ALDH18A1 silencing inhibits lung adenocarcinoma progression through mitochondrial dysfunction and PPAR signaling pathway activation

Abstract

Background

Lung adenocarcinoma (LUAD) remains a prevalent malignant tumor characterized by a dismal prognosis. This study aimed to reveal potential mitochondria-related biomarkers and explore possible mechanisms.

Methods

Mitochondria-related differentially expressed genes (mitoDEGs) were identified by intersecting LUAD-associated DEGs with mitochondria-associated genes. Hub mitoDEGs were determined via protein-protein interaction (PPI) network analysis and machine learning. Aldehyde dehydrogenase 18 family member A1 (ALDH18A1) was silenced in LUAD cells to investigate its effects on proliferation, invasion, migration, and mitochondrial function. The potential involvement of the peroxisome proliferator-activated receptor (PPAR) signaling pathway in ALDH18A1-associated effects was evaluated in LUAD cells and tumor-bearing mice using the PPARγ antagonist GW9662.

Results

ALDH18A1 was identified as a hub mitoDEG and was highly expressed in LUAD.ALDH18A1 silencing promoted apoptosis, suppressed invasion and migration, and inhibited the proliferation of A549 cells. ALDH18A1 silencing induced mitochondrial dysfunction, as evidenced by elevated reactive oxygen species (ROS), diminished mitochondrial membrane potential, decreased ATP production, increased dynamin related protein 1 (DRP1) and mitofusin 2 (MFN2) expression, and decreased optic atrophy 1 (OPA1) in vitro and in vivo. ALDH18A1 silencing increased PPARγ and fatty acid-binding protein 4 (FABP4) expression, whereas GW9662 notably attenuated the antitumor and mitochondrial dysfunction-related effects of ALDH18A1 knockdown.

Conclusion

ALDH18A1 silencing inhibited LUAD progression and induced mitochondrial dysfunction, which may be partly associated with activation of the PPAR signaling pathway. These findings suggest that ALDH18A1 may serve as a candidate mitochondria-related molecular target for LUAD.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13062-026-00882-5.

Introduction

Globally, lung cancer leads in both incidence and mortality among cancers, with approximately 2.5 million new diagnoses and 1.8 million deaths [1]. It is projected that by 2050, the number of new cases and deaths will increase to 4.62 million and 3.55 million, respectively [2]. Non-small cell lung cancer (NSCLC) constitutes the majority (over 85%) of these cases [3], with lung adenocarcinoma (LUAD) being its most common subtype, particularly in Eastern Asia (including China) [1]. While tobacco consumption is a primary risk factor for LUAD, genetic predisposition, occupational exposures or air pollution in lung carcinogenesis-acting also contribute to lung carcinogenesis independently or in concert with tobacco [4]. Recent clinical and translational studies have further emphasized the molecular heterogeneity and complex regulatory mechanisms underlying lung cancer progression. For example, transforming growth factor beta receptor 3 (TGFBR3) has been reported to be downregulated in different types of lung cancer and is closely associated with tumor invasion, metastasis, angiogenesis, and apoptosis [5]. In addition, the microRNA-451 (miR-451)/ETS variant transcription factor 4 (ETV4)/matrix metallopeptidase 13 (MMP13) signaling axis has been shown to promote epithelial-mesenchymal transition and tumor progression based on clinically collected NSCLC samples [6]. Pharmacological evidence also indicates that sanguinarine suppresses lung cancer cell invasion and migration by promoting reactive oxygen species (ROS) accumulation, inhibiting B-cell lymphoma 2 (Bcl-2), and enhancing Bcl-2-associated X protein (Bax) expression [7]. These studies suggest that lung cancer progression is regulated by multiple molecular events. Therefore, identifying novel molecular biomarkers and regulatory pathways remains essential for improving lung cancer diagnosis and treatment.

