As shown in Figure 1 , the data were downloaded from the GEO database (https://www.ncbi.nlm.nih.gov/geo/): 23 tissues (10 hippocampus tissues with AD and 13 control hippocampus tissues) in GSE5281 (GPL570 platform) and 30 tissues (22 hippocampus tissues with AD and 8 control hippocampus tissues) in GSE28146 (GPL570 platform). The 38 rosacea tissues and 20 control tissues in GSE65914 were also downloaded from GEO. The data were standardized and the batch effects were removed using “limma” and “sva” ( Figures S1A , S2A ). Differentially expressed genes (DEGs) between the disease and control tissues were identified using the R “limma” package with the threshold of |logFC| >0.5 and FDR <0.05.

The Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were carried out using R “clusterProfiler”, “enrichplot”, and “ggplot2” packages.

The protein–protein interaction (PPI) network was constructed using the STRING database (https://string-db.org/) with a combined score >0.7. “cytoHubba” was used to calculate the weight of each gene and explore the hub genes of the PPI network, and “MCODE” was used to identify the representative modules using Cytoscape software 3.8.1 (https://cytoscape.org).

The TF–targets were downloaded from the TRRUST database (https://www.grnpedia.org/trrust/). The differently expressed TFs in AD and rosacea, which could regulate DEGs in AD/rosacea, were identified as key TFs. The differently expressed TFs and target genes were used to construct the TF–target networks using Cytoscape software.

To prioritize the list of drugs for AD/rosacea, key TF and target genes were used for potential candidate drugs using the DGIbd database (24, 25), and then the Connectivity Map (CMap) database was also used to verify the candidate drugs for AD and rosacea (26).

The potential pharmacological targets of MLT were obtained from accessible online tools, including the Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP), SwissTargetPrediction, and TargetNet (27). The candidate genes were corrected and identified using the UniProt database (28).

The molecular structure of MLT (CID_4091) was downloaded from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/). The protein structures of MMP9_6ESM were downloaded from the PDB database (https://www.rcsb.org/). The maestro software was used for molecular docking analysis as previously described.

The synthetic peptide LL37 (LLGDFFRKSKEKIGKEFKRIVQRIKDFLRNLVPRTES) used in this study was purchased from Sangon Biotech (Shanghai, China). All peptides were HPLC purified (purity >95%) and characterized by mass spectrometry. Melatonin was purchased from Sigma Company (USA). Human recombinant human TNF-α was purchased from PeproTech (USA).

Specific pathogen-free female BALB/c mice (6 to 7 weeks old; weight, 20–25 g) were purchased from SLAC Laboratory Animal Co., Ltd. (Shanghai, China). Before the experiment, all the animals were acclimated to experimental conditions for 1 week. The mouse model of rosacea was constructed as described before (29). All the 8-week-old BALB/c mice (7 to 8 weeks old, weight, 20–25g) were divided into four groups (control group, LL37 group, MLT group, LL37 + MLT group) randomly. Mice were sacrificed 12 h after the final injection, and mice dorsal lesion skins were collected. The inflammation severity of the rosacea mice model was evaluated according to the area of skin lesions, erythema, and thickness (30), and the skin tissues were divided into three parts for RNA extraction, HE staining, and immunofluorescence staining, respectively. Once the experiments were not to be carried out immediately, tissues were stored at −80°C until use. All mice were housed in sterile conditions at 23°C ± 2°C (12 h light/dark). All experimental protocols followed the guidelines of animal experimentation and were approved by the Ethical Committee of Xiangya Hospital of Central South University (Approval No. 201611610).

Fresh mouse dorsal skin tissues were fixed in 4% paraformaldehyde (PFA), embedded in paraffin, and sectioned at 6 µm. Subsequently, hematoxylin and eosin (H&E) staining was performed on paraffin sections for observation under a light microscope (Olympus, Japan), and the number of dermis-infiltrating cells was then averaged and assessed statistically.

