Nanopore analysis of salvianolic acids in herbal medicines

Abstract

Natural herbs, which contain pharmacologically active compounds, have been used historically as medicines. Conventionally, the analysis of chemical components in herbal medicines requires time-consuming sample separation and state-of-the-art analytical instruments. Nanopore, a versatile single molecule sensor, might be suitable to identify bioactive compounds in natural herbs. Here, a phenylboronic acid appended Mycobacterium smegmatis porin A (MspA) nanopore is used as a sensor for herbal medicines. A variety of bioactive compounds based on salvianolic acids, including caffeic acid, protocatechuic acid, protocatechualdehyde, salvianic acid A, rosmarinic acid, lithospermic acid, salvianolic acid A and salvianolic acid B are identified. Using a custom machine learning algorithm, analyte identification is performed with an accuracy of 99.0%. This sensing principle is further used with natural herbs such as Salvia miltiorrhiza, Rosemary and Prunella vulgaris. No complex sample separation or purification is required and the sensing device is highly portable.

Fig. 1. Identification of salvianolic acid B using MspA-90PBA.

a The structure of MspA-90PBA. MspA-90PBA is a hetero-octameric MspA modified with a single phenylboronic acid (PBA) adapter at its pore constriction (Methods). The mechanism of salvianolic acid sensing is described on the right. Briefly, the PBA at the pore constriction can react reversibly with a cis-diol group of the analyte to produce a nanopore event. b A cartoon of the herb Salvia miltiorrhiza Bunge. Salvia miltiorrhiza (Danshen, dotted box), which is the root of Salvia miltiorrhiza Bunge, and contains rich levels of salvianolic acids. c The chemical structure of salvianolic acid B (SalB). SalB is a type of salvianolic acids. The 1, 2-diol groups on SalB are marked in red. SalB, which is the most abundant salvianolic acid in Salvia miltiorrhiza, is widely applied in the treatment of cardiovascular and cerebrovascular diseases. d A representative trace containing nanopore events of SalB. The measurement was carried out using MspA-90PBA in a buffer of 1.5 M KCl, 100 mM MOPS, pH 7.0. A + 100 mV voltage was continually applied. SalB was added to cis with a final concentration of 0.1 mM. Three types of events were observed from the trace. For the ease of demonstration, each event was respectively marked with different roman numerals to show their identities. e The scatter plot of $\Delta I/I_o$ versus $S.D.$ for data acquired as described in (d). To remove background noises, the data in the scatter plot was treated by cluster analysis using DBSCAN (Supplementary Fig. 4). 672 events were demonstrated in the scatter plot. Source data are provided as a Source Data file.

Fig. 2. Discrimination of eight salvianolic acids using MspA-90PBA.

a–h The chemical structures of eight types of salvianolic acids and their corresponding nanopore events. The salvianolic acids include caffeic acid (CA), protocatechuic acid (PCA), protocatechualdehyde (PA), salvianic acid A (SAA), rosmarinic acid (RA), lithospermic acid (LSA), salvianolic acid A (SalA) and salvianolic acid B (SalB). The 1, 2-diol groups of each compound are marked in red. The abbreviations of each analytes are also marked with color bands, including black (CA), pink (PCA), red (PA), green (SAA), blue (RA), wine-red (LSA), lavender (SalA) and orange (SalB). All measurements were carried out using MspA-90PBA in a buffer of 1.5 M KCl, 100 mM MOPS, pH 7.0 (Methods). CA (1 mM), PCA (2 mM), PA (0.5 mM), SAA (0.5 mM), RA (0.3 mM), LSA (0.2 mM), SalA (0.03 mM) and SalB (0.1 mM) were separately added to cis. A + 100 mV bias was continually applied. CA (a), PCA (b) and PA (c) contain a single 1, 2-diol group and only one type of event was reported for each type of analyte. SAA (d), RA (e) and LSA (f), which contain two 1, 2-diol groups, report two types of events. SalA (g) and SalB (h), which contain three 1, 2-diol groups, report three types of events. (i) Top: The scatter plot of $\Delta I/I_o$ versus $S.D$. of events acquired from all eight types of salvianolic acids. 500 events acquired with each type of analyte were included in the scatter plot (n = 4000). To remove background nomises, all events were treated by cluster analysis using DBSCAN, as described in Supplementary Figs. 4–11. Bottom: the zoomed-in view of the area marked with a dashed box in the top. Source data are provided as a Source Data file.

