Brucea javanica seeds were purchased from Guangxi Zhuang Autonomous Region, People’s Republic of China, and authenticated by Prof. Guo-Yue Zhong (Jiangxi University of Chinese Medicine). BD and BH were extracted from the seeds of B. javanica and identified on the basis of nuclear magnetic resonance (NMR) data (the separation method is described in detail in the Supplementary Material ); their purity was ≥95% based on high-performance liquid chromatography (HPLC) analysis. Gefitinib was provided by Meilun (ZD1839, Dalian, China). The primary antibodies NOTCH3 (mouse), β-catenin (mouse), EGFR, phospho-EGFR (Tyr1068), FOXO3a (mouse), CBP (Lys1535)/p300 (Lys1499), and β-actin were from Cell Signaling Technology (CST, Danvers, MA, USA). Bax, Bcl-2, caspase-3, and cleaved caspase-3 were provided by Santa Cruz Biotechnology (SCBT, Dallas, TX, USA). The secondary antibodies were HRP-conjugated anti-rabbit IgG and anti-mouse IgG (Abcam, Cambridge, UK).

The human NSCLC cell lines (A549, PC-9, and PC-9/GR) and other cell lines (16HBE and LO2) were purchased from the American Type Culture Collection (ATCC, Manassas, VA, USA). A549 and LO2 cells were cultured in RPMI-1640 (Solarbio, Beijing, China) containing 10% fetal bovine serum (FBS, Gibco, Carlsbad, CA, USA), and PC-9 and 16HBE cells were cultured in DMEM (Solarbio) with 10% FBS. The PC-9/GR cells, a gefitinib-resistant strain of human lung adenocarcinoma cell, were grown in DMEM with 10% FBS and gefitinib (800 ng/ml). All cell lines were cultivated at 37°C under 5% CO₂.

The Cell Counting Kit-8 assays (Beyotime, Shanghai, China) were performed to assess cell viability. A549 (5 × 10³/100 μl/well), PC-9 (6 × 10³/100 μl/well), or PC-9/GR cells (6 × 10³/100 μl/well) were cultured overnight in 96-well plates (NEST, Wuxi, China) and were then treated with several concentrations of BD, BH, and/or gefitinib at the indicated concentration for 24 or 48 h. Then, the cells were incubated for an additional 2 h with 100 μl of RPMI-1640/DMEM and 10 μl of CCK-8 solution at 37°C. Microplate readers (Infinite F500, Tecan, Mannedorf, Switzerland) were used to measure the absorbance (A) at 450 nm. All samples were assessed in triplicate.

A549 cells (3 × 10⁵/ml/well) were seeded into a six-well plate. After overnight culture, when the fusion rate reached 95%–100%, the sterile pipette tip was perpendicular to the plate plane, and a straight wound was drawn on the cell monolayer. Following treatment with BD or BH, the migration of the cells was examined under a microscope (×4, Nikon, Tokyo, Japan), and pictures were obtained at 0 and 48 h. The computation of the cell migration formula is as follows: relative migration area (%) = (0 h wound area – 48 h wound area)/0 h wound area × 100%.

The Transwell insert in a 24-well plate (8 μm, Corning, NY, USA) was taken out, and a 600-μl medium containing 10% FBS was added to the bottom chamber. The A549 cells were resuspended with 10% FBS culture medium, and the diluted cell suspension (3 × 10⁵/300 μl) was cultured in the top chamber and then incubated with or without Matrigel (BD BioCoat). After 12 h, the medium in the top chamber was changed to RPMI-1640 without FBS, and BD or BH with different concentrations was added (31). After an additional 12 h, the cells that migrated or invaded the bottom side of the polycarbonate film were fixed with methanol for 30 min and then stained for 15 min with 0.1% crystal violet (Solarbio), after which their number was assessed under a microscope (×4, Nikon).

