Identification of DEGs after weight loss
After standardizing gene sets (Fig. 1), 1011 DEGs (|logFC|> 1, p < 0.05) were screened out from GSE103766, GSE35411, GSE112307, GSE43471, and GSE35710 based on the above method. The results included 513 downregulated and 498 upregulated genes, as shown in the volcano plot (Fig. 2 and Supplementary Table S1). The abscissa in the volcano plot is log2 (fold change) value, and the ordinate is log10 (p-value).
PPI network analysis and identification of hub genes
As shown in Supplementary Fig. S1, the PPI network of DEGs, based on the Search Tool for the Retrieval of Interacting Genes (STRING) database, includes 584 nodes and 1417 edges. Using the MCODE plugin in Cytoscape software, the most significant modules (score = 6.667) were recognized from the PPI network as comprising 27 hub genes, including ACP5, CETP, COL1A1, COL1A2, CSF1, DNMT3B, EED, HIST1H2AI, HIST1H2BB, HIST1H2BD, HIST1H4B, HIST1H4H, HIST2H3C, HP, LCN2, LIPC, LPA, MMP2, MMP7, MMP9, MSR1, MUC1, PLA2G7, SPP1, THBS1, THBS2, and VLDLR (Table 1 and Fig. 3).
GO enrichment analysis of hub genes
An enrichment analysis bubble chart was drawn under GO level 2 classifications using Omicshare tools (Fig. 4 and Supplementary Table S2). As shown in the figure, hub genes were significantly enriched in regulating plasma lipoprotein particle levels, lipid transport, extracellular matrix (ECM) organization, response to reactive oxygen species, and the oxygen-containing compound for biological process (BP). The hub genes were significantly enriched for cell composition (CC) in lipoprotein particles, extracellular regions, ECM, extracellular exosomes, and secretory granules. For molecular function (MF), the hub genes were significantly elevated in lipoprotein particle binding, glycosaminoglycan binding, ECM structural constituents, and peptidase activity.
KEGG pathway enrichment analysis of hub genes
KEGG pathway enrichment analysis showed that the hub genes were primarily enriched in ECM–receptor interaction, cholesterol metabolism, PI3K-Akt, IL-17, and TNF signaling pathways, endocrine resistance, and leukocyte transendothelial migration (Fig. 5 and Supplementary Table S3).
Screening of active ingredients
We converted 27 gene names of the hub genes into protein names that could be recognized through the TCMSP database using the Universal Protein Resource (Uniprot). Moreover, the hub genes can be input in the required format to identify potential herbs with anti-obesity effects from the TCMSP database. After excluding the genes that were not present in the databases or those that had no related ingredients, nine were screened for further research, namely, COL1A1, MMP2, MMP9, SPP1, DNMT3B, MMP7, CETP, COL1A2, and MUC1. These genes corresponded to 16 ingredients [(-)-epigallocatechin-3-gallate (EGCG), arachidonic acid, arctiin, baicalein, beta-carotene, capillarisin, deoxypodophyllotoxin, ellagic acid, fisetin, irisolidone, luteolin, matrine, nobiletin, quercetin, rutaecarpine, tanshinone IIa] showing adequate OB and DL values (OB ≥ 30%, DL ≥ 0.18) (Supplementary Table S4).
Screening and annotation of key herbs
There were 254 herbs with active ingredients in the databases. The top 10 herbs were Aloe, Portulacae Herba, Mori Follum, Silybum Marianum, Phyllanthi Fructus, Pollen Typhae, Ginkgo Semen, Leonuri Herba, Eriobotryae Folium, and Litseae Fructus. These were associated with more DEGs (related genes = 6) and were, therefore, selected as crucial herbs in our study and annotated using Chinese pharmaceutical properties (CMPs), including characters, tastes, and meridian tropisms (Table 2).
Construction of ingredients-targets, herbs-ingredients-targets, and herbs-taste-meridian tropism networks
We screened the key ingredients in treating obesity using an Ingredients-Targets network containing 25 nodes and 27 edges (Fig. 6). The nine orange nodes represent the target genes and 16 green nodes represent the active ingredients. As most genes could be linked (degree = 4), quercetin and EGCG were considered the most critical components in the treatment of obesity.
As shown in Fig. 7a, the Herbs-Ingredients-Targets network containing 24 nodes and 43 edges was constructed to demonstrate the relationship between them: the 10 green nodes represent the key herbs and the six yellow nodes represent the active ingredients in them; the eight blue nodes depict the target genes. By analyzing the network, Phyllanthi Fructus and Portulacae Herba were associated with the most ingredients (degree = 4). Moreover, quercetin was the most frequent active ingredient (degree = 23) found in all herbs. Regarding gene targets, MMP2 was targeted by most ingredients (degree = 5) followed by MMP9 (degree = 4). Other genes were only acted upon by one component (degree = 1).
We also established the Herbs-Taste-Meridian tropism network containing 24 nodes and 40 edges to clarify the distribution of CMPs (Fig. 7b). Five yellow nodes represent tastes and eight purple nodes represent meridian tropisms. To indicate different characters, we presented 10 nodes of herbs having different greens (light green, medium green, and dark green). Regarding characters, cold-cool herbs like Mori Follum were the most frequent (nodes = 7), followed by herbs having calm (nodes = 2) and warm (nodes = 1) characters. In terms of taste, herbs were mostly bitter (edges = 6), followed by sweet (edges = 4), acid (edges = 2), symplectic (edges = 2), and astringent (edges = 2). Regarding meridian tropism, most herbs belonged to the liver meridian (edges = 6), followed by the stomach and lung (edges = 4), large intestine (edges = 2), bladder (edges = 2), kidney (edges = 2), pericardium (edges = 2), spleen (edges = 1), and gallbladder (edges = 1) meridians.