Elucidating direct kinase targets of compound Danshen dropping pills employing archived data and prediction models

Research on direct targets of traditional Chinese medicine (TCM) is the key to study the mechanism and material basis of it, but there is still no effective methods at present. We took Compound Danshen dropping pills (CDDP) as a study case to establish a strategy to identify significant direct targets of TCM. As a result, thirty potential active kinase targets of CDDP were identified. Nine of them had potential dose-dependent effects. In addition, the direct inhibitory effect of CDDP on three kinases, AURKB, MET and PIM1 were observed both on biochemical level and cellular level, which could not only shed light on the mechanisms of action involved in CDDP, but also suggesting the potency of drug repositioning of CDDP. Our results indicated that the research strategy including both in silico models and experimental validation that we built, were relatively efficient and reliable for direct targets identification for TCM prescription, which will help elucidating the mechanisms of TCM and promoting the modernization of TCM.

Research on direct targets of traditional Chinese medicine (TCM) is the key to study the mechanism and material basis of it, but there is still no effective methods at present. We took Compound Danshen dropping pills (CDDP) as a study case to establish a strategy to identify significant direct targets of TCM. As a result, thirty potential active kinase targets of CDDP were identified. Nine of them had potential dose-dependent effects. In addition, the direct inhibitory effect of CDDP on three kinases, AURKB, MET and PIM1 were observed both on biochemical level and cellular level, which could not only shed light on the mechanisms of action involved in CDDP, but also suggesting the potency of drug repositioning of CDDP. Our results indicated that the research strategy including both in silico models and experimental validation that we built, were relatively efficient and reliable for direct targets identification for TCM prescription, which will help elucidating the mechanisms of TCM and promoting the modernization of TCM.

CDDP
Compound Danshen dropping pills TCM Traditional Chinese medicine CAD Coronary artery disease SEA Similarity ensemble approach IC 50 Half-maximal inhibitory concentration ABPP Activity-based protein profiling IMPDH2 Inosine monophosphate dehydrogenase 2 HSP70 Heat shock protein 70 VSMCs Vascular smooth muscle cells IPA Ingenuity pathway analysis CHD Coronary heart disease auROC Area under the receiver operating characteristic curve Traditional Chinese medicine (TCM) prescriptions are the characteristics of Chinese medicine. They have been practiced for thousands of years and have been proved to be effective in modern clinical practice. These prescriptions embody the dialectical thought of Chinese medicine and the medication holistic view. In recent years, the reductionist research model has accumulated a lot of data, and also provided illuminating research results, such as the discovery of artemisinin 1 . It was discovered by Youyou Tu, a Chinese traditional medicine scientist, which is an effective and quick acting antimalarial drug. However, there is still a lack of effective approaches to systematically study its mechanism. The research of reductionism is not capable of answering the essential question of the overall efficacy of TCM. It may lead to deviate from the system theory of TCM, so it needs to be combined with the system theory. In recent years, a variety of "omics" techniques based on system theory have been widely used in TCM research [2][3][4][5] , for elucidating the pharmacological characteristics of TCM better [6][7][8][9] , but still cannot fully reveal the nature of it. Comprehensively understanding the mechanism of synergism among the effective components, drug targets and metabolic pathways remains highly demanded. One key to break this dilemma

