The Strategy of Studying the Herb-Pair Inter-Reinforcement Mechanism Mapping on Chemical Space Between Sparganii Rhizoma and Curcumae Rhizoma

With the ever-increasing acceptance of combination therapy and the use of traditional Chinese medicine (TCM) has become an emerging trend. The multi-herb formulae therapy is one of the most important characteristics of TCM, but the modernization drive of this conventional wisdom has faced many obstacles due to its unimaginable complexity. Herb pairs, the most fundamental and the simplest form of multi-herb formulae, are a centralized representative of Chinese herbal compatibility. A systematic search of herb pair related research was carried out using multiple online literature databases, books and monographs published in the past 20 years. The herb-pair has an important characteristics---inter-reinforcement (Xiangxu in TCM terms) in TCM basic theory. The inter-reinforcement mechanisms of herb pairs are extremely complicated and their exact molecular mechanisms are also still not well elucidated. This research integrated the to the inter-reinforcement mechanisms in with (Sparganii at the inter-reinforcement

herb-pair inter-reinforcement mechanism mapping on chemical space is a certain scienti c and rational method to illustrate material basis of Chinese herbal medicine at multi-dimension.

Background
Chinese herbal medicine (CHM) is not only medical treasure trove of traditional Chinese medicine (TCM), but also the vastness of natural and biologically relevant chemical space [1,2]. Chemical space is a term often used for 'multi-dimensional descriptor space' [1], geographical map to illustrate the distribution of molecules and their properties [3][4][5]. Additionally, traditional Chinese medicine (TCM) has been practiced to treat disease from ancient time and it has been recognized as an interesting alternative to conventional medicine [6,7]. Long-term applications and their experiences have demonstrated that herb compatibility is not only a better choice, but also a necessary choice [8]. Herb-pair is the most basic form of herb compatibility in traditional Chinese medicine and is the ancient medical experts of long-term medical practice summary. Herb-pairs are a combination of two relatively xed herbs, which have been proven effective in clinical practice. The herb-pair has their inherent prescriptions and it is a bridge between single herb and prescriptions. Traditional Chinese medicine (TCM)-herbal mixtures developed though observation and experience accumulated over thousands of years, but with unknown mechanisms of action [9].
In this context, we take herb-pair Sparganii Rhizoma (tuber of Spargranium stloniferum Buch.-Ham.) and Curcumas Rhizoma (rhizome of Curcuma phaeocaulis Val.) (SR-CR) as our research object to study the herb-pair inter-reinforcement (Xiang Xu in Chinese ancient medical term) mechanism from holistic perspective in chemical space. The rein-reinforcement is the combination of herbs with similar property and e cacy which can signi cantly reinforce some kind or several kinds of therapeutic effects [10]. The herb-pair Sparganii Rhizoma and Curcumae Rhizoma (SR-CR) has been increasingly recognized to possess multiple therapeutic activities for treatment of tumor [11][12][13], thrombosis [14], in ammation [15], blood stasis [16,17], amenorrhea, functional dyspepsia [16]. The details of SR-CR main chemical ingredients see the supplementary Table S1.
In a Chinese herbal mixture, the paired-herb is referring to a combined usage of two herbs as to optimize the pharmacological property [18]. Mutual enhancement of herb-pair can be achieved by using pair of herbs with ingredients of similar therapeutic actions, or by using a principal herb with main therapeutic actions and another one acting to enhance or assist the effects of the principle one [19]. Many ingredients of herbs are inactive individually but become active in combination, called coalistic combinations, which are quite common in herb pairs [20,21]. However, the exact molecular mechanisms of these TCM herb- pairs are yet unclear [19]. The herb-pair contains many combinations among the chemical space. Though there are many reports on pharmaceutics, pharmacokinetics, and pharmacodynamics of TCM, there remains a serious lack of summarization and systemic analyses of these reported data to help uncover the compatibility rationale of TCM herb pairs based on their chemical spaces [22]. Chemical spacewhich encompasses all possible small organic molecules, including those present in biological systemsis vast [1,23]. The vastness of chemical space will guide us to discover biologically active compounds [2], natural products [24], and the synthesis of new ligands [25,26]. Although herb-pair inter-reinforcement is not entirely clear, the explorations of herb-pair chemical space will greatly enhance our understanding of herb-pair inter-reinforcement mechanism in biology, and lead to new strategies to treat disease [27,28].

