Anti-inammatory Properties of Fangji Huangqi Tang: Discovery Based on Network Pharmacology, Molecular Docking and Arrowsmith Tools

Fangji Huangqi Tang (FHT) is a classical formula widely used in Chinese clinical practice. In this study, a creative application of FHT for inammation has been identied by network pharmacology-based framework. Specically, a total of 17 bioactive compounds including 42 potential targets of FHT, and 205 related targets involved in inammation were retrieved from mainstream databases and subjected to network analysis. 13 intersection targets indicated the principal elements linked to inammation therapy. Top terms of Gene Ontology (GO) analysis were identied, while 7 related signaling pathways were revealed by Kyoto Encyclopedia of Genes and Genomes (KEGG) results. Subsequently, Calycosin-PTGS2 with tight binding anity (BA) was manifested by molecular docking as the critical compound-target couple. Meaningful links between Calycosin and inammation were implied through Arrowsmith, which overlapped with the ndings in enrichment analysis, such as MAPK, NF-κB that could be regulators of PTGS2. In summary, the present study explored the potential targets and signaling pathways of FHT against inammation, which may help to illustrate the mechanisms responsible for the action of FHT and provide a better understanding of its anti-inammatory effects.

Semen (Yiyiren, YYR) [1][2][3] . As a representative prescription for damp clearing of Traditional Chinese medicine (TCM), it possesses the features of Yiqi Qufeng, Lishui and Jianpi, and is usually used for the curation of chronic nephritis, rheumatoid arthritis and osteoarthritis, etc [4][5][6] . Thus, we speculated that FHT may play roles as an anti-in ammatory formula with broad spectrum.
In current times, researchers did lots of work on the modernization of FHT. Wang et al. developed and validated an ultrahigh performance liquid chromatography method coupled with triple quadrupole mass spectrometry (UHPLC-QqQ-MS) for simultaneous determination of bioactive components in FHT [7][8][9] . Chen et al. found that FHT could improve the immunity of mice 10 . The results of these studies laid a good foundation for the application of modern science and technology to explain the rationality and compatibility theory of FHT. But, the depth of these studies was limited to the advanced molecular pharmacological methods, while the theoretical research on effective material has not been really carried out.
Network pharmacology is intensively applied in drug research because of its holistic, comprehensive and e cient characteristics for the systematic study of the internal relationship among 'drugs-targetspathways-diseases' using TCM databases, bioinformatics tools and molecular biology concepts [11][12][13] .
TCM-oriented network pharmacology sheds light on revealing the mechanisms of TCM with multiple compounds through pharmacokinetic evaluation, pathway analysis as well as target prediction, becoming an indispensable method for development of TCM [14][15][16][17] . In many previous studies, network pharmacology has successfully unveiled potential targets and pathways of TCM [18][19][20] . Meanwhile, molecular docking simulation is frequently employed to further validate the potential bioactive components and core targets [21][22][23] .
Non-interactive literature-based discovery (NILBD) with Arrowsmith is a practical tactic performing two PubMed searches to retrieve biomedical articles from the Medline database and generating a common list of words and phrases found in the titles of both literatures, which could be utilized to deeply discover the commonness between the subjects that have never been studied [24][25][26] . The results should be integrated and cross veri cated with network pharmacology ndings, making the conclusion much more convincing.
In general, this study aims to provide an insight into decoding the possible bioactive compounds and potential mechanism of FHT in in ammation treatment via network pharmacology and Arrowsmith interpretation. Work ow of this work is shown in detail in Figure 1.

Functional annotation of FHT targets
To investigate the potential pharmacological mechanism of FHT, we analyzed GO term and KEGG pathway by DAVID. The top 10 signi cant GOBP, GOCC and GOMF terms (P ≤ 0.05) based on counts were listed in Figure 3A-3C, respectively. The BPs of FHT targets may relate to 'positive regulation of transcription from RNA polymerase II promoter', 'transcription, DNA-templated', 'signal transduction', and so on. GOCC analysis showed that 'nucleus', 'plasma membrane' and 'cytoplasm' accounted for the top 3 proportion (28, 26, and 19 targets, respectively). Besides, 'protein binding', 'enzyme binding' and 'ATP binding' were mainly enriched terms indicated by GOMF. In addition, 34 KEGG pathways were recognized as P < 0.05. Top 10 enrichment results were shown in Figure 3D, delineating the critical KEGG pathways (including Cancer, Calcium, and cAMP signaling pathways, etc.).
T-D network construction and Common targets discovery 205 targets related to in ammation were gained through a double screening of TTD and BATMAN-TCM. Based on previous strategy, STRING database was employed to reveal the interactions among disease targets ( Figure 4A). The T-D network of in ammation with 205 nodes and 2993 edges was visualized by Cytoscape 3.6.0 ( Figure 4B). IL-6, TNF, IL-10, ALB and TLR4 are top 5 predominant targets with maximum degree in network. At last, on account of 42 putative targets of FHT and 205 targets related to in ammation, 13 common targets were screened out by intersection. Active compounds associated with selected overlapping targets are listed in Table 2.

