Therapeutic targets for UC were mined from GeneCard (https://www.genecards.org/) (Safran et al., 2010), Searching Online Mendelian Inheritance in Man (OMIM, http://www.omim.org/) (Amberger and Hamosh, 2017), and DisGeNET (http://www.disgenet.org) (Piñero et al., 2017) databases with “ulcerative colitis” as index word and limiting to “Homo sapiens” species. The overlapped targets from these three databases were removed.

The active compounds of GXD were screened from public databases, including TCM Integrated Database (TCMID, http://www.bidd.group/TCMID/) (Huang et al., 2018), TCM system pharmacology technology platform (TCMSP, https://tcmsp-e.com/) (Ru et al., 2014), and Herb Ingredient’ Targets (HIT, http://hit2.badd-cao.net/) (Ye et al., 2011) and deleted some active compounds, for which no targets information is available.

Subsequently, the potential targets of GXD active ingredients were retrieved from TCMID, TCMSP, HIT, and Search Tool for Interacting Chemicals (STITCH, http://stitch.embl.de) (Szklarczyk et al., 2016) databases. Then, the targets with compound-target associated scores >400 in STITCH databases remained and were normalized the name with the Gene module of the NCBI database.

To further screen the active compounds and potential targets of GXD, several fishing methods were implemented. Drug-likeness (DL) is usually used to signify whether the compound has acceptable ADME (absorption, distribution, metabolism, and elimination) properties, and DL assessment of compounds helps to identify the active ingredients in TCM formulas (Yang et al., 2018). The quantitative estimate of DL (QED) (Bickerton et al., 2012) was used to further screen pharmaceutically active compounds in GXD and set 0.2 as the threshold value in this work. Then, the potential targets of GXD were evaluated using binomial statistical mode (Yang et al., 2018; Liang et al., 2014).

The common targets related to GXD and UC were retrieved by Venny 2.1.0 database (https://bioinfogp.cnb.csic.es/tools/venny/) and then analyzed using the STRING database (https://string-db.org/). Subsequently, files of the PPI network were input into Cytoscape 3.8.0 software to visualize it.

Gene ontology (GO) and Kyoto Encyclopedia of Gene and Genomes (KEGG) pathway enrichment analysis was carried out by DAVID 6.8 (https://david.ncifcrf.gov/tools.jsp). In this study, the GO and KEGG analysis results were thought as significant using adjusted p < 0.01 corrected by the Bonferroni algorithm for each term.

Chemical structures of the active compounds were derived from the Zinc website (http://zinc15.docking.org) (Sterling and Irwin, 2015). Crystal structures of the hub proteins were obtained from the RCSB Protein Data Bank (PDB, http://www.rcsb.org/) (Burley et al., 2017). Then, molecular docking was analyzed using PyMol 2.3.0 (Seeliger and de Groot, 2010), AutoDockTools 1.5.6 (Morris et al., 2009), and AutoDock Vina 1.1.2 (Trott and Olson, 2010).

All animal protocols were approved by the Animal Experimental Ethics Committee of Xiaoshan Third People’s Hospital. Thirty female C57BL/6 mice (6–8 weeks old, 18–20 g, from GemPharmatech Co., Ltd., Nanjing, China) were housed at room temperature (24°C–26°C) with free access to food and the water cycle.

After 7 days, mice were randomized into six groups (n = 5), including control, model, low-dose GXD, middle-dose GXD, high-dose GXD, and positive control-sulfasalazine (SASP) groups. Mice in the control group received sterile water, while mice in other groups were treated with 2.5% dextran sodium sulfate (DSS, MP Biomedicals, CA, USA) for 7 days. Meanwhile, the mice in the GXD and SASP administer groups were given GXD at 3.7 g/kg (low-dose), 7.4 g/kg (middle-dose), 14.8 g/kg (high-dose), and SASP at 125 mg/kg, respectively. In comparison, mice in the control and model groups received 0.5% CMC-Na.