Cells (4 × 10⁵) were seeded in six-well plates, incubated overnight, and treated for 48 h with BH or gefitinib alone or combined before staining before examining with an Annexin V-FITC Apoptosis Detection Kit (Yeasen, Shanghai, China). After being digested with trypsin without EDTA, the cells were washed gently with culture medium and twice with cold PBS. The pellet was resuspended in 400 μl of 1× binding buffer, the resuspended cells were transferred to flow tubes, then Annexin V-FITC (5 μl) and propidium iodide (10 μl) were added, and the mixture was incubated for 15 min at room temperature in the dark. The percentage of apoptotic cells was determined by flow cytometry (Backman, CA, USA) and FlowJo V10 software.

The ligand molecule was downloaded from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/) and it was further converted to a PDB file using Open Babel GUI. The crystal structures of EGFR and Notch3 corresponding to PDB ID (3G5Z, 4ZLP) were obtained from the PDB database (https://www.rcsb.org/). The crystal water and other small molecules were removed from the protein structure by PyMOL (https://pymol.org), then hydrogens and charges were added using the AutoDock tool (http://autodock.scripps.edu/). Discovery Studio (http://dstudio19.csc.fi:9944) was used to predict the hydrophobic interaction between BD, BH, and gefitinib with the active site of the residue, along with the force of binding.

The Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geoprofiles/) was searched to download two microarray datasets (GSE115864 and GSE33532), and then their gene expression level and differential expression were analyzed. The detailed clinical information and complete mRNA expression profile data of 552 lung adenocarcinoma (LUAD) samples and 504 lung squamous cell carcinoma (LUSC) samples were obtained from The Cancer Genome Atlas database (TCGA, https://tcga-data.nci.nih.gov/tcga/) (accessed on May 12, 2021). We used the limma (linear models for microarray data) tool (32) to screen DEGs between PC-9 cells resistant to gefitinib and PC-9/GR treated with Notch3 inhibition and gefitinib, overlapped with the most significant module identified by weighted gene co-expression network analysis (WGCNA) (33) to obtain the hub genes, and then we subsequently carried out Gene Ontology (GO) functional annotation analysis and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis. Then, the key genes were predicted by Cytoscape plugin cytoHubba, and six key genes were selected according to the expression patterns in GEO and TCGA databases. Finally, the diagnostic value of the top six key genes was further verified by survival analysis and expression level in mutant lung cancer.

The total proteins of cells and tissues were extracted with RIPA cleavage buffer containing a protease inhibitor cocktail, phenylmethylsulfonyl fluoride (PMSF, Solarbio). The cell lysates (20 μg) were run on SDS-polyacrylamide gel (6%–10%) and transferred to the polyvinylidene fluoride membrane (PVDF, Millipore, MA, USA) by electrophoresis. The membrane was blocked with 5% skim milk in TBST for 2 h, incubated overnight with the corresponding primary antibody at 4°C, and then incubated with the corresponding secondary antibody at room temperature for 2 h. The reaction products were detected by chemiluminescence by ECL Western Blotting Detection System. Finally, quantitative analysis was carried out with ImageJ software.

The heart, liver, spleen, lung, kidney, and tumor tissues were embedded in paraffin, and the sections (4 µm) were stained with hematoxylin–eosin (H&E, Solarbio) and then examined and analyzed under a Leica DM6B microscope (Leica Microsystems, Wetzlar, Germany).

A TUNEL Assay Kit (Yeasen) was used to detect apoptosis. Briefly, the tumor section was dewaxed and incubated with proteinase K at 37°C for 20 min, followed by incubation with labeling buffer containing terminal deoxynucleotidyl transferase (TdT) and digoxigenin-labeled deoxyuridine triphosphate (dUTP). Then, the tissues were stained with 4ˊ,6-diamidino-2-phenylindole (DAPI) to stain the cell nucleus. TUNEL-positive cells were quantified by counting positively stained cells.

