We queried sepsis from GeneCards and DisGeNET with a total of 12 sepsis-related studies. The CUI IDs in DisGeNET were as follows: C0243026, C1719672, C0684256, C3164780, C0036686, C0152965, C0152964, C1141927, C0342959, C0036690, C0242966, and C0036685. The data obtained from these 12 studies were treated as pathogenetic genes of sepsis.

In order to construct comprehensive weight gene network of XBJI, the PPI data were derived from public web servers CMGRN and PTHGRN (Guan et al., 2014a, 2014b). Pathogenetic genes from DisGeNET (Piñero et al., 2017) were mapped to the PPI network and used in constructing the weighted gene regulatory network of XBJI. Cytoscape (Version 3.7.2) was utilized to visualize the network.

The chemical ingredients of XBJI were collected from a published natural product data sources, TCMSP database (Ru et al., 2014). Besides, considering that chemical analysis plays an important role in the study of substance basis and mechanism of botanical drugs in the formulas, we also regarded some ingredients which had been detected in XBJI and reported in literatures as potential active ingredients.

The potential active ingredients of XBJI in treating sepsis were selected from all chemical ingredients of XBJI based on ADME models, including Caco-2 permeability (Caco-2) and drug-likeness (DL) (Xu et al., 2012). The chemical ingredients with the criteria that Caco-2 > −0.4 and DL ≥ 0.18 were selected as potential active ingredients (Tao et al., 2013). According to literatures, some chemical ingredients which have high content and high biological activities were also selected as potential active ingredients of XBJI in treating sepsis.

Targets of potential active ingredients of XBJI were predicted in three published databases, HitPick (Liu et al., 2013), similarity ensemble approach (Appearance Research Collaboration., 2014) (Tao et al., 2013), and SwissTargetPrediction (Gfeller et al., 2014). In order to evaluate the importance of targets, we conducted an ingredients–targets (I-T) network, investigated the node importance of ingredients and targets, and conducted KEGG and GO enrichment analysis of targets.

Node importance could reflect the effect of node in a complex network. Here, we designed a node importance calculation method to figure out the quantified influence of each node. The detail of the method can be described as follows: N i m ( s ) = ( ∑ s   ≠ a   ≠ b   ∈ V   σ a b   ( s ) σ a b )   ×   [ |   U   ( C ( s ) ) | |   U   |   ×   ∑ w ∈ C ( s ) ( Δ C ( s ) + 1 − d i s t ( s , w ) ) max   { d i s t ( s , w ) : w ∈ C ( s ) } ] 2 F R S =   { N i m i   ≥   M i d   { min ( N i m )   : max ( N i m ) } ,   i   ≤ | U | } where, Nim represents the node of importance. σ a b represents the number of shortest paths between nodes a and b, σ a b ( s ) is the number of shortest paths passing through node s. C_(s) represents the genes which contain node s. U represents the collection of nodes within the network. |U| represents the number of nodes. Δ C ( s ) is the maximum distance between gene C and the other genes passing through node s. dist_(s,w) represents the length of the shortest path between nodes s and w. The dist_(s,w) is equal to infinite if C_(s) ≠ C_(w), and it makes methods of this category inapplicable to networks with disconnected genes. Functional response space (FRS) represents the set of genes in the FRS; Med, Min, and Max represent median, minimum, and maximum, respectively. The genes in the FRS are defined as the key response proteins.

To optimize active ingredients and get the critical functional ingredients group (CFIG), which could be used to illustrate the molecular mechanism of XBJI in the therapy of sepsis, we constructed an ingredients–contribution–proportion (ICP) model, which could value the node importance in the network and contribute to the selection of the critical ingredients in representing the importance of the effect of all potential active ingredients.

We set m as the number of ingredients, I as the set of ingredients, I = [I₁, I₂, I₃ … … I_m], T as the set of targets of ingredients, T = [T1, T2, T3 … … T_m]. After sorting the target number of ingredients in the descending order, we got the descending order set ST of the targets and the rearranged component set SI. SI was the new order of I after T in the descending order. At the same time, we set the set of optimized ingredients as Q and the set of optimized targets as W. CFIG was equal to Q. The ICP model can be described as:

For analyzing the main function of the targets, the clusterProfiler package and R Bioconductor (https://www.bioconductor.org/) were used to perform GO analysis (Yu et al., 2012) and KEGG pathway enrichment analysis (p-values = 0.05) (Draghici et al., 2007). The graphs creations and genes annotations were conducted with ggplot2 package and the Pathview (Luo et al., 2017), respectively.

The process of sepsis is related to a series of complex changes in phenotype and gene expressions. These different phenotypic changes are accompanied by a number of alterations of genes expression. These genes may be labeled as pathogenetic genes both at diagnostic and intervention levels. The collection and analysis of pathogenetic genes are the foundational and critical ways in understanding the pathogenesis of sepsis and providing intervention strategies. For obtaining more comprehensive pathogenetic genes, we extracted pathogenetic genes from the proved sepsis-related literatures in DisGeNET and GeneCards databases (Supplementary Tables S1, S2).

