Transcriptomic count data and matched clinical information from 594 LUAD patients were obtained from TCGA (https://portal.gdc.cancer.gov/). LUAD-specific gene mutation profiles were also obtained from the same source. Raw RNA-seq count were normalized into Reads Per Kilobase Per Million (RPKM) and Transcripts Per Million (TPM) using R software (version 4.1.1; https://cloud.r-project.org/) for downstream analysis. A curated list of ferroptosis-related genes was extracted from the FerrDb database (http://www.zhounan.org/ferrdb/).

For clinical validation, six LUAD tumor samples harboring confirmed EGFR mutations and matched adjacent normal tissues were collected from treatment-naïve patients undergoing surgical resection at our institution. All samples were flash-frozen on collection and kept at –80 °C in pre-cooled containers to preserve molecular integrity. EGFR mutation status was verified postoperatively by institutional pathologists. All patients were treatment-naive, with no history of radiotherapy, chemotherapy, targeted agents, or immunotherapy.

The study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Ethics Committee at Sun Yat-sen University’s Seventh Affiliated Hospital (Authorization Number: 2024-152). Informed consent was secured from each participant before collecting samples.

A random forest model was developed using the “Boruta” (https://cloud.r-project.org/web/packages/Boruta/index.html) R package, focusing on ferroptosis-related genes signatures. This ensemble learning method integrates multiple decision trees built on randomly selected data subsets to improve prediction robustness and reduce overfitting. The cohort was randomly split into training and validation sets at a 3:2 ratio. Feature importance was assessed and visualized using the “ggrepel” (https://cloud.r-project.org/web/packages/ggrepel/index.html) package . Model performance was evaluated by plotting the receiver operating characteristic (ROC) curve and calculating the area under the curve (AUC) using the “ggplot2” (https://cloud.r-project.org/web/packages/ggplot2/index.html) package. The ROC curve illustrated the sensitivity–specificity trade-off, with a higher AUC reflecting stronger discrimination capacity.

Differential expression between LUAD tumor and adjacent normal tissues was assessed using the “edgeR” (https://www.bioconductor.org/packages/release/bioc/html/edgeR.html) R package to identify significantly altered genes. A heatmap was generated using the “pheatmap” (https://cloud.r-project.org/web/packages/pheatmap/index.html) package to visualize expression patterns of differentially expressed genes (DEGs).

To evaluate prognostic value, patients were dichotomized into high- and low-expression groups based on median gene expression. Kaplan–Meier survival curves were generated using the “survival” (https://cloud.r-project.org/web/packages/survival/index.html) and “survminer” (https://cloud.r-project.org/web/packages/survminer/index.html) packages, and statistical differences were assessed to identify potential prognostic biomarkers.

Patients were stratified by EGFR mutation status (mutant vs. wild-type). Expression differences of model genes were compared using the Wilcoxon rank-sum test. Genes with p-values 0.05 were considered significant, and results were visualized using boxplots.

Immune cell composition in tumor and normal tissues was estimated using the “CIBERSORT” algorithm. Patients were further grouped by median expression of selected genes to evaluate immune cell infiltration differences across subgroups.

To explore pathways associated with AURKA expression, GSEA was performed on expression data from 594 LUAD cases. Patients were divided into high- and low-expression groups based on the median AURKA level. A total of 1000 permutations were conducted to compute p-values. Significant pathways were defined by p    0.01 and false discovery rate (FDR) 0.25.

A high-confidence protein–protein interaction (PPI) network of differentially expressed ferroptosis-related genes was constructed using the STRING database (https://string-db.org), applying a minimum interaction score threshold at 0.70 to ensure high-confidence associations. Network visualization and topological analysis were conducted in Cytoscape (version 3.10.2; https://cytoscape.org/) [25]. Functional enrichment analysis was performed through STRING using multiple ontologies and pathway databases, including Gene Ontology (GO: biological process [BP], cellular component [CC], molecular function [MF]), Reactome, Kyoto Encyclopedia of Genes and Genomes (KEGG), and WikiPathways. The enrichment analysis results were graphically presented using the “ggplot2” (https://cloud.r-project.org/web/packages/ggplot2/index.html) package in R. KEGG mapping identified significantly enriched signaling pathways [26], while GO terms highlighted key biological functions [27]. Reactome pathways provided expert-curated mechanistic insights [28], and WikiPathways served as a collaborative source for community-curated networks [29].

To detect densely connected regions within the network, the molecular complex detection (MCODE) plugin in Cytoscape (version 3.10.2; https://cytoscape.org/) was used [30]. MCODE identifies molecular complexes by assigning weights to nodes based on local connectivity and expanding from high-density seed proteins. Clustering was performed in directed mode to improve specificity. Modules and hub genes were identified using the following parameters: degree cutoff of 2, node score threshold of 0.2, k-core value of 2, and a maximum depth setting of 100.