Discussion
TCM prescriptions are the characteristics of Chinese medicine, which embody the dialectical thought of Chinese medicine and the medication holistic view. As accumulating evidences have proved that the ingredients entering the blood, main metabolites, bioequivalence components compared to the prescription, and active components reported in literatures contribute more to the effects and mechanisms of TCM [16][17][18] , we raised the hypothesis that the potential targets of all the important components mentioned above should be more likely to become the direct targets of the whole prescriptions. In addition, we included another component reported the most in single herbs, quercetin as important component to finalize the list. This method improves the credibility of the data, which is different from most used network pharmacology research flowchart [19][20][21] . In addition, in order to obtain the potential target data for the important components in a more accurate way, we integrated the recorded data and predicted data. On one hand, the existing research results from open data have been fully utilized. On the other hand, algorithm models were used to predict potential targets to avoid missing some important targets.  www.nature.com/scientificreports/ Moreover, kinase targets predicted by KinomeX platform were used to filter the kinase targets obtained by the above two methods, which can further improve the success rate of further verification. Most research focused on active components extracted from Chinese herbals and other natural products at present. Among them, small molecule affinity chromatography and activity-based protein profiling (ABPP) are the most widely used target identification technologies for active ingredients, such as target fishing technology [22][23][24][25][26][27] . Using this strategy, a series of targets for active components of TCM have been successfully identified, including the targets identification of sumitone 28 and chrysanthema lactone 29 . However, target fishing technology suits more for further in-depth analysis as low-through put experiment due to its excessive cost, which is not an efficient method to obtain direct targets broadly, especially for a whole prescription.
In this study, 106 potential kinase targets of CDDP were tested, and finally 30 active targets were obtained, with an accuracy of 28.3%. As expected, the success rate of the known kinase targets is higher than that of the Table 3. Kinase targets verified by activity test at 25 µg/mL concentration of CDDP. *The kinase activity inhibition rate of the sample compared to the blank group. The kinase activity of the blank was 100%. Generally speaking, the residual enzyme activity below 30% is strong inhibition, 30-70% is moderate inhibition. Considering the characteristics of TCM, we take 80% as the screening threshold. The lower the value is, the stronger the kinase activity is inhibited.   Table 4). The filter by KinomeX predictive results enables a higher success rate, which suggested that the strategy we built may serve as an efficient direct target predicting system for the other TCM prescriptions. However, in this study, only the algorithm based on structural similarity is used to predict the component target relationship. This method cannot distinguish the molecules with very similar structure, and the prediction results are often the same. However, the potency of a pair of molecules with similar structure will vary greatly 30,31 . In further study, a variety of state-of-art algorithms based on different principles should be utilized to predict component-target relationship for improving the accuracy of predicting results [32][33][34][35][36] , such as in silico models based on network topology parameters 33 , drug and target structure similarity 34 , clustering multi-dimensional drug target data 35 , deep learning and heterogeneous network 36 etc. In addition, molecular docking technology can also be used to gain more reliable targets for further experimental verification 37,38 . Among the 14 targets retested at the concentration of 250 µg/mL, the inhibitory activity of four kinases (CAMK2G, CSF1R, FYN and RET) did not decrease but increased at high concentration. The possible reasons may as follow: firstly, the components with high molecular weight in TCM form great stereo-hindrance effect when the concentration increases, which may hinder the combination between active molecules and targets. The second possibility is potential positive effectors involved in CDDP. The synergy weakens the affinity and internal effectiveness of the ligand on the receptors [39][40][41][42][43] . For example, both CDDP 12 (Rosmarinic acid) and CDDP 37 (Catechol) contained in CDDP could act on the common target FYN. However, the binding site for the two components may be different, which may bring the allosteric effect, weakening the inhibition effect under the condition of high concentration. These components may not directly bind to protein active sites, but to the allosteric sites, outside the active sites of the protein, causing the conformational change of proteins and their activity.
Three kinase targets (AURKB, MET and PIM1) of CDDP, that have been finally validated on cellular level, could provide basis for further elucidating the mechanism of CDDP in treating cardiovascular diseases 44,45 . For example, AURKB positively correlates with platelet aggregation and acute myocardial infarction (MI) 46 . MET shows repair function in cardiomyocytes and blood vessels through pro-angiogenesis, anti-inflammation and preventing fibrosis 47 . PIM1 plays a role in vascular smooth muscle cells (VSMCs) proliferation, which is closely related to the pathogenesis of atherosclerosis 48 . Besides, these targets are closely related to some other diseases [49][50][51][52][53][54][55] , indicating the potential function of CDDP against other indications, especially cancers. Actually, it has been reported the anti-tumor activity of several significant components of CDDP, including Danshensu 56 , Tanshinone I 57 , Cryptotanshinone 58 , Tanshinone IIA 59 , Rosmarinic acid 60 , and Ginsenoside Rg1 61 , suggesting the potential anti-tumor effect of CDDP.
Comparing with the above three kinases, it is worthy of note that CDDP promoted SYK activity in several cell lines (Figs. 3, 4), which showing an opposite trend with the kinase assays result (Fig. 2). One possible reason Table 5. Activity data for 14 kinase targets tested at different concentration of CDDP. a The lower the value, the stronger the binding activity. b The kinase with potential dose-dependent effect. Activity data 1 and 2 is the activity data of kinase targets tested at 25 µg/mL and 250 µg/mL concentration of CDDP, respectively. # Kinase Gene Gene ID Activity data 1 (% a ) Activity data 2 (% a )  www.nature.com/scientificreports/ of such inconsistency could be the complexity of TCM prescriptions when treating with cells. When some components with weak affinity/activating effect on SYK entering the cells preferentially, while those with strong affinity/inhibitory effect on SYK being obstructed by cell membrane, CDDP exerted activating effect on SYK as a whole prescription on cellular level. All the kinase targets obtained in this study need to be verified in a variety of disease models in the follow-up studies, which can help to explain the mechanism of CDDP on the existing main indications, or expand the new indications of CDDP.
In conclusion, 30 direct targets of CDDP were obtained in this study by the strategy we built, which is independent of any specific disease model and can provide a series of potential direct targets of TCM efficiently. Moreover, this strategy takes TCM as a whole research object, which is in line with the holistic view and systematic theory of TCM, conforming to the guiding principles of pharmacology theory of TCM. The direct targets not only provide the theoretical basis for elucidating the mechanism of action and the material basis, but also indicating rationales for the research of drug repositioning, which is of great significance for promoting TCM modernization.