Date collection and visualizationDate collection and digitalization of compounds
We collected compounds of Sparganii Rhizoma and Curcumae Rhizoma from four typical TCM databases, namely TCMID [53], TCMSP [36], TCM Database@Taiwan [54], BANMAN-TCM [55]. All of these compounds collected were normalized to the canonical SMILES format [56]. Duplicates and compounds without structures were excluded. Data set S was collected from Sparganii Rhizoma chemical space and contained 47 molecules. Similarly, data set E was from Curcumae Rhizoma chemical space and contained 91 molecules. Chemical space was used to discuss the similarities and differences between the molecule sets of Sparganii Rhizoma (data set S) and Curcumae Rhizoma (data set E). The data set S and E can be mapped to two-dimensional space while maintaining a lot of geometric structure. A total of 47 chemical components of Sparganii Rhizoma were obtained from the databases and were numbered from S001 to S047 (SupplementaryTable S1). The 91 chemical components of Curcumae Rhizoma were numbered from E001 to E093, and we excluded the no clear structure chemical E071, E075 in this research (SupplementaryTable S1).
All of the related targets were normalized to the o cial gene symbols using the UniProt database (http://www.uniprot.org/uploadlists/). Duplicates from different sources were excluded. ngerprints for each of the data sets. The MDL 166-key ngerprint is also known as the "Molecular Access System (MACCS)" key in the literature and is a no hashed ngerprint consisting of 166 bits. The MDL 166-key ngerprint is one of the very few available that offers a 1-1 mapping [57]. These ngerprints were assembled into the ngerprint array.
The data sets used to evaluate the models' ability to cluster compounds by molecular structure were given in Table 1. The PaDEL-Descriptor [58] was used to generate a set of MDL 166-key ngerprints for each of the data sets. The MDL 166-key ngerprint is also known as the "Molecular Access System (MACCS)" key in the literature and is a no hashed ngerprint consisting of 166 bits. The MDL 166-key ngerprint is one of the very few available that offers a 1-1 mapping [57]. These ngerprints were assembled into the ngerprint array. The diversities were computed using Tanimoto coe cient [57,59], which is a statistic used for comparing the similarity and diversity of sample sets. Tanimoto coe cient is also known as the Jaccard coe cient, and, when used to measure dissimilarity rather than similarity [60]. Turning then to the similarity coe cient that is used for comparing ngerprints, the most popular is the Tanimoto coe cient (Equation 1). If two molecules have a and b bits set in their fragment bit-strings, with c of these bits being set in both of the ngerprints, then the Tanimoto coe cient is de ned to be: The Tanimoto coe cient gives values in the range of zero (no bits in common) to unity (all bits the same); and is usually used for Boolean vector, that is, when the components are only 0 or 1. In our research, the compounds of herb pair were normalized to MDL 166-key ngerprints.
Visualization for herb pair's chemical space Generative Topographic Mapping (GTM) is nonlinear and works by training an Radial Basis Function (RBF) neural network to produce a mapping from two-dimensional latent (i.e., visualization) space to ndimensional data (i.e. ngerprint) space [57]. Essentially, GTM models was used to reduce the dimensionality of the ngerprint array from 166-dimensional space to two-dimensional space where it could be plotted and visualized (Fig. 2). It has reported that greater the structural similarity, the greater the biological similarity [61]. As can be seen from Fig. 2, the chemical compounds in the chemical space of SR-RC could be divided into two groups. Nevertheless, there are many compounds with similar structure (Fig. 2).
The work ow of studying the herb-pair inter-reinforcement mechanism We calculated the human intestinal absorption (HIA properties of the compounds from Sparganii Rhizoma and Curcumae Rhizoma using a model integrated within Pipeline Pilot (version 7.5.2, BIOVIA, San Diego, CA, USA) [62] to rank the molecules into the following four classes: good (0), moderate (1), poor (2) and very poor (3). The very poor class of 10 compounds was discarded. The remaining compounds were further analyzed (for further details of four classi cations for these compounds, see Supplementary Table S1).
Subsequently, we collected experimentally determined compound-target pairs from four databases, namely STITCH (version 5.0) [63], BindingDB [64], PubChem [65] and DrugBank [66,67]. No target records among these four typical TCM databases of these compounds were predicted using BATMAN-TCM bioinformatics tool [55] and chemical-protein interactions (CPI) predictor [68]. For each compound, the score above 80 predicted targets were stored as putative targets when using the bioinformatics tool-BATMAN-TCM; the probability above 0.6 predicted targets were stored as putative targets when using CPI predictor [68]. The calculating score method of the BATMAN-TCM is to rank potential drug-target interactions based on their similarity to the known drugtarget interactions [69]. On this basis, the Compound-Target Network were constructed, indicating the interaction between compounds and targets.
We will explain the inter-reinforcement mechanism for herb pair of from the pathway level and the disease level. On this basis, the target-pathway network and target-disease network on the basis of Compound-Target network were constructed. In order to study the contribution of common compounds and non-common compounds for the herb-pair inter-reinforcement in the herb pair, we examined compound-target-pathway and compound-target-disease with common compounds and the noncommon compounds. Finally, the CTDIs and TPAs were constructed with the common/non-common compounds to comparative analysis the inter-reinforcement mechanism of the herb pair (Fig. 3).