Weight assessment: no loss was scored as 0 points, 1%–5% loss: 1-point, 6%–10% loss: 2-points, 11%–15% loss: 3-points, more than 15% loss: 4-points. Stool assessment: good stool pattern was scored as 0-points, a loose stool: 2-points, and diarrhea: 4-points. Bleeding assessment: no rectal bleeding was scored as 0-point, no rectal bleeding: 2=points and heavy bleeding: 4-points.

Colon tissues were fixed in 4% paraformaldehyde, dehydrated in a different concentration of ethanol, embedded in paraffin, and cut into 5 μm thick slices. Slices were stained with H&E (Beijing Solarbio Science & Technology, Beijing, China). The pathological morphology of colon tissue was observed under a microscope (Olympus, Tokyo, Japan).

Total RNA from colon tissues was isolated and reverse-transcribed into cDNA by TRIzol reagent (Invitrogen, Carlsbad, CA, USA) and PrimeScript RT Master Mix (TaKaRa, Dalian, China), respectively. Each well (20 μL PCR reaction volume) included 10 μL of SYBR Green Mix (Applied Biosystems, CA, USA), 1 μL of cDNA, 2 μL of primer pair mix, and 7 μL of ddH2O. PCR was carried out on a system (MX3000P, Agilent Stratagene, Santa Clara, CA, USA) with the thermocycling conditions as follows: Initial denaturation at 95°C for 3 min, followed by 40 cycles of 95°C for 12 s and 62°C for 40 s. The levels of specific genes were calculated by the 2^−∆∆CT method. GAPDH worked as a reference gene. Primer sequences are listed in Suppl. Table S1.

Total proteins were extracted from colon tissues using RIPA lysis buffer (Beyotime, Shanghai, China), separated by gel electrophoresis, and transferred to membranes. Membranes were blocked by 5% fat-free milk for 1 h and incubated with p65-nuclear factor-(NF)-ĸB (1:1000, ab207297, Abcam, Cambridge, UK), phosphorylated Signal transducer and activator of transcription 3 (p-STAT3; 1:1000, ab76315, Abcam), STAT3 (1:1000, ab68153, Abcam), and glyceraldehyde 3-phosphate dehydrogenase (GAPDH; 1:10000, ab181602, Abcam) antibodies at 4°C overnight, followed by incubation of secondary antibody (1:2000, ab6721, Abcam) for 1 h. Finally, the bands were visualized using an enhanced chemiluminescent reagents kit (Millipore, Darmstadt, Germany).

The Prism 8 (GraphPad, San Diego, CA, USA) software was used for statistical analyses. All data were exhibited with mean and standard deviation and analyzed using one-way ANOVA, followed by Tukey’s test. p < 0.05 was considered to reflect a significant difference.

We found 63, 322, and 521 therapeutic targets for UC from GeneCard, OMIM, and DisGeNET databases, respectively. Then, 804 UC-related targets were finally retrieved after removing duplicate targets.

From TCMID, TCMSP, and HIT databases, 371 active compounds of GXD were obtained. A total of 7154 potential targets of GXD active compounds were retrieved from TCMID, TCMSP, HIT, and STITCH databases. Next, 346 active compounds were suitable for DL screening (QED ≥ 0.2). Then, from 285 active compounds, 692 potential targets were collated by binomial statistical mode. Subsequently, UC-related targets and GXD potential targets were entered into Venny 2.1.0, and after eliminating the redundancy, 89 common targets between UC and GXD were obtained (Fig. 1). Finally, 213 active compounds were mapped to 89 common targets. These 213 active compounds and 89 common targets are detailed in Tables 1 and 2, respectively. Furthermore, we constructed a Drug-compounds-targets (D-C-T) network using Cytoscape software with 213 active compounds and 89 common targets. The D-C-T network includes 309 nodes and 1545 edges (Fig. 2).

Interaction relationships of common targets were obtained using the STING database. The PPI network comprised 89 nodes and 1367 edges, with an average degree of 30.7; the larger sizes of the target nodes indicate that nodes have more degrees in the PPI network (Fig. 3A). Then, the top 10 hub degree genes, including ALB, AKT1, IL6, TNF, VEGFA, TP53, CXCL8, MAPK1, PTGS2, and IL1β, were screened out by Cytoscapehubba, a Cytoscape plug-in. The hub gene network consisted of 10 nodes and 45 edges (Fig. 3B).

