HPLC Feature Generator
Chemical characteristics can be used to describe and evaluate medicinal substances as a whole. The HPLC method has good accuracy, sensitivity and reproducibility, and can be used to quickly and specifically identify different herbs based on overall chemical composition. The HPLC chromatograms of COR and CGR are shown in Fig. 2. Component identities were confirmed based on retention time and UV spectra (285 nm) of chemical markers. The main chemical components of COR and CGR were similar. As expected, the content of curculigoside (peak 1), the indicator component of COR, was not significantly different from that of CGR. Interestingly, CGR contains a unique compound that was detected in HPLC chromatograms, but was not found in the COR profile. Therefore, this unique compound was specifically separated and purified, and the structure was identified by modern spectroscopic techniques. This was a new compound determined to be 5-(3′,4′-dihydroxyphenyl)-1-(4″-hydroxyphenyl)pentane-1,4-dione, 1D and 2D NMR spectra were available at Figs. S1 – S5. However, the low content of this compound was not sufficient to accurately distinguish between two plant sources.
Variations in the stable isotope ratios of COR and CGR
The variations of the stable isotopic compositions between COR and CGR have been presented in Fig. 3. Mean N% values in COR and CGR samples were 1.898% and 0.720%, N% values in COR were significantly higher (Fig. 3a). The mean C% values of the COR and CGR samples were 40.052% and 39.998%, respectively (Fig. 3b). The average δ15The COR N value was -3.157‰, which was significantly lower than the CGR value, with a mean value of -0.173‰ (Fig. 3c). The average δ13The COR C value was -28.678‰, which was significantly higher than the CGR value, with a mean value of -31.487‰ (Fig. 3d). There were significant differences in the mean value of N%, δ15N and δ13C according to botanical origins (all PJ -test).
The 3D point cloud of N%, δ15N and δ13The C values have been presented in Fig. 4, and they showed excellent ability to predict COR and CGR. Overall, the COR had a high N% and δ13C, and a low δ15N value, so they gathered in the upper section of the 3D chart. However, the CGR, on the other hand, mainly appeared at the bottom. The stable isotope ratio shows a good effect to distinguish different sources of Curculigo Rhizome.
Analysis of mineral elements
The mineral element contents in the COR and CGR samples are shown in Table 1. The results appeared to be significantly different between the two source species except for B, Mg, K, Ca, Cu, Se, Ba. K and Ca were the most abundant inorganic elements in COR and CGR. The contents of Li, Al, Mn, Co, Ni, Zn and Cd were higher in the COR than in the CGR, while the concentrations of the elements Na, Ti, Fe, Sr and Mo were present at a lower level in the COR samples.
Principal component analysis of COR and CGR
Multivariate assessment is needed to improve the overall accuracy of COR and CGR. Based on the chemical analysis of the stable isotope ratios combined with the concentrations of 19 mineral elements, the result of the PCA analysis has been presented in Fig. 5a. The vectors and the cumulative contribution of the variance of the first three PCs (PC1-3) have been presented in Table S3. A three-factor model (the top three PCs with eigenvalues >1) can explain 88.0% of the total variability in the original data, which showed that the top three PCs can reflect most of the information in the samples. PC1, PC2 and PC3 contributed 61.0%, 19.8% and 7.2% of the total variance respectively. The result showed that 10 COR samples grouped together and 9 CGR samples grouped into another category. It was presented that COR and CGR samples can be well distinguished by PCA. Notably, three batches of COR from Yunnan tend to be distinguished from Sichuan.
The PCA biplot of PC1 and PC2 has been shown in Fig. 5b. PC1 was mainly correlated with N% intensity, δ13C, Li, K, Mn, Co, Cu, Zn, Se, Cd and negatively correlated with δ15Signal N, Sr, Mo35.36. The intensity of B, Al, Fe, Ni, Ba was significant in PC2. COR samples (1–7) from Sichuan were mainly affected by the content of N% and elements Li, K, Mn, Zn, Co, Cd, while COR samples from Yunnan (8–10) were isolated. PC1 had a better ability to discriminate COR samples. However, CGR samples (11–19) were clustered with δ15N, Ti, Sr, Mo. The classification of RGCs was related to the content of these elements and can be distinguished by them. Plant metabolic activities were found to have a greater impact on the δ content13C that environmental factors24.26. Therefore, the difference between COR and CGR samples may be due to the different elements accumulated in plant metabolism.
Identification of COR and CGR by OPLS-DA
To further utilize the potential discrimination capability of stable isotope and phased array analysis, OPLS-DA was used to process the data of COR samples and counterfeit CGR samples, and the result was shown in Fig. 6. Genuine COR samples and counterfeit CGR samples were significantly differentiated, indicating that stable isotope ratios and element contents combined with OPLS-DA analysis was an effective method to separate COR and CGR samples . The number of important components is determined by calculating the explained variance X (R2X), deviation Y (R2Y), and the predictive ability of cross-validation (Q2) 37. The evaluation parameters of the OPLS-DA prediction models were: R2X = 0.800, R2Y = 0.993, Q2 = 0.991. Generally, the model has good fit when these values are close to 1.0, the point of intersection of R2 and Q2 with the Y axis must be less than 0.3 and 0.05 respectively, and the difference between R2 and Q2is less than 0.338.39. Therefore, the results showed that this OPLS-DA model was reliable. In addition, VIP > 1 was considered a good identification marker27,34,40and OPLS-DA provided 13 effective potential markers (δ15N, Cd, Sr, δ13C, N%, Co, Se, Ti, Zn, Li, Cu, Mn, K) to determine the authenticity of COR samples and counterfeit CGR samples (Fig. S7). Notably, the three CORs from Yunnan were also separated from the COR samples from Sichuan based on their stable isotope ratios and element contents by the OPLS-DA model. The results indicated that stable isotope ratios combined with element contents may have the potential ability to predict the geographic origin of CurculigoRhizome. Based on these advantages, stable isotope ratios and element contents combined with OPLS-DA analysis provide an excellent method for discriminating COR and CGR samples.
Classification of Curculigo Rhizoma using LDA
To verify the reliability of the classification model, LDA was performed using a cross-validation procedure to calculate the classification and probability of the COR and CGR samples.23.28. The cross-validation result was displayed in Table 2. The LDA model gave a good classification rate (100%) and the cross-validation rate (100%), COR and CGR were identified successfully. So the predictive model worked well, LDA analysis combined with stable isotopes and elements could be used to discriminate the two source species of CurculigoRhizome.