Results
Networks can be used to view global drug-targets-pathway-disease relationships between nodes [29,30].
Network pharmacology approach have been applied in drug discovery [31][32][33][34][35]. In this research, we utilized the network pharmacology method to illustrate SR-CR inter-reinforcement mechanism by mapping compound combinations into SR-CR chemical space.
Target-Pathway association pro le of Common compounds for the herb pair The herb-pair shared 2 common compounds and they have 24 targets from four public databases. We obtained 24 targets associated 15 KEGG pathways and 22 REACTOME pathways in total and the interreinforcement mechanisms of herb-pair were mapped to the target-pathway network according to their targets information (Fig. 4, see Supplementary Table S2 for further details of these pathways). The degree of PIK3CG among the 24 targets is the largest one. We also obtained 27 targets associated 87 diseases in total and an average degree of 14 targets per compound (Fig. 5a, see Supplementary Table   S3 for further details of these 87 diseases). The two common compounds were connected to multiple targets and they have 6 common targets, namely PTGS2, PGR, NCOA2, ADH1C, PTGS1, LYZ. The 6 targets associated 35 diseases and the maximum degree of 32 diseases is PTGS2 in Fig. 5b. (See SupplementaryTable S4 for further details of these 35 diseases)

Compound-Target-Pathway Associations (CTPAs) network
To understand the complex interaction of herb-pair's targets at system level, we constructed a Compound-Target-Pathway Associations (CTPAs) network. The CTPAs network embodied 327 nodes (100 compounds, 104 targets, and 128 pathways) (Fig. 6, see Supplementary Table S5 for further details of these pathways associated targets). The mean degree value (the number of associated common targets) of candidate compounds was 1.08, and 24 compounds had a degree value >10, indicating that most of the compounds regulated multiple targets to exert various therapeutic effects in the herb-pair's chemical space. The top 5 degree of Compound-Target-Pathway network were Neuroactive ligand-receptor interaction, Morphine addiction, Retrograde endocannabinoid signaling, Calcium signaling pathway, Nicotine addiction.

Compound-Target-Disease interactions (CTDIs) network
The Compound-Target-Disease interactions (CTDIs) network was built by targets and diseases [37][38][39][40][41]. Node size was proportional to its degree. The light green (green) ellipses indicated Sparganii Rhizoma (Curcumae Rhizoma) compounds, the pink ellipses indicates the targets of herb-pair's compounds, and the cyan ellipses indicated the disease types (Fig. 7, see SupplementaryTable S6 for further details of these diseases). The involved diseases were type 2 diabetes, hypercholesterolemia, hypertension, cardiovascular diseases, etc.
All of these targets were shown by the gene name. All of the targets and diseases were selected to construct the Compound-Target-Disease interactions (CTDIs) network by linking the targets to the corresponding diseases. All visualized network graphs were constructed by Cytoscape 3.2.1 (http://www.cytoscape.org/), an open software package project for visualizing, integrating, modeling and analyzing the interaction networks [42,43].

Target-Pathway Association Interactions (TPAs)
We analyzed Target-Pathway Association Interactions (TPAs) and screened the combinatorial compounds which their targets were in the same pathway [21]. Then we combined the two compounds which one from Sparganii Rhizoma and the other from Curcumae Rhizoma. Finally, 14370 pairs of pathway-guided combination compounds, including 1267 pairs of compound combinations (Supplementary Table S7) were obtained. After setting compounds similarity to 0.7 in each compound combinations (Fig. 8), 497 pairs of compound combinations associated 128 pathways (SupplementaryTable S8) were obtained.

