated. By comparing the samples with the 1H-NMR spectra of standard compounds, the addition of relevant reference compounds directly to the pH-adjusted NMR sample, and examining the published literature, a total of 22 resonance signals were revealed to be consistent with specific structures. Detailed information for the assigned peaks can be found in Table 2, Fig. 1B and Supplementary Fig. 2.
(A) Representative 1H-NMR spectra of rhizome sample extracts from three Rhodiola species (0.0–9.5 ppm). (a) R. crenulata (b) R. kirilowii, (c) R. fastigiate. (B) Representative 1H-NMR spectra of the expanded regions and signal assignments.
| No. | Metabolites | Chemical shift (ppm), J (Hz) |
|---|---|---|
| 1 | Formic acid | 9.72(s) |
| 2 | Salidroside | 2.84 (d, J = 6.8 Hz), 3.26 (m), 4.36 (d, J = 6.8 Hz), 6.72 (d, J = 4.8 Hz), 7.13 (d, J = 8.2 Hz) |
| 3 | Tyrosol | 2.75 (t), 3.24 (t), 7.13 (d, J = 8.2 Hz) |
| 4 | Gallic acid | 7.08 (s) |
| 5 | Ethyl gallate | 1.36 (t), 4.33 (q), 7.09 (s) |
| 6 | Maleic acid | 6.02 (s) |
| 7 | Sucrose | 5.40 (d, J = 3.8 Hz) |
| 8 | Saturated fatty acid | 1.32 (m), 2.27 (t), 5.36 (t) |
| 9 | α-Glucose | 5.15 (d, J = 3.8 Hz) |
| 10 | β-Glucose | 4.54 (d, J = 6.4 Hz) |
| 11 | Quinic acid | 1.98 (m), 4.17 (m) |
| 12 | Fructose | 3.62 (m), 3.69 (m), 4.04 (m) |
| 13 | Alanine | 1.47 (d, J = 7.3 Hz), 3.78 (m) |
| 14 | L-chiro-Inositol | 3.48 (s) |
| 15 | Unsaturated fatty acid | 1.32 (m), 2.27 (t) |
| 16 | Pyruvic acid | 2.48 (s) |
| 17 | Glycine | 3.56 (s) |
| 18 | Valine | 1.06 (d, J = 4.6 Hz), 1.08 (d, J = 4.6 Hz) |
| 19 | Sterol | 0.66 (s) |
| 20 | p-Coumaric acid | 6.84 (m), 6.06 (d, J = 12.2 Hz) |
| 21 | Catechin | 3.37 (m), 6.29 (m), 7.48 (d, J = 4.9 Hz), 7.58 (m) |
| 22 | Caffeic acid | 6.84 (m), 6.94 (m), 7.10 (s) 7.50 (m) |
PCA of the 1H-NMR spectral data was used to compare interpretations of different variations in Rhodiola species, visualize the underlying trend, and understand the metabolic distinction of different Rhodiola species. PCA was first performed on the 1H-NMR spectral data for all 41 samples. However, noticeable overlaps were observed among different species, indicating that these samples could not be well separated. The PCA score plot (PC1 = 44.1%, PC2 = 13.2%) of the mean-centred dataset in Fig. 2A shows that R. crenulate could be distinguished from the other two species; however, R. kirilowii and R. fastigiate were not clearly classified.
(A) PCA score plot of the remaining three Rhodiola species, (B) PLS-DA score plot of the remaining three Rhodiola species. (C) HCA of three Rhodiola species (calculated with ward and sorted by size; the relationship between numbers and species are shown in Supplementary Table 1).
PLS-DA score plots provided good agreement with the PCA results (Fig. 2B, 2C). The PLS-DA and HCA score plot of Rhodiola species showed that three species could be clearly separated (R2X(cum) = 0.634, R2Y(cum) = 0.774, Q2(cum) = 0.651). The PLS-DA model of the remaining three Rhodiola species was also validated with 200 permutations of 3 components and had a proper R2Y-intercept of 0.24 and Q2Y-intercept of −1.33.