Mitochondria are oval-shaped organelles responsible for energy generation calcium homeostasis maintenance [8]. Also, mitochondria regulate various biological processes, comprising proliferation, cell death, and metabolic adaptation [9, 10]. Mitochondrial dynamics preserve mitochondrial structure and function through a coordinated balance between fusion and fission. Fusion is regulated by mitofusin 1 (MFN1), mitofusin 2 (MFN2), and optic atrophy 1 (OPA1), whereas fission is controlled by mitochondrial fission 1 (FIS1) and dynamin-related protein 1 (DRP1). Disruption of this balance impairs mitochondrial homeostasis and contributes to disease pathogenesis [11]. Increased expression of mitochondrial fusion mediator-MFN1 and decreased expression of mitochondrial division regulator-FIS1 are observed in NSCLC patients [12]. FIS1-mediated mitochondrial fission enhances mitophagy in lung cancer stem cells stemness [13]. Additionally, the imbalance of DRP1 and MFN2 promotes mitochondrial fission in lung cancer cell lines, whereas complementary approaches to restore mitochondrial network formation inhibit cell proliferation and induce spontaneous apoptosis [14]. These findings indicate that mitochondrial dynamics are closely involved in lung cancer progression. The peroxisome proliferator-activated receptor (PPAR) signaling pathway is involved in lipid metabolism, mitochondrial biogenesis, and energy homeostasis. This pathway includes three major subtypes: peroxisome proliferator-activated receptor alpha (PPARα), peroxisome proliferator-activated receptor beta/delta (PPARβ/δ), and peroxisome proliferator-activated receptor gamma (PPARγ) [15]. PPARα has been reported to selectively restrict mitochondrial oxidative phosphorylation (OXPHOS) in tumor cells and increase ROS production, thereby inducing mitochondrial oxidative stress in melanoma cells with limited effects on non-tumor cells [16]. In LUAD, treatment with the PPARγ agonist pioglitazone induces metabolic reprogramming by suppressing pyruvate oxidation and reducing glutathione levels, which markedly increases ROS accumulation and promotes cell-cycle arrest [17]. In addition, increased expression of OXPHOS-related genes and PPARγ is associated with worse survival outcomes in patients with LUAD [18]. However, the precise molecular mechanisms linking PPAR signaling to mitochondrial dysfunction and LUAD progression, remain incompletely understood. Aldehyde dehydrogenase 18 family member A1 (ALDH18A1), an ALDH family member, is an enzyme implicated in the conversion of glutamine to proline through glutamate [19]. ALDH18A1 is identified as a prognostic hub gene associated with glutamine metabolism, and its higher expression contributes to poor clinical outcomes and poor survival of LUAD patients [20]. Although ALDH18A1 is linked to mitochondrial energy metabolism and is upregulated in lung cancer [21], its specific role in the regulation of mitochondrial homeostasis in LUAD is not well defined. This study aimed to identify mitochondria-related differentially expressed genes (mitoDEGs) and investigated the effects of silencing ALDH18A1 in LUAD cell proliferation, apoptosis, aggression, and mitochondrial function. Its potential downstream pathways were determined and the role of silencing ALDH18A1 in the downstream pathway was further studied in both LUAD cells and tumor-bearing mice.

Methods

Data collection and preprocessing

Gene expression profiles of LUAD were collected from the Gene Expression Omnibus (GEO, https://www.ncbi.nih.gov/geo/) database by retrieving datasets associated with “lung cancer”, “homosapiens”, and “array”. Four GEO datasets, including GSE7670, GSE10072, GSE32863, and GSE27262, were included in this study. RNA-seq expression profiles and corresponding clinical information of TCGA-LUAD were downloaded from the UCSC Xena database (https://xenabrowser.net/datapages/). In this study, GSE10072 dataset, GSE32863 dataset, and TCGA-LUAD dataset were used as training datasets, and GSE27262 dataset and GSE7670 dataset were used as validation datasets. The GSE10072 dataset (GPL96 [HG-U133A Affymetrix Human Genome U133A Array) comprised 58 LUAD samples and 49 control samples. The GSE32863 dataset (GPL6884, Illumina HumanWG-6 v3.0 expression beadchip) contained 59 matched LUAD and non-tumor lung samples. TCGA-LUAD dataset included 515 LUAD samples and 59 normal samples. The GSE27262 dataset (GPL570 HG-U133_Plus_2 Affymetrix Human Genome U133 Plus 2.0 Array) contained LUAD and normal tissues from 25 stage I LUAD patients. The GSE7670 dataset (GPL96 HG-U133A Affymetrix Human Genome U133A Array) included 33 LUAD samples and 33 adjacent normal tissues.

After removing genes/samples with ≥ 50% missing value, the impute.knn function was applied for imputation, with the number of neighbors of K to complete the missing data. Expression matrices were Log2 + 1 transformed, and training datasets were merged using the inSilicoMerging R package (version 1.0.8) to correct for batch effects.

Differential analysis and enrichment analysis

The limma package (version 3.58.1) was used to identify DEGs (adj.p < 0.05 and |log foldchange| (|log FC|) ≥ 1). Human mitochondrial genes were sourced from the MitoCarta3.0 database (https://www.broadinstitute.org/mitocarta/mitocarta30-inventory-mammalian-mitochondrial-proteins-and-pathways). A Venn diagram was generated using the VennDiagram R package (version 1.8.2) to screen out mitoDEGs by intersecting DEGs and mitochondria-associated genes.

The org.Hs.eg.db R package (version 3.18.0) provided Gene Ontology (GO) annotations, and the Kyoto Encyclopedia of Genes and Genomes (KEGG) rest API (https://www.kegg.jp/kegg/rest/keggapi.html) provided KEGG annotations. Enrichment analysis was performed using the clusterProfiler package (version 4.10.0). GO enrichment analyses were categorized into biological process (BP), molecular function (MF) and cellular component (CC). The gene set size was restricted to 5–5,000 genes, with p < 0.05 and false discovery rate (FDR) < 0.25 considered statistically significant. KEGG enrichment was measured by - log10 (p-value), count value, and enrichment multiple. The most significantly enriched pathway term entries were determined based on p-value values.