LUAD cell lines NCI-H1975 (harboring the EGFR L858R/T790M mutation conferring intrinsic gefitinib resistance), A549, NCI-H1650, and PC‑9 were obtained from the American Type Culture Collection (ATCC) (https://www.atcc.org/). Cell authentication was verified via short tandem repeat (STR) profiling, and all lines were confirmed to be mycoplasma-free. Cells were cultured in Roswell Park Memorial Institute (RPMI) 1640 medium supplemented with 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin and maintained at 37 °C in a 5% CO₂ humidified incubator (Thermo Scientific, MA, USA, cat#51033582).

Before seeding, cells were confirmed to be 80–90% confluent by microscopy (Olympus,Tokyo, Japan, cat#CX23-1RH2) . Cells were washed with PBS, digested with trypsin, and resuspended in complete medium. A 10 µL aliquot of the suspension was used for cell counting with a hemocytometry (Hausser Scientific, Horsham, PA, USA, cat#15170-172). Cells (8 $\times$ 10⁴ per well) were plated in 12-well plates with 1 mL of 1640 medium supplemented with 10% FBS and incubated under standard conditions.

Gene knockdown was achieved using small interfering RNA (siRNA) transfection. Specific siRNAs targeting AURKA mRNA were synthesized (GEMA Genetics, Shanghai, China). A negative control (NC) siRNA was included to control for off-target effects.

Cells were plated in 12-well plates and transfected after a 24-hour incubation. A mixture of siRNA, transfection reagent, and PolyJet (as per manufacturer’s protocol) was prepared and incubated for 15 minutes. Culture medium was replaced with 1 mL of antibiotic-free 1640 medium with 10%. Subsequently, 100 µL of the transfection mixture was added to each well, yielding a final siRNA concentration of 20 nM. Cells were incubated for 72 hours before harvesting protein analysis.

siRNA sequences:

si-NC: 5^′-UUCUCCGAACGUGUCACGUTT-3^′; 5^′-ACGUGACACGUUCGGAGAATT-3^′.

si-AURKA: 5^′-GCAAUUUCCUUGUCAGAAUTT-3^′; 5^′-AUUCUGACAAGGAAAUUGCTT-3^′.

Forty-eight hours after siRNA-induced AURKA silencing, total RNA was isolated from NCI-H1975 cells using TRIzol reagent (Sangon Biotech, Shanghai, China, cat#B511311). RNA sequencing and primary data processing was carried out by Sangon Biotech (Shanghai, China, https://www.sangon.com/). Raw read counts were converted to TPM for downstream analysis using R software (v4.1.1; https://cloud.r-project.org/). Differential expression analysis was performed using the “DESeq2” (https://www.bioconductor.org/packages/release/bioc/html/DESeq2.html) package in R. The Benjamini-Hochberg method was applied to adjust p-values for multiple comparisons, and genes with an adjusted p-value (q-value) 0.05 were considered significantly differentially expressed. DEGs were visualized using volcano plots created with the “ggplot2”, “ggrepel”, and “ggthemes” R packages. Gene expression heatmaps were constructed with the ‘pheatmap’ (https://cloud.r-project.org/web/packages/pheatmap/index.html) package in R.

To assess the long-term proliferative capacity, colony formation assays were performed. NCI-H1975 cells were seeded into 12-well plates at 2000 cells per well and cultured at 37 °C with 5% CO₂ for 2 weeks. Colonies were then washed with PBS, fixed with 4% paraformaldehyde (MACKLIN, Shanghai, China, cat#P804536) for 20 minutes, and stained with 1% crystal violet (Solarbio, Beijing, China, cat#C8470) at room temperature for 10 minutes. After washed off excess dye, the plates were air-dried and imaged.

Cell migration following AURKA silencing was assessed using the Transwell assay. After 48 hours of siRNA transfection, NCI-H1975 cells were washed, digested with trypsin, and resuspended in serum-free 1640 medium. Approximately 30,000 cells were added to the upper compartment of Transwell inserts, which were positioned in 24-well plates filled with 700 µL RPMI-1640 medium containing 15% FBS. After 48 hours of incubation at 37 °C with 5% CO₂, cells on the upper surface of the membrane were removed, and migrated cells on the underside were fixed in formaldehyde, stained with 1% crystal violet, and imaged under a stereomicroscope (Olympus,Tokyo, Japan, cat#SZX16).

Intracellular Fe²⁺ levels were assessed by treating siRNA-transfected NCI-H1975 cells with 1 µM FerroOrange (Dojindo, Kumamoto, Japan, cat#F374) at 37 °C for 30 minutes. Fluorescence signals indicating ferrous iron accumulation were visualized using a fluorescence microscope (BioTek Cytation 5, BioTek, Winooski, VT, USA).