The PC-9/GR cells were added to glass slides in six-well plates and treated with BH and gefitinib in different concentrations, and DMSO served as an untreated control. The primary FOXO3a antibody (CST) was diluted at 1:100 and incubated at 4°C overnight. Afterward, the goat anti-rabbit IgG secondary antibody (dilution ratio: 1:400; Abcam) was added to the fixed PC-9/GR cells in a dark condition for 1 h and the nuclei were stained with DAPI (CST) for 5 min. Finally, fluorescence images were captured using the LAS AF Lite software and Leica DM6B microscope (Leica Microsystems).

The tumor section was dewaxed and repaired with antigen, 3% H₂O₂ was added, and then the section was sealed with 10% goat serum at room temperature for 30 min. The first antibody against Ki-67 and Notch3 (CST) was incubated overnight at 4°C, and the second antibody was incubated at room temperature for 30 min. After DAB staining, the sections were counterstained with hematoxylin for 2 min and then observed and analyzed under a Leica microscope.

The animal procedures were approved by the Ethics Committee of the Experimental Animal Center of Jiangxi University of Chinese Medicine (Nanchang, China, SCXK 2016-0006) and conducted following the Guide for the Care and Use of Laboratory Animals. Four-week-old male BALB/c nude mice (weighing 16–17 g, SPF grade) were obtained from Shanghai Shrek Experimental Animal Co., Ltd. (Shanghai, China). The xenotransplantation model was established by subcutaneous injection of A549 or PC-9/GR cells in nude mice, which were collected with PBS and mixed with an equal volume of Matrigel at a final concentration of 1 × 10⁷/ml. Approximately 100 mm³ of tumor volume was reached, and mice were randomly divided into four groups: vehicle control, gefitinib alone, BH alone, or a combination of gefitinib and BH. Gefitinib (25, 50 mg/kg) and BH (2.5, 5 mg/kg) were dissolved in 0.5% CMC-Na containing 5% DMSO and administered every day by oral gavage. Measurements of tumor size and body weight were conducted every 2 days with a vernier caliper and an electronic balance. Nodule growth was quantified by using the following formula: tumor volume (mm³) = (short-diameter)² × large-diameter × π/6.

At the end of the experiment, blood samples were taken and analyzed by blood routine. The remaining blood samples were precipitated, followed by a 15-min centrifugation (3,000 rpm) at 4°C, and the serum was taken for biochemical analysis. The lungs and spleen were weighed and recorded. H&E staining of the heart, liver, spleen, lungs, and kidneys was performed to evaluate the cytotoxic effects of BH.

We administered BH by gavage and intravenous administration to SD rats (male, 200–220 g, n = 6) after they had fasted for 12 h. For oral gavage, the BH was dissolved in DMSO/0.5% CMC-Na (5/95, v/v/), and for intravenous administration, it was dissolved in DMSO/0.9% NaCl (5/95, v/v/), and then the blood samples were collected at intervals to determine the pharmacokinetic (PK) profiles of SD rats.

All the experimental data were analyzed by GraphPad Prism 8.0 software (USA). The results were expressed as the average ± SEM of at least three independent experiments. Double-tailed t-test and ANOVA were used; when *P <0.05, **P <0.01, or ***P <0.001, it was considered to be statistically significant.

The structures of BH and BD are presented in Figure 1A , which were identified by ¹H and ¹³C nuclear magnetic resonance (NMR) ( Figures S6–S9 ). HPLC analysis was performed in our study to confirm that the purity of BH and BD was ≥95% ( Figure 1B ). Bruceine H and gefitinib showed very good binding affinity (−31.9, −20.0 kcal/mol) with the amino acid residues PHE98, TRP48 and GLU123, ARG216 of the target protein EGFR (3G5Z), respectively. Furthermore, the Notch inhibitors BH and BD were also docked to find the interaction and binding affinity (LibDockScore:169.0, 165.2 >120) with the target protein Notch3 (4ZLP). BH and BD bind to the same sites on the Notch3 HD domain in modeling, despite the different chemical formula ( Figure 1C ). This shows that both of these compounds have similar affinity to Notch3 protein, and in the subsequent study of the cell model, BH and BD have also been proven to have similar antitumor activity.

