Abstract

   Identifying the molecular targets for the beneficial effects of active
   small-molecule compounds simultaneously is an important and currently
   unmet challenge. In this study, we firstly proposed network analysis by
   integrating data from network pharmacology and metabolomics to identify
   targets of active components in sini decoction (SND) simultaneously
   against heart failure. To begin with, 48 potential active components in
   SND against heart failure were predicted by serum pharmacochemistry,
   text mining and similarity match. Then, we employed network
   pharmacology including text mining and molecular docking to identify
   the potential targets of these components. The key enriched processes,
   pathways and related diseases of these target proteins were analyzed by
   STRING database. At last, network analysis was conducted to identify
   most possible targets of components in SND. Among the 25 targets
   predicted by network analysis, tumor necrosis factor α (TNF-α) was
   firstly experimentally validated in molecular and cellular level.
   Results indicated that hypaconitine, mesaconitine, higenamine and
   quercetin in SND can directly bind to TNF-α, reduce the TNF-α-mediated
   cytotoxicity on L929 cells and exert anti-myocardial cell apoptosis
   effects. We envisage that network analysis will also be useful in
   target identification of a bioactive compound.
     __________________________________________________________________

   Bioactive compounds exert their biological activities through direct
   physical binding to one or more cellular proteins[36]^1. The detection
   of drug-target interactions is therefore necessary for the
   characterization of compound mechanism of action[37]^2. There are two
   fundamentally different approaches to identify molecular targets of
   bioactive molecules: direct and indirect[38]^3. The direct approach
   utilizes affinity chromatography often with compound-immobilized beads.
   Many compounds cannot be modified without loss of binding specificity
   or affinity[39]^4. Moreover, because of above characteristics, this
   approach is only suitable to identify targets of one drug once and
   cannot afford target identification of many compounds simultaneously,
   such as active components in herbs. With the indirect approach, such as
   system biology approaches, including proteomics, transcriptomics and
   metabolomics, are the major tools for target identification and have an
   unbiased attitude towards all active compounds[40]^5. A proteomic or
   transcriptomics approach for identification of binding proteins for a
   given small molecule or compounds in herbs involves comparison of the
   protein expression profiles for a given cell or tissue in the presence
   or absence of the given molecule(s). These two methods have been proved
   successful in target identification of both many compounds and one
   drug[41]^6,[42]^7,[43]^8,[44]^9. Whereas metabolomics has been mainly
   developed to identify drug(s)-affected pathways[45]^10,[46]^11, the
   “readout”, such as proteins in the pathway, is often far downstream
   from the drug targets. Therefore using metabolomics for target
   identification run into the bottleneck.

   As bioactive molecules exert their effects through direct physical
   association with one or more cellular proteins[47]^1, these target
   proteins will then act on related proteins, above proteins eventually
   affect the content of related metabolites. With the advent of the era
   of big data, now there are large amounts of data about known and
   predicted protein interactions[48]^12. Once we use network pharmacology
   to predict potential targets of active components in Traditional
   Chinese Medicine (TCM) formula[49]^13, a component-target
   protein-related protein-metabolite network can be constructed with the
   combination of network pharmacology and metabolomics. As a combination
   of approaches is most likely to bear fruit, the combination of network
   pharmacology and metabolomics called network analysis could increase
   the degree of accuracy of target identification of network
   pharmacology. In addition, metabolomics and network pharmacology
   employed global profiling methods for the comprehensive analysis of
   altered metabolites and target proteins, providing insights into the
   global state of entire organisms, which are well coincident with the
   integrity and systemic feature of TCM formula. Thus apart from target
   identification of a bioactive compound, this network analysis method is
   more beneficial in identifying unknown targets of active compounds in
   TCM formula simultaneously in an unbiased fashion.

   Here, we introduce a new, potentially widely applicable and accurate
   drug target identification strategy based on network analysis to
   identify the interactions of active components in TCM formula and
   target proteins. Our previous studies have confirmed that SND, composed
   of three medicinal plants: Aconitum carmichaelii, Zingiber officinale
   and Glycyrrhiza uralensis, can treat heart failure[50]^14. Metabolomics
   researches have also been conducted to demonstrate its
   effectiveness[51]^14,[52]^15. Chemome[53]^16, serum
   pharmacochemistry[54]^16 and xenobiotic metabolome[55]^17 of SND were
   also characterized. Thus in this study, we took SND as an example to
   test the potential of network analysis in target identification. Active
   components in SND against heart failure were identified by serum
   pharmacochemistry, text mining and similarity match. Their potential
   targets were then identified by network analysis. At last, the most
   possible target was validated experimentally to demonstrate the
   potential of network analysis. Above results will be helpful to
   investigate the action mechanisms of SND and promote the development of
   Chinese Drug modernization. More importantly, network analysis will not
   only conferred a unique advantage to identify targets of active
   compounds in TCM formula simultaneously, but also provided a new method
   for the target identification of a bioactive compound. Detailed
   procedures can be seen in [56]Fig. 1.

Figure 1. The flowchart of network analysis approach.