Lipid peroxidation, a key marker of ferroptosis, was assessed using the fluorescent probe C11-BODIPY 581/591 (10 µM; ABclonal, Wuhan, China, cat#RM02821). NCI-H1975 cells were treated with siRNA for 72 hours and incubated with the probe for 1 hour at 37 °C in 5% CO₂. After washing with PBS, cells were harvested and resuspended for flow cytometry analysis. To investigate ferroptosis-specific responses, cells were pretreated 24 hours in advance with ferroptosis inhibitors or inducers, including Ferrostatin-1 (Fer-1, 10 µM, cat#S7243), Liproxstatin-1 (LIP-1, 10 µM, cat#S7699), Deferoxamine (DFO, 10 µM, cat#S5742), Necrostatin-1 (10 µM, cat#S8037), and Sorafenib (20 µM, cat#S7397) —all from Selleck (TX, USA).

Following 72 hours of siRNA treatment in NCI-H1975 cells, Fer-1 (10 µM, Selleck, TX, USA, cat#S7243) was added 24 hours prior to the assay. Mitochondrial membrane potential was assessed using Mito-Tracker Red FM (Beyotime Biotechnology, Shanghai, China, cat#C1032) following the supplier’s instructions. Fluorescence intensity was recorded using the Cytation5 cell imaging multi-mode reader (BioTek Cytation 5, BioTek). A reduction in mitochondrial membrane potential was interpreted as a hallmark of ferroptosis.

After 72 hours of AURKA knockdown and 24-hour Fer‑1 pre-treatment (10 µM), NCI‑H1975 cells were fixed in 2.5% glutaraldehyde in 0.1 M Sorenson buffer (pH 7.2) for 1 hour, followed by postfixation in 1% osmium tetroxide for another hour. Samples were dehydrated, embedded, and sections into 60 nm ultrathin slices using an MT-7000 ultramicrotome (RMC, Tucson, AZ, USA, cat#MT-7000). Sections were stained with uranyl acetate and lead citrate, and mitochondrial ultrastructure was observed using a JEOL JEM-1200 EXII transmission electron microscope (JEOL, Tokyo, Japan, cat#JEM-1200 EXII).

To assess lipid peroxidation and antioxidant status, malondialdehyde (MDA) and glutathione (GSH) levels were measured in NCI-H1975 cells after 72 hours of siRNA transfection, with Fer-1 (10 µM) pre-treatment for 24 hours. Commercial kits (Beyotime Biotechnology, cat#S0131M for MDA and cat#S0053 for GSH) were used following the supplier’s instructions. Elevated MDA levels were indicative of increased lipid peroxidation, while decrease GSH levels reflected consumption during oxidative stress.

ROS were quantified using the MitoSOX™ Red mitochondrial superoxide indicator (Beyotime Biotechnology, cat#S0061S). After 72 hours of AURKA knockdown and 24-hour Fer‑1 treatment, cells were incubated with the probe following the manufacturer’s protocol. ROS levels were analyzed via flow cytometry. A marked increase in mitochondrial ROS was considered evidence of ferroptotic oxidative injury.

Western blotting was conducted to assess protein expression following pharmacologic and genetic interventions. NCI-H1975 cells were pretreated with the AURKA inhibitor LY3295668 (500 nM, Selleck, TX, USA, cat#S8782) for 72 hours and the NRF2 inhibitor ML385 (10 µM, Selleck, cat#S8790) for 24 hours before lysis. Cell lysates were prepared using PMSF-supplemented RadioImmunoPrecipitation Assay (RIPA) buffer (Solarbio, China, cat#R0010), and protein concentrations were determined using a bicinchoninic acid (BCA) assay kit (Sangon Biotech, China, cat#C503021). Subsequently, 10 µg of total protein per sample was separated on 10% SDS-PAGE gels and transferred to nitrocellulose membranes. Membranes were blocked in 5% BSA for 1 hour, incubated overnight at 4 °C with primary antibodies, washed with Tris-buffered saline with Tween®20 (TBST), and probed with horseradish peroxidase (HRP)-conjugated secondary antibodies (Cell Signaling Technology, Danvers, MA, USA, cat#7074P2/cat#7076P2, 1:4000) for 1 hour at room temperature. Chemiluminescent detection was performed using BeyoECL Moon (Beyotime Biotechnology, Shanghai, China, cat#P0018FS).

