To identify the potential targets of DMDD, the drug target predictions database PharmMapper (http://lilab-ecust.cn/pharmmapper/) was carried out. The PharmMapper database (Liu et al., 2010; Wang et al., 2016; Wang et al., 2017) was designed to identify potential targets for small molecules using pharmacophore mapping approach. DMDD MOL2 format file was uploaded into the web-server and the DMDD potential targets were obtained and sorted by normalized fit score value. Only the Homo sapiens protein targets were selected.

GEPIA2 (Gene Expression Profiling Interactive Analysis 2, http://gepia2.cancer-pku.cn/#index) is a web-based tool to deliver fast and customizable functionalities based on TCGA datasets. 1,078 non small cell lung cancer (NSCLC) samples form TCGA datasets were selected, including 483 lung adenocarcinoma (LUAD) and 59 corresponding normal, 486 lung squamous cell carcinoma (LUSC) and 50 corresponding normal samples. Differentially expressed genes (DEGs) between the NSCLC and normal samples were analyzed using GEPIA2 by LIMMA methods. Using |log2 fold change| ≥ 1 and Q-value < 0.01 as cutoff conditions.

To explore the interaction between DMDD targets or DEGs in lung cancer, the DMDD targets or DEGs were analyzed using STRING database (https://string-db.org/) to generate PPI network. PPI pairs with a combined score >0.4 were extracted. The PPI network was visualized using Cytoscape 3.7.2 software (Shannon et al., 2003) and the most important module was performed using the MCODE plug-in in Cytoscape software. In addition, Metascape web-based tool (https://metascape.org/gp/index.html) (Zhou et al., 2019) was used for functional enrichment analysis (GO analysis and KEGG pathway analysis) of these genes.

Survival information of the DEGs in lung cancer patients were obtained from the Kaplan-Meier Plotter (http://kmplot.com/analysis/), which was established based on the microarray and RNA-seq datasets for several cancer types including lung cancer (Gyorffy et al., 2013). The information of 1,715 lung cancer patients were collected from the Gene Expression Omnibus (GEO, http://www.ncbi.nlm.nih.gov/geo/) (GSE4573, GSE14814, GSE8894, GSE19188, GSE3141, GSE31210, GSE29013 and GSE37745), the Cancer Biomedical Informatics Grid (caBIG, http://cabig.cancer.gov/) caArray and The Cancer Genome Atlas (TCGA, http://cancergenome.nih.gov) NSCLC cohort. Among the 1,715 NSCLC samples, overall survival (OS) data were available for 1,577 samples. Kaplan–Meier analysis was used to plot the OS survival curves. Survival differences were assessed by the log-rank test using the median of DEGs as cutoff value. The hazard ratio (HR) with 95% confidence intervals and log-rank p-values were calculated.

The 3D formats of DMDD was obtained from ZINC15 drug database (http://zinc15.docking.org/) (Sterling and Irwin, 2015). The X-ray crystal structure of DEGs were obtained from the RCSB PDB protein databank (http://www.rcsb.org/). The water molecules and ligands molecules were removed from the crystal structures by Discovery Studio 4.5 Client before the docking simulation began. Docking simulations of DMDD and DEGs were carried out by PyRx software. Interaction energies were calculated for predicting docking positions and selecting the binding pose that had the lowest binding energy (kcal mol⁻¹). Results was visualized and analyzed using Discovery Studio 4.5 Client.

2-Dodecyl-6-Methoxycyclohexa-2, 5-Diene-1,4-Dione (DMDD) was purified from the roots of Averrhoa carambola L. as described in our previous study (Wen et al., 2012). The plant material Averrhoa carambola L. was collected from Lingshan, Guangxi, China and was identified by Prof Maoxiang Lai (Traditional Chinese Medicine Research Institute of Guangxi). A 20 mM stock solution of DMDD was prepared using dimethyl sulfoxide (DMSO).