Fig. 3. The machine learning workflow.

a The training dataset. 500 events acquired with each type of analytes, including CA, PCA, PA, SAA, RA, LSA, SalA and SalB, were collected to form the training dataset (top). Two event features, including the relative blockage depth ($\Delta I/I_o$) and the standard deviation ($S.D.$), were extracted from each event to form a feature matrix (bottom). b Training accuracies. Six commonly used models including K-Nearest Neighbor (KNN), Extreme Gradient Boosting (Xgboost), Classification and Regression Tree (CART), Support Vector Machine (SVM), Gradient Boost Decision Tree (GBDT) and Random Forest (RF) were evaluated. KNN, which reports the highest validation accuracy, was selected for all subsequent prediction tasks. c The confusion matrix result of salvianolic acids classification performed by the trained KNN model. d A representative trace acquired by simultaneous sensing of all eight salvianolic acids. The measurement was carried out using MspA-90PBA in a buffer of 1.5 M KCl, 100 mM MOPS, pH 7.0 (Methods). All analytes were added to cis to reach the desired final concentrations and a + 100 mV bias was continually applied. Specifically, the final concentrations of CA and SAA were 40 μΜ, that of PCA was 100 μΜ and that of PA, RA, LSA, SalA and SalB were 20 μΜ. All events were automatically predicted by machine learning and labeled with corresponding labels. e The scatter plot of $\Delta I/I_o$ versus $S.D.$ of results acquired by simultaneous sensing of all eight salvianolic acids using the same nanopore (n = 4268). Each event was identified by the previously trained KNN model and is color labeled. Source data are provided as a Source Data file.

Fig. 4. Rapid analysis of salvianolic acids in salvianolate injection.

a The workflow of salvianolate injection analysis. Left: The powder of salvianolate injection was dissolved in Milli-Q water to reach a 5 mg/mL concentration. Center: 4 μL dissolved salvianolate injection was added to the cis chamber of a nanopore device. The measurement was carried out using MspA-90PBA in a buffer of 1.5 M KCl, 100 mM MOPS, pH 7.0 (Methods) and a bias of +100 mV was continually applied. Right: Corresponding nanopore events observed immediately. (b) A representative trace acquired during salvianolate injection analysis. The events were identified by the trained KNN model and are labeled accordingly. (c) The scatter plot of $\Delta I/I_o$ versus $S.D.$ of events acquired with the salvianolate injection. The events in the scatter plot were taken from a 30 min continually recorded trace and a total of 846 events were collected. The events were labeled according to the prediction results performed by the previously trained KNN model. (d) The proportion of salvianolic acid events in the salvianolate injection. Data were presented as mean ± standard deviation values derived from results of three independent measurements (N = 3) (Supplementary Fig. 20). The error bars represent standard deviation values. Clearly, SalB is the main component of the salvianolate injection. However, other salvianolic acid components were also detected by nanopore. Source data are provided as a Source Data file.

Fig. 5. Rapid identification of salvianolic acids in natural herbs.

a A workflow of nanopore identification of salvianolic acids directly from natural herbs. The gray timeline stands for the time of the whole procedure and red bars represent the time of human operation. Phase I: Sample pretreatment. Natural herbs were crushed and soaked in Milli-Q water for 12 h at 4 °C. Human operations: herb crushing and soaking (5 min). Phase II: Liquid collection. The soaking liquid was centrifuged at 4 °C and 1500 g for 10 min and the supernatant was collected. Human operations: centrifugation preparation (1 min) and supernatant collection (1 min). Phase III: Ultrafiltration. The collected supernatant was treated with a 3 kDa ultrafiltration tube at 4 °C and 1900 g for 30 min and the filtrate was collected. Human operations: ultrafiltration preparation (1 min) and filtrate collection (1 min). Phase IV: Nanopore sensing. 20 μL filtrate was added to the cis chamber of a nanopore device. Human operations: sample addition (10 s). Phase V: Data analysis. Human operations: automatic data analysis by machine learning (2 min). A more detailed workflow was also described in Methods and Supplementary Fig. 23. b, e, h Three types of commercially available natural herbs including (b) Salvia miltiorrhiza, (e) Rosemary and (h) P. vulgaris and their corresponding soaking liquids. c, f, i Representative nanopore traces acquired with different herb samples. All events were identified by the trained KNN model and correspondingly labeled as RA (blue), LSA (wine-red), SalB (orange), PA (red), SalA (lavender) and others (black). The ‘other’ events represent events that don’t belong to any previously trained salvianolic acid model compounds, based on results of outlier analysis (Supplementary Figs. 24, 26, 27). d, g, j The proportion of salvianolic acid events from results acquired with (d) Salvia miltiorrhiza, (g) Rosemary and (j) P. vulgaris (Supplementary Figs. 25, 28, 29). Data were presented as mean ± standard deviation values derived from results of three independent measurements (N = 3). The error bars represent standard deviation values. All above described results were acquired by nanopore measurement using MspA-90PBA in a buffer of 1.5 M KCl, 100 mM MOPS, pH 7.0 and a + 100 mV bias, which was continually applied. Source data are provided as a Source Data file.