The NSCLC cell lines A549 (EGFR wild-type KRAS mutation) and PC-9 (EGFR exon 19 deletion) were treated with BH and BD at different concentrations for 24 and 48 h, respectively. The results showed that gefitinib, BH, and BD could inhibit the proliferation of A549 and PC-9 cells in a dose-dependent manner. The IC₅₀ (48 h) values of gefitinib, BH, and BD against A549 cells were 23.60, 25.54, and 26.80 μM, and the IC₅₀ values against PC-9 cells were 11.44, 12.57, and 21.84 μM, respectively ( Figure 2A ). The inhibitory effect of BH on cell proliferation was similar to that of BD, especially on A549 cells.

Given the role that migration and invasion of cancer cells play in cancer metastasis and patient mortality, further investigation of BH and BD for their effect on migration and invasion was conducted. Wound healing assay showed that the migration of A549 cells was impaired by 4 μM BH ( Figure 2B ). Transwell cell invasion assay results demonstrated that BH and BD can exert significant anti-invasive and antimigratory actions toward A549 in a dose-dependent manner ( Figures 2C, D ). In an attempt to determine whether the proliferative inhibition by BH was related to apoptosis, cells from A549 and PC-9 were treated with BH for 48 h to determine the percentage of apoptotic cells. Figure 2E shows that the proportion of apoptotic cells increased after treatment with various concentrations of BH, as well as the early apoptosis rate in A549 cells, while the late apoptotic rate of PC-9 cells increased, relative to the control conditions. It is noteworthy that BD shows higher sensitivity in 16HBE and LO2 cell lines than BH ( Figure S1 ). Taken together, BH exhibited not only potent antitumor activities but also no apparent toxicity to normal human cell lines and was selected for further studies. Then, we further determined whether BH would synergize with gefitinib to decrease the PC-9/GR cell viability.

Lung cancer with EGFR exon 19 deletion is known to have the greatest prevalence, which is present in 1.57% of AACR GENIE cases (34). As compared with PC-9 cells, gefitinib-resistant PC-9 cells (PC-9/GR) contain the same exon 19 deletions of the EGFR gene but show low sensitivity to gefitinib and lower expression of FOXO3a ( Figure 3A ). PC-9 cells harboring EGFR activation mutation were sensitive to gefitinib, IC₅₀ at 11.44 μM, while PC-9/GR was resistant to gefitinib under long-term low concentration of gefitinib, IC₅₀ at 51.30 μμ ( Figure 3B ). Figures 3C, D show that drug synergism in PC-9/GR was calculated using CalcuSyn 2.0 software when BH and gefitinib are combined according to their IC₅₀ concentrations. It was discovered that drug synergy increased as drug concentration was increased, and the drug synergy was the strongest at an inhibition rate of 80%. As a result of gefitinib treatment alone, PC-9/GR is no longer sensitive to gefitinib, and BH and gefitinib combined improved the inhibitory effect compared with a single treatment. To explore the effects of BH and gefitinib on the apoptosis of PC-9/GR cells, the results of flow cytometry showed that low-dose BH or gefitinib alone had little effects, while a combination of BH and gefitinib obviously induced apoptosis in PC-9/GR cells compared with gefitinib monotherapy ( Figures 3E, F ).