Human lung cancer cell lines H1975 (ATCC, CRL-5908) were obtained from American Type Culture Collection and maintained using standard media and conditions. PC9 was kindly provided by Dr. Guoan Chen (School of Medicine, Southern University of Science and Technology). H1975 and PC9 were maintained in RPMI 1640 supplemented with 10% FBS and 1% penicillin/streptomycin at 37°C in a 5% CO2 cell culture incubator. Mycoplasma contamination was excluded in these cell lines.

Cell viability was detected by CellTiter 96^® AQueous One Solution Cell Proliferation Assay (Promega) according to manufacturer's instructions. H1975 and PC9 cells (4 × 10³ cells/well) were treated with DMDD at different time and different concentration, and then incubated with CellTiter 96^® AQueous One Solution reagent at 37°C in a humidified with 5% CO₂ for 1h. The absorbance at 490 nm was recorded using an ELISA plate reader. The cell viability rates were calculated by normalizing to the control group.

Human lung cancer cell lines H1975 and PC9 (100 cells/well) were seeded into 12-well plates and cultured overnight. Cells were then treated with DMDD at different concentration (0, 10, 20, 40 µM). After 2 weeks, cells were fixed with 4% paraformaldehyde and stained with 0.1% crystal violet. The number of clones was counted in each well.

H1975 and PC9 (4.5 × 10⁵ cells/well) were seeded into 12-well plates and cultured overnight. Cells were grown to confluency in a monolayer and treated with DMDD at different concentration (0, 10, 20, 40 µM). A scratch was made with a pipette tip to create an incision-like gap. The scratch area was photographed immediately after wounding and at 24 h time points thereafter, and cell migration was quantified and expressed as average percentage of closure of the scratch area.

For Western blot analysis, cell lysates were boiled in sample buffer at 95°C for 5 min. The samples were separated by polyacrylamide gel electrophoresis at 90 V for 2 h and transferred onto polyvinylidene difluoride membranes. After blocking for 1 h with 5% non-fat milk, the membranes were incubated with primary antibodies against human CCNE1 (1:1,000 dilution), E2F1 (1:1,000 dilution), CDK2 (1:1,000 dilution), p21 (1:1,000 dilution), GAPDH (1:1,000 dilution) at 4°C overnight. After incubation with secondary antibody (1:2000 dilution) for 1 h at room temperature, the membranes were developed using enhanced chemiluminescence (ECL), and the signals were detected by MiniChemi 610 Plus (SageCreation Science Co.).

Forty Eight hours after treatment with DMDD at various concentration (0, 5, 20 µM), H1975 and PC9 cells were harvested and fixed with 70% ice-cold ethanol for 2 h at 4°C, washed with PBS, re-suspended in 1 ml of propidium iodide (PI) staining solution (40 μg/ml RNaseA and 10 μg/ml PI in PBS), and then incubated in the dark at room temperature for 30 min. Samples were transfered to the flow cytometer and used to measure the cell cycle by FACScan Flow Cytometry (BD FACSCalibur).

Data were analyzed using GraphPad Prism 5 and SPSS 19.0 software. To identify gene expression patterns, an unsupervised hierarchical cluster analysis using un-centered average linkage was performed using Cluster v3.0 after median—centering genes and arrays. Heat maps were visualized using Tree View software. The other data such as proliferation were evaluated by unpaired Student's t-test. A two-tailed p value <0.05 was considered significant. To determine potential underlying biological processes associated with DMDD potential targets or DEGs in lung cancer, Gene Ontology enrichment analysis was performed using Metascape web-based tool (https://metascape.org/gp/index.html).