GEP ResultsThe parameters used per run are summarized in Supplementary Table 3. Good solutions with R2 values of 0.719 and 0.875 were obtained for the training and test sets, respectively. The C++ function was converted into the following equation, in which d and its subscript represent descriptors of the integral position:
 |
The integration results at d166, d56, d236, and d128 of the R. crenulata, R. kirilowii and R. fastigiata 1H-NMR spectra were imported and using this formula, if the y value is less than 1.37, the Rhodiola may be R. crenulata. If the value of y is more than 1.37, the Rhodiola is R. kirilowii or R. fastigiate. Most of the predicted values were very close to the experimental values, but a few compounds were more poorly fitted to the experimental values. In this formula, d166 and d56 are the integral areas of caffeic acid and tyrosol. The GEP results that caffeic acid and tyrosol may important chemical components distinguishing R. crenulata and the other two Rhodiola species. For random initialization which can affect GEP results, 20 evolution runs were performed, and the fitness and accuracy were 1000 and 100%, as shown in Supplementary Table 3 and Fig. 3.
The corresponding loading plots for PLS-DA elucidated that signals from alanine, fatty acids, fructose, salidroside, and glycine were the dominant discriminators at the species level (Fig. 4A). To obtain a better understanding of the different individual metabolite levels among rhizome samples from the three Rhodiola species, eight selected metabolites were analyzed using a qNMR method. The characteristic signals of gallic acid (7.08 ppm, s), salidroside (6.72 ppm, s), α-glucose (5.15 ppm, d), glycine (3.56 ppm, s), fatty acids (2.29 ppm, t), and alanine (1.47 ppm, d) were selected for quantification, and the results are shown in Fig. 4B. Significant differences were observed for the levels of all selected metabolites across most of the Rhodiola species (p < 0.05). The gallic acid concentration in R. crenulata was significantly higher than in the other. Rhodiola species as was the level of salidroside. R. crenulata also exhibited substantially higher levels of α-glucose. The fatty acid level in R. kirilowii was lower than the other species. These findings were in good agreement with the PLS-DA results.
Data are given as the means ± S.D. Different letters above the bars indicate significant differences between species, and the same letters above the bars indicate no significant differences between species based on Tukey’s multiple comparison tests (p < 0.05).
It was significant to verify the results of 1H-NMR metabonomics and to validate the importance of differences in metabolite levels between different Rhodiola species. So, HPLC was used for simultaneous determination of four metabolites in the three species (Supplementary Fig. 3). As shown in Fig. 5, gallic acid and salidroside were significant different in three species of Rhodiola, and these compounds in R. crenulate were higher than other species of Rhodiola. It was similar with the results of qNMR. R. crenulata also has higher levels of tyrosol and caffeic acid. It was in good agreement with the results of GEP. In addition, although the results of HPLC showed R. crenulate had more tyrosol and caffeic acid, qNMR indicated that their differences were not significant. So, caution was needed when explaining the results of qNMR. It also demonstrated that the 1H-NMR and GEP analysis should be combined to supply more information for the selection of metabolic markers.
Data are given as the means ± S.D. Different letters above the bars indicate significant differences between species, and the same letters above the bars indicate no significant differences between species based on Tukey’s multiple comparison tests (p < 0.05).
LC-GC/MS and 1H-NMR are most usually used in metabolomics. Although LC-GC/MS was sensitive and specific, compared with the two methods, 1H-NMR combined GEP can detect both primary and secondary metabolites at the same time in a single operation, produce unbiased plentiful metabolic profile. Furthermore, the method can also be used for qualitative and quantitative analysis of various chemical components, with simple and reproducible sample preparation. Thus, 1H-NMR combined with multivariate data analysis has recently demonstrated its suitability for metabolomics studies for the discrimination, authentication, and quality assessment of food, crops and herbal medicines.
Lacking of effective species identification and quality assessment devalue the clinical application of Rhodiola.32) In this study, chemical profiling and species discrimination of three Rhodiola species native to China were achieved using 1H-NMR-based metabolomics coupled with multivariate statistical analysis and GEP analysis, and the quantitation of major metabolites to discriminate species was performed using TMSP as a reference. Moreover, differences in metabolite signal intensities were observed among Rhodiola species. Additional PLS-DA loading plots and a qNMR method coupled with analysis of variance proved that these metabolites were important species discriminators.