The following primary antibodies were used: 4-HNE (Biosynthesis Biotechnology, Beijing, China, cat#bs6313R, 1:1000), GPX4 (Cell Signaling Technology, cat#52455, 1:1000), solute carrier family 7 member 11 (SLC7A11; ABclonal, China, cat#A13685, 1:1000), NRF2 (Cell Signaling Technology, cat#12721, 1:1000), KEAP1 (Cell Signaling Technology, cat#8047, 1:1000), HO-1 (Cell Signaling Technology, cat#86806, 1:1000), AURKA (Cell Signaling Technology, cat#91590, 1:1000), Cyclin D1 (Cell Signaling Technology, cat#55506, 1:1000), glyceraldehyde-3-phosphate dehydrogenase (GAPDH; Sangon Biotech, Shanghai, China, cat#D110016, 1:1000). Secondary antibodies included anti-rabbit IgG (Cell Signaling Technology, cat#7074, 1:3000) and anti-mouse IgG (Cell Signaling Technology, cat#5127, 1:3000).

Paraffin-embedded tissue sections were baked at 60 °C for 1, deparaffinized in xylene, and rehydrated through graded ethanol. Antigen retrieval was performed in citrate buffer. Endogenous peroxidase activity was quenched with 3% H₂O₂ for 30 minutes, followed by blocking with 5% goat serum for another 30 minutes. Sections were incubated overnight at 4 °C with anti-AURKA (Cell Signaling Technology, cat#91590, 1:300), followed by 30-minute incubation with HRP-conjugated secondary antibody at 37 °C. Immunoreactivity was visualized with the 3,3^′-diaminobenzidine (DAB, Sangon Biotech, Shanghai, China, cat#A600140), and nuclei were counterstained with hematoxylin.

For gene expression analysis, NCI-H1975 cells were plated in 6-well plates and treated with ML385 (10 µM) for 24 hours and AURKA siRNA for 72 hours. Total RNA was extracted using TRIzol reagent (Solarbio, China, cat#10296010) and reverse transcribed with M-MLV Reverse Transcriptase (Invitrogen, Carlsbad, CA, USA, cat# 28025013) from 1 µg of RNA in a 20 µL reaction. qPCR was performed with SYBR® Green (Promega, Madison, WI, USA, cat# A6001) on the ABI 7500 system. GAPDH was used as an internal control, and relative gene expression was calculated via the 2^{-Δ⁢Δ⁢Ct} method [20]. All reactions were performed in triplicate. Primer sequences for GAPDH, AURKA, and NRF2, are listed in Table 1.

Cells were seeded on coverslips in 6-well plates and treated with ML385 (10 µM) and siRNA as described. After fixation with 4% paraformaldehyde for 20 minutes, cells were permeabilized with 0.5% Triton X-100 buffer (Beyotime, Shanghai, China, cat#P0096) and blocked with 5% BSA for 30 minutes. Primary antibody against NRF2 was applied overnight at 4  °C. After PBS washes, fluorescent secondary antibody (CST, USA, cat#7074, 1:800) were added for 1 hour at room temperature. Nuclei were counterstained with DAPI (Cell Signaling Technology, Danvers, USA, cat#4083). Images were acquired using the Cytation 5 system and quantified using ImageJ.

All data are presented as mean $\pm$ standard deviation (SD). Statistical analyses were conducted using GraphPad Prism. Normality was verified using the Shapiro-Wilk test. For comparisons between two groups Student’s t-test was applied. For multi-group comparisons, one-way analysis of variance (ANOVA) followed by Tukey’s post hoc test was used. p-values 0.05 were considered statistically significant. Significance levels are indicated as follows: *p 0.05; **p 0.01; ***p 0.001; ****p 0.0001.

To explore genetic factors related to LUAD mutations, we analyzed the somatic mutation profiles of individual samples from the TCGA-LUAD cohort. As shown in Supplementary Fig. 1A, the top frequently mutated genes included titin (TTN), mucin 16 (MUC16), ryanodine receptor 2 (RYR2), CUB and Sushi multiple domains 3 (CSMD3), tumor protein p53 (TP53), low density lipoprotein receptor related protein 1B (LRP1B), usherin (USH2A), zinc finger homeobox 4 (ZFHX4), xin actin binding repeat containing 2 (XIRP2), and KRAS proto-oncogene (KRAS). Further assessment of commonly altered oncogenic drivers in lung cancer revealed high mutation frequencies in KRAS, EGFR, and B-Raf proto-oncogene (BRAF) (Supplementary Fig. 1B).

Considering the higher frequency of EGFR mutations in Asians populations [3, 4] and the widespread clinical use of EGFR‑targeted therapies—despite rising resistance rates [31, 32]—we developed a ferroptosis gene—based random forest model to predict EGFR mutation status. The model’s convergence is illustrated in Supplementary Fig. 2A, and the ranked variable importance is displayed in Supplementary Fig. 2B. The model demonstrated strong predictive power, with AUC values of 0.84 in the training set (Fig. 1A) and 0.83 in the validation set (Fig. 1B), indicating robust performance and generalizability.