We predicted the DMDD potential target proteins and found 60 candidates (Supplementary Table S1). PPI networks of these 60 potential targets was subsequently analyzed. The PPI network contained 59 nodes and 47 edges which indicated that there was a wide range of interactions among these proteins (Figure 1A). MCODE algorithm results showed that MTOR (MCODE Score = 3.47), E2F1 (MCODE Score = 3.47), AR (MCODE Score = 2.7), CCNE1 (MCODE Score = 2.7), CUL1 (MCODE Score = 2.7) and EP300 (MCODE Score = 2.7) were densely connected (Figure 1A). GO enrichment analysis and KEGG pathway analysis were applied to 60 DMDD potential target proteins to further investigate their functions. These proteins were found to be highly involved in specific cellular processes such as protein binding (p = 0.00037), regulation of response to stimulus (p = 0.000047) (Figure 1B). Pathway enrichment showed that various pathways especially pathways in cancer (p = 0.00016), cell cycle (p = 0.015) and small cell lung cancer (p = 0.049) were involved in DMDD potential target proteins (Figure 1C). These data suggested that cancer related genes and pathways were abundant in DMDD potential targets.

We analyzed the DEGs in NSCLC dataset of TCGA and resolved 3,545 candidates, among which 787 were up-regulated and 2,758 were down-regulated in tumor compared with normal tissues (Supplementary Table S2). To further analyses upregulated and downregulated DEGs in lung cancer, we performed KEGG pathway enrichment analysis by using Metascape respectively and use p < 0.05 as the cut-off criterion. For upregulated DEGs, the top 3 KEGG pathway were cell cycle, p53 signaling pathway and biosynthesis of amino acids (Figure 2A). Among them, cell cycle and pathways in cancer overlapped with DMDD potential targets KEGG pathway (Figure 2A). For downregulated DEGs, the top 3 relevant KEGG pathway were vascular smooth muscle contraction, focal adhesion and Ras signaling pathway (Figure 2B). Eight overlapping genes were found between potential therapeutic targets of DMDD and lung cancer related DEGs (Figure 2C). The up regulated genes were FUT8 (FC = 2.10 in LUSC, FC = 2.81 in LUAD, q < 0.01), CCNE1 (FC = 4.46 in LUSC, FC = 2.29 in LUAD, q < 0.01), E2F1 (FC = 2.88 in LUSC, FC = 2.13 in LUAD, q < 0.01), ARHGAP11A (FC = 4.40 in LUSC, FC = 2.02 in LUAD, q < 0.01), and the down regulated genes were ARHGEF7 (FC = 0.40 in LUSC, FC = 0.40 in LUAD, q < 0.01), RUNX1T1 (FC = 0.28 in LUSC, FC = 0.34 in LUAD, q < 0.01), ACADVL (FC = 0.26 in LUSC, FC = 0.33 in LUAD, q < 0.01), FES (FC = 0.17 in LUSC, FC = 0.31 in LUAD, q < 0.01) (Figures 2D,E).

To explore the prognostic value of these 8 DEGs which were also the potential targets of DMDD in lung cancer, the Kaplan Meier plotter online database (Nagy et al., 2018) was used. The results showed that CCNE1, E2F1, ARHGAP11A, RUNX1T1 and FES were significantly associated with patient survival in lung cancer (n = 1925). We found that high expression of CCNE1 (low vs. high: 96.0 vs. 46.2, p < 0.01) (Figure 3A), E2F1 (low vs. high: 91.0 vs. 50.0, p < 0.01) (Figure 3B) and ARHGAP11A (low vs. high: 93.0 vs. 47.9, p < 0.01) (Figure 3C) and low expression of RUNX1T1 (low vs. high: 60.7 vs. 104.0, p < 0.01) (Figure 3D), FES (low vs. high: 57.0 vs. 79.9, p < 0.01) (Figure 3E) were significantly related to poor patient survival (p < 0.01). Then we analyzed the correlation between transcription level of CCNE1, E2F1, ARHGAP11A, RUNX1T1, FES and tumor stage in TCGA NSCLC cohort by GEPIA 2 online website. The results demonstrated that the expression levels of CCNE1, E2F1, ARHGAP11A, RUNX1T1 and FES displayed significant correlation with the tumor stage in patients with NSCLC (Supplementary Figures S5A–E). These results suggest that CCNE1, E2F1, ARHGAP11A, RUNX1T1 and FES were significantly associated with the tumor stage and can be used as prognostic biomarkers for lung cancer. It's worth noting that E2F1 and CCNE1 had high MCODE score in DMDD targets PPI and both of them were demonstrated to be associated with cell cycle and pathways in cancer and small cell lung cancer (Figure 1A).