Salidroside, tyrosol, gallic acid and caffeic acid have been reported to be the essential primary metabolites in Rhodiola.33) These compounds are the key determinant of officinal quality of Rhodiola. The PCA and PLS-DA score plots showed that most of the Rhodiola species assessed were clearly classified. Hence, it can be concluded that metabolites such as gallic acid, salidroside, α-D-glucose, glycine, and alanine can be used as discriminators. Different ecological factors may be important for influencing primary metabolite concentrations.34) Although fatty acids were revealed in the PCA and PLS-DA loading plots as species discriminators, the extraction solvent used in this study favoured polar compounds. Therefore, the quantification of fatty acids may not reflect the actual raw material content, and this metabolite cannot be regarded as a metabolic marker for species differentiation in this study.35)
As a traditional Tibetan medicine, Rhodiola species have been widely used for therapies of cardiovascular disease, hypobaric hypoxia, microbial infection, tumour and muscular weakness. Organic acids, flavonoids, and phenylpropanoids, such as gallic acid, salidroside, tyrosol, and caffeic acid were the major effective constituents in Rhodiola and can be used as markers to evaluate the quality of Rhodiola. In this paper, we analyzed the concentration of gallic acid, salidroside, tyrosol and caffeic acid in three species via qNMR and HPLC methods. As shown in the Figs. 4B and 5, the results indicated that four secondary metabolites above in R. crenulata was significantly higher than in the others. Interestingly, gallic acid and salidroside markedly showed the difference between R. kirilowii and R. fastigiata. Therefore, the four compounds can be used as metabolic markers to distinguish different species of Rhodiola. Meanwhile, we could consider them as sensitive indicators for the evaluation of Rhodiola quality.
GeneXproTools is a powerful computing software package used to develop the classification model.36) The method provides transparent modelling solutions and delivers to users the mathematical equation describing the classification model.37) The results revealed that combined analysis using the classification algorithm was successful. The GEP approach creates randomly formed functions, evolves them using genetics and natural selection principles, and then selects the one that best fits the results.38) A sample produces the results of multiple data points, which are then used in the formula to distinguish the sample from other data sets in the GEP process. 1H-NMR piecewise integral areas are an example of such a data set. Therefore, 1H-NMR of Rhodiola data is suitable for the application of GEP. In this research, a formula was calculated using GEP to distinguish R. crenulata from other Rhodiola species. The model was built successfully. Therefore, GEP could also identify the important chemical compounds distinguishing different Rhodiola species, such as tyrosol and caffeic acid. Consequently, GEP has ability to establish different Rhodiola 1H-NMR data formula sets and define the integral key areas.
In previous studies, data from 1H-NMR metabolomics have been primarily processed with PLS, PCA, HCA, and other methods. Although these methods can be used to rapidly analyse the metabolite content of different plants, they do not establish a formula for the rapid discrimination of several known medicinal materials. This study is the first to combine PCA, PLS-DA, HCA, and other chemometrics methods with GEP and other neural network algorithms to analyse 1H-NMR metabolomics data. The study demonstrates that different Rhodiola species can be discriminated by GEP. A formula was generated by GEP to discriminate R. crenulata and the two other Rhodiola species included, and variations in the metabolites from R. crenulata and other two Rhodiola species were observed. These results provide a new neural algorithm to handle metabolomics data processing. Finally, in order to make a reference data, the samples were analyzed by HPLC. The results of HPLC were similar with the observations obtained by the combination of 1H-NMR and GEP. It demonstrated the results of present method.
In conclusion, 1H-NMR-based metabolomics combined with multivariate statistical analysis is a fast and reliable method for species differentiation and quality control. The specific analysis of individual metabolites is crucial for the quality assessment of natural products made from different Rhodiola species relative to their specific therapeutic and health care effects. Gallic acid, salidroside, α-D-glucose, glycine, alanine, caffeic acid, and tyrosol were metabolic markers for discrimination of Rhodiola. The PCA, PLS-DA, and GEP models established in this study can be used to recognize and authenticate unknown Rhodiola species. 1H-NMR metabolomics combined with chemometrics and GEP analysis have potential for other traditional Chinese medicine identification.
We would like to acknowledge funding from National Natural Science Foundation (No. 81203000) and the National Key Research and Development Program of China (No. 2017YFC1703900) for supporting this work.
The authors declare no conflict of interest.
The online version of this article contains supplementary materials.