Fig. 1.

Construction of the ferroptosis-based random forest model and prognostic analysis of AURKA. (A,B) Receiver operating characteristic (ROC) curves illustrating the performance of the random forest model in the training (A) and validation (B) datasets. (C) Heatmap showing differential expression of model-related genes in LUAD versus normal tissues. (D) Box plot illustrating the association between AURKA expression and EGFR mutation status. (E) Kaplan–Meier survival curve showing that high AURKA expression correlates with poor prognosis (p = 0.011). (F–H) Correlation of AURKA expression with gender (p = 0.0058), tumor stage (p = 0.068) and survival status (p = 0.022). AURKA, Aurora kinase A; LUAD, adenocarcinoma of Lung; EGFR, epidermal growth factor receptor; EIF2S1, eukaryotic translation initiation factor 2 subunit 1; STMN1, stathmin-1; ALOX15B, arachidonate 15-lipoxygenase type B; ZNF419, zinc finger protein 419; FANCD2, fanconi anemia group D2 protein; AUC, area under the curve.

Heatmap analysis comparing tumors with adjacent normal tissues showed downregulation of arachidonate 15-lipoxygenase type B (ALOX15B), alongside upregulation of fanconi anemia group D2 protein (FANCD2), eukaryotic translation initiation factor 2 subunit 1 (EIF2S1), stathmin-1 (STMN1), zinc finger protein 419 (ZNF419), and AURKA in tumor samples (Fig. 1C).

To identify genes linked to EGFR status, we examined their expression in EGFR-mutant versus wild-type subgroups. Notably, AURKA expression was significantly elevated in EGFR wild-type (p = 0.035; Fig. 1D), suggesting a potential inverse association between AURKA activity and EGFR mutations.

To evaluate AURKA’s prognostic relevance, LUAD patients were categorized into high- and low-expression cohorts using the median expression as a cutoff. Kaplan–Meier analysis revealed that elevated AURKA levels were significantly linked to reduced overall survival (p = 0.011; Fig. 1E). Additionally, high AURKA expression correlated with more advanced pathological stage, male sex, and increased mortality (Fig. 1F–H), supporting its potential role as a negative prognostic biomarker in LUAD.

To investigate how AURKA expression correlates with immune cell infiltration in LUAD, we utilized the CIBERSORT algorithm to estimate the proportions of 22 distinct immune cell types. Comparative analysis between tumor and adjacent normal tissues revealed that 17 immune cell types exhibited significant differences. Specifically, M2 macrophages, M0 macrophages, resting CD4 memory T cells, resting mast cells, activated dendritic cells, monocytes, neutrophils, eosinophils, and resting NK cells showed decreased infiltration in tumor (p 0.01). In contrast, increased infiltration was observed for plasma cells, M1 macrophages, resting dendritic cells, regulatory T cells (Tregs), follicular helper T cells, activated CD4 memory T cells, memory B cells, and $\gamma$$\delta$ T cells (p 0.01; Supplementary Fig. 3A).

Patients were divided into high- and low-AURKA expression cohorts according to the median transcript level. Among the 22 immune cell types, 14 showed significant infiltration differences between the two groups. High AURKA expression was associated with elevated infiltration of M0 and M1 macrophages, follicular helper T cells, activated CD4 memory T cells, activated mast cells, resting NK cells, and $\gamma$$\delta$ T cells (p 0.05; Supplementary Fig. 3B).

Conversely, resting CD4 memory T cells, resting mast cells, resting dendritic cells, Tregs, activated dendritic cells, monocytes, and memory B cells were more abundant in the low AURKA expression group (p 0.05; Supplementary Fig. 3B).

Further analysis showed that AURKA expression was positively correlated with programmed Cell Death 1 (PDCD1) (p = 0.022), while no significant association was observed with CD274 (p = 0.066) or cytotoxic T-lymphocyte associated protein 4 (CTLA4, p = 0.22; Supplementary Fig. 3C–E). These results indicate that AURKA may play a pivotal role in shaping the immune microenvironment of LUAD, potentially affecting tumor progression through immune modulation. Collectively, the data support the classification of AURKA as a high-risk gene in LUAD, given its associations with EGFR mutation status, immune infiltration patterns, and unfavorable prognosis.

To uncover the biological pathways associated with AURKA, GSEA was performed using TCGA-LUAD transcriptomic data. Among 177 pathways analyzed, 124 showed a positive correlation with AURKA expression, with 42 significantly enriched (p 0.01). Meanwhile, 53 pathways were negatively correlated, 31 of which reached statistical significance (p 0.01). Notably, pathways enriched in the high AURKA group included those involved the cell cycle, the tricarboxylic acid (TCA) cycle, glutathione metabolism, biosynthesis of unsaturated fatty acids, and alanine aspartate and glutamate metabolism (Fig. 2A, Table 2), suggesting potential roles in tumor progression and ferroptosis regulation.