GO enrichment analysis results contained 1107 biological process (BP) terms, 43 molecular function (MF) terms, and 18 cellular component (CC) terms when adjusted at p < 0.01. For BP terms, the common targets were mainly related to response to oxygen levels, lipopolysaccharide, the molecule of bacterial origin, etc. For MF terms, the common targets were mainly phosphatase, cytokine receptors, protein phosphatase binding, etc. For CC terms, the common targets were primarily located in the membrane microdomain, membrane raft, membrane region, etc. (Figs. 4A–4C). KEGG analysis remarkably enriched 144 pathways (p < 0.01), of which the top 15 pathways are presented in Fig. 4D. Most of the common targets were enriched in the AGE-RAGE signaling pathway in diabetic complication (hsa04933), HIF-1 signaling patIy (hsa04066), endocrine resIance (hsa01522), etc. Additionally, to explain that GXD could treat UC, we used the key targets, top 15 KEGG pathways, and top 15 BP terms to construct a targets-pathways-BP network, containing 106 nodes and 673 edges (Fig. 5). Furthermore, to further illustrate the ability of GXD to ameliorate UC, we also evaluated the relationship between the hub target and BP, KEGG, as shown in Fig. 6, according to which, AKT1, PTGS2, IL1β, TNF, and MAPK1 may play a vital role during UC treatment.

The interactions between the active compounds and hub targets were investigated to further explore the action mechanism of the GXD compounds. Nine active compounds (maslinic acid, baicalin, betulinic acid, protopine, liquiritin, baicalein, glycyrrhizic acid, isoliquiritigenin, and gingerol; the detailed information is elucidated in Table 3) and eight hub genes (MAPK1, PTGS2, AKT1, TP53, IL6, ALB, and TNF) were selected for molecular docking analysis. After setting the threshold affinity <−5 kcal/mol, we obtained 59 pairs of docking results (Table 4). The top 10 pairs of docking results are exhibited in Fig. 7. Among them, the binding affinity of maslinic acid with tumor necrosis factor (TNF) was −9.7 kcal/mol, which was contributed by bonding with SER-60 and LEU-120 residues. The binding affinity of baicalin with PTGS2 was −9.7 kcal/mol, which was contributed by bonding with HIS-388, TYR-385, THR-212, and HIS-214 residues. The binding affinity of betulinic acid with TNF was −9.6 kcal/mol, which was contributed by bonding with HIS-388, TYR-385, THR-212, and HIS-214 residues.

To verify the effect of GXD on UC, we established a UC mouse model by administering 2.5% DSS. DAI scores were used to assess the severity of colitis in mice. As shown in Fig. 8A, compared with the control group, DAI scores were significantly increased in the model group. While GXD and SASP administration decreased the DAI scores. Additionally, the colon length was significantly shorter in the model group than in the control group. GXD and SASP administration notably suppressed the shortening of colon length in UC mice (Figs. 8B and 8C). The colon in the model group showed tissue damage with mucosal architecture and apparent inflammatory cell infiltration. Likewise, GXD provided a therapeutic effect in a dose-dependent manner (Fig. 8D).

To further verify whether GXD could act against UC, we detected the levels of inflammatory cytokines and potential targets using qRT-PCR and western blotting. The mRNA levels of inflammatory cytokines interleukin (IL)-1β, IL-17, IL-6, TNF-α, and cyclooxygenase (COX)-2 were markedly increased in the model group; then they were weakened by dose-dependent addition of GXD (Fig. 9A). Meanwhile, the protein expression of p65-NF-ĸB, p-STAT3, and STAT3 using western blotting assay revealed that oral administration of 2.5% DSS obviously enhanced p65-NF-ĸB and p-STAT3 expression levels, while GXD and SASP addition reversed this enhancement of DSS. Besides, the expression of STAT3 did not change in all groups (Fig. 9B).