Capillary electrophoresis in the profiling and authentication of herbal medicines

Herbal medicines are very complex, containing various numbers of different plants. In order to try and authenticate the medicines it is necessary to generate a chromatographic profile of the herbal substance. This is normally performed using HPLC but other methods are also available. The usefulness of capillary electrophoresis (CE) in providing the required profile was investigated using a model system in which most of the components were known and readily available. The model systems used were opium samples from different locations and poppy straw samples from differing plants. A novel CE method for the quantitative determination of the main components in the opium and poppy straw samples was developed using a variation of CE known as sweeping. The sweeping method developed allowed reduced detection levels and better resolution of components than methods published previously. The main components found in opium and poppy straw are morphine, codeine, thebaine, papaverine, noscapine, narceine, reticuline and oripavine. The assay method was validated with regards to migration time and peak area reproducibility. Calibration graphs for each of the components using dihydrocodeine as internal standard (IS) were performed. The linearity obtained was good with r2 values > 0.99 for all components. The calibration data was validated for intra and inter day reproducibility. Extractions of the opium samples from 4 different countries and poppy straw from 5 different plants were performed (n=10). Samples were injected into the CE in triplicate (n=3). The concentration of individual components was calculated using calibration data obtained on the same day. Multivariate statistical analysis (MSA) is used to visualise large amounts of data from samples. Three methods routinely used in MSA are hierarchical cluster analysis (HCA), principal component analysis (PCA) and soft independent modelling of class analogy (SIMCA) and each method was applied to differentiate the opium and poppy straw samples. HCA and PCA are used to see if the data forms clusters (or classes) and SIMCA is used to predict which class an unknown sample belongs. HCA analysis of the opium samples produced 4 clusters (in 2D) which was representative of the countries of origin. In the poppy straw samples 5 clusters were also achievable using HCA although it should be noted that 2 of the samples were from similar plant species. The clusters were reduced to 4 to align with the plant types. PCA was used to show the data in a 3D environment, by reducing the dimensionality. Principal components were extracted from the data which was ordered to retain most of the variance in the first few components. PCA of the opium samples showed 4 clearly defined spatial areas. The 3D plot of the straw samples only showed 4 spatial areas with poorer differentiation of similar plant species. For SIMCA analysis the data for the opium and poppy straw samples were divided into "training" and "unknown" sets. The training set was used to create a model to predict the classification of the unknown samples. The models created were based on various options available for pre-processing the data. For the prediction of the unknown opium samples two of the SIMCA models gave 100% correct answers. For the poppy straw samples the % correct classification was 98%. These results show that using quantitative data it was possible to differentiate opium samples from different locations and poppy straw samples from various plants. The main objective was to eliminate the need for quantitative data, and to use only migration times and peak areas without the requirement to identify and quantify individual components. The chromatograms generated above were used to create a fingerprint trace of all the peaks found in the opium and poppy straw samples. The opium samples produced 21 peaks while the poppy straw samples produced 15 peaks. The peaks were used in conjunction with the internal standard to produce tables of relative migration times and relative peak areas. Once again HCA and PCA analysis produced the correct number of clusters, with good differentiation. The data was once again divided into training and unknown sets to enable models to be built for SIMCA predictions. SIMCA predictions for the unknown opium samples produced an average of 97% correct results, while the predictions for the unknown poppy straw samples produced an average of 96% correct results. The results show that it was possible to use the fingerprint trace with only information on migration times and peak areas to differentiate sample types. The methods developed previously (CE and MSA) were used to try and differentiate Scutellaria plant varieties. Samples of Scutellaria baicalensis and Scutellaria lateriflora along with a single Scutellaria amoena sample were purchased from various suppliers. A separation method using CE with sweeping was developed to generate a fingerprint trace. The method developed was validated. No attempt was made to identify individual flavonoid components. Extractions of the different samples (n=10) were completed and analysed using the CE separation with injections performed in triplicate. The relative migration times and relative peak areas were calculated for 28 peaks which was representative of the fingerprint for Scutellaria samples. The data was firstly subjected to HCA analysis which showed 3 main clusters. The 3 clusters were representative of the 3 Scutellaria plant types. One sample of baicalensis was placed in the lateriflora cluster. Similar groupings were obtained using PCA. Prior to SIMCA analysis the data was split into two sets, training and unknown. The training set was used to create different models which would be used to classify the unknowns into one of the three classes. The % predictions correct for amoena was 92.8% while for lateriflora it was 97.4%. For baicalensis the % correct was 93.5%. The developed separation and multivariate analysis shows that samples of Scutellaria baicalensis can be successfully authenticated with greater than 90% correct predictions.