Fig. 2.

AURKA signaling pathway analysis in LUAD. (A) Gene set enrichment analysis (GSEA) showing AURKA-associated pathways in TCGA LUAD data. (B,C) Volcano plot (B) and heatmap (C) of differential expressed ferroptosis-related genes following AURKA knockdown in NCI-H1975 cells. (D) GO enrichment of differentially expressed genes. (E) KEGG, Reactome, and WikiPathways analyses indicating ferroptosis- and NRF2-related pathway enrichment. (F) PPI network constructed from 83 ferroptosis-related genes using STRING. (G) Identification of key gene modules via MCODE analysis in Cytoscape. TCGA, The Cancer Genome Atlas; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; NRF2, NF-E2-Related Factor 2; PPI, protein–protein interaction; MCODE, molecular complex detection; TCA, the tricarboxylic acid.

To validate these findings, RNA sequencing was performed on NCI-H1975 cells following AURKA knockdown via siRNA. Differential expression analysis revealed 61 downregulated and 22 upregulated ferroptosis-related genes in the knockdown group (Fig. 2B,C). Functional enrichment analyses were conducted using GO, Reactome, KEGG, and WikiPathways databases to investigate the biological relevance of these genes. GO analysis indicated significant enrichment in oxidative stress and ROS-related processes (Fig. 2D). KEGG pathways included ferroptosis and FoxO signaling pathways—both linked to ROS homeostasis (Fig. 2E). Reactome analysis highlighted the KEAP1/NRF2 axis, and regulation of HO-1 activity, while WikiPathways emphasized enrichment in the ferroptosis and NRF2 signaling pathways, reinforcing the connection between AURKA and ferroptosis through the NRF2 pathway (Fig. 2E).

A PPI network of 83 differentially expressed ferroptosis-related genes was generated using STRING. MCODE clustering analysis identified five functional modules with scores of 4, 3.5, 3.429, 3.333, and 3, respectively (Fig. 2F). Four key hub genes— HO-1 (cluster 1), SP1 (cluster 3), GABARAPL1 (cluster 4), and SRXN1 (cluster 5) —were identified as seed genes (Fig. 2G). These data strongly implicate AURKA in the regulation of ferroptosis, potentially via modulation of the KEAP1/NRF2/HO-1 signaling axis.

To evaluate AURKA expression in EGFR-mutant LUAD, Western blotting and immunohistochemical staining were performed on tumor specimens from six patients. Clinical and demographic characteristics—including sex, age, histological subtype, and TNM stage (classified according to the 9th edition of the AJCC Staging Manual) [33]—are detailed in Table 3. The results demonstrated strong expression of AURKA, along with ferroptosis-associated proteins GPX4 and SLC7A11, in tumor tissues (Fig. 3A–C), suggesting that LUAD with EGFR mutations may exhibit an upregulated ferroptosis-related protein profile, potentially indicating ferroptosis sensitivity.

Fig. 3.

AURKA expression and its functional impact on ferroptosis and proliferation in NCI-H1975 cells. (A) Immunohistochemical showing AURKA expression in six LUAD tumor–normal tissue pairs (red arrows indicate elevated staining), scale bar = 100 µm. (B,C) Western blot analysis (B) and quantification (C) of AURKA, GPX4, and SLC7A11 in LUAD tumor vs. adjacent tissues. (D,E) AURKA expression across LUAD cells lines, with quantification (E). (F) Morphological changes in NCI-H1975 cells after AURKA knockdown (20$\times$, scale bar = 100 µm). (G,H) Flow cytometric (G) and quantification (H) showing increased cell death upon AURKA silencing. (I,J) Clone formation assay and corresponding quantification following AURKA knockdown. (K,L) Transwell assay result and quantification, showing reduced migration and invasion after AURKA knockdown, scale bar = 100 µm. NCI‑H1975, Human non-small cell lung adenocarcinoma cell line NCI-H1975; GPX4, glutathione peroxidase 4; SLC7A11, solute carrier family 7 member 11; NC, negative control; *p 0.05; **p 0.01; ***p 0.001; ****p 0.0001, student’s t-test.

