[Establishment of chromatographic fingerprint of and its applications in animal source identification and quality grade discrimination].

, or "Chuanshanjia" in Chinese, is a traditional Chinese medicine (TCM) for promoting blood circulation and reducing swelling and discharge; the only animal source used in TCM is the scales of . However, in today's pharmaceutical market, there are many scales from other species of the same genus that are difficult to distinguish from . High-quality and low-quality scales are also severely confused. To solve the above problems, various analytical methods have been developed, such as thin-layer chromatography, mass spectrometry and DNA detection. Owing to their low resolving ability, high equipment cost, and inconvenient operation, none of these methods are appropriate for routine identification of . A chromatographic fingerprint can comprehensively reflect the synergic action of multiple chemical compositions in TCM and has been widely used for the quality control of TCM. In the present study, we established a fingerprint of and explored its feasibility in identifying the origin and quality grade of scales. First, powder was hydrolyzed by hydrochloric acid (1 mol/L). Next, the extract was analyzed on a Symmetry 300 C18 column by linear gradient elution, using 0.1% trifluoroacetic acid (v/v) in water and 0.1% trifluoroacetic acid (v/v) in acetonitrile as the mobile phase and 280 nm as the detection wavelength. The established method was systematically validated, demonstrating good precision, repeatability and sample stability (relative standard deviation (RSD)<5%). Subsequently, samples of different sources and quality grades were distinguished by similarity evaluation and discrimination analysis based on the fingerprint data. In the similarity evaluation, the reference fingerprint was defined as the average fingerprint of twelve first-class samples, and seventeen chromatographic peaks were identified as common peaks. Similarities between the reference fingerprint and fingerprints with different base sources and quality grades were calculated using the absolute area of common peaks as original data. The similarities between and scales from other animals were all less than 0.776, while the similarities between of different grades overlapped significantly, varying from 0.988 to 0.996 for first-class samples and 0.950 to 0.995 for general samples. The results reflected the feasibility of similarity evaluation for discriminating base source and its limitation in the distinguishing between quality grades. Nonetheless, first-class scales showed higher average similarity and lower RSD than general scales, which indicates some level of revelation between fingerprint similarity and quality grade. Thus, a better algorithm or discriminant model is required to distinguish between quality grades. Therefore, a supervised chemometric technique, kernel-based support vector machine (SVM), was applied to construct predictive models. The SVM is a common discriminant model that classifies samples by constructing a separate hyperplane in -dimensional space, maximizing the margin between classes. Combination with a kernel function can effectively avoid "dimension disaster" when dealing with nonlinear data. In the model, the quality grade was defined as a sample label, and the absolute peak areas constituted the data matrix. Verified by 10-fold cross-validation, the unbiased prediction accuracy was up to 95.83%. The predicted results were highly consistent with the actual classifications. The results indicate the high feasibility of the established model for determining quality grade, as it performed significantly better than the similarity evaluation. Samples from batches A and B were completely discriminated and only two samples from batch S were incorrectly classified. Given the batch bias, we believe that model error may have been caused by man-made tag errors rather than the model itself. In conclusion, we established a chromatographic fingerprint for quality analysis and demonstrated its feasibility in animal source identification and quality determination by combining different data analysis methods. The established strategy may provide a new method for improving the the validity and accuracy of in clinical use.