Systems biology approaches to identify potential targets and inhibitors of the intestinal microbiota to treat depression

Depression is a common mental disease, with some patients exhibiting ideas and behaviors such as self-harm and suicide. The drugs currently used to treat depression have not achieved good results. It has been reported that metabolites produced by intestinal microbiota affect the development of depression. In this study, core targets and core compounds were screened by specific algorithms in the database, and three-dimensional structures of these compounds and proteins were simulated by molecular docking and molecular dynamics software to further study the influence of intestinal microbiota metabolites on the pathogenesis of depression. By analyzing the RMSD gyration radius and RMSF, it was finally determined that NR1H4 had the best binding effect with genistein. Finally, according to Lipinski's five rules, equol, genistein, quercetin and glycocholic acid were identified as effective drugs for the treatment of depression. In conclusion, the intestinal microbiota can affect the development of depression through the metabolites equol, genistein and quercetin, which act on the critical targets of DPP4, CYP3A4, EP300, MGAM and NR1H4.

org/), and the protein interaction relationship was obtained after hiding the free nodes. The results were imported into Cytoscape 3.7.2 for visualization, and network topology analysis was performed using the CytoNCA function to identify the key target genes 24,25 .
Construction of the "intestinal microbiota-metabolites-target genes" network relationship. We used Cytoscape 3.7.2 to construct a "microbiota-metabolites-target genes" network to help determine the pharmacological mechanism. The degree value reflects the importance of nodes in the network. The higher the value, the more influential the node is. The core metabolites were identify by degree.
Molecular docking. The core metabolites were downloaded from the PubChem database. The small molecules were hydrogenated, and the charge was calculated using Autodock 4.2.6 software 26 . The core target structures were obtained using the AlphaFold Protein Structure Database 27,28 ; they were then imported into Autodock, and an appropriate grid box was set for molecular docking. Finally, the conformation with the lowest docking binding energy was chosen as the final docking result, which was visualized using PyMOL. If the docking binding energy is less than − 5 kcal mol −1 , the receptor and ligand can bind spontaneously. The lower the binding energy is, the greater the possibility of binding, the more stable the binding conformation, and the greater the possibility of reaction. We set the appropriate grid box and we presented the the values as follows:  29 . The protein topology file was generated using the AMBER99SB-ILDN force field 30 , whereas the ligand topology file was generated by the AnteChamber PYthon Parser interfacE (ACPYPE) script using the AMBER force field. MD simulations were carried out in a triclinic box filled with TIP3 water molecules and periodic bounding conditions. The system was neutralized with NaCl counter ions. Before MD simulation, the complex was minimized for 1000 steps and equilibrated by running canonical ensemble, constant-pressure (NVT) and constant-temperature (NPT) for 100 ps. Then, MD simulation was performed for 100 ns for each system under periodic boundary conditions at 310 K temperature and 1.0 bar pressure.
Drug similarity and toxicity profile assessment. Similarities and toxicity characteristics were determined using SwissADME 31 (http:// www. swiss adme. ch/) and validation ADMETlab 32 network tools (https:// admet mesh. scbdd. com/). Because these two factors are critical in the promotion of new agents, we evaluated their physicochemical properties and side effects.

Results
Intersection of the Venn diagram. After obtaining human intestinal microbiota data from the GutMgene database, the target genes that metabolites could act on were predicted via the SEA and STP databases (Tables S1 and S2, Supplemental Digital Contents, which demonstrate the prediction results of SEA and STP). A total of 790 targets were obtained by the intersection of the genes obtained from the two databases ( Fig. 2) (Table S3, Supplemental Digital Content, which demonstrates the result of the intersection of the STP and SEA results).
Intersection between the target genes of the intestinal microbiota and depression in the GEO database. Seven hundred ninety metabolite target genes were intersected with 1908 depression-related differentially expressed genes screened in the GEO database. Accordingly, 44 differentially expressed genes were obtained, which can be considered the crucial genes of intestinal microbiota metabolites that affect depression ( Fig. 5) (Table S5, Supplemental Digital Content, which demonstrates the intersection of metabolite targets and GEO data).
GO enrichment analysis and KEGG analysis of core genes. The 44 differentially expressed genes were analyzed and visualized by GO and KEGG analyses based on the Metascape database (Fig. 6).
Construction and analysis of the PPI network. The PPI network consisted of 28 nodes and 32 edges.
According to the degree values, equol, genistein, quercetin and glycocholic acid were considered the key metabolites affecting depression (Fig. 8).