Functional enrichment analysis of common targets
To further validate whether the candidate targets as mentioned above was correlated with FHT against in ammation, functional annotation was performed via DAVID. Interestingly, the results were similar to that of FHT targets, indicating the main function of FHT was to treat in ammation. Ultimately, 29 BPs, 7 CCs and 25 MFs enriched for these targets had a P value less than 0.05. The representative results were shown in Figure 5A-5C. Meanwhile, in terms of KEGG, a total of 7 pathways ( Figure 5D) were screened, including 'Pathways in cancer', 'Leishmaniasis', 'Prolactin signaling pathway' and 'Retrograde endocannabinoid signaling' when corrected by a P value ≤ 0.01.

Major bioactive compound-target and mechanism speculation
To explore the potential bioactive components and targets of FHT more accurately, a compound-targetfunction module was built by combing the networks above. Based on topology to nd densely connected protein complexes, MCODE divided C-T network into 4 function modules ( Figure 6). The cluster with highest MCODE Score (10.471) had 18 nodes and 89 edges, verifying PTGS2, TOP2A, NR3C1 as main targets, while Calycosin as a pivotal compound.
Molecular docking technology was applied to verify the a nity between Calycosin and the crystal structure of PTGS2 which was recognized as the most promising pair of compound-target. The obtained docking result indicates the receptor-ligand interaction. According to Figure 7, Calycosin small molecule forms hydrogen bonds with GLU524 residue on PTGS2. The BA in this couple was -8.1 kJ/mol, which was less than that of Dihomo-gamma-linolenic acid used as a positive control (-6.2 kJ/mol), further proving the strong binding ability of Calycosin to PTGS2.
No published data concerning Calycosin and in ammation could be retrieved through PubMed. Thus, meaningful links were further implied by Arrowsmith through a two-step query, yielding a list containing terms that appeared in both Calycosin and in ammation literature. Finally, there were 105 terms on the current list (26 are predicted to be relevant with Probability ≥ 0.33), which was shown ranked according to predicted relevance (Table S1). There was a consistency to the ndings in enrichment analysis, such as MAPK, NF-κB involved in 'Pathways of cancer' and 'VEGF signaling pathway'.

Discussion
In ammation is a chronic and progressive pathogen or immune-mediated pathological process causing consecutive symptoms including pain, fever, fatigue and so on 27 . Despite a positive defense reaction, dysregulated in ammation could eventually cause disorders and conditions, such as rheumatoid arthritis, atherosclerosis, periodontitis, and even cancers 28 .
Widely applied therapeutic agents include antibiotics, non-steroidal anti-in ammatory drugs (NSAIDs) and glucocorticoids, which have achieved good therapeutic effects 29 . However, some side effects have resulted because of their extensive use 30 . Meanwhile, based on continuous breakthroughs in studies for compatibility of TCM, increasing attention has been paid because of its signi cant therapeutic effects and reduced side effects 31 .
A series of clinical practices demonstrated that FHT could be used to treat acute or chronic glomerulonephritis, rheumatoid arthritis and ankylosing spondylitis, relieving typical symptoms as well as suppressing in ammatory reaction 4 . By observing the effects of FHT on immune function of normal mice, it was found that the immunoregulatory activity of FHT was signi cant 10,32 .
Although researches on the treatment of in ammation with FHT continue, no systematic and comprehensive understanding of the relationships between targets and pathways involved has emerged.
Fortunately, network pharmacology, which is especially suitable for multi-compound and multi-target analysis, provides a prospective method to solve this problem with bioinformatics, molecular biology and databases [33][34][35] .
In current study, we predicted and analyzed the potential mechanisms of FHT via systematic network pharmacology tools. After preliminary screening based on databases in combination with ADME properties, 17 bioactive ingredients of FHT were disclosed. Thereafter, 43 candidate targets from 16 potential compounds were assembled for further analysis. Ranked by connectivity, 3'-Methoxyglabridin, Gancaonin B, Calycosin were the top 3 bioactive compounds with the effect of mitigating in ammations, while PTGS2, ESR1, NOS2 were the top 3 targets frequently connected with in ammation-related pathways, such as NF-κB and PI3K/AKT. These results suggested that FHT formula could probably regulate anti-in ammatory action and the immune system.
Subsequently, 13 common targets were screened out by intersecting from C-T and T-D networks, among which 8 genes were derivated from Calycosin. Functional enrichment analyses suggested these targets were closely related with in ammatory-and immune-related molecules or signaling pathway. For example, 'Pathways in cancer' and 'cAMP signaling pathway' precisely implied the extensive characteristics of in ammation and its potential harm of carcinogenesis. Overall, PPI and constructed networks indicated that the therapeutic effects of FHT on in ammation may be achieved through the synergistic and additive effects on multiple molecules and multiple pathways.
Finally, module analysis uncovered the most promising cluster. Calycosin-PTGS2 was recognized as the pivotal compound-target couple. Then, molecular docking technology was utilized to verify the interaction of PTGS2 with Calycosin. The results showed that the binding free energy of Calycosin with the core target was much less than -5 kJ/mol, which indicated that the ligand molecule could spontaneously bind to the receptor protein and had a strong binding force 36 . Finally, we put Calycosin and in ammation into Arrowsmith, generating a string of terms consisted of NF-κB, MAPK, etc., which was consistent with what we found in the enrichment analysis. Reportedly, several in ammatory stimuli have been shown to affect PTGS2 expression in a NF-κB or ERK dependent manner, which implies the underlying role of Calycosin-NF-κB/MAPK-PTGS2 axis in FHT treatment against in ammation 37,38 .