Human immortalized keratinocyte (HaCaT) cells were cultured in complete calcium-free DMEM media (Gibco, USA) containing 10% fetal bovine serum (FBS) (Gibco, USA), 1% L-glutamine (Gibco, USA), and 1% penicillin/streptomycin (Gibco, USA), at 37°C in 5% CO₂ in an incubator. When HaCaT cells reached 70% confluence (logarithmic phase HaCaT cells were inoculated into a 24-well plate with cell-climbing slices at the concentration of 1 × 10⁴/well, if HaCaT cells were prepared for immunofluorescence experiments), the culture medium was replaced with serum-free calcium-containing DMEM medium. After HaCaT cells were starved for 12 h, they were then treated with melatonin (1 mM) for 24 h and stimulated with 8 μM LL37 or 100 ng/ml TNF-α for 12 h, and then RNA extraction or immunofluorescence was performed. For the immunoblot experiments, after HaCaT cells reached 70% confluence, they were then treated with melatonin (1 mM) for 6 h and then stimulated with TNF-α (100 ng/ml) for 15 min, and then the cell protein was collected (the other conditions were the same as above).

Human umbilical vein endothelial (HUVEC) cells were cultured in complete RPMI 1640 medium (Thermo Fisher Scientific, USA) containing 10% FBS (Gibco, USA) and 1% penicillin/streptomycin (Gibco, USA) in an incubator (37°C, 5% CO₂). For the detailed research methods of chemotaxis and migration assay, refer to the following specific section.

Total RNA in cells and tissue samples was extracted using the standard RNA extraction method with TRIzol (Invitrogen Life Technologies, USA). After a NanoDrop spectrophotometer determined RNA concentration, reverse transcription was performed on 2 μg RNA, using the Maxima H Minus First-Strand cDNA Synthesis Kit with dsDNase (Thermo Scientific, K1682, USA) to synthesize the cDNA strand, and gene expression was analyzed using an Applied Biosystems ® 7500 machine (Life Technologies, USA) with the qPCR SYBR Green Master Mix (Vazyme Biotech Co., Ltd., Nanjing, China). The primers are shown in Table S11 .

Fresh mouse dorsal skin tissues were embedded in optimal cutting temperature compound and 6- to 8-μm frozen sections were cut. Then, the frozen tissues or HaCaT cells were fixed in 4% PFA for 15 min. Afterwards, the samples were permeabilized and blocked using blocking buffer (0.2% Triton-X/5% donkey serum) for 1 h. Next, the skin sections were incubated with different antibodies: rat anti-mouse CD4 antibody (1:100), rat anti-mouse CD31 antibody (1:100), and rat anti-mouse F4/80 antibody (1:100), and these antibodies were purchased from eBioscience. HaCaT cells were incubated with rabbit anti-phospho-NF-kB p65 antibody (1:200, Cell Signaling, USA) at 4°C overnight. Anti-rat Alexa 488 (1:200; Invitrogen, USA) and anti-rabbit Alexa 594 (1:200; Invitrogen, USA) were used as the secondary antibody for staining samples 1 h at room temperature. Finally, all samples were stained with 4′,6-diamidino-2-phenylindole (DAPI) for 5 min to visualize nuclei. Fluorescence images were acquired under a fluorescence microscope (Zeiss, Germany).

Proteins of cells were extracted with RIPA lysis buffer containing protease and phosphatase inhibitor cocktail. Then, the protein concentration was measured by BCA protein assay kits (Thermo Fisher Scientific, USA). Twenty micrograms of total protein was resolved on 10% SDS polyacrylamide gels and transferred onto a polyvinylidene fluoride (PVDF) membrane. After transferring, membranes were blocked with 5% non-fat milk and then the PVDF was incubated with the primary antibodies (rabbit anti-p65, 1:1,000; rabbit anti-phospho-p65, 1:1,000; and rabbit anti-GAPDH, 1:5,000) at 4°C overnight. After washing with TBST, the membranes were incubated with secondary horseradish peroxidase-conjugated antibodies (1:5,000, room temperature for 1 h). Finally, the immunoreactive bands were visualized using a chemiluminescent HRP substrate (Millipore, USA) and imaged using a ChemiDoc^TM XRS+ System (Bio-Rad, CA, USA). GAPDH expression was used as the endogenous control.

HUVEC chemotaxis was performed with transwell chambers (8 μm, Millipore, Billerica, MA, USA). In a 24-well culture plate, after being treated with 1 mM MLT or equal volumes of the vehicle solution for 24 h, HUVEC cells were seeded into each upper chamber at a density of 2 × 10⁴ cells with 100 μl serum-free RPMI 1640 medium. Then, the plate was placed in an incubator (37°C, 5% CO₂). Non-migratory cells on the upper surface of the membrane were wiped off gently with a cotton swab after the following 24 h. Afterwards, the chamber was fixed with 4% paraformaldehyde for 15 min, and 0.1% crystal violet solution was used to stain the cells on the lower layer for 30 min. Finally, the number of invaded cells was counted under a light microscope (Olympus, Japan).