Molecular docking results.
The key metabolites screened were molecularly docked with depressionrelated targets. The binding energies of the main active constituents and main targets were all < − 7.0 kcal/mol (Table S8, Supplemental Digital Content, which demonstrates the molecular docking results). The smaller the binding energy is, the higher the binding activity is, and the easier the compound is to bind to the target. Our molecular docking results show that key genes and metabolites can form stable conformations. Among them, the key gene MGAM had the best molecular docking results with the metabolite glycocholic acid, and its value was − 9.7 kcal/mol (Table 1). All molecular docking results were visualized (Fig. 9).
Molecular dynamics simulation results. The root mean square deviation (RMSD) curve represents the fluctuation of protein conformation. It can be seen from the figure that in the beginning, RMSD increases because of the interactions between the complex and the solvent. Therefore, RMSD has certain fluctuations in the early stage. However, MGAM-glycocholic acid, NR1H4-equol, NR1H4-genistein and NR1H4-quercetin all    www.nature.com/scientificreports/ increased briefly and tended to be stable, which indicated that the conformation of proteins would not change significantly after the combination of small molecular ligands with proteins, and the combination was relatively stable (Fig. 10a). The gyration radius is often used to describe the change in the overall structure of a protein and to show the compactness of the overall structure. It can be seen from the figure that MGAM-glycocholic acid, NR1H4-equol, NR1H4-genistein and NR1H4-quercetin all have very stable gyration radii. This result is consistent with the RMSD curve reaction, which proves the stability of the protein conformation (Fig. 10b).
Root mean square fluctuation (RMSF) represents the fluctuation of the protein amino acid residues. This reflects the protein's flexibility in the molecular dynamics simulation process. Usually, after the drug is combined with the protein, the flexibility of the protein is reduced to stabilize the protein and play the role of enzyme. As seen from the figure, the simulation results of NR1H4 protein and equol drug, NR1H4 protein and genistein drug, and NR1H4 protein and quercetin drug show that the protein has good flexibility (Fig. 10c).
Despite an acceptable therapeutic value, a drug is still not acceptable as a final product if it exhibits unintended toxicity. Therefore, drug candidates should exceed toxicity limits for further validation. Therefore, equol, genistein, and quercetin were evaluated by the ADMETlab platform for hERG blockers, acute oral toxicity in rats, eye corrosion, and respiratory toxicity (including LD50 [5.238 mg/kg]). The results showed that these substances could play an important role in the treatment of depression (Table 3).