Subsequently, the extracts of PC-9/GR cells treated with BH and/or gefitinib for 48 h were prepared for immunoblotting. The Western blot analysis showed that BH combined with gefitinib treatment deregulated the anti-apoptotic protein (Bcl-2) and significantly increased the expression of Bax. Additionally, compared with the vehicle and gefitinib or BH alone, the expression of cleaved caspase-3 was remarkably increased and caspase-3 was inhibited ( Figure 4A ). Together, these results indicate that bruceine H in combination with gefitinib results in a significant induction of PC-9/GR apoptosis. Additionally, the level of FOXO3a was activated and β-catenin levels were suppressed by a combination of gefitinib and BH ( Figure 4B ). We found that, through inhibiting the activity of β-catenin and activating FOXO3a in drug-resistant cells, BH enhanced the sensitivity of PC-9/GR cells to gefitinib. To clarify the role of FOXO3a, we confirmed the increase of FOXO3a protein accumulation in PC-9/GR cells by immunofluorescence staining ( Figures 4C, D ). These results indicated that BH significantly enhanced gefitinib sensitivity in PC-9/GR cells by β-catenin activating the FOXO3a signaling pathway.

Based on the analysis of the differentially expressed genes between EGFR inhibition and combined EGFR and Notch inhibition in EGFR mutant of NSCLC, a total of 936 DEGs were identified, consisting of 366 upregulated and 570 downregulated DEGs ( Figure 5A ). On the other hand, the difference of gene expression between 80 non-small cell lung cancer samples and 20 marginal samples yielded 2,817 DEGs, consisting of 1,097 high-expression genes and 1,720 low-expression genes ( Figure 5B ). Figures 5C, D represent the heatmap of the expression levels of these DEGs. The 259 NSCLC samples from the TCGA database ( Table S1 ), with a clear record of a canonical mutation in KRAS, were used to establish a co-expression module for WGCNA to evaluate the correlation between potential genes and EGFR mutations in patients with non-small cell lung cancer. Cluster analysis demonstrated that the mutation in the sample has a certain relationship with smoking history, and the longer the smoking period, the more likely it is that the mutation of lung cancer would occur ( Figure S2A ). Besides, the power value of β = 4 (scale-free R ² = 0.91) was set to ensure low mean connectivity and high-scale independence ( Figures S2B, C ). A total of nine modules ( Figures S2D, E ) were identified through the average linkage hierarchical clustering and segmented the clustering results under the set criteria (dissimilarity = 0.25). The correlation of the modules with clinical traits, including tumor-normal and EGFR-TKI mutant, is shown in Figures 5E–G . Finally, the 742 gene sequences in the blue module were intersected with mutant-specific DEGs to identify 89 genes that regulated drug persistence in patients with NSCLC with EGFR mutations ( Figure 5H ).

GO function and KEGG pathway enrichment analyses were performed to assess the function of 89 genes, which intersected the blue module genes and the two differential overlapping genes. The top 10 results of GO that include terms of biological process (BP), cellular component (CC), and molecular function (MF) are shown in Figure 6A . The most significant enrichment terms of BP, CC, and MF for EGFR-TKI drug persistence were “sterol biosynthetic process”, “clathrin-coated endocytic vesicle membrane”, and “nuclear receptor binding”, respectively. Twenty significantly enriched KEGG pathways were identified in these overlapping genes ( Figure 6B ). The most significant KEGG pathways, including the FOXO and Notch signaling pathways, might be related to the EGFR resistance response of NSCLC patients.

The survival analysis of the genes in the above network was conducted to determine the prognostic value of the six hub genes in NSCLC. In the relapse-free survival (RFS) analyses based on sequencing data of 1,925 NSCLC patients, high NOTCH3 and NCBP2 expression levels were associated with shorter NSCLC survival times ( Figure 6C , P < 0.01). Furthermore, individuals expressing high levels of FOXO3a had a more favorable prognosis than those expressing low levels of the gene. We analyzed the expression levels of FOXO3a, EGFR, NCBP2, CTNNBIP1, NFKBIA, and NOTCH3 at different resistance statuses in LUAD and LUSC in the TCGA database ( Figure 6D ). The result indicated that the Notch3 and β-catenin genes were overexpressed in EGFR resistance processes in NSCLC.

