Information of expression profiling dataBased on the inclusion criteria, four datasets (GSE32924, GSE107361, GSE121212, and GSE230200) were finally selected and their gene expression profiles were acquired from the GEO database. With the criteria of P value < 0.05 and |log2 FC| > 1.0, the following DEGs were obtained: 1739 DEGs from GSE32924, with 782 upregulated genes and 957 downregulated genes; 2440 DEGs from GSE107361, with 1177 upregulated genes and 1263 downregulated genes; 1848 DEGs from GSE121212, with 1031 upregulated genes and 817 downregulated genes; 1631 DEGs from GSE230200, with 879 upregulated genes and 752 downregulated genes. The volcano plots of the DEGs are shown in Fig. 1A, and the top 25 significant upregulated and downregulated genes were represented by heatmaps in Fig. 1B. In addition, the PCA plots demonstrated the separation between the DEGs of AD and healthy tissues (Fig. 1C). Therefore, the identification of co-DEGs may provide important data for the diagnosis of AD.Fig. 1Determination of the DEGs. (A) Volcano plots of the DEGs. The upregulated DEGs (red dots) and downregulated DEGs (blue dots) of each dataset were identified with the use of criteria of P value < 0.05 and |log2FC|> 1.0. (B) Expression heatmaps of the top 50 DEGs in each dataset, as determined based upon P value. (C) PCA plots of DEGs in each dataset. DEGs differentially expressed genes, PCA principal component analysis, HC healthy control, AD atopic dermatitis.Identification of co-DEGs and functional enrichment analysisTo elucidate the correlation of DEGs in the four datasets, a total of 146 co-DEGs were identified from the intersection of Venn diagram (Fig. 2A). GO and KEGG pathway enrichment analyses were used to investigate the biological functions and pathways associated with the co-DEGs. Biological processes, cellular components and molecular functions are included in GO annotation enrichment terms. The most highly enriched cellular components were extracellular matrix and external encapsulating structure (Fig. 2B). With regards to biological processes and molecular functions, co-DEGs were substantially enhanced in cytokine-mediated signaling pathway and calcium ion binding, respectively (Fig. 2C,D). The KEGG enrichment analysis revealed that co-DEGs were mainly associated with cytokine–cytokine receptor interaction, viral protein interaction with cytokine and cytokine receptor, JAK-STAT signaling pathway, and IL-17 signaling pathway (Fig. 2E). These data suggested that chemokines and cytokines are jointly involved in the development of AD.Fig. 2Determination of co-DEGs and functional enrichment analysis. (A) Venn diagram illustrating the co-DEGs screened from the intersection of the DEGs in four datasets. (B–D) GO functional analysis showing enrichment of co-DEGs in cellular component, biological process, and molecular function. (E) KEGG pathway enrichment analysis of co-DEGs. Co-DEGs common differentially expressed genes, GO gene ontology, KEGG kyoto encyclopedia of genes and genomes.PPI network construction and module analysisThe PPI network of the co-DEGs was obtained from STRING and analyzed by Cytoscape, which contained 93 nodes and 292 interaction pairs (Fig. 3A). Four closely connected gene modules (module 1, score = 13.18; module 2, score = 7.00; module 3, score = 5.00; module 4, score = 4.00) were identified through MCODE plug-in of Cytoscape, including 34 nodes and 149 interaction pairs (Fig. 3B–E). The GO enrichment analysis showed that these genes were predominantly related to cytokine-mediated signaling pathway, cellular response to cytokine stimulus, and innate immune response (Fig. 3F). The KEGG pathway enrichment analysis showed that they were mainly involved in cytokine–cytokine receptor interaction, viral protein interaction with cytokine and cytokine receptor, and chemokine signaling pathway (Fig. 3G). The results suggested the essential role of inflammatory and immunological response in AD.Fig. 3PPI network, significant gene modules and enrichment analysis of the modular genes. (A) The PPI network of co-DEGs. (B–E) Four significant gene clustering modules. (F,G) GO and KEGG enrichment analysis of the modular genes. PPI protein–protein interaction, co-DEGs common differentially expressed genes, GO gene ontology, KEGG kyoto encyclopedia of genes and genomes.Identification and validation of hub genesThrough the eight algorithms of plug-in cytoHubba, we calculated the top 20 hub genes and identified eight common hub genes, including CCL2, CCR7, GZMB, IL7R, CD274, IL10RA, IRF7, and CCL22 (Fig. 4A). Subsequently, we analyzed the networks and functions of these genes via GeneMANIA. These genes exhibited a complex PPI network with a co-expression of 75.48%, shared protein domains of 16.43%, and co-localization of 8.09% (Fig. 4B). GO analysis showed that these genes are mainly involved in response to cytokine, cytokine-mediated signaling pathway, and response to biotic stimulus (Fig. 4C). In addition, KEGG pathway analysis showed that they are mainly involved in cytokine–cytokine receptor interaction, viral protein interaction with cytokine and cytokine receptor, chemokine signaling pathway, and JAK-STAT signaling pathway (Fig. 4D). To verify the reliability of these genes, GSE130588 and GSE16161 were employed to validate common hub genes. The expression levels of seven hub genes (CCL2, CCR7, GZMB, IL7R, CD274, IRF7, and CCL22) were significantly up-regulated in AD patients compared with healthy controls in the two validation datasets (Fig. 5A,B). These hub genes were further subjected to ROC analysis. Hub genes with AUC values > 0.70 suggested that all hub genes had diagnostic value in AD (Figs. 6 and 7). Among these genes, CCR7 exhibited the best specificity and sensitivity for the diagnosis of AD in the GSE130588 dataset (Fig. 6B), while CD274, IRF7, and CCL22 proved to have the highest specificity and sensitivity for AD diagnosis in the GSE16161 dataset (Fig. 7E–G).Fig. 4Determination and co-expression network of common hub genes. (A) Venn diagram showed that eight algorithms have screened out common hub genes. (B) Common hub genes and their co-expression genes were analyzed via GeneMANIA. (C,D) GO and KEGG enrichment analysis of the common hub genes. The outermost circle on the right represented terms, and the inner circle on the right represented the significant P value of the corresponding pathway of the gene. GO gene ontology, KEGG kyoto encyclopedia of genes and genomes.Fig. 5The relative expression levels of hub genes were validated using (A) GSE130588 and (B) GSE16161. HC healthy control, AD atopic dermatitis.Fig. 6The diagnostic effectiveness of hub genes ((A) CCL2, (B) CCR7, (C) GZMB, (D) IL7R, (E) CD274, (F) IRF7, and (G) CCL22) was validated using GSE130588.Fig. 7The diagnostic effectiveness of hub genes ((A) CCL2, (B) CCR7, (C) GZMB, (D) IL7R, (E) CD274, (F) IRF7, and (G) CCL22) was validated using GSE16161.Prediction of miRNAs and TFsTo gain insight into the relationship between miRNAs and TFs and hub genes during transcriptional repression or abrogation of protein translation, we utilized Cytoscape to develop gene regulatory networks. Based on the TarBase database, we found that eight miRNA molecules may regulate the expression of hub genes. The hub genes and their associated regulatory miRNAs were shown in Fig. 8A. The seven candidate miRNAs were listed in Table 1, and hsa-mir-24-3p, hsa-mir-146a-5p, and hsa-let-7a-5p were found to be more closely associated with AD. We further predicted targeted TFs of the identified hub genes by applying the JASPAR database. A total of 17 nodes and 28 interaction pairs were obtained in the TF-target gene network. FOXC1 was predicted to regulate CCL2, CCR7, GZMB, and CD274 (Fig. 8B). Subsequently, we performed a correlation analysis between FOXC1 and its targeted hub genes. The results indicated that the expression level of FOXC1 was negatively correlated with those of CCL2, GZMB, and CD274 (Fig. 8C). Therefore, these key miRNAs and TFs are crucial for understanding the physiopathological processes and molecular mechanisms of AD.Fig. 8Key miRNAs and TFs regulatory network with hub genes. (A) MiRNA regulatory network. MiRNAs were marked in yellow, and the hub genes were marked in red. (B) TF regulatory network. TFs were marked in yellow, and the hub genes were marked in red. (C) Correlation analysis between FOXC1 and CCL2, GZMB, and CD274. MiRNAs microRNA, TFs transcription factors.Table 1 Candidate miRNAs (Degree ≥ 3) targeting hub genes in AD.Analysis of immune cell infiltrationTo explore immune cell infiltration between healthy controls and AD patients, we employed CIBERSORT to evaluate the immune cell infiltration. The constitutions of 22 types of immune cells in each sample were exhibited in Fig. 9A. Compared with healthy controls, there were more T cells CD4 naive, T cells CD4 memory activated, T cells gamma delta, NK cells resting, dendritic cells resting, and dendritic cells activated in AD patients, but fewer B cells naive, NK cells activated, macrophages M0, and mast cells resting (Fig. 9B). Correlation analysis among the 22 types of immune cells indicated that T cells CD4 memory activated was significantly positive correlation with dendritic cells activated (R = 0.55), and mast cells resting was significantly negative correlation with dendritic cells activated (R = − 0.68) (Fig. 9C). These findings suggested that active specific inflammatory cells have positive implications for the evolution of AD pathogenesis process.Fig. 9Immune cell infiltration analysis. (A) Relative fraction of 22 sub-populations of immune cells. (B) The differences of 22 sub-populations of immune cells between the HC and AD samples. (C) Correlation between 22 sub-populations of immune cells. HC healthy control, AD atopic dermatitis. *P value < 0.05, **P value < 0.01, ***P value < 0.001, and ****P value < 0.0001.Relationship between identified hub genes and immune cellsTo explore the relationship between the hub genes and 22 types of immune cells, we conducted a correlation analysis to reveal their interactions and potential synergistic effects. As shown in Fig. 10, the expression levels of CCL2, CCR7, GZMB, CD274, and CCL22 were significantly positively correlated with the number of dendritic cells activated (R > 0.6), with CCL22 exhibiting the most significant correlation (R = 0.84). Additionally, the expression levels of CCL2, GZMB, CD274, and CCL22 were significantly negatively correlated with the number of mast cells resting (R < − 0.6). These results may provide a more detailed understanding of AD pathogenesis.Fig. 10Relationship between hub genes ((A) CCL2, (B) CCR7, (C) GZMB, (D) IL7R, (E) CD274, (F) IRF7, and (G) CCL22) and immune cells.