Discussion
Studies have shown that metabolites of intestinal microbes can play an essential role in a variety of psychiatric diseases via the "GMB" axis 33,34 .
In the PPI network, DPP4, CYP3A4, EP300 MGAM and NR1H4 showed higher degrees and were identified as essential genes of intestinal microbiota influencing the occurrence of depression through the brain-gut axis. The primary mechanism of action of DPP4 is the selective cleavage of cytokines and glucagon-like peptide-1 35 . Previous studies have found that DPP4 inhibitors can improve cognitive function and mitochondrial function in the brain 36 . Clinical studies have also confirmed the hypothesis that low plasma DPP4 activity is a characteristic marker of major depression and that changes in DPP4 enzyme activity play a role in the pathophysiology of major depression 37 . CYP3A4 is a metabolic enzyme widely found in human tissues and organs, is involved in approximately 50% of drug metabolism and is an essential factor affecting drug metabolism and efficacy in vivo 38 . Recent studies have confirmed that the mechanism of action of most antidepressants is related to the regulation of CYP3A4 [39][40][41][42][43] . MGAM was associated with an increased risk of Alzheimer's disease (AD) and major depressive disorder (MDD) 44,45 . NR1H4 is closely related to cholestasis, which can cause depression 46,47 . Cytochrome P450 oxidoreductase (POR) is involved in the biosynthesis of endogenous substances, such as bile acids and other steroids, as well as in the oxidative metabolism of xenobiotics, and POR knockdown resulted in the downregulation of NR1H4 (FXR) and the deregulation of bile acid and cholesterol biosynthesis 48 .
GO and KEGG enrichment analyses of hub genes showed that genes were primarily enriched in carbohydrate digestion and absorption, bile secretion, drug metabolism-cytochrome P450, and other pathways. Cytochrome P450 is one of the essential enzymes in drug oxidation metabolism because it can oxidize and metabolize many exogenous substances, including drugs. Secondary bile acids correspond to the metabolism of these products by the gut microbiome. The two primary BAs are cholic acid and deoxycholic acid, which are often secreted into bile in combination with taurine or glycine 49 . Activation of farnesoid X receptor (FXR) may play a central role in the onset of depression under pathological conditions 50 . During prenatal brain development, synapses form between neurons, resulting in neural circuits that support complex cognitive functions. Selective serotonin reuptake inhibitors are commonly used throughout pregnancy to treat depression 51 . Dysregulation of the serotonergic system has been reported to have a significant role in several neurological disorders, including depression, autism and substance abuse disorders 52 . Cholestasis can impair social motivation behavior and induce depression-like behavior. Cholestasis can also affect anxiety and pain behaviors in mice 53 . Pharmacotherapy for neuropsychiatric disorders, such as anxiety and depression, has been characterized by significant interindividual variability in drug response and the development of side effects. Pharmacogenetic research on depression and anxiety has focused on genetic polymorphisms affecting metabolism via cytochrome P450 (CYP) 54 . In antidepressant drug treatment, most drugs are metabolized via the polymorphic cytochrome P450 enzyme CYP2D6 55 . Activating cAMP-PKA signaling could prevent the behavioral changes and hippocampal synaptic deficits elicited by posttraumatic stress disorder (PTSD) 56 . Restoring hippocampal cAMP and BDNF levels could be an antidepressant treatment 57 . Decreases in cAMP and ERK1/2 phosphorylation could reduce the immobility time of chronic restraint stress (CRS) mice in the FST 58 . Research has found that regulating the HIF-1 signaling pathway can improve LPS-induced depressive behavior 59 , and CPSP with comorbid anxiety and depression can be improved by increasing cerebral blood flow and inhibiting HIF-1α/NLRP3 inflammatory signaling 60 .
According to the network analysis of "intestinal microbiota-metabolites-substrate-target genes", equol, genistein, quercetin and glycocholic acid were found to be the key metabolites affecting depression. Studies have shown that intestinal microbiota such as Enterococcus casseliflavus and Bacteroides sp. 45 can metabolize the  www.nature.com/scientificreports/ flavonoid quercetin-3-glucoside to produce quercetin 61,62 . Quercetin acts as an antidepressant by regulating neurotransmitter levels, promoting hippocampal neuron regeneration, reducing inflammation and antioxidant stress, and increasing serotonin levels 63,64 . Equol, a key metabolite of isoflavone with estrogenic and antioxidant activities 65 , can decrease body weight, abdominal WAT, and depression-related behaviors 66 . Experimental studies have found that S-equol can help reduce depression and anxiety in individuals 67 . Genistein, which is produced by the strains "Hugonella massiliensis" DSM 101782 T and "Senegalimassilia faecalis" KGMB04484 T68 , treats depression by suppressing the expression of miR-221/222 by targeting connexin 69 . Molecular docking showed that the key metabolites had good binding activity with the hub genes, and the binding sites were hydrogen bonded to form a stable conformation, indicating that the combination of intestinal microbiota metabolites and depression targets may help in treating depression. Additionally, the molecular dynamics simulation results showed that MGAM-glycocholic acid, NR1H4-equol, NR1H4-genistein and NR1H4-quercetin bind stably.
Drug similarity and toxicity evaluations of equol, genistein, quercetin, and other metabolites revealed that they have antidepressant effects. Genistein can be found in Pueraria lobata, Tempeh tempeh, and Cistanche deserticola.

Conclusions
In this study, we developed a comprehensive strategy to analyze the metabolites of the intestinal microbiota and the target genes of the intestinal microbiota affecting depression through systems biology. We explored the potential targets and inhibitors of the intestinal microbiota in treating depression. We found that intestinal microbiota metabolites such as quercetin, equol, and glycocholic acid can affect the course of depression by acting on targets such as MGAM and NR1H4. The mechanisms of action are related to carbohydrate digestion and absorption, bile secretion, drug metabolism-cytochrome P450, and other pathways. The mechanism of action has multitarget and multipathway results. Subsequently, we further verified these results by molecular docking and molecular dynamics simulations. Finally, we also evaluated the critical metabolites' drug similarity and toxicity characteristics to further confirm their potential drug possibilities. However, accumulating microbiome information has certain limitations, and we plan to conduct further preclinical or clinical trials to provide more references for its clinical application and development.