Evaluation of the water soluble extractive of astragali radix with different growth patterns using ¹H-NMR spectroscopy

2.1 Plant materials

Astragalus membranaceus plants were collected in two locations of North Shanxi Province, China, Hunyuan (39°72′N, 113°68′E) and Wuzhai (40°01′N, 113°21″E). All plant materials were authenticated by Prof. Xue-Mei Qin of Shanxi University, with the voucher specimens deposited in the herbarium in the Center of Modern Research for Traditional Chinese Medicine of Shanxi University, China. All samples were air-dried and ground into fine powder with a pestle and mortar for analysis. Three biological replicates were used in all experiments.

2.2 Solvents and chemicals

Deuterium oxide (D₂O) was obtained from Merck (Darmstadt, Germany). Sodium 3-trimethlysilyl [2, 2, 3, 3-d₄] propionate (TSP) was obtained from Cambridge Isotope Laboratories Inc. (Andover, MA, USA), and deuterated sodium hydroxide (NaOD) was purchased from Armar (Döttingen, Switzerland).

2.3 WSE value determination

The WSE values of AR were determined according to the procedures described in the appendix XA of Chinese Pharmacopoeia [10]. Four grams of air-dried, coarsely powdered material were transferred to a glass-stoppered conical flask. One hundred milliliters of H₂O was added, and the flask was weighed. The flask was shaken for 6 h and then allowed to stand for 18 h. The contents were then filtered rapidly through a dry filter. Twenty milliliters of the filtrate was transferred to an evaporating dish of pre-determined weight, and evaporated in a water bath to dryness. The filtrate was dried at 105 °C for 3 h, cooled in a desiccator for 30 min, and then weighed as soon as possible. The percentage of WSE value was calculated with reference to the air-dried drug. Three replicates were used for all experiments.

2.4 UV absorption spectroscopy

Filtrates of the extract were diluted 25-fold for UV detection. Absorption spectroscopy was performed using a UV-Vis Spectrophotometer (Varian Cary 50, Agilent, Santa Clara, CA, USA) in the wavelength range from 250 to 800 nm. The raw data were transferred to a microcomputer and peaks were analyzed with software provided by the manufacturer.

2.5 Sample preparation for NMR and measurements

The above-obtained WSE (25 mg) was transferred into a 1.5 mL centrifuge tube, and 800 μL KH₂PO₄ buffer in D₂O (adjusted to pH 6.0 by 1 N NaOD) containing 0.05% TSP was added. After centrifuging for 10 min at 16,060 g, the supernatant (600 μL) was transferred into a 5 mm NMR tube for NMR analysis. ¹H-NMR spectra were recorded at 25 °C on a Bruker 600-MHz AVANCE III NMR spectrometer 600.13 MHz proton frequency (Billerica, MA, USA). D₂O was used for internal lock purposes. ¹H-NMR spectra were acquired using the 1D NOESY pulse sequence, and each ¹H-NMR spectrum consisted of 64 scans, requiring 5 min acquisition time with the following parameters: 0.18 Hz/point, pulse width (PW) 12.7 μs, and relaxation delay (RD)=5.0 s. Free induction decay was Fourier transformed with LB=0.3 Hz. The resulting spectra were manually phased and baseline-corrected, and calibrated to TSP at 0.00 ppm.

2.6 Data analysis and quantification

The ¹H-NMR spectra were processed using MestReNova (version 8.0.1, Mestrelab Research, Santiago de Compostella, Spain). The spectra were scaled to TSP and reduced into integrated regions of equal width (0.04 ppm) for the region of 0.00–10.00. The regions of 4.70–5.02 were excluded from the analysis because of the residual signal of H₂O. Principal component analysis (PCA) and orthogonal projections to latent structures discriminant analysis (OPLS-DA) was performed with the SIMCA-P software (version 13.0, Umetrics, Umeå, Sweden) using the Pareto scaling method. A 7-fold cross validation of the OPLS-DA was applied to test and verify the model. Six of seven samples were used as a training set and the remaining were used as a testing set. To test the model, test samples were defined as unknown samples. This process was conducted seven times. Every sample was defined as a training and testing set, then R² and Q² were used for evaluating this model’s fitting and prediction ability [15–17]. Hierarchical cluster analysis (HCA) was performed in SPSS (version 11.5.0, USA). The imported data were then Pareto-scaled for the multivariate analysis. Recursive support vector machine (R-SVM) and heatmap were performed by MetaboAnalyst 2.0 (Edmonton, Alberta, Canada; www.metaboanalyst.ca), a comprehensive tool suite for analyzing metabolomic data. For the relative quantification, the peak areas of the TSP were integrated manually for all the samples as an internal standard for quantification of metabolites.

3.1 WSE values

The water soluble extractive of AR should be no less than 17%, as stated in the Chinese Pharmacopoeia [10]. As can be seen in Table 1, the WSE value of cultivated AR is significantly higher than that of the wild AR (p<0.01).

3.2 UV-absorbance

Figure 1 illustrates the filtrates obtained from the preparation of WSE, with the color of wild AR very different from that of the cultivated AR. Comparison of their UV- absorbance in the range of 250–800 nm suggests that the UV-absorbance of wild AR between 250 and 400 nm is stronger than that of the cultivated AR (Figure 2), which can be used as a simple method to distinguish the wild and cultivated ARs.