To further investigate the functional relevance of AURKA in LUAD, we examined its expression in the human bronchial epithelial-like cells 16HBE and four LUAD cell lines: A549, NCI-H1975, NCI-H1650, and PC-9. Notably, NCI-H1975 cells harbor the EGFR T790M mutation, which confers primary resistance to first-generation EGFR-TKIs such as gefitinib. This makes NCI-H1975 an ideal model for investigating therapeutic strategies in TKI-resistant LUAD. The results revealed the highest AURKA expression in NCI-H1975 cells (Fig. 3D,E), leading us to select these cells for further experiments.

siRNA-mediated AURKA knockdown was then performed in NCI-H1975 cells. Morphological assessment under a 20$\times$ microscope revealed substantial cell death in the AURKA-silenced group (Fig. 3F). Flow cytometry confirmed a significant increase in cell death following knockdown (p 0.0001; Fig. 3G,H). To assess changes in proliferative and invasive potential, colony formation and Transwell assays were conducted. AURKA knockdown significantly impaired clonogenic capacity (p 0.001; Fig. 3I,J), and also led to marked reductions in both migration and invasion (p 0.01; Fig. 3K,L).

Collectively, these results highlight the critical role of AURKA in promoting LUAD cell viability, proliferation, and metastatic behavior, underscoring its function as a key driver of malignancy in EGFR-mutant LUAD.

To determine whether AURKA regulates ferroptosis in lung cancer cells, we performed siRNA-mediated gene silencing and pharmacologic inhibition using the AURKA-specific inhibitor LY3295668. Both approaches resulted in decreased expression of the ferroptosis suppressors SLC7A11 and GPX4, along with elevated level of 4-Hydroxynonenal (4HNE) increased a marker of lipid peroxidation (Fig. 4A,B). These findings suggest that AURKA activity may negatively regulate ferroptosis in LUAD cells.

Fig. 4.

Ferroptosis induction following AURKA knockdown or inhibitor in NCI-H1975 cells. (A,B) Western blot and quantification of ferroptosis-related proteins (SLC7A11, GPX4, 4HNE) after AURKA knockdown or LY3295668 treatment. (C,D) Western blot and quantification of ferroptosis markers after AURKA knockdown with or without Fer-1 treatment. (E) Confocal imaging of intracellular Fe²⁺ using FerroOrange (red intensity indicates iron levels, scale bar = 20 µm). (F) Quantification of intracellular MDA levels. (G–I) Lipid peroxidation analysis using C11-BODIPY: dot plots (G), peak plots (H), and statistical summary (I). (J–L) Mitochondrial ROS detection: dot plots (J), peak plots (K), and quantification (L) following AURKA knockdown $\pm$ Fer-1. (M) Mitochondrial membrane potential detection using Mito Red Tracker Red FM (redder color indicates higher mitochondrial membrane potential, scale bar = 100 µm). (N) Quantification of intracellular GSH levels. (O) TEM showing mitochondrial shrinkage and cristae loss (red arrows) after AURKA knockdown $\pm$ Fer-1, scale bar = 1 µm. 4HNE, 4-Hydroxynonenal; GSH, glutathione; MDA, malondialdehyde; ROS, reactive oxygen species; *p 0.05; **p 0.01; ***p 0.001; ****p 0.0001, student’s t-test; ns, not significant.

To further validate this hypothesis, NCI-H1975 cells were treated with siAURKA in the presence or absence of ferroptosis inhibitor Fer-1. AURKA knockdown led to reduced protein levels of SLC7A11 and GPX4and upregulated NRF2 expression (Fig. 4C,D). Importantly, co-treatment with Fer-1reversed these changes, indicating that the observed molecular effects were ferroptosis-dependent.

Given that ferroptosis involves disrupted iron homeostasis, reduced GSH levels, and increased ROS, we subsequently assessed these key features. AURKA silencing led to increased intracellular ferrous iron (Fe²⁺; Fig. 4E), elevated MDA content (Fig. 4F), enhanced lipid peroxides as evidenced by C11-BODIPY staining (Fig. 4G–I), and increased mitochondrial ROS levels (Fig. 4J–L). Additionally, a significant decrease in mitochondrial membrane potential (Fig. 4M) and GSH levels (Fig. 4N) was observed. Transmission electron microscopy further confirmed classic ferroptotic morphological features, including mitochondrial shrinkage and cristae loss, in AURKA-silenced cells (Fig. 4O). All of these alterations were effectively reversed by Fer-1 co-treatment (Fig. 4E–O), confirming that AURKA knockdown induces ferroptosis in NCI-H1975 cells.

To further explore the interaction between AURKA and ferroptosis in the context of pharmacologic induction, we assessed the effects of sorafenib, a multi-targeted anti-cancer drug known to trigger ferroptosis [34, 35]. Compared to sorafenib alone, AURKA knockdown significantly enhanced sorafenib-induced lipid peroxidation (p 0.01; Fig. 5A–C), indicating a synergistic effect. Additional phenotypic assays confirmed that AURKA deletion potentiated sorafenib-induced ferroptosis (Fig. 5D–F). Notably, these effects were reversed by Fer-1, but not by other ferroptosis inhibitors LIP-1, DFO or the necroptosis inhibitor necrostatin‑1 (NEC‑1), suggesting a specific ferroptosis-mediated mechanism. Collectively, these data indicate that AURKA acts as a critical suppressor of ferroptosis in EGFR-mutant LUAD cells, and that its inhibition enhances susceptibility to ferroptotic cell death, particularly in response to sorafenib.