The HUVEC suspension with a 6 × 10⁵/ml density was inoculated on a six-well plate at 2 ml per well. At a confluency of 95%, the 200-μl pipetting tip was drawn vertically on the culture plate, and the non-adherent cells were rinsed with PBS three times. Then, the images were captured by using an optical microscope (Olympus, Japan) to determine the distance of scratch for 0 h. Following HUVEC incubation with RPMI 1640 medium with 3% FBS co-incubated with 1 mM melatonin or its vehicle in an incubator at 37°C for 24 h, photographs were retaken under the same conditions. The measurement tool of Image-Pro Plus software was used to measure the distance of the cell scratch boundary before and after treatment, and the distance before treatment was subtracted from the distance after treatment, which was the migration distance of the cells at 24 h.

All data were analyzed with SPSS 17.0 statistical software and were presented as means ± SEM. One-way ANOVA analyzed the data between groups, and the LSD method was used for pairwise comparison. *, P < 0.05; **, P < 0.01; and ***, P < 0.001 are considered significant.

After integrated bioinformatics analyses for GSE5281 and GSE28146, a total of 5,091 DEGs were identified in AD tissues compared with normal tissues ( Figure S1B ). The GO analysis of DEGs was mainly related to cellular respiration, ATP metabolic process, and ATP metabolic process ( Figure S1C and Table S1 ). Furthermore, KEGG enrichment analysis disclosed that these DEGs were significantly associated with pathways of neurodegeneration-multiple diseases, tight junction, virus infection, and so on ( Figure S1D and Table S2 ).

Next, we constructed the PPI network and used “cytoHubba” to explore the hub genes ( Figure S2 ). The top 5 representative modules were extracted from the PPI network using “MCODE”, and the module genes were enriched in mRNA splicing, clathrin-mediated endocytosis, G2/M transition, translation, G alpha (q) signaling events, and so on ( Table 1 ).

For rosacea, 3,016 DEGs were identified in rosacea tissues compared with normal tissues ( Figure S3B ). GO enrichment indicated that rosacea was associated with immune-related biological processes ( Figure S3C and Table S3 ). KEGG enrichment results showed immune-related pathways, the PPAR pathway, virus infection, bacterial infections, metabolism-related pathways, and so on ( Figure S3D and Table S4 ). “cytoHubba” was used to explore the hub genes of the PPI network ( Figure S4 ), and the top 5 representative modules were extracted from the PPI network using “MCODE” ( Figure S4 ). The module genes were enriched in G alpha (i) signaling events, interferon signaling, neutrophil degranulation, cornification, neutrophil degranulation, and so on ( Table 2 ).

To reveal the shared regulatory network, 747 overlapped DEGs (ol-DEGs) were detected in AD and rosacea ( Figure 2A ). GO enrichment analysis showed that the ol-DEGs were related to immune-related processes and lipid metabolic processes ( Figure 2B ). The KEGG enrichment results showed that ol-DEGs were related to cytokine/chemokine pathways, infectious/inflammatory disease, and metabolism-related pathways ( Figure 2C ). Consistent with these results, inflammation-, metabolism-, and apoptosis-related pathways were enriched using the Metascape database ( Figure 2D ). Moreover, the enrichment analysis showed that ol-DEGs were related to inflammation-, vascular-, and infection-related diseases in DisGeNET ( Figure 2E ). Transcription factor–target analysis showed that many ol-DEGs were regulated by TFs in the TRRUST database ( Figure 2F ), indicating that the TF regulatory network may play an essential role in the progression of AD and rosacea.

To further reveal the TF regulatory network in both AD and rosacea, we analyzed the differently expressed TFs with potential targets differently expressed in AD and rosacea. Twenty-four hub TFs are identified in Figure 3A . The TF regulatory network in AD and rosacea is shown in Figure 3B and Table S5 . The TFs and targets were enriched in inflammatory-, blood vessel development-, and apoptosis-related signaling pathways ( Figure 3C ). In Figure 3D , four models were observed to be enriched in atherosclerosis-, IL-17-, proteoglycan-, and inflammatory response-related signaling pathways ( Figure 3D and Table 3 ). These results indicated that inflammatory- and blood vessel-related pathways are crucial during the pathogenesis of AD and rosacea.