   [57]Figure 1
   [58]Open in a new tab

Results

The rationality of components in SND in absorption and metabolism

   Results considering the known metabolism of components in SND have been
   concluded in [59]Supplementary Table S1. Total flavones (H) and total
   saponins (Z) were major active components in Glycyrrhiza uralensis.
   From [60]Table S1, we can conclude that many flavones and saponins in
   Glycyrrhiza uralensis were known CYP450 inhibitor, while alkaloids in
   Aconitum carmichaelii are mostly not. Conclusions can be made that
   Glycyrrhiza uralensis can inhibit the metabolism of alkaloids and
   improve their bioavailability. Researchers also demonstrated that
   Glycyrrhiza uralensis can improve bioavailability of diester
   diterpenoid alkaloids in Aconitum carmichaelli, which coincided with
   results above. And researchers found that Aconitum carmichaelli can
   also improve the bioavailability of glycyrrhizic acid which is a major
   component in Glycyrrhiza uralensis[61]^18, so we can conclude that the
   combination of Aconitum carmichaelli and Glycyrrhiza uralensis can
   enhance efficacy of each medicinal materials. In addition, Zingiber
   officinale could promote the elimination of diester diterpenoid
   alkaloids and enhance the absorption of monoester diterpenoid
   alkaloids[62]^19. As diester alkaloids are the chief toxic components
   in Aconitum carmichaelli. The results might be helpful in explaining
   the mechanism of combination of Aconitum carmichaelli− Zingiber
   officinale to decrease toxicity and increase efficacy. Researchers also
   proved that compared with Aconitum Carmichaeli, the bioavailability of
   three monoester-diterpenoid alkaloids increased in SND[63]^20. And
   compared with Aconitum Carmichaeli, the bioavailability of
   hypaconitine,i.e. diester diterpenoid alkaloids decreased in SND. The
   SND formula can decrease toxicity and increase efficacy. The
   information above demonstrated the rationality of components of SND in
   Absorption and metabolism.

Compound families and chemical space properties of active components in SND

   To give an overview of the compound families in SND, chemical
   clustering was conducted ([64]Supplementary Fig. S1A). The areas of
   overlap may show that active components in SND and anti- heart failure
   drugs are physicochemical property similar, resulting in similar
   pharmacological actions against heart failure[65]^21. In addition, the
   active components are also clustered in three independent areas (region
   B, C and D), indicating that the active components may be structurally
   or pharmacologically different in three herbs.

   The physicochemical characteristics of a compound are important for its
   drug likeness. Comparing the physicochemical characteristics of active
   components in SND with FDA-approved oral drugs will provide insight
   into the drug likeness of these components. Here, seven physicochemical
   characteristics of active components in SND were compared with approved
   orally administered drugs ([66]Supplementary Fig. S1B–H). The overall
   shapes of the distributions of these characteristics are similar
   between active components in SND and approved oral drugs, which
   indicates that many ingredients in herbs have drug potential. The
   proportion of compounds with more than 10 rotatable bonds in SND is
   more than in approved drugs ([67]Supplementary Fig. S1G), which means
   the structures of ingredients in SND are more flexible. There are
   statistically significant differences between drug and herb of all
   variables in the aspect of distribution by Kolmogorov-Smirnov test in
   [68]Supplementary Table S2. As all variables in both drug and herb do
   not follow normal distributions, we conducted wilcoxon test to evaluate
   the difference of all variables in drug and herb. The results showed
   that there are significant differences between drug and herb of all
   variables in the aspect of median except for Polar Surface Area in
   [69]Supplementary Table S2. To make a conclusion, physiological
   characteristics of active components in SND are special while compared
   with approved oral drugs in the median and distribution.

Prediction analysis of pharmacological mechanism based on network
pharmacology

   We constructed the component-target network ([70]Fig. 2A) based on text
   mining and docking. This network had 109 nodes and 556 edges, in which
   red circles and hexagons correspond to active components and target
   proteins, respectively. Many targets in the middle of [71]Fig. 2A are
   targeted by components in three medicinal herbs, which meant that these
   targets are main potential targets. According to the data from CHEMBL,
   BindingDB and PubMed database, 13 out of 61 potential targets were
   validated to be exact targets of active components in SND
   ([72]Supplementary Table S3), which proved the reliability of molecular
   docking and text mining.

Figure 2.

   [73]Figure 2
   [74]Open in a new tab

   (A) The Component-Target network. The red circles represent the 48
   active components in SND, S, J, H and Z refers to alkaloids, gingerols,
   flavones and saponins in Aconitum carmichaelii, Zingiber officinale and
   Glycyrrhiza uralensis. The blue hexagons represent the gene names of
   targets of the three herbs found by text mining, while the green
   hexagons are the targets found by dock. The yellow hexagons represent
   the gene names of targets found both by text mining and dock. Targets
   in the center of network represent the common targets of three herbs,
   and targets in the curve of S, J or H and Z represent the targets of
   each kind of active components respectively. (B) The enrichment
   analysis in biological processes, cellular components and molecular
   functions of 61 identified target proteins by STRING database.

   The STRING database (version 10.0) ([75]http://string-db.org/) was used
   to elucidate biological processes, Cellular components, molecular
   functions and pathways of target proteins. And we only choose
   meaningful pathways, biological processes, Cellular components and
   molecular functions with a p value < 0.05 as key pathways, processes,
   components and functions. Functional classification of target proteins
   is detailed in [76]Fig. 2B. Molecular function of the target proteins
   is classified to two categories: binding and receptor activity
   ([77]Fig. 2B). The binding activities that appeared are mainly
   associated with receptor binding and enzyme binding. And the receptor
   activity is adrenergic receptor activity. According to the
   classification of cellular component, the proteins are located in
   cytoplasmic part, extracellular region and membrane region. The
   biological processes that the target proteins are involved can be
   summarized in [78]Fig. 2B. The results firstly demonstrated that SND
   exerted its protective effects by regulation of blood
   circulation[79]^22, response to oxidative stress[80]^23,[81]^24,
   regulation of apoptotic process[82]^25 and inflammatory
   response[83]^26, which coincided with previous researches. In addition,
   the results also indicated that active components in SND could also
   exert anti- heart failure effect through regulation of blood pressure,
   regulation of vasodilation, regulation of muscle contraction,
   regulation of heart contraction, blood coagulation and regulation of
   angiogenesis. Although large amounts of references showed that the