Identification of hub genes

A protein-protein interaction (PPI) network for mitoDEGs was constructed using the STRING (version 12.0; https://www.string-db.org/) database, with a confidence interaction score at 0.400. The network was then visualized using Cytoscape software (version 3.10.1; www.cytoscape.org). The degree algorithm was applied to identify the top candidate genes with the highest scores in CytoHubba (version 0.1). In order to further determine key genes, the Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis was performed using the glmnet R package (version 4.1-8) and the support vector machine-recursive feature elimination (SVM-RFE) algorithm was used to select the best feature genes from the candidates using the e1071 R package (version 1.7–14). The overlapping genes shared by LASSO regression and SVM-RFE analysis among the PPI-prioritized candidate genes were defined as hub genes.

Independent-dataset DEG analysis for sensitivity validation

As a sensitivity analysis, DEGs were also identified independently in the GSE10072, GSE32863, and TCGA-LUAD datasets to evaluate whether the merged-dataset strategy introduced potential bias. The same statistical significance threshold of adjusted p < 0.05 was applied, and |logFC| ≥ 0.9 was used to account for potential attenuation of fold-change values across different platforms. Upregulated and downregulated DEGs were analyzed separately. Genes showing consistent differential expression direction across all three datasets and overlapping with MitoCarta3.0 genes were defined as consensus mitoDEGs. These consensus mitoDEGs were compared with the mitoDEGs obtained from the merged-dataset analysis and with the final hub mitoDEGs to assess the robustness of candidate gene identification.

Validation of hub mitoDEG expression, diagnostic value, and prognostic relevance

The expression patterns of the hub mitoDEGs were further validated in the GSE7670, GSE27262, and TCGA-LUAD cohorts. The expression levels of the hub mitoDEGs between LUAD and normal tissue samples were compared and visualized using boxplots. Statistical differences between the two groups were assessed using the Wilcoxon rank-sum test, and p < 0.05 was considered statistically significant. To evaluate the diagnostic performance of the hub mitoDEGs, receiver operating characteristic (ROC) curve analysis was performed in the GSE7670, GSE27262, and TCGA-LUAD cohorts using the R package pROC (version 1.18.5). The area under the curve (AUC) was calculated to assess the ability of each hub gene to distinguish LUAD from normal samples. The prognostic value of the hub mitoDEGs was evaluated by Kaplan–Meier survival analysis. Patients were divided into high- and low-expression groups according to the median expression level of each gene. Overall survival differences between the two groups were assessed using the log-rank test. Hazard ratios (HRs) with 95% confidence intervals (CIs) were calculated to estimate the association between gene expression and patient prognosis.

Bioinformatics analysis of the downstream pathways of target gene

To identify potential downstream pathways associated with ALDH18A1, ALDH18A1-related genes were retrieved from the GeneCards database (https://www.genecards.org/) using “ALDH18A1” as the search term. The GeneCards relevance score was used to rank genes according to their association with ALDH18A1. All genes with available relevance scores were retained without applying an additional relevance-score cutoff. Duplicate gene symbols were removed, and gene names were standardized according to official gene symbols. Finally, 5962 ALDH18A1-related genes were obtained for subsequent analysis. Shared genes were then identified by intersecting ALDH18A1-related genes with mitoDEGs. Functional enrichment analysis of the shared genes was performed using DAVID database (https://davidbioinformatics.nih.gov/). GO and KEGG enrichment analyses were conducted, and terms with p < 0.05 were considered significantly enriched.

Cell culture and cell transfection

Human BEAS-2B cells and LUAD cell lines (A549, H358, and HCC827) were obtained from the iCell Bioscience Inc. (Shanghai, China). BEAS-2B cells were cultured in DMEM (cat. D6429, Sigma-Aldrich, St. Louis, MO, USA), whereas LUAD cells were cultured in RPMI-1640 medium (Sigma-Aldrich). Both media were supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin, and cells were maintained at 37 °C in a humidified atmosphere with 5% CO₂. All cell lines were authenticated by short tandem repeat profiling.

Plasmid transfection was used to silence ALDH18A1. SiRNAs targeting ALDH18A1 (si-ALDH18A1-1, si-ALDH18A1-2, and si-ALDH18A1-3) were transfected into LUAD cells, and non-targeted siRNA (si-NC) was served as a control. The culture medium containing siRNAs was mixed with Lipo3000™ transfection reagent (Thermo Fisher, Massachusetts, USA) for 5 min, and 50 µL of mixture was added into 24-well plates at 37 °C for 48 h. Next, cells were treated with the peroxisome proliferator-activated receptor (PPAR) pathway inhibitor GW9662 (20 µM) for 24 h. The sequences of si-ALDH18A1-1, si-ALDH18A1-2, and si-ALDH18A1-3 were shown in Table S1.