Subcutaneous mouse xenografts of the NSCLC cell line A549 were further used to evaluate BH’s antitumor effect. BH significantly inhibited the growth of human lung cancer (A549) xenografts after 16 days of treatment with 2.5 or 5 mg/kg dose given via intragastric administration, and the inhibitory effect of 2.5 mg/kg brucine H was similar to that of the effective drug gefitinib ( Figures 7A, B ). For A549 xenograft tumors, tumor volumes and tumor weights decreased dose-dependently in the 2.5- and 5-mg/kg treatment groups ( Figure S3A ) without significant changes in body weight. Meanwhile, there were no significant changes in organ indexes of the heart, kidney, liver, lung, and spleen tissues ( Figure S3B ), indicating that BH treatment did not cause harmful side effects in tumor-bearing mice. Moreover, tumor cell apoptosis was evaluated by HE staining, and the mitotic index (Ki-67) and Notch3 signal were detected through immunohistochemistry (IHC) and Western blotting. In contrast to the vehicle group, BH caused significant cancer cell apoptosis, and the number of Ki-67-positive cells in cancer tissues was decreased ( Figures 7C, D, F ). Notch3 expression in tumor xenografts decreased in a dose-dependent manner in the BH treatment group compared with that in the model group, according to the IHC analysis ( Figure 7E ). Additionally, in the BH treatment group, the expression of FOXO3a increased in the tumor tissues, whereas Notch3, β-catenin, and Bcl-2 protein levels decreased ( Figure 7F ). The results showed that BH significantly suppressed excessive Notch3 signaling and displayed potent antitumor activity against human lung cancer cell xenograft models.

To further determine the therapeutic effect of BH combined with gefitinib, we established a nude mouse xenograft model of PC-9/GR cells. The results revealed that treatment with BH at 2.5 mg/kg alone or a low dose of gefitinib (25 mg/kg) alone showed a slight inhibitory effect on tumor growth of PC-9/GR by intragastric administration, respectively ( Figure 8A ), while co-administration of BH (2.5 mg/kg) and gefitinib (25 mg/kg) significantly suppressed xenograft tumor growth ( Figures 8B, C ). There were no significant changes in body weights and HE staining of the heart, kidney, liver, lung, and spleen tissues ( Figures S4A, B ). In addition, BH and gefitinib combination treatment caused significant cancer cell apoptosis compared with a single treatment, increased significantly the number of TUNEL-positive cells in cancer tissues, and reduced the number of Ki-67-positive cells in cancer tissues ( Figures 8D–F ). Also, IHC staining demonstrated that the level of Notch3 was inhibited by BH and gefitinib combination ( Figure 8G ). Western blotting further proved that the expression of Notch3, as well as β-catenin downstream genes CBP (Lys1535)/p300 (Lys1499) and p-EGFR (Tyr1068) in the tumor tissues, was reduced ( Figure 8H ). Taken together, combining a Notch3 inhibitor with gefitinib inhibits the growth of TKI-resistant lung cancer xenografts more effectively than either treatment separately.

To prove that BH had no significant toxicity in vivo, the organ ( Table S2 ), blood routine ( Table S3 ), and biochemical indexes ( Table S4 ) of nude mice were determined. Compared with the model group or positive drug group, we did observe no significant changes in the coefficients of lung and spleen and in the levels of white blood cells, lymphocytes, and neutrophils in the whole blood and in the levels of total cholesterol, triglycerides, low-density lipoprotein, and high-density lipoprotein in the serum. As a result, we examined the preliminary pharmacokinetic parameters ( Table 1 and Tables S5–S7 in the Supplementary Material ) of BH after intravenous and oral administration in SD rats. The preliminary pharmacokinetic parameters for a single dose of the compound BH after intravenous (2 mg/kg) and oral administration (10 mg/kg) were as follows: the half-life (T _1/2) was 9.694 and 3.353 h, maximum concentration (C _max) was 477.4 and 146.6 ng/ml, and the area under the plasma concentration–time curve (AUC_0–last) was 943.5 and 961.2 ng h/ml, respectively. The results showed that oral bioavailability (F) was 20.38%.