Conclusion
In general, our study was noteworthy that network pharmacology measures de nitely shed light on investigating the complicated action of FHT. Certain limitations also existed rstly due to lack of animal or in vitro experiments. Second, in light of the di culty in data acquisition, certain databases may not include all known or unknown targets and PPI. This situation could be improved in the future when more data were available.
Prediction of the candidate targets of bioactive compounds The selected compounds were appointed for targets exploration. Potential drug targets were identi ed through Traditional Chinese Medicine Systems Pharmacology (TCMSP) and Bioinformatics Analysis Tool for Molecular Mechanism of Traditional Chinese Medicine (BATMAN-TCM) by a similarity-based method. Thereafter, candidate targets in common were mapped to Uniprot (http://www.uniprot.org/) for annotation and normalization.

Prediction of targets for in ammation treatment
The targets related to in ammation were predicted by integrating multisource databases, including Therapeutic Target Database (TTD, http://db.idrblab.net/ttd/) and Online Mendelian Inheritance in Man (OMIM, http://www.omim.org/). These two databases illustrate the relationship between targets and disease from diverse perspectives. TTD is comprehensive and freely available, from which the user can obtain information in form of disease-target-drug on the known and explored therapeutic protein and nucleic acid targets and the corresponding drugs 41 . OMIM focuses on the relationship between phenotype and genotype on all known mendelian disorders and over 15,000 genes 42 . By searching for items with su x 'itis' or containing 'in ammation'/'in ammatory' among the downloaded le 'Targets to disease mapping with ICD identi ers' or under Gene Map, the targets were retrieved and integrated via TTD and OMIM, respectively.

Protein-Protein Interaction (PPI) data collection
Search Tool for the Retrieval of Interacting Genes (STRING, https://string-db.org/) was employed to collect possible PPI by uploading putative targets that related to in ammation and active compounds, respectively 43 . Species was restricted to 'Homo sapiens' with a con dence score > 0.90. The interactions among nodes with layout degrees could be inferred for further network construction.

Network construction and analysis
Two kinds of networks in this study were established using Cytoscape 3.6.0 software: compound-target network (C-T network) and target-disease network (T-D network) 44 . C-T network was composed of bioactive compounds and their potential targets, which was built to reveal the drug-target interactions. T-D network was built based on the potential targets and their related diseases. In the networks, the nodes represented compounds and targets, while the edges stood for the interactions between two nodes.
Furthermore, the signi cance of each node in the networks was assessed via one crucial topological parameter 'degree', which was de ned as the total of edges associated with a node. The protein complexes present as clusters were analyzed via plugin MCODE 1.6 of Cytoscape 3.6.0.

Functional Enrichment Analyses
GO terms of biological process (GOBP), cellular component (GOCC) and molecular function (GOMF) as well as KEGG pathways of the candidate targets deduced from bioactive compounds and in ammation were analyzed via the (DAVID) database 45 . The enrichment results re ecting top items ranked by gene counts were downloaded. Then, R software 3.6.1 was applied for data visualization with 'ggplot2' package.

Molecular Docking
Molecular docking is used to verify the binding activity between target proteins and active components. The protein structure of the key target was obtained from the RCSB PDB database (https://www.rcsb.org/), and the 3D structure of the pivotal compound was obtained from the ZINC database (http://zinc.docking.org/) 46 . Discovery Studio 2016 and PyRx 0.9.8 was used to hydrogenate the protein, nd active pockets, and conduct molecular docking veri cation with FHD 47 . The total-score function of AutoDock/Vina module was used to score the interaction between active components and target proteins. The visualization of binding geometry and interaction was present via PyMol 1.7.x 48 .
Potential mechanism discovery by NILBD NILBD was conducted by Arrowsmith (http://arrowsmith.psych.uic.edu/) for deep validation of the pharmacological mechanism on key component of FHT 25,49,50 . Information retrieval with keywords was done on a two-node search interface, looking for links that may be present in common between two distinct sets of articles in MEDLINE. The rst search de nes 'literature A' and the second de nes 'literature C'. The program then generates a "B-list" of words and phrases found in the titles of both literatures, which assisted us exploring a crucial molecule or purported mechanism involved in the role of FHT on in ammation.

Data availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.     Supplementary Files