Cell viability assessment

A total of 100 µL of cell suspension (3 × 10³/mL) of A549 cells was added into 96-well plates and incubated at 37 °C until 80% confluence. Then, culture medium was replaced by maintenance medium (2% FBS and 98% DMEM). After incubation for 24 h, 90 µL of maintenance medium and 10 µL of Cell Counting Kit-8 (CCK-8) solution (Beyotime, Shanghai, China) were used for 2 h. The absorbance was read at 450 nm under a microplate reader.

Cell proliferation assessment

A549 cells were seeded into 12-well plates (2 × 10⁴ cell/well) and incubated overnight until 80% confluency. A total of 500 µL of 5-ethynyl-2’ -deoxyuridine (EdU) reagent (10 µM, Beyotime) was used for 2 h. After three times washing with PBS, cells were fixed in 4% paraformaldehyde for 15 min, permeabilized with 1 mL of Triton X-100 (GenStar, VA11410) for 5 min. Click additive solution (0.5 mL) was added into each cell for 30 min at room temperature in the darkness. Nuclear counterstaining was conducted using 4’,6-diamidino-2-phenylindole for 5 min. The positive EdU cells were observed using a fluorescence microscope (Olympus, Tokyo, Japan).

Cell wound scratch assay

A549 cells (1 × 10⁵cell/well) were digested by trypsin and seeded into 6-well plates and incubated for 24 h until 80% confluency. A straight line was drawn vertically with the pipette tip. Cells were washed three times with PBS and incubated with serum-free culture medium at 37 °C with 5% CO₂. Photos were taken at 0 h and 24 h of cultivation, and the ability of migration was assessed using NIH Image-J software.

Transwell assay

The invasion ability of LUAD cells was assessed by Transwell assay. Matrigel was diluted (1:1) with chilled serum-free growth medium and placed directly onto the Transwell cell at 37 °C for 30 min. A549 cells were detached with trypsin and the process was stopped in 10% serum-containing medium. A total of 100 mL cell suspension containing 5 × 10⁵ cells was added to the upper chamber of replicate, and 600 mL of medium containing 20% FBS was added to the lower chamber for 24 h. The residual cells were gently removed with a cotton swab. Invasion cells were fixed with 4% paraformaldehyde for 30 min and stained with 600 µL of 0.1% crystal violet solution for 20 min. Five fields of view were observed and the invasion cells were counted using NIH ImageJ software.

Cell apoptosis

A549 cells (5 × 10⁴ to 1 × 10⁵) were resuspended using PBS and centrifuged at 1000 g for 5 min at 4°C. Annexin V-FITC solution (195 µL) was added to the cells for resuspension, and then 5 µL of Annexin V and 10 µL of propidium iodide were added into A549 for incubation in the dark for 15 min. Apoptotic rates were assessed using a flow cytometer (Cytoflex, BD Biosciences, USA).

Detection of intracellular ROS

A549 cells (5 × 10⁵ cell/well) were seeded into 12-well plates and incubated for 24 h until 80% confluence. After washing with PBS, cells were incubated with 10 µM of 2’,7’-Dichlorodihydrofluorescein diacetate (DCFH-DA) at 37 °C for 15 min. Then, cells were treated with 200 µL of Hoechst 33,258 staining solution at 37 °C for 10 min. Fluorescence was assessed using a fluorescence microscope, and the fluorescence intensity was analyzed by ImageJ software.

The production of ATP

The cellular ATP concentrations were measured using the ATP assay kit (S0026, Beyotime). A549 cells were collected for lysis using the lysis buffer. Tumor tissues were treated with 100–200 µL of lysis buffer per 20 mg of tissue and homogenized. Supernatant was harvested after centrifugation at 12,000 g for 5 min at 4 °C. The ATP standard solution was diluted with the ATP lysis buffer, and a working solution was prepared at a ratio of 100 µL of ATP per sample or standard. A total of 100 µL of working solution was added into each well for 5 min, and 20 µL of supernatant was added into each well. The absorbance was read with a multifunctional microplate reader, and the concentration of ATP was calculated based on the standard curve.

Mitochondrial membrane potential analysis

The changes in mitochondrial membrane potential of LUAD cells were assessed using a JC-1-based assay kit (Beyotime). A549 cells (1.0 × 10⁴ cell/well) were placed in 6-well plates at 37 °C for 24 h. Thereafter, 1 mL of JC-1 staining solution was introduced into each well and incubated at 37 °C for 20 min. For the staining procedure, JC-1 staining buffer (1 ×) was prepared by adding 4 mL of distilled water per 1 mL of JC-1 staining buffer (5 ×), followed by adding 2 mL of culture medium. Fluorescence microscope was utilized to evaluate mitochondrial membrane potential.