Construction of important component set for CDDP.
In order to review the literatures related to CDDP as comprehensively as possible, we used "Danshen Dropping Pills" as the keyword to obtain the Chineselanguage literatures through CNKI (https:// www. cnki. net/). Similarly, through PubMed (https:// pubmed. ncbi. nlm. nih. gov/), "Compound Danshen drilling pills", "Fufang Danshen Diwan", "T89", "dantonic" and "Cardiotonic Pills" were used to get the English-language literatures (time to April 15, 2020). Then, the components contained in CDDP were extracted manually and standardized through PubChem database (https:// pubch em. ncbi. nlm. nih. gov/) 62 . Besides the ingredients entering the blood, main metabolites, bioequivalence components compared to the prescription, active components of CDDP reported in literatures, in order to avoid missing critical components included in CDDP, we selected the most extensively studied component in the three single herbs but still unconfirmed in the whole prescription, through retrieving TCM related databases, such as TcmSP 63   www.nature.com/scientificreports/ Prediction of potential direct kinases targets of CDDP. Based on the hypothesis and research strategy, we followed the steps below to obtain the potential direct kinase targets of CDDP, as described in the flowchart (Fig. 1).
Targets of 40 compounds obtained by retrieving public databases. The known activity data of 40 important components in CDDP were obtained from three authoritative public databases, namely, ChEMBL 68 , PubChem 62 , BindingDB 69 . The targets with definite activity information were standardized by annotating the basic informa- Each cell lines were divided into four groups as follows: control group, CDDP group (0.25 mg/mL), CDDP group (0.5 mg/mL), and CDDP group (1.0 mg/mL). The treatment time was 12 h for MCF7 and T47D cells. The samples derive from the same experiment and that gels/blots were processed in parallel. Statistical significance was determined by a two-tailed, unpaired Student t-test (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001 vs control). CDDP, Compound Danshen dropping pills. Full-length blots/gels are presented in Supplementary  Fig. S2 www.nature.com/scientificreports/ tion, such as Gene Symbol, Entrez Gene Name, Location, and Type(s) through ingenuity knowledge base in Ingenuity Pathway Analysis (IPA) and subsequently the kinase targets were screened. It is a professional database of functional annotation and biological interaction, which collects millions of detailed annotation information about proteins, genes, compounds, cells, tissues, drugs and diseases, as well as their interaction information. All information was collected from the original literatures and reviewed by hundreds of doctoral experts to ensure its accuracy. Each cell lines were divided into four groups as follows: Control group, CDDP group (0.25 mg/mL), CDDP group (0.5 mg/mL), and CDDP group (1.0 mg/mL). The treatment time was 6 h for TPC1 and BCPAP cells. The samples derive from the same experiment and that gels/blots were processed in parallel. Statistical significance was determined by a two-tailed, unpaired Student t-test (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001 vs control www.nature.com/scientificreports/ Targets of 40 compounds predicted by multi-voting SEA algorithm. Avoiding missing some important targets, multi-voting SEA algorithm 70 was utilized to predict potential targets of important components. In this algorithm, prediction models, namely Topological SEA, Morgan SEA, MACCS SEA, Atom Pair SEA and Pharmacophore SEA, were integrated to calculate potential targets of components, which could take advantages of different models and improve the robustness and the success rates of the models. By combining the five models, a flexible forecasting scheme was obtained with precision range from 71 to 90.6%, F 0.5 -Measure range from 0.663 to 0.684 and F 0.25 -Measure range from 0.696 to 0.817. Finally, all potential targets of each component were normalized by IPA and the kinase targets were selected. Kinase targets by predicted by KinomeX. KinomeX system (https:// kinome. dddc. ac. cn/ en/) 71 is a prediction and analysis platform of single compound regulated kinase spectrum. It enables users to predict its potential kinase targets based on the structure of a given molecule with the average 0.75 area under the receiver operating characteristic curve (auROC), which is significantly higher than other prediction methods [72][73][74][75][76][77][78] . Therefore, we used the KinomeX to predict the potential protein kinase targets of 40 important components in CDDP.
Potential direct kinase targets of CDDP. To obtain the kinase target set of CDDP with high reliability, the prediction results from KinomeX were used to screen the targets obtained from public databases and Multivoting SEA algorithm mentioned above. The screened kinase targets were regarded as potential targets of CDDP and subsequently to conduct following experimental verification.  ). Firstly, the filter-binding radioactive kinase activity assays were performed at a concentration of 25 µg/mL of CDDP. The kinase activity inhibition rate of the sample was expressed as the percentage of the result of sample compared to the blank group. The kinase activity of the blank was 100%. Generally speaking, if the residual enzyme activity is less than 30%, it is considered to be strongly inhibited. And if the residual enzyme activity is between 30 and 70%, it is considered as moderate inhibition. Considering the weak interaction superposition characteristic and synergistic effect of TCM ingredients 79,80 , the threshold value in this study was defined as 80%. In order to get the dose-dependent kinase targets, the kinase targets with activity value less than 70 were retested at a concentration of 250 µg/mL of CDDP.