To explore the transcription levels of CCNE1, E2F1 in other types of cancer, we profiled the tissue-wise expression of these genes in different cancer types by GEPIA 2 based on TCGA datasets. We found that high expression (log2FC > 1, q < 0.01) of CCNE1 was presented in 18 of 31 cancer types. The top 3 were uterine carcinosarcoma (FC = 45.16, q < 0.01), lymphoid neoplasm diffuse large B-cell lymphoma (FC = 43.90, q < 0.01) and uterine corpus endometrial carcinoma (FC = 30.08, q < 0.01) (Supplementary Figure S1). E2F1 was significantly high expressed (log2FC > 1, q < 0.01) in 22 of 31 cancer types in TCGA data set. Top 3 are lymphoid neoplasm diffuse large B-cell lymphoma (FC = 21.26, q < 0.01), liver hepatocellular carcinoma (FC = 19.61, q < 0.01), cervical squamous cell carcinoma and endocervical adenocarcinoma (FC = 18.42, q < 0.01) (Supplementary Figure S2). Then the correlation analysis of DNA methylation level and transcript level of CCNE1 and E2F1 in 370 lung squamous cell carcinoma and 456 lung adenocarcinoma tissues from TCGA cohort was performed to identify the possible mechanism underlying CCNE1 and E2F1 DNA methylation in NSCLC by cBioPorta. The results showed that the level of DNA methylation was negatively related to the decreased CCNE1 and E2F1 transcription in 370 lung squamous cell carcinoma tissues (p < 0.05, Supplementary Figures S3A and S4A) and 456 lung adenocarcinoma tissues (p < 0.05, Supplementary Figures S3B and S4B). Indicated that decreased methylation levels of CCNE1 and E2F1 might associated with the upregulation of CCNE1 and E2F1 in NSCLC tissues.

These results suggest that CCNE1, E2F1 are involved in the development of various cancers and are closely related to cell cycle and pathways in cancer, etc. Furthermore, decreased methylation levels might associated with the upregulation of CCNE1 and E2F1 in NSCLC tissues. Based on the above analysis, we predict that CCNE1 and E2F1 holds the potential to be diagnostic/prognostic markers of NSCLC and DMDD might inhibit the development of lung cancer by targeting CCNE1 and E2F1.

A molecular docking model was constructed to further explore the mechanism of interaction between DMDD and its potential target proteins. Processing of the DMDD included energy minimized. The refinement of structure of DMDD was used for the dock. PyRx was used for the docking studies. The docked conformation corresponding to the lowest binding energy was selected as the most probable binding conformation. We found that DMDD could enter into the N-terminal cyclin box of CCNE1 (PDB code: 1W98) which leads to pCDK2 kinase activation (Honda et al., 2005), and stayed in the binding pocket surrounded by key residues (His147, Ile345, Val337 and Lys145) (Figure 4A). DMDD could bound with the marked-box domain of E2F1 (PDB code: 2AZE) which leads to Rb and E2F1 binding (Hallstrom and Nevins, 2003), and stayed in the binding pocket surrounded by key residues (Met263, Pro292, Ile293 and Val295) (Figure 4B). Our investigations revealed that DMDD exhibited significant binding affinities within the active site of CCNE1 and E2F1. DMDD may be used as an effective anti-lung cancer drugs by targeting CCNE1 and E2F1.