Totally, 1,606 genes were obtained and were treated as pathogenetic genes, including 578 genes from DisGeNET and 1,273 genes from GeneCards. Among these 578 genes from DisGeNET, 186 genes were reported in more than 2 published reports (Figure 2A) and 21 genes were reported in more than 10 published reports. Among the 1,273 genes from GeneCards, the relevance scores of the 612 genes were equal or more than 2.00 but less than 3.00, and the relevance scores of the 81 genes were equal or more than 10.00 but less than 40.00. The top 10 genes with high numbers of supported reports in 578 genes from DisGeNET were TNF, IL-6, TLR4, IL-10, HMGB1, TLR2, MBL2, CD14, CXCL8, and IL-1β (Figure 2G). The top 10 genes with high relevance scores in 1,273 genes from GeneCards were TNF, IL-6, IL-10, CRP, ELANE, CXCL8, TLR4, IL-1B, CD14, and TLR2. TNF and IL-1β were typical pro-inflammatory factors and could reduce the body's immune function (Kamisoglu et al., 2015), while IL-6 and IL-10 were some anti-inflammatory cytokines (Chaudhry et al., 2013). Furthermore, many of these pathogenetic genes were reported to be related to sepsis closely. For example, TNF-α, a member of TNF, was usually detected in the blood of patients with sepsis (Fujishima, 2016) and was reported to induce NO production significantly (Ahn et al., 2012) and increase the death rate in LPS-induced sepsis mice (Li et al., 2020). IL-6 could mediate the inflammatory response of LPS-induced SIRS mice (Silva et al., 2019). TLR4 was a subtype of TLRs and could identify pathogenic bacteria in the innate immune system. It could activate IκB kinase (IKKs), cause the ubiquitination and degradation of IκB, activate NF-κB, and increase TNF-α expressions, thereby mediating hepatic inflammation, lung injury, and heart injury (Bieghs and Trautwein, 2014; Yin et al., 2018; Shen et al., 2019; Chang et al., 2021). HMGB1 was typically found in the nucleus of a variety of cells (immune, endothelial, epithelial cells, etc.). After the activation of TLR, HMGB1 could acetylate and trigger its translocation from the nucleus to the circulation and interact with a variety of target cell receptors (RAGE, TLR2, TLR4, etc.), stimulating the release of pro-inflammatory cyto-/chemokines and leading to inflammations and sepsis (Yang et al., 2015; Abdulmahdi et al., 2017; Andersson et al., 2018). CD14 was a receptor of LPS and was reported to mediate the occurrence and development of sepsis through the TLR4-NFκB pathway (Skirecki and Cavaillon, 2019).

To verify whether the genes supported by more literature in DisGeNET or by high score in GeneCards had a wide range of functions, we performed KEGG and GO enrichment to all pathogenetic genes and found that the number of most of the genes in the 578 genes from DisGeNET was directly proportional to the number of pathways and GO-terms. Genes supported by more literatures mostly had more pathways and GO-term correlations (Figure 2C,E). In addition, the relevance scores of all 1,273 genes from GeneCards were directly proportional to the number of pathways and GO-terms (Figure 2D,F). These results showed that the 1,606 genes from DisGeNET and GeneCards could represent the current occurrence and research of sepsis and could be used for pathogenetic genes in subsequent studies.

1,606 pathogenetic genes were enriched into 152 KEGG pathways and 3,446 GO-terms (Figure 2M). Among the 152 KEGG pathways, 8 pathways were reported to be associated with sepsis and were reported as cytokine–cytokine receptor interaction and some signaling pathways of JAK-STAT, TLR, NF-κB, T-cell receptor, PI3K-Akt, TNF, and IL-17. Among the 3,446 GO-terms, 8 GO-terms were reported to be associated with sepsis and were reported as positive regulation of cytokine production, leukocyte migration, T-cell activation, regulation of lymphocyte activation, regulation of peptide secretion, response to LPS, regulation of inflammatory response, and regulation of protein secretion.

The JAK/STAT signaling pathway had been proved to mediate the myocardial injury of septic rats with distinctly higher relative mRNA expression levels of JAK and STAT3 (Jin et al., 2020). Furthermore, JAK2/STAT3 signaling pathway had been demonstrated to induce the expression of miR-181b, which could downregulate sphingosine-1-phosphate receptor 1 (S1PR1) and decrease BBB cell adhesion, thereby promoting BBB impairment in rats with sepsis (Chen et al., 2020). TLR signaling was reported to be involved in cardiac redox signaling, calcium handling, energy metabolism, and affecting cardiac structure and gap junction proteins in sepsis mice (Lackner et al., 2020). The high expression of some proteins, such as TLR2, TLR3, and TLR4, have been demonstrated to play a crucial role in sepsis-induced acute injury of the lungs, kidneys, and intestine (Krivan et al., 2019; Kumar, 2020a). NF-κB and TNF-α were considered to be the links in regulating apoptosis (Zhao et al., 2015). Some studies showed that the cell damage caused by LPS was closely related to NF-κB, such as inflammatory apoptosis (Chand et al., 2018). The phosphorylation level of NF-κB was significantly increased in the lung tissues of sepsis mice (Wang et al., 2020). The activation of NF-κB would destroy intestinal barrier function during sepsis (Cao et al., 2021). Furthermore, the T-cell receptor signaling pathway was also demonstrated to be related to sepsis by network biology approach (Kim et al., 2021). The NFATC2 was upregulated in sepsis, which was also known as NFAT1, the first member of the nuclear factor of activated T cells (NFAT) family and had an important function in inducing gene transcription during an immune response (Qin et al., 2020). T-cell activation of both CD4+ and CD8+ T cells, as well as T-cell cytokine production, was suppressed acutely and persistently after burn injury, while sepsis could account for 47% of post burn mortality (Beckmann et al., 2020). Additionally, the activation of PI3K-Akt signaling pathway and the phosphorylation of Akt could attenuate apoptosis and improve survival in animal models of sepsis (Liu et al., 2019; Mizuta et al., 2020). The expression level of NF-κB and TNF-α was significantly higher in LPS-induced H9c2 cardiomyocytes, suggesting that the expression of NF-κB/TNF-α might be related to inflammatory response and apoptosis of H9c2 cells during sepsis (Zheng et al., 2020). The secretion of pro-inflammatory cytokine TNF-α was increased significantly in LPS-stimulated bone marrow–derived macrophages (BMDMs) (Liu et al., 2020). In the regulation of cytokine production, some cytokines had been proved to be beneficial for the control of sepsis, such as the anti-inflammatory cytokines of IL-1Ra, IL-4, IL-6, IL-10, IL-11, IL-13, IL-35, and TGF-β (Chaudhry et al., 2013). Studies showed that most of the differentially expressed proteins were mainly involved in leukocyte migration (Luo et al., 2021). The trans-endothelial leukocyte migration could be facilitated by miR-887-3p, thereby leading to an increase in leukocyte trafficking and dysregulation of inflammation and enhanced lung injury in sepsis-related acute respiratory distress syndrome (LeVine et al., 2000) (Goodwin et al., 2020). The latest study with a septic mouse model demonstrated that leukocyte transendothelial migration might be related to some downregulated lncRNAs (Huang et al., 2020). Septic patients were found to have a reduction in lymphocytes count (Coakley et al., 2019), which was because lymphocyte activation was negatively regulated by PD-1 ligands and PD-L1 (Kawamoto et al., 2019).

The weighted gene regulatory network can contribute to understand the pathogenetic genes and provide intervention strategies for sepsis. In order to construct weighted gene regulatory network, the base network which could show the relationship and interactions of genes was downloaded and combined from two published databases, including CMGRN and PTHGRN (Guan et al., 2014a, 2014b). 1,606 pathogenetic genes were constructed into a weighted gene regulatory network (PPI network) by mapping to the base network. The degree, relevance score, and number of literatures were multiplied to get the weight of the node in the regulatory network (Figure 3).

The weighted gene regulatory network contains 1,274 nodes and 22,560 edges (Figure 3). The number of edges represent the importance of the node in the network. Genes IL-6, TNF, TLR4, HMGB1, and CD14 were found to have higher weight in the top 100 weighted gene regulatory network. Some genes with high degrees have high supported literatures, such as genes TNF, TLR4, HMGB1, and TLR2 have scores of 37, 35, 46, and 25, respectively, larger than the average scores of 17.71 of all genes in the network, and also have the number of supported literatures of 85, 53, 32, and 27. Furthermore, some genes having a higher score in the PPI network also have a higher relevance score in the GeneCards database, such as genes TNF and ELANE have degrees of 37 and 23, respectively, higher than the average degree of 17.71 of all genes in the PPI network, and these also have high relevance scores of 31.51 and 22.01, respectively, in GeneCards. These genes with high degrees in the PPI network have been reported to participate in some signaling pathways associated with sepsis, such as the cytokine–cytokine receptor interaction (hsa04060), NF-κB signaling pathway, and MAPK3 signaling pathway. Among these pathways, some genes were reported to be associated with sepsis closely. For example, IL-6 could improve the stimulation and enhance the functions of T lymphocytes, B lymphocytes, and other cells involved in cell proliferation and differentiation, playing a role in anti-inflammation and regeneration. IL-10 was a powerful regulatory cytokine with anti-inflammatory effects and could inhibit the synthesis of a variety of cytokines, such as interferon c (IFN-c), TNF, and other pro-inflammatory cytokines. TNF-α was a pleiotropic cytokine which could trigger pro-inflammatory pathways by activating NF-κB to initiate the transcription of a variety of genes related to the production of inflammatory cytokines and could regulate the activity of inflammatory response promoters (AP-1) and other transcription factors through the MAPK3 signaling pathway. These results indicated that genes which were related to sepsis or to the activity of inflammation played important roles in the interaction in the PPI network and also that the PPI network could be used to further study FRS building.

By a systematic search of ingredients from public databases of five botanical drugs in sepsis, including CS, CX, DS, DG, and HH, we obtained 728 ingredients (Supplementary Table S3). Meanwhile, we obtained 36 ingredients of XBJI according to literatures (Table 1).

We obtained 205 potential active ingredients from these 728 ingredient base on two ADME-related models, including Caco-2 and DL (Figure 4). Considering that chemical analysis plays important roles in the study of substances basis and mechanism of botanical drugs in the formulas, we regarded the 36 ingredients collected from literatures as potential active ingredients. As a result, we collected 241 potential active ingredients of XBJI (Table 2, Supplementary Table S4).

A total of one ingredient, palmitic acid (MOL000069), is contained in five botanical drugs. Four ingredients were contained in four botanical drugs, and they are EIC (MOL000131), sitogluside (MOL000357), moslene (MOL000202), and oleic acid (MOL000675). Nine ingredients were contained in three botanical drugs, and these are baicalin (MOL002776), beta-sitosterol (MOL000358), stigmasterol (MOL000449), nonanal (MOL000116), cymol (MOL000117), adenine (MOL001788), caffeic acid (MOL000223), stearic acid (MOL000860), and succinic acid (MOL000346). In addition, 62 ingredients were contained in two botanical drugs (Figure 4, Supplementary Table S5). Among them, palmitic acid was reported to increase serum levels of TNF-α, IL-6, IL-10 and to activate TLR4 and downstream inflammatory responses by binding to TLR4 accessory protein MD2 directly, thereby regulating the TLR4 signaling pathway, MAPKs/NF-κB signal pathway, and the level of pro-inflammatory molecules (Wang et al., 2017; Panpetch et al., 2020). Oleic acid was reported to be higher in septic patients (Mecatti et al., 2018) and to decrease plasma-free FA concentration and increase the levels of CPT1A, UCP2, and AMPK, thereby decreasing the levels of reactive oxygen species (ROS) in septic mice (Gonçalves-de-Albuquerque et al., 2016). Baicalin could significantly reduce IL-1β in the sera of bacterial infected mice and could inhibit NLRP3 inflammasome activation through augmenting PKA signaling, thereby improving mouse survival in bacterial sepsis (Li et al., 2017) and could reduce the level of LPS, TNF-α, and IL-6 in the blood of sepsis mice (Chen et al., 2016). Besides CS, DG, and HH, beta-sitosterol was also found to exist in Hechtia glomerata Zucc., which was used as a source in ethnomedicine to treat various diseases derived from bacterial infections, such as sepsis, bronchitis, and laryngitis (Stefani et al., 2019). Beta-sitosterol was predicted to mediate NF-κB signaling pathway (Kasirzadeh et al., 2021) and was found to significantly reduce the pro-inflammatory cytokines in the sera of cecal ligation and puncture-induced septic rats, such as TNF-α and NF-κB mRNA expressions. In addition, stigmasterol was proved to decrease the levels of cyclooxygenase-2 (Cox-2) and NF-κB (p65) in the brain of rats, thereby attenuating inflammation and improving antioxidant defenses (Liang et al., 2020). Caffeic acid could reduce pro-inflammatory cytokines IL-1β, IL-6, and TNF-α and attenuate iNOS and COX-2 in LPS-stimulated macrophages (Chen et al., 2019).

Except the shared ingredients, most of the botanical drugs have its specific ingredients (Supplementary Table S5). For example, the number of unique ingredients of CS, CX, DS, DG, and HH are 91, 142, 173, 91, and 155, respectively (Figure 4).

According literatures, these unique ingredients in different botanical drugs have special therapeutic effects. For example, paeonol (MOL000874), paeoniflorin (MOL001924), galloylpaeoniflorin (MOL001932), albiflorin (MOL007004), and oxypaeoniflorin (MOL007006) are the specific ingredients of CS. Among them, paeonol was reported to promote phagocytosis of macrophages through confining HMGB1 to the nucleus, thereby promoting the immune response during sepsis (Miao et al., 2020) and attenuate the inflammation mediated by HMGB1 and IKK-β by upregulating miR-339-5p expression, thereby protecting the kidneys and improving the survival rate of sepsis mice (Mei et al., 2019). Paeoniflorin was a critical ingredient of monoterpene glycoside in CS and XBJI (Wang et al., 2020). It was reported to reduce inflammation, prevent MODS, and protect organ functions by reducing the plasma levels of aspartate aminotransferase (AST), creatine kinase-MB, and soluble triggering receptor expressed on myeloid cells-1 (Liu et al., 2016). Furthermore, it has been reported to improve myocardial function in septic mice by attenuating LPS-induced myocardial inflammatory cytokines production, such as TNF-α, IFN-γ, IL-1β, IL-6, IL-12, and MCP-1 (Zhai and Guo, 2016). Galloylpaeoniflorin had the same structural characteristics of paeoniflorin, which had significant NF-κB inhibitory effects (Lu et al., 2019). Oxypaeoniflorin was proved to be an important ingredient in XBJI with a significant effect in inhibiting NF-κB, IL-6, and IL-8 in LPS-induced human monocyte THP-1 cells (Jiang et al., 2013). Protocatechuic acid (MOL000105) exists in CX and was proved to be an important ingredient in XBJI (Li et al., 2016; Li et al., 2019). Its derivative, protocatechuic acid isopropyl ester, had been proved to protect murine model against sepsis by inhibiting TNF-α production, NO production, and the augmentation of IL-6 and IL-10 (Yan et al., 2004). Salvianolic acid B (MOL007074) is a water-soluble ingredient in DS and had been detected to be an important ingredient in XBJI (Huang et al., 2011; Li et al., 2016; Li et al., 2019; Xing et al., 2020). It was demonstrated to exert an antiviral effect (Chung et al., 2015) and was supposed to be useful in treating COVID-19 (Xing et al., 2020). Danshensu (MOL007134), an ingredient in DS, had been detected in XBJI and was reported to inhibit the overexpression of NF-κB in human embryonic kidney 293 (HEK 293) cells which was significantly stimulated by TNF-α (Jiang et al., 2013). Ferulic acid (CIS) (MOL000389), an ingredient in DG, had been reported to play a protective role on sepsis-induced oxidative damage by significantly decreasing malondialdehyde levels (Salomão et al., 2008), significantly increasing glutathione (GSH) levels and superoxide dismutase (Deng et al., 2013), and glutathione peroxidase (GSH-Px) enzyme activities in Wistar albino rats (Bacanlı et al., 2014). Furthermore, it was reported to positively modulate the inflammatory response to septic liver injury through the GSK-3β/NF-κB/CREB pathway, including decreasing the activities of MPO, AST, and alanine aminotrasferase (ALT), and including decreasing the levels of some pro-inflammatory factors, IL-1β, IL-6, IL-10, IL-12, and TNF-α (Cao et al., 2021).

For exploring the therapeutic mechanism of XBJI in treating sepsis, 241 potential active ingredients and 1,105 targets were used in constructing the I-T network (Supplementary Table S6). The top 50 ingredients and top 50 targets valued by the degree in the I-T network are shown in Figure 5. Some of these potential active ingredients are related multiple targets, resulting in that 8,598 I-T associations between all the potential active ingredients and targets in the I-T network. The average number of targets of per ingredient is 35.68, indicating that there are multi-ingredients and multi-targets characteristics of XBJI in treating sepsis. Among these potential active ingredients, ferulic acid (CIS) (MOL000389) has the highest number of targets as 185, followed by ingredients ethyl ferulate (COM10) and coniferyl ferulate (MOL008288) with the number of targets as 151 and 131, respectively. Among these ingredients, ferulic acid (CIS) had been proved to decrease sepsis-induced oxidative damage (Bacanlı et al., 2014) and protect the liver during sepsis through the GSK-3β/NF-κB/CREB pathway (Cao et al., 2021). Ethyl ferulate was proved to inhibit the inflammatory responses in RAW264.7 cells and acute lung injury of mice by activating the Nrf2/HO-1 pathway and inhibiting the NF-κB pathway (Wang et al., 2021). Coniferyl ferulate was proved to inhibit the activation of NMDAR-CaMKII-MAPKs and mitochondrial apoptotic pathways, including inhibiting the generation of ROS, decreasing the activity of SOD, thereby significantly attenuating the decrease of cell viability (Gong et al., 2020). These results demonstrated that the potential active ingredients played important roles in treating sepsis.

Here, we constructed a disease–targets network based on the weighted gene regulatory network and I-T network. This network contains 2,110 nodes and 58,258 edges (Supplementary Table S7).

Node importance is an important topological property and can be used to evaluate the influence of nodes in the network. The nodes whose node importance are larger than the average node importance of all nodes are treated as critical roles and hub nodes in the network (Liu et al., 2016). Here, we designed a novel node importance calculation method to figure out the importance and the influence of genes. According to this rule, these nodes with higher importance scores than average importance score were kept and integrated with their edges to form FRS.

In order to evaluate the effectiveness of the FRS, we defined the intersection of pathogenetic genes and XBJI targets as the un-optimized effective targets (UETs) of XBJI and defined the genes which were included in the FRS as the key response proteins. Then, we evaluated the key response proteins with three evaluation indicators: 1) the proportion of key response proteins in the number of UETs, 2) the proportion of key response proteins in the number of UETs enriched pathways, and 3) the proportion of the key response proteins in the number of UETs enriched GO-terms.

Before the optimization, the number of XBJI targets and pathogenetic genes of sepsis were 1,105 and 1,606, respectively (Figure 6C); the numbers of targets and pathogenetic genes enriched pathways were 185 and 152 (p < 0.05), respectively; while, the numbers of targets and pathogenetic genes enriched GO-terms were 2,605 and 3,446, respectively (p < 0.05). We used four methods in the optimization of targets and pathogenetic genes. Our proposed node importance calculation method was better than three other traditional methods: Degree, Closeness Centrality, and Clustering Coefficient (Figure 6D). Based on our method, 1,055 nodes were filtrated as important nodes and were defined as the key response proteins. The key response proteins and their interactions were used to construct the FRS. There were three subtypes of key response proteins in the FRS (Figure 6K). The first subtype represents essential common targets, which directly linked pathogenetic genes and ingredients' targets. The second subtype represents disease-specific targets. The third subtype represents ingredient-specific targets.

The number of key response proteins enriched pathways and GO-terms were 189 and 3,932, respectively (Figures 6L,M). The enriched pathways accounted for 95.83% of UETs enriched pathways (Figure 6F). The GO-terms accounted for 98.59% of UETs enriched GO-terms (Figure 6H). These high coverages confirmed the accuracy and reliability of our method in constructing the FRS.

Furthermore, in order to validate whether the key response proteins stored in the FRS could represent the importance of pathogenetic genes of sepsis, KEGG, and GO enrichments, analysis was conducted for the key response proteins.

KEGG enrichments showed that the key response proteins could be involved in 189 pathways, and mainly involved in some signaling pathways of PI3K-Akt (hsa04151), MAPK (hsa04010), Rap1 (hsa04015), human cytomegalovirus infection (hsa05163), and T-cell receptor (hsa04660), etc. Many of these pathways had been reported to have close association with sepsis. For example, the activation of PI3K-Akt signaling pathway and the phosphorylation of Akt could attenuate apoptosis and improve survival in animal models of sepsis (Liu et al., 2019; Mizuta et al., 2020). The activation of the MAPK signaling pathway was reported to promote sepsis-induced myocardial injury by overexpressing the level of angiotensin II type 1 receptor (AT1R). The signaling cascades of TLR4-MyD88-NF-κB/MAPK could be effected by DS, thereby significantly ameliorating LPS-challenged acute kidney injury, inhibiting dimethylbenzene-induced mouse ear edema, and rescuing LPS-induced sepsis in mice (Yuan et al., 2019). The activation of the PI3kinase/RAP1 signaling pathway could lead to the activation of integrin receptor GPIIbIIIa and the release of dense granules, contributing to the thrombotic complications of sepsis (Keane et al., 2010). Furthermore, the T-cell receptor signaling pathway was reported to release some activated T cells and nuclear factors, CD4⁺, CD8⁺, and NFATC2, to take the immune response in sepsis (Beckmann et al., 2020; Kim et al., 2021).

GO enrichments showed that the key response proteins could be enriched into 3,932 GO-terms (Figure 6H), including leukocyte migration (GO:0050900), response to oxidative stress (GO:0006979), response to LPS (GO:0032496), cellular response to drugs (GO:0035690), neutrophil activation (GO:0042119), etc. Many of these GO-terms had been reported to have a close association with sepsis. For example, some of the differentially expressed proteins in plasma of septic patients acted on leukocyte transendothelial migration (Luo et al., 2021), and leukocytes from septic patients could reverse glucose metabolism to oxidative phosphorylation from glycolysis and display multiple energy-metabolizing defects (Cheng et al., 2016). Some proteins and receptors had been proved to be related with oxidative stress, oxygen species metabolic process, etc., during sepsis. For example, heat shock protein 70 in the intracellular environment was reported to have chaperone activity, correcting damaged proteins and modulating inflammatory response (Sulzbacher et al., 2020); P2X7 receptor blockers were reported to limit oxidative damage and inflammations in sepsis (Larrouyet-Sarto et al., 2020); the overexpression of protein kininogen-1 (KNG1) could strengthen inflammation and oxidation in sepsis-induced acute lung injury (Hu et al., 2020). LPS was reported to effect diminishing the intracellular alkalization and change neutrophil size induced by platelet-activating factor (PAF), which was an important mediator of the systemic inflammatory response (Hug et al., 2021), and have an effect on the phagocytosis and reactive oxygen species (ROS) production of neutrophils (Messerer et al., 2020). The depolarization of membrane potential of neutrophils could be significantly impaired by PAF, thereby inducing neutrophil dysfunctions during sepsis (Hug et al., 2021). During the late stage of sepsis, CD4⁺ and CD8⁺ and other T cells become apoptotic and responsible for the development of lymphopenia and immunosuppression, thereby exacerbating the deterioration of the sepsis condition (Rimmelé et al., 2016; Kumar, 2018).

The CFIG of XBJI in treating sepsis was united with the effective ingredients which were established from the FRS with the ICP model. According to the contribution accumulation score ranking, targets of the top nine ingredients including ferulic acid (MOL000389), ethyl ferulate (COM10), evofolin B (MOL007022), quercetin (MOL000098), guanosine (MOL002687), methyl tricosanoate (MOL005368), paeonol (MOL000874), luteolin (MOL000006), and apigenin (MOL000008) could cover 50.63% of the key response proteins. For further analysis, targets of the top 60 ingredients could cover 90.16% of the key response proteins, and these 60 ingredients were selected as the CFIG (Figure 7; Table 3, Supplementary Figure S1). The high targets coverage of key response proteins of 90.16% proved that the CFIG of XBJI played the leading role in the therapy of sepsis.

Most of the CFIG had been proved to state and be high enough in XBJI or have a close association with sepsis and anti-inflammatory responses. For example, three ingredients in the CFIG, namely, ferulic acid, guanosine, and paeonol, had been reported to state and be high enough, with concentrations of 35.12, 27.15, and 0.04 μg/ml, respectively, in XBJI (Ji et al., 2010; Zuo et al., 2017). Ferulic acid is a characteristic ingredient in DG and has the highest number of targets as 185 in our I-T network. It had been proved to protect Wistar albino rats from sepsis-induced oxidative damage (Bacanlı et al., 2014) and positively modulate the inflammatory response to septic liver injury through the GSK-3β/NF-κB/CREB pathway (Cao et al., 2021). Ethyl ferulate has the secondary number of targets as 151, following ferulic acid in the I-T network and has been proved to inhibit the inflammatory responses in RAW264.7 cells and acute lung injury in mice by activating the Nrf2/HO-1 pathway and by inhibiting the NF-κB pathway (Wang et al., 2021). Furthermore, some ingredients had been proved to inhibit the inflammatory effectively. For example, evofolin B had been proved to exhibit potent inhibitory activities on human neutrophil O₂ generation (Chen et al., 2009). Quercetin was reported to protect mice from LPS-induced sepsis by inhibiting TNF-α and IL-1β expressions, NF-κB activation, and apoptosis (Wei et al., 2018) and proposed to prevent myocardial dysfunction through the TLR4/NF-κB signaling pathway during sepsis (He et al., 2020). Paeonol was reported to have protective effects on endotoxin-induced kidney injury by attenuating inflammatory responses and suppressing the TLR4 and NF-κB signaling pathway (Fan et al., 2016), to improve immune response during sepsis by promoting phagocytosis of macrophages through confining HMGB1 to the nucleus (Miao et al., 2020) and protecting the kidneys of sepsis mice through attenuating inflammation by targeting HMGB1 by upregulating miR-339-5p, which was a cancer-related molecule and could participate in multiple cell processes, such as proliferation, migration, and invasion (Jansson et al., 2015; Mei et al., 2019).

For further exploring the potential mechanism of the effect of XBJI in treating sepsis, we conducted a pathway analysis to CFIG targets. The number of CFIG targets enriched pathways was 182 (p < 0.05) which could cover 81.58% of the pathogenetic genes enriched pathways (Figure 8).

Based on CFIG targets annotations, some signaling pathways which could not be clearly found out in pathogenetic gene annotations and ingredient target annotations were signed out, which had been reported to be closely related with sepsis according to literatures, such as nonalcoholic fatty liver disease (NAFLD) (hsa04932), retrograde endocannabinoid signaling (hsa04723), proteoglycans in cancer (hsa05205), the sphingolipid signaling pathway (hsa04071), and the Ras signaling pathway (hsa04014).

According to some literatures on NAFLD, some immune-related loci associated with NAFLD could exhibit some level of pleiotropy influencing sepsis with genes of CD14, IL-6, MIF, TLR4, and TNF (Sookoian and Pirola, 2019). NAFLD was reported to upregulate 11β-HSD1, 11β-HSD2, and G6PD, thereby enhancing the production of corticosterone, which could contribute to combat sepsis shock (Huang et al., 2020). Retrograde endocannabinoid signaling could effect the secretion of endocannabinoids, which is a lipid-based retrograde messenger with the effect of regulating the strength of neuron communications and regulating the blood–brain integrity in the restriction of systemic inflammation, such as septic encephalopathy (Kasatkina et al., 2021). A latest study showed that a high abundance of miRNA in community-acquired pneumonia (CAP), a major cause of sepsis, revealed a significant overrepresentation for proteoglycans in cancer, demonstrating that some small non–coding RNA might be related to CAP (Khan et al., 2021). Before this study, Sdc1 had been proved to be a member of the syndecans, which belong to the family of type I transmembrane heparan sulfate proteoglycans in vertebrates and are related to sepsis (Holzmann et al., 2018; Ikeda et al., 2018). The results of CFIG target annotations showed that our model in selecting CFIG was useful in uncovering the potential mechanism of XBJI in treating sepsis.

In order to further explore the potential mechanism of XBJI in treating sepsis, we constructed a comprehensive pathway with five signaling pathways, including MAPK (hsa04010), NF-κB (hsa04064), PI3K-Akt (hsa04151), TNF (hsa04668), and TLR (hsa04620) (Figure 9). These five signaling pathways repeated in the signaling pathway of KEGG enrichments of pathogenetic genes, targets of ingredients, and targets of the CFIG, indicating that these four signaling pathways might be important in XBJI in treating sepsis.

In this comprehensive pathway, some proteins could exist in more than two signaling pathways. For example, TAK1, TNF-α, and IL-1β could exist in three signaling pathways: TLR, TNF, and NF-κB; MEK1 could exist in three signaling pathways: TNF, MAPK, and PI3K-Akt. Most of these proteins had been reported to be closely associated to sepsis and anti-inflammatory responses. For example, 1) TAK1: it was proved to play important roles in the inflammatory response mediated by TLR4, which had been regarded as a method in controlling sepsis (Opal et al., 2013; Chen et al., 2015). It was a signal transducer upstream of p38 MAPK and NF-κB, while the inhibition of the interaction between TAK1 and TAB1 could attenuate TAK1-mediated MAPK activation, thereby alleviating sepsis-induced multiple organ dysfunction in mice (Zhang et al., 2020). Some other factors, such as the IKK complex, p38, JNK, and EKR, could also be functionalized and phosphorylated by activated TAK1, NF-κB, and MAPK signaling pathways, thus being activated afterward (Ajibade et al., 2013). Some studies have proved that the activation of TAK1 could lead to the activation of IKK complex—NF-κB and MAPKs, playing as an essential signaling factor of both NF-κB and MAPK signaling pathways (Kawasaki and Kawai, 2014; Seumen et al., 2021). A latest study showed that TAK1 could be targeted by ubiquitin-specific peptidase 18, negatively regulating and inhibiting LPS-induced sepsis (Hu et al., 2021). 2) TNF-α: This is a typical pro-inflammatory factor, which could promote T cells to further produce a large number of inflammatory mediators, which could reduce the body's immune function and destroy the barrier function of various tissues and organs, leading to the spread of systemic inflammatory response (Kamisoglu et al., 2015). It has been considered to be one of the links in regulating apoptosis (Zhao et al., 2015). Some latest studies showed that the expression level of TNF-α was significantly higher in LPS-induced H9c2 cardiomyocytes, suggesting that the expression of NF-κB/TNF-α might be related to inflammatory response and apoptosis of H9c2 cells during sepsis (Zheng et al., 2020). The level of TNF-α was reported to be increased significantly in LPS-stimulated BMDMs (Liu et al., 2020). Comparatively, TNF-α could be reduced by baicalin (MOL002776) in the blood of sepsis mice (Chen et al., 2016) and was reduced by beta-sitosterol [MOL000358 (CS, DG, and HH)] (Kasirzadeh et al., 2021), caffeic acid [MOL000223 (CX, DS, and HH)] (Chen et al., 2019). A Chinese formula, RDN, had been proved to ameliorate LPS-induced acute lung injury by downregulating TNF-α (Yang et al., 2021). 3) IL-1β: This is also a typical pro-inflammatory factor, which could promote T cells to further produce a large number of inflammatory mediators, which could reduce the body's immune function and destroy the barrier function of various tissues and organs, leading to the spread of systemic inflammatory response (Kamisoglu et al., 2015). Comparatively, it was proved to be downregulated by RDN and to ameliorate LPS-induced acute lung injury (Yang et al., 2021). It was indicated to be decreased by CX (Lv et al., 2012) and reported to be reduced by baicalin (MOL002776) in the sera of bacterial infected mice, significantly, thereby improving mouse survival in bacterial sepsis (Li et al., 2017), and to have been reduced by caffeic acid [MOL000223 (CX, DS, and HH)] in LPS-stimulated macrophages (Chen et al., 2019). IL-1β was also reported to be attenuated by paeoniflorin (MOL001924), a critical ingredient of monoterpene glycoside in CS and XBJI (Wang et al., 2020), thereby improving myocardial function in septic mice (Zhai and Guo, 2016), and reported to be inhibited by quercetin (MOL000098), thereby protecting mice from LPS-induced sepsis apoptosis (Wei et al., 2018); it was also reported to be decreased by ferulic acid (CIS) (MOL000389), an ingredient in DG, thereby positively modulating the inflammatory response to septic liver injury (Cao et al., 2021). 4) MEK: this is a key protein in the MEK signaling pathway, of which it could inhibit apoptosis induced by LPS (Chen et al., 2020). The upregulated phosphorylation of MEK1 was reported to stimulate the syntheses of TLR2 and surfactant protein-A (SP-A) in human alveolar epithelial A549 cells (Wu et al., 2011), where TLR2 existed in human pulmonary epithelial cells and participated in the response to lung injury (Droemann et al., 2003; Slevogt et al., 2008), and SP-A could help the lungs defend bacterial infections (LeVine et al., 2000). Some studies showed that MEK1 could participate in the activation of NADPH oxidase, which was a multi-ingredient enzyme with a function of generating oxygen-dependent antimicrobial arsenals, thereby improving the neutrophil immune function during sepsis (Huber-Lang et al., 2002).

Furthermore, 23 proteins could exist in two pathways in the comprehensive pathway, such as AKT, NF-κB, and p38. Most of these proteins were proved to be related with sepsis. The comprehensive pathway proved that the CFIG of XBJI could affect sepsis by multigenes and multichannel biological processes, suggesting that we need to consider the interactions of targets and ingredients when treating sepsis, and also indicating that our model could be used in selecting the CFIG of XBJI in treating sepsis effectively.

For selecting some potential ingredients which could be effectively used in the in vitro experiment, we conducted molecular docking by 59 ingredients with 3D conformer and 160 protein structures responding to 56 genes in comprehensive pathways and obtained 86,475 binding relationships in the docking results (Supplementary Table S8).

In the molecule docking, the effectiveness of the interactions between proteins and ligands could be evaluated by the binding energy, such that a lower value of affinity represented better binding energy (Fu et al., 2018; Elhenawy et al., 2019; Cui et al., 2020). The information of binding relationships divided with binding affinity values showed that all the 59 ingredients of the CFIG could bind the 56 genes of the comprehensive pathways, whose affinities were equal or lower than −6 kcal/mol, indicating that the 160 proteins searched from PDB could well represent the effectiveness of the 56 genes in the comprehensive pathways, and confirming that the ingredients of the CFIG could well target the proteins involved in the comprehensive pathways (Figures 10A–C). Three ingredients used in the in vitro experiments showed good combinations with proteins. Apigenin (MOL000008) could bind the best with protein structure 3zs5 responding to gene MAPK14 with a binding affinity of −10.70 kcal/mol, while baicalein (MOL002714) and luteolin (MOL000006) could bind the best with protein structure 3zs5 responding to gene MAPK14 (with a binding affinity of −10.40 and −10.40 kcal/mol, respectively) (Figure 10D₁–F₂ ).

Among the 86,475 binding relationships, the ingredient MOL000286 could bind the best with protein structure 3ri1 responding to gene FGFR2 with a binding affinity of −11.50 kcal/mol (Supplementary Figure S2A₁–A₂ ), followed by binding relationships of MOL007111-3zs5 (MAPK14, −11.20 kcal/mol), MOL000286-3iw4 (PRKCA, −11.20 kcal/mol), MOL000357-3iw4 (PRKCA, −11.20 kcal/mol), MOL007154-6q7b (EPHA2, −11.10 kcal/mol), MOL007068-3zs5 (MAPK14, −11.10 kcal/mol), and MOL004495-5ikr (PTGS2, −11.10 kcal/mol) (Supplementary Figure S2B₁–G₂ ). We analyzed in-depth the binding relationships whose binding affinity values were equal to or lower than −6.00 kcal/mol and obtained 52,276 bindings, including 56 genes, 59 ingredients, and 160 proteins, and focused on two questions: 1) which ingredient could bind the most number of genes and 2) which genes could bind the most number of ingredients. The results were as follows: 1) a total of 26 ingredients could bind 56 genes, eight ingredients could bind 55 genes, and 3 ingredients could bind 54 genes, indicating that CFIG could bind the genes in the compressive signaling pathway effectively. Among these 26 ingredients related with 56 genes, and according to the average affinities of each ingredient, three ingredients had the highest value of binding affinities and were COM12, ZINC8234227 MOL004495, and ZINC56874950 MOL002764. The average affinities of these three ingredients were −7.87, −7.85, and −7.64 kcal/mol. The optimal docking bindings of these three ingredients were COM12-5f1a (PTGS2, −10.70 kcal/mol), MOL004495-5ikr (PTGS2, −11.10 kcal/mol), and MOL002764-5f19 (PTGS2, −10.90 kcal/mol) (Supplementary Figure SG₁–G₂ ). 2) Gene MAPK14 could bind the largest number of ingredients with the 59 ingredients, followed by genes EPHA2, FGFR1, and KDR with 58 ingredients, whose average affinities were −7.43, −7.86, −7.28, and −7.32 kcal/mol, respectively. The optimal docking bindings of these four genes were MOL007111-3zs5 (MAPK14, −11.20 kcal/mol) (Supplementary Figure SB₁–B₂ ), MOL007154-6q7b (EPHA2, −11.10 kcal/mol) (Supplementary Figure SE₁–E₂ ), MOL004495-5ew8 (FGFR1, −9.80 kcal/mol) (Supplementary Figure SJ₁–J₂ ), MOL000286-3vhe (KDR, −10.30 kcal/mol) (Supplementary Figure SK₁–K₂ ), and MOL004495-3vhe (KDR, −10.30 kcal/mol) (Supplementary Figure SL₁–L₂ ). These above results indicated that the CFIG could effectively bind with the proteins involved in the comprehensive pathway and is worthy to be studied further for effects in treating sepsis. Here, three ingredients, apigenin, baicalein, and luteolin, were selected into in vitro experiments.

LPS-induced RAW264.7 cells model was utilized to assess the potential anti-inflammatory effect of pivotal ingredients in vitro. The results of cell viability showed that apigenin at 1–20 μM, baicalein at 0.5–8 μM, and luteolin at 1–4 μM had no obvious cytotoxicity to RAW264.7 cells (Figures 11A–C). As shown in Figure 11D–F, the level of NO production was markedly increased after LPS treatment, and apigenin (1–20 μM), baicalein (1–4 μM), and luteolin (1–4 μM) significantly decreased the level of NO, which suggests that apigenin, baicalein, and luteolin inhibited NO production in LPS-induced RAW264.7 cells.