Next, we analyzed the potential drugs for AD and rosacea. Using the DGIbd database (https://dgidb.org/), 12 of the 37 hub genes were observed as targets of the 113 predicted drugs ( Table S6 ). We compared the 74 drugs for AD, the 44 drugs for rosacea, and the 113 predicted drugs for hub genes. Three drugs (aspirin, thalidomide, and hydroxychloroquine) overlapped in AD and rosacea, and two (aspirin and thalidomide) of these were observed in the predicted drugs ( Figure 3E ), proving that the candidate drugs for AD/rosacea have high credibility. Moreover, three predicted drugs (MLT, olanzapine, and citalopram) have been used for AD treatment. The Sankey diagram revealed the correlation between disease, drugs, and targets ( Figure 3F ). These results indicated that the three predicted drugs (MLT, olanzapine, and citalopram) could be effective therapeutic strategies for rosacea. Among these potential drugs, MLT was co-associated both with retinoid-related orphan receptor alpha (RORA) as well as IFN-γ. MLT was predicted as a candidate drug using the CMap database ( Table S7 ). So, MLT was selected for further function assay.

Five hundred twenty-nine pharmacological targets of MLT were predicted using the TCMSP, TargetNet, and SwissTargetPrediction databases, and then the repetitive genes were deleted using the UniProt database. An overlap of 128 AD/rosacea TF–targets with MLT pharmacological targets identified 19 intersection genes of MLT against AD/rosacea ( Figure 4A and Table S8 ). GO analyses of the 19 genes indicated that MLT could affect the regulation of inflammatory response, collagen metabolic process, response to oxygen levels, vascular-associated smooth muscle cell proliferation, and so on ( Figure 4B and Table S9 ). Additionally, 51 KEGG pathways were significantly enriched (P-adjusted < 0.05), including the AGE-RAGE signaling pathway in diabetic complications, lipid metabolism and atherosclerosis, NF-kappa B signaling pathway, IL-17 signaling pathway, PPAR signaling pathway, and TNF signaling pathway ( Figure 4B and Table S10 ). Next, the PPI network of the 19 intersection targets was analyzed using STRING ( Figure 4C ), cytoHubba was used to identify eight core gene targets, namely, CCND1, EGFR, ICAM1, MMP2, MMP9, PTGS2, SERPINE1, and TNF ( Figure 4C ).

Next, molecular docking was performed to identify the possible binding between MLT and eight core targets. The result showed that MLT could bind to CCND1, EGFR, ICAM1, MMP9, PTGS2, and SERPINE1 with docking scores of −5.86, −7.16, −6.02, −5.95, −6.52, and −6.81, respectively ( Figure 4D ).

Studies showed that keratinocyte-mediated secretion of inflammatory chemokines and cytokines plays a pivotal role in the progression of rosacea (29). Here, we explored the effects of MLT on chemokine and cytokine expression in LL37-treated HaCaT cells. The results showed that 1 mM MLT significantly reduced LL37-induced CCL2, CCL20, MMP9, KLK5, IL-6, IL-8 and VEGF-c expression ( Figure 7A ). What is more, the expression of NF-κB downstream genes (IL-1α and IL-1β) was significantly repressed by MLT. To further assess the precise anti-inflammatory mechanism of MLT, we stimulated the HaCaT cells with TNF-α (100 ng/ml), one of the most potent inducers of NF-κB activation (36). We found that TNF-α induced the mRNA levels of chemokines and cytokines, including CCL2, CCL20, IL-8, KLK5, TGF-β1, TLR2, IL-1α, and IL-1β, which were abrogated by MLT treatment obviously ( Figure 7B ). Moreover, immunofluorescence results also revealed the inhibition of MLT on TNF-α-induced p65 translocation ( Figures 7C, D ) and phosphorylation of p65/NF-κB ( Figure 7E ). Collectively, these data suggest that MLT ameliorated HaCaT-mediated inflammation partly by modulating NF-κB signaling.

We next analyzed the effects of MLT on angiogenesis in rosacea using immunostaining. As shown in Figures 8A, B , the number of CD31⁺ vessels was significantly reduced by MLT treatment. MLT suppressed LL37-induced HUVEC chemotaxis ( Figures 8C, D ) and migration ( Figures 8E, F ). Collectively, these data provide novel evidence that MLT may be a promising therapeutic strategy for vascular dysfunction in rosacea.