Figure 1:

The filtrates of water soluble extractives (WSE) obtained from four groups of astragali radix (AR). 1, Wild AR of Shanxi Wuzhai (WZW); 2, wild AR of Shanxi Hunyuan (HYW); 3, cultivated AR of Shanxi Wuzhai (WZC); 4, cultivated AR of Shanxi Hunyuan (HYC).

Figure 2:

Absorbance of the filtrates of WSE in the range of 250–800 nm.

3.3 Metabolite identification

The representative ¹H NMR spectra of four groups of WSE is shown in Figure 3. The compounds were identified by analysis of the chemical shift, coupling constant, and comparison with the authentic standards, as well as other reported data [18, 19]. Most of the metabolites identified in the WSE were primary metabolites such as sugars, amino acids, organic acids, choline, and adenine. Sucrose was the major compound in the WSE of four groups of AR. The identified compounds as well as the characteristic chemical shifts are listed in Table 2.

Figure 3:

Representative 600 MHz ¹H NMR spectra of WSE from the four groups of AR.

3.4 Multivariate analysis

Because visual inspection indicated that the WSE of four groups of AR had very similar ¹H NMR spectra, multivariate analysis was used to uncover any subtle differences among them. Principal component analysis (PCA), one of the most common multivariate analysis methods, was used first to identify differences or similarities among the samples. As seen in the score plot of PC1 versus PC2 (Figure 4) with a total 70.7% of variance explained, most of the wild AR were located in the negative range of the PC1, while most of the cultivated samples were located in the positive range of PC1. The cultivation patterns may have greater impact on the chemical composition of WSE than the growing regions. In addition to PCA, all samples were further analyzed by hierarchical cluster analysis (HCA), another unsupervised clustering method. An HCA dendrogram of all the samples was calculated using squared Euclidean distance methods, which grouped the samples into two clusters according to their growth patterns, and the results were consistent with that of the PCA analysis.

Figure 4:

Score (PC 1 vs. PC 2) plot of PCA (A) and HCA (B) results obtained from ¹H NMR spectra of the WSE from four groups of AR. 1–3, Wild AR of Shanxi Wuzhai (WZW); 4–6, wild AR of Shanxi Hunyuan (HYW); 7–9, cultivated AR of Shanxi Wuzhai (WZC); 10–12, cultivated AR of Shanxi Hunyuan (HYC).

Supervised methods, including partial least squared discriminant analysis (PLS-DA), orthogonal partial least squares discriminant analysis (OPLS-DA) and R-SVM were then applied to the same bucketed ¹H NMR data to find the differential metabolites between the WSE of wild and cultivated AR. In the score plot of PLS-DA, clear separation between the two groups of AR was observed. In order to evaluate the validity of the model, permutation tests (permutation number: 200) were performed. All Q² and R² values were lower in the permutation tests than those in the real model, revealing great predictability and goodness of fit. OPLS-DA was then applied, and the R² (0.709) and Q² (0.96) values proved the good model quality (Figure 5A). Cross-validated analysis of variance further confirmed the validity of the model (p=0.024). The S-plot of OPLS-DA (Figure 5B) showed that the WSE of the wild AR contained higher levels of arginine, valine, threonine, asparagine, succinate, glutamine, and taurine, while the WSE of the cultivated AR contained a higher sucrose content.

Figure 5:

Scores plot (A), loadings plot (B), R-SVM classification plot (C) and variable importance plot (D) of the OPLS-DA and R-SVM analysis of NMR data of astragali radix.

R-SVM is a classification of different feature subsets, which are selected by their relative contribution to the classification using cross-validation error rates. This process creates a series of R-SVM models. The features used by the best model are plotted [20–22]. Figure 5C shows that when the number of feature subsets was 228, the error rate was 0%. The error rate increased to 0.03 when there were 46–91 feature subsets, suggesting that an NMR fingerprint can correctly distinguish WSE of wild and cultivated AR. As seen in Figure 5D, integrated regions were mostly occupied by the characteristic peaks of valine, threonine, arginine, glutamine, succinate, asparagine, and sucrose, and unknown peaks in integrated regions 6.52, 6.92, 7.6, 7.68 were important in distinguishing the two different varieties of AR. Most of the metabolites identified by R-SVM were consistent with those of OPLS-DA, indicating that different statistical methods should be used to obtain accurate results.

3.5 Relative quantification of metabolites

The seven differential compounds revealed above, i.e. arginine, valine, threonine, asparagine, succinate, glutamine, and sucrose, were relatively quantified based on the measurement of the average peaks of the characteristic resonances. The t-test confirmed several compounds contributing to the separation of wild and cultivated AR. The results were generated as a heat map (Figure 6), with the red indicating high intensity and the green indicating low intensity. It is obvious that arginine, valine, threonine, asparagine, succinate, glutamine, and sucrose were the metabolites differing most significantly between the WSE of the two kinds of AR.

Figure 6:

Heatmap of relative qualification of key metabolites. 1–3, Wild AR of Shanxi Wuzhai (WZW); 4–6, wild AR of Shanxi Hunyuan (HYW); 7–9, cultivated AR of Shanxi Wuzhai (WZC); 10–12, cultivated AR of Shanxi Hunyuan (HYC). *p<0.05, **p<0.01.