Fig. 5.

AURKA regulates sorafenib-induced ferroptosis in NCI-H1975 cells. (A–C) Flow cytometry of lipid peroxidation in AURKA knockdown + sorafenib vs. sorafenib alone. (D–F) Lipid peroxidation levels following treatment with sorafenib alone or in combination with AURKA knockdown, Fer-1, Lip-1, DFO, or necrostatin-1 (NEC-1); includes dot plots (D), peak plots (E), and quantitative analysis (F). Fer-1, Ferrostatin-1; Lip-1, Liproxstatin-1; DFO, Deferoxamine; Deferoxamine; *p 0.05; **p 0.01; ***p 0.001; ****p 0.0001, student’s t-test; ns, not significant; “+”, with the indicated drug; “–”, without the indicated drug.

AURKA has been widely recognized for its essential function in driving mitosis and controlling the cell cycle [36, 37]. Consistent with this, GSEA analysis in our study identified a significant enrichment of the cell cycle pathway in association with AURKA expression (Fig. 2A; Table 2). To investigate whether AURKA knockdown disrupts cell cycle progression in LUAD cells, flow cytometry was employed to assess the cell cycle distribution, and Western blotting was used to evaluate cyclin expression.

Flow cytometric analysis showed that silencing AURKA resulted in significant G0/G1 phase enrichment, suggesting induction of cell cycle arrest (Fig. 6A). Additionally, Western blot analysis demonstrated reduced expression of cyclin D1 following either AURKA knockdown or treatment with the selective AURKA inhibitor LY3295668 (Fig. 6B,C). These findings collectively support that AURKA suppression induces G1 phase arrest in NCI-H1975 cells.

Fig. 6.

AURKA knockdown induces cell cycle arrest and activates KEAP1/NRF2/H0-1 signaling in NCI-H1975. (A) Cell cycle analysis by flow cytometry showing G0/G1 arrest after AURKA knockdown. (B,C) Western blot and quantification of Cyclin D1 following AURKA knockdown or LY3295668 treatment. (D,E) Western blot and quantification of KEAP1, NRF2, and HO‑1 expression after AURKA knockdown $\pm$ ML385. (F) qRT-PCR analysis of NRF2 mRNA levels following AURKA knockdown $\pm$ ML385. (G,H) Immunofluorescence staining and quantification of NRF2 after AURKA knockdown $\pm$ ML385, scale bar = 100 µm. KEAP1, Kelch-like ECH-associated protein 1; HO‑1, heme oxygenase 1; qRT-PCR, quantitative real-time PCR; *p 0.05; **p 0.01; ***p 0.001; ****p 0.0001, student’s t-test; ns, not significant.

To elucidate the molecular mechanism underlying AURKA-mediated ferroptosis regulation, we performed Western blot analysis in NCI-H1975 cells following AURKA knockdown. Compared to controls, AURKA knockdown increased the protein expression of NRF2 and HO-1, while concurrently reducing KEAP1 levels (Fig. 6D,E). Co-treatment with the NRF2 inhibitor ML385 reversed these effects, suggesting that AURKA regulates ferroptosis, at least in part, through the NRF2/HO‑1 pathway.

Interestingly, qRT-PCR analysis revealed a decrease in NRF2 mRNA levels following AURKA knockdown (Fig. 6F), despite the observed increase in NRF2 protein level. This discordance implies that AURKA may regulate NRF2 post-transcriptionally through KEAP1, a negative regulator of NRF2 stability, thus supporting a mechanism involving protein-level modulation [38].

To further investigate NRF2 subcellular localization, immunofluorescence staining was performed, with DAPI used to label nuclei. Localization was inferred based on fluorescence patterns: signal peripheral to DAPI was considered membrane-associated, perinuclear distribution suggested cytoplasmic localization, and complete colocalization with DAPI indicated nuclear localization. In control NCI-H1975 cells, NRF2 was primarily cytoplasmic (Fig. 6G). AURKA knockdown promoted nuclear accumulation of NRF2 proteins, a change that was effectively reversed by ML385 (Fig. 6G,H). These findings collectively demonstrate that AURKA knockdown activates the NRF2/HO-1 axis by promoting NRF2 nuclear translocation, thereby inducing ferroptosis in EGFR-mutant LUAD cells.