Experimental validation in a high throughput way by
Kinase assays for AURKB, MET, PIM1, SYK Kinase analysis. To further obtain a mean IC 50 value and its standard deviation, we chose four targets showing obvious inhibitory action at the concentration of 250 µg/mL to carry out the kinase assay. Pharmaron (Beijing) was commissioned to perform in vitro kinase assays for AURKB, MET, PIM1, SYK. The detailed information about the assays, such as the reagents, instruments, assay procedure, data analysis, and calculation of IC 50 , for AURKB, MET, PIM1, SYK can be referred in the attachment (see Supplementary Table S1). Ten concentration points were obtained by 3 dilution fold.
Cell experiments in vitro for AURKB, MET, PIM1, SYK in four cell lines. Cell lines and treatments. The human breast cancer cell lines MCF7, T47D and thyroid cancer cell lines BCPAP, TPC1 used in this study were purchased from American Type Culture Collection (Manassas, VA, USA). These four cell lines were maintained in Dulbecco's modified Eagle medium (1640) (HyClone, UT, USA) supplemented with 10% fetal bovine serum (FBS) (Gibco, Gaithersburg, MD). All of them were placed in a 5% CO 2 and humidified atmosphere at 37 °C. For treatments, each cell lines were divided into four groups as follows: Control group, CDDP group (0.25 mg/mL), CDDP group (0.5 mg/mL), and CDDP group (1.0 mg/mL). Cells at a density of 2 × 10 5 cells/well in 6-well plates were treated with CDDP according to the groupings. The treatment time was 12 h for MCF7, T47D and 6 h for TPC1, BCPAP.

Data availability
All data generated or analyzed during this study are included in this published article (and its Supplementary Information). www.nature.com/scientificreports/