Animals and construction of a mouse tumor-bearing model

Nude BALB/c mice (female, 6 weeks old, 18–20 g) were procured from SPF (Beijing) biotechnology Co. Ltd, and fed under 22 ± 2°C and 50–60% humidity with standard food and tap water. Animals were randomly allocated into the sh-NC, sh-ALDH18A1, and sh-ALDH18A1 + GW9662 groups (n ≥ 6 per group). Tumor-bearing models were induced by subcutaneously injection of 0.1 mL of cell suspension of A549 cells (5 × 10⁷cells/mL) obtained from si-NC group, si-ALDH18A1 group, and si-ALDH18A1 + GW9662 group aforementioned. Four weeks after injection, the animals were euthanized by 2% isoflurane inhalation and tumor samples were harvest immediately. Tumor dimensions (major axis a and minor axis b) were recorded using a vernier caliper. Tumor volumes were computed as follows: V = 0.52 × a × b² (a was the major axis and b was the minor axis). All animal experimental procedures were approved from the Animal Care and Use Committee and complied with guidelines for laboratory animal use and care.

Hematoxylin-eosin (HE) staining

Tumor tissues of mice were fixed with 4% paraformaldehyde (Beyotime) for 24 h and dehydrated in gradient ethanol (50%, 70%, 80%, 90%, 95%, and anhydrous ethanol). After dehydration, tissue samples were placed in a solution of anhydrous ethanol and xylene (1:1) for 30 min. Slices with 4-µm thickness were prepared, stained with hematoxylin for 10 min and 0.5% eosin for 30 s, and dehydrated through an alcohol-xylene series. Slides were sealed with neutral resin, and pathological changes of tumor tissues were examined under a microscope (Olympus).

Immunohistochemistry (IHC)

The expression of Ki67 in tumor tissues was detected using IHC. Briefly, paraffin-embedded section of tumor tissues was dewaxed using xylene for 40 min and hydrated using gradient ethanol (anhydrous ethanol, 95%, 85%, and 75%) for 5 min, respectively. After rinsing with double-distilled (DD) water twice, sodium citrate solution (10 mmol/L) was added for antigen retrieval. Slides were immersed in PBS-Tween buffer solution (10 ×, pH7.4, Biosharp, Hefei, China) for 5 min. Endogenous peroxidase was inactivated using a mixture of 0.3% hydrogen peroxide andmethanol solution at a ratio of 1:9 for 30 min at 4 °C. Then, 3% bovine serum albumin was added on slides for 30 min, and the primary rabbit monoclonal antibody (1:400, ab16667) was used overnight at 4 °C. The secondary goat anti-rabbit antibody (1:200, Beyotime) was directly dropped onto the tissue at room temperature for 80 min. Color development was achieved with 3,3’-Diaminobenzidine for 3 min, and counterstaining was performed with 80–100 µL hematoxylin staining solution for 5 min. Gradient ethanol (75%, 85%, 95%, and anhydrous ethanol) and xylene were used for dehydration, and slides were then sealed with neutral resin. IHC images were captured using a microscope (Olympus).

Enzyme linked immunosorbent assay (ELISA)

ROS levels in tumor tissues from tumor-bearing mice were determined using an ELISA kit (Esebio, Shanghai, China). Standard substances (50 µL) of varying concentrations were added to standard sample wells. Then, 10 µL of the sample and 40 µL of sample diluent were added in test wells. Horseradish peroxidase (HRP)-labeled antibodies (100 µL) were introduced and plates were incubated at 37 °C for 60 min. After washing by scrubbing solution, 50 µL of substrate A and B were added, and plates were incubated at 37 °C in the dark for 15 min. Absorbance was measured at 450 nm within 15 min.

Real-time quantitative polymerase chain reaction (RT-qPCR)

TRIZOL reagent (Invitrogen, California, USA) was used to extract total RNA from LUAD cells and tumor tissues. RNA purity was assessed by measuring the OD260/OD280 ratio (1.9-2.0) after 20-fold dilution. The synthesis of cDNA was completed using the Hiscript II QRT Supermix for qPCR reverse transcription kit (Vazyme, Nanjing, China) under the following conditions: 25 °C for 5 min, 42 °C for 30 min, and 85 °C for 5 s. RT-qPCR program was performed on an ABI7500 instrument (Applied Biosystems, Foster City, CA, USA) with denaturation at 95 °C for 30 s, followed by 40 cycles of denaturation at 95 °C for 10 s and annealing at 60 °C for 30 s. GAPDH served as the endogenous control. Gene expression was quantified via the 2^−ΔΔCt method [22]. Primers used for RT-qPCR are provided in Table S1.

Western blot analysis

Lysis buffer was used to extract total proteins from LUAD cells and tumor tissue samples, and protein concentration was determined with a bicinchoninic acid (BCA)protein assay kit (Beyotime). Proteins were electrophoresed using 10% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto polyvinylidene fluoride (PVDF) membranes for 2 h at 4 °C. Membranes were blocked with 5% skimmed milk for 2 h and incubated overnight with primary rabbit monoclonal antibodies against ALDH18A1 (1:1000, ab315817), DRP1 (1:1000, ab184247), MFN2 (1:1000, ab124773), OPA1 (1:1000, ab157457), peroxisome proliferator-activated receptor γ (PPARγ, 1:1000, ab45036), and fatty acid binding protein 4 (FABP4, 1:1000, ab92501) at 4 °C. Membranes were then incubated with HRP-labeled secondary goat anti-rabbit IgG (1:5000, A0208, Beyotime) for 1 h at room temperature. The membrane was transferred to the Tanon 5200 chemiluminescence imaging system, and enhanced chemiluminescence reagent (BeyoECL Plus, Beyotime) was used for color development. The expression of proteins was analyzed using NIH ImageJ software.