Pathway guided compound combinations
The pathway of platelet activation is one of 128 pathways, and platelets play a key role in preventing blood loss in response to injury, thrombosis, atherosclerosis, in ammation and metastasis, but they are also responsible for the formation of pathogenic thrombi that cause acute manifestations of vascular atherothrombotic disease [44][45][46]. In this research, four compound combinations and their associated targets in platelet activation pathway (Fig 9.) were obtained. The compound combination (S039, E063) acted on Akt pathway, and the compound combinations ((S039, E085), (S039, E074), (S039, E005)) targeted on the PTGS1 (the numbered chemical ingredients see the supplementary Table S1). The mechanism of compound combinations is alike drug combinations. As we knew, drug combinations can improve over single therapeutic agents in two ways [47]. In additionally, drug combinations have been used for treating diseases and reducing suffering in Chinese herbal medicine [48].
Synergy between two drugs may result in a better response than the two drugs independently [47]. In this research, we picked out another two pathways (Calcium signaling pathway, vascular smooth muscle contraction) closely related to cardiovascular diseases (CVDs) and the 3 pathways were consolidated into a "CVDs pathway" under the pathological and clinical data [49]. In order to illustrate the material basis of herb-pair inter-reinforcement mechanism, we mapped compound combinations to pathways associated compounds' targets (supplementary Table S9).
Across all kingdoms in the tree of life, Calcium (Ca2 + ) is an essential element used by cells to respond and adapt to constantly changing environments [50]. Signaling occurs when the cell is stimulated to release calcium ions (Ca2 + ) from intracellular stores, and/or when calcium enters the cell through plasma could induce the development of atherosclerosis [49].
This strategy of studying herb-pair inter-reinforcement membrane ion channels [51]. Many diseases, such as cardiovascular diseases, type 2 diabetes, malignant hyperthermia, central core disease, associated with calcium signaling pathway [50,52]. Vascular smooth muscle contraction also plays an important role in the regulation of CVDs, there is due mechanism mapping on chemical space will lead to the development of many today's drug discovery from Chinese herb-pairs. Future research in our group will focus on investigating compound combinations' pathological effect on speci c diseases, such as, CVDs.

Discussion
This research proposed to explore the inter-reinforcement mechanism of Sparganii Rhizoma and Curcumae Rhizoma based on the chemical space for the rst time. We adopted the chemoinformatics methods to screen the chemical compounds that can be absorbed and utilized by the human body from the chemical space of Sparganii Rhizoma and Curcumae Rhizoma. And then, the network of Compound-Target-Disease interactions (CTDIs) network, Compound-Target-Disease interactions (CTDIs) network, Target-Pathway Association Interactions (TPAs), Pathway guided compound combinations were constructed with network pharmacology method.
It explains inter-reinforcement mechanism of Sparganii Rhizoma and Curcumae Rhizoma with the combination compounds in cardiovascular disease-related pathways. Each group of combination compounds was derived from Sparganii Rhizoma and Curcumae Rhizoma. This strategy could provide research strategies for further research on the mechanism of interaction of TCM pairs. However, this approach to screen out combination compounds on speci c disease-related pathways required further animal or cell experiments to con rm its combined effect.

Conclusions
This strategy will provide methodology to study the herb-pairs inter-reinforcement mechanism on compound combinations level. In this context, several compound combinations with potential e cacy to treat diseases were discovered, especially cardiovascular disease.
The compound combinations with the common compounds (hederagenin and alexandrin) plays an important role in the cardiovascular diseases (CVDs) pathways from Supplementary Herb-pair mapping into chemical space to explain the inter -reinforcement mechanism at molecular level. The work ow of studying the herb-pair inter-reinforcement mechanism mapping on chemical space to guide compound combination.

Figure 4
The two common compound (alexandrive and hederagenin) of target-pathway association pro le from KEGG database. A target node and a pathway node were linked if the target was associated with the pathway.

Figure 5
The two common compound (alexandrive hederagenin) of Compound-Target-Disease Interaction (CTDIs) network. (a) contains 28 targets of the two compound, (b) exacting the 6 common target from the 28 target to construct target node and a disease node were linked if the target was associated with the disease.

Figure 7
The non-common compound of target-disease association pro les. A targets node and a node were linked if the target was associated with the disease.