Statistical analyses

Bioinformatics statistical analyses were performed using R software (version 4.3.2). Differential expression was assessed using the limma package with Benjamini–Hochberg correction, and group differences in validation datasets were analyzed using the Wilcoxon test. ROC curves were used to evaluate diagnostic performance, and Kaplan–Meier survival analysis with log-rank tests was used to assess prognostic value. All experimental data were analyzed using GraphPad Prism software (version 9.0; GraphPad Software, San Diego, CA, USA), and measurement data are presented as mean ± standard deviation. Each in vitro experiment was independently repeated at least three times. For animal experiments, at least six mice were included in each group. Two-group comparisons were made using the t-test, while multiple groups were compared via one-way analysis of variance (ANOVA), followed by Tukey’s post hoc test. p < 0.05 indicated statistically significant.

Results

Identification, enrichment analysis, and hub gene screening of mitoDEGs

Before correction, the expression distributions of the GSE10072, GSE32863, and TCGA-LUAD datasets differed markedly, and UMAP analysis showed clear dataset-specific clustering (Fig. S1A). After correction, the density curves became more consistent, and samples from different datasets were more uniformly intermixed in the UMAP plot (Fig. S1B). A total of 631 LUAD samples and 166 normal tissue samples were obtained for differential expression analysis after merging the GSE10072, GSE32863, and TCGA-LUAD datasets. Subsequently, 1,279 DEGs were identified between LUAD and normal samples, including 644 upregulated and 635 downregulated genes (Fig. 1A). By intersecting these DEGs with 1,136 mitochondria-associated genes from MitoCarta3.0, 44 mitoDEGs were obtained (Fig. 1B). Functional enrichment analyses via KEGG displayed that mitoDEGs were primarily involved in energy metabolism pathways including carbon metabolism, citrate cycle (TCA cycle), and pyruvate metabolism, as well as in several amino acids related pathways such as biosynthesis of amino acids, arginine and proline metabolism, glycine, serine and threonine metabolism, and tryptophan metabolism. In GO-BP category, mitoDEGs were predominately enriched in some metabolic processes such as small molecule metabolic process, oxidation-reduction process, carboxylic acid metabolic process, oxoacid metabolic process, and organic acid metabolic process. These mitoDEGs were significantly abundant in mitochondrion-related CC such as mitochondrion, mitochondrial envelope, mitochondrial membrane, mitochondrial matrix, and mitochondrial inner membrane. In MF category, mitoDEGs were mainly enriched in catalytic activity, small molecule binding, anion binding, nucleotide binding, and cofactor binding (Fig. S2).

Fig. 1.

Identification of hub mitochondria-related differentially expressed genes (mitoDEGs) in lung adenocarcinoma (LUAD). (A) Volcano plots of 1279 DEGs between LUAD and normal samples, including 644 upregulated genes and 635 downregulated genes. Red points indicate DEGs. (B) Venn diagram shows 44 mitoDEGs at the intersection of DEGs and mitochondria-associated genes from MitoCarta3.0. (C) A highly interconnected gene cluster is identified based on the degree algorithm analysis. The node color gradient indicates the degree score, with red representing higher connectivity and yellow representing lower connectivity. (D) Cross-validation curve of least absolute shrinkage and selection operator (LASSO) regression analysis. The x-axis represents log(λ), and the y-axis represents binomial deviance. The numbers at the top indicate the number of candidate genes retained at each λ value. (E) LASSO coefficient profiles of candidate genes. The x-axis represents log(λ), and the y-axis represents the regression coefficient of each candidate gene. Each colored line represents one candidate gene. (F). Support vector machine-recursive feature elimination (SVM-RFE) analysis showing the selection of optimal feature genes. The x-axis represents the number of variables retained, and the y-axis represents the root mean square error (RMSE) of cross-validation. The point with the lowest RMSE indicates the optimal number of feature genes

To further prioritize hub genes among the 44 mitoDEGs, a PPI network was constructed (Fig. S3). Using the degree algorithm, we screened nine top hub genes, including serine hydroxymethyltransferase 2 (SHMT2), ALDH18A1, pyruvate carboxylase (PC), aldehyde dehydrogenase 2 family member (ALDH2), isocitrate dehydrogenase 2 (IDH2), methylenetetrahydrofolate dehydrogenase 2 (MTHFD2), acetyl-CoA carboxylase beta (ACACB), phosphoribosylaminoimidazole carboxylase and phosphoribosylaminoimidazolesuccinocarboxamide synthase (PAICS), and acyl-CoA dehydrogenase long chain (ACADL) (Fig. 1C). LASSO regression analysis confirmed these nine genes as candidate genes (Fig. 1D and E), while SVM-RFE algorithm identified eight feature genes (Fig. 1F). According to the Venn diagram, overlap analysis indicated eight common hub genes (SHMT2, ALDH18A1, PC, ALDH2, IDH2, ACACB, PAICS, and ACADL) (Fig. 1G). To evaluate whether the merged-dataset strategy introduced potential bias, we performed an independent-dataset DEG analysis as a sensitivity analysis. DEGs were identified separately in the GSE10072, GSE32863, and TCGA-LUAD datasets and then intersected with mitochondria-associated genes from MitoCarta3.0. This analysis identified 10 consensus mitoDEGs, including seven downregulated genes and three upregulated genes (Fig. S4). The 10 genes included ACADL, catalase (CAT), cytochrome c oxidase subunit 7A1 (COX7A1), pyruvate dehydrogenase kinase 4 (PDK4), ALDH2, monoamine oxidase A (MAOA), monoamine oxidase B (MAOB), PAICS, ALDH18A1, and aldo-keto reductase family 1 member B10 (AKR1B10). Notably, all 10 consensus mitoDEGs were fully contained within the 44 mitoDEGs identified by the merged-dataset analysis. In addition, four of the eight hub mitoDEGs identified in the merged analysis, including ACADL, ALDH2, PAICS, and ALDH18A1, were retained in the independent analysis. These results indicate that the merged-dataset analysis is generally consistent with the independent-dataset analysis and support the robustness of hub gene identification.

Expression validation and prognostic assessment of hub mitoDEGs

The expression patterns and clinical relevance of the eight hub mitoDEGs were evaluated in the GSE7670, GSE27262, and TCGA-LUAD cohorts. In the GSE7670 dataset, ACACB, ACADL, and ALDH2 were significantly downregulated in lung cancer tissues compared with normal tissues, whereas ALDH18A1, IDH2, PAICS, PC, and SHMT2 were significantly upregulated (Fig. 2A). ROC curve analysis showed that these eight hub mitoDEGs exhibited potential diagnostic performance in the GSE7670 dataset, with AUC values ranging from 0.848 to 0.988 (Fig. 2B). In the GSE27262 dataset, ACACB, ACADL, and ALDH2 showed lower expression in LUAD samples, while ALDH18A1, IDH2, PAICS, PC, and SHMT2 showed higher expression (Fig. 2C). The AUC values of these genes ranged from 0.830 to 0.994 (Fig. 2D). In the TCGA-LUAD cohort, expression levels of ACACB, ACADL, and ALDH2 were decreased, whereas ALDH18A1, IDH2, PAICS, PC, and SHMT2 were increased in LUAD tissues compared with normal tissues (Fig. 2E). The eight hub mitoDEGs also showed potential discriminatory ability in TCGA-LUAD, with AUC values ranging from 0.802 to 0.936 (Fig. 2F). Notably, all eight hub mitoDEGs showed consistent expression patterns across the three independent cohorts: ACACB, ACADL, and ALDH2 were consistently downregulated, whereas ALDH18A1, IDH2, PAICS, PC, and SHMT2 were consistently upregulated in tumor tissues. Kaplan-Meier survival analysis demonstrated that high expression levels of ALDH18A1, IDH2, PAICS, and SHMT2 were significantly associated with poorer overall survival in patients with LUAD. In contrast, high expression levels of ACACB, ACADL, and ALDH2 were associated with better survival outcomes. PC expression was not significantly associated with overall survival (Fig. 2G). ALDH18A1 encodes a rate-limiting enzyme involved in proline biosynthesis and links glutamine metabolism to mitochondrial function. Compared with other hub genes, such as SHMT2 and PAICS, the role of ALDH18A1 in LUAD remains relatively underexplored [23–25]. Moreover, ALDH18A1 was consistently upregulated in GSE7670, GSE27262, and TCGA cohorts, and its high expression was significantly associated with poor prognosis (HR = 1.22, 95% CI: 1.08–1.37, log-rank P = 0.0012). Based on its mitochondrial metabolic relevance, consistent upregulation, prognostic significance, and limited characterization in LUAD, ALDH18A1 was selected as the primary candidate for further investigation.

Fig. 2.

Validation of expression patterns and diagnostic and prognostic value of hub mitoDEGs in LUAD. (A) The expression patterns of eight hub genes in normal and tumor samples from the GSE7670 dataset. The x-axis represents individual hub mitoDEGs, and the y-axis represents gene expression levels. (B) Receiver operating characteristic (ROC) curves and area under the curve (AUC) values of eight hub genes in the GSE7670 dataset. The x-axis represents 1 − specificity, and the y-axis represents sensitivity. (C) The expression patterns of eight hub genes in the GSE27262 dataset. The x-axis represents individual hub mitoDEGs, and the y-axis represents gene expression levels. (D) ROC curves and AUC values of eight hub genes in the GSE27262 dataset. The x-axis represents 1 − specificity, and the y-axis represents sensitivity. (E) Boxplots showing the expression levels of eight hub mitoDEGs in normal and tumor samples from the TCGA-LUAD dataset. The x-axis represents individual hub mitoDEGs, and the y-axis represents gene expression levels. (F) ROC curves were used to evaluate the diagnostic performance of the eight hub mitoDEGs in the TCGA-LUAD dataset. The x-axis represents 1 − specificity, and the y-axis represents sensitivity. (G) Kaplan-Meier survival curves of the association between hub mitoDEG expression and overall survival in patients with LUAD. Patients were divided into high- and low-expression groups according to the median expression level of each gene. The x-axis represents survival time in months, and the y-axis represents survival probability. Statistical significance in boxplots was determined using the Wilcoxon test. Prognostic significance was evaluated using the log-rank test. ^****p < 0.0001 vs. normal group

ALDH18A1 silencing promotes apoptosis and inhibits invasion and migration in LUAD cells

We then validated the expression of ALDH18A1 in LUAD cells. RT-qPCR and western blot analysis showed that ALDH18A1 mRNA and protein expression levels were higher in A549, H358, and HCC827 cells than in BEAS-2B cells (p < 0.001), with the highest expression observed in A549 cells (Fig. 3A and B). Therefore, A549 cells were selected for subsequent functional experiments. Transfection with si-ALDH18A1-1, si-ALDH18A1-2, and si-ALDH18A1-3 significantly decreased its mRNA and protein expression levels in A549cells relative to the si-NC group (p < 0.001) (Fig. 3C and D), which indicated good silencing efficiency of si-ALDH18A1. The si-ALDH18A1-1 and si-ALDH18A1-2 were used for further analysis. Transfection of si-ALDH18A1-1 and si-ALDH18A1-2 notably inhibited cell viability at (p < 0.001) and increased apoptosis (p < 0.001) compared with the si-NC group (Fig. 3E and F). Additionally, silencing ALDH18A1 significantly suppressed the invasion ability of A549 cells compared with the si-NC group (p < 0.001) (Fig. 3G). As displayed in Fig. 3H, silencing ALDH18A1 reduced the migration ability of A549 cells compared with the si-NC group (p < 0.01). EdU assays indicated that silencing ALDH18A1 inhibited the proliferation of A549 cells compared with the si-NC group (p < 0.01) (Fig. 3I). Collectively, silencing ALDH18A1 increased apoptosis and inhibited invasion and migration abilities of LUAD cells.

Fig. 3.

Silencing ALDH18A1 promotes apoptosis and inhibits invasion and migration in LUAD cells. (A) Real-time quantitative polymerase chain reaction (RT-qPCR) quantifies the expression of ALDH18A1 in A549, H358, and HCC827 cell lines. (B) Western blot analysis of ALDH18A1 protein expression in BEAS-2B, A549, H358, and HCC827 cells. (C-D) Cell transfection efficiency was detected using RT-qPCR and western blot after transfection with si-ALDH18A1-1, si-ALDH18A1-2, and si-ALDH18A1-3 in A549 cells. (E) Cell viability of A549 cells after transfection with si-ALDH18A1-1 and si-ALDH18A1-2 was measured by Cell Counting Kit-8 (CCK-8) assay. (F) The apoptosis rate of A549 cells after transfection with si-ALDH18A1-1 and si-ALDH18A1-2 was detected by flow cytometry. (G) The transwell assay was used to detect the invasion ability of A549 cells after ALDH18A1 silencing (Scale bar = 100 μm). (H) The wound scratch assay was used to assess the migration ability of A549 cells after ALDH18A1 silencing. (I) Representative images and quantitative analysis of 5-ethynyl-2′-deoxyuridine (EdU) staining in A549 cells after ALDH18A1 silencing. 4′,6-Diamidino-2-phenylindole (DAPI) was used to stain nuclei, EdU-positive cells indicate proliferating cells, and the y-axis represents the percentage of EdU-positive cells (Scale bar = 100 μm). Data are presented as the mean ± standard deviation. Statistical significance was determined by one-way analysis of variance (ANOVA) followed by Tukey’s post hoc test. N ≥ 3. ^*p < 0.05, ^**p < 0.01, and ^***p < 0.001 vs. si-NC group

ALDH18A1 silencing induces mitochondrial dysfunction and alters mitochondrial dynamics in LUAD cells

We next investigated whether ALDH18A1 silencing affects mitochondrial function in LUAD cells. DCFH-DA staining showed that both si-ALDH18A1-1 and si-ALDH18A1-2 significantly increased intracellular ROS levels in A549 cells compared with the si-NC group (p < 0.001) (Fig. 4A). Mitochondrial membrane potential was then assessed using JC-1 staining. Compared with the si-NC group, ALDH18A1 silencing markedly reduced the JC-1 red/green fluorescence ratio, suggesting loss of mitochondrial membrane potential (p < 0.01) (Fig. 4B). Consistently, ATP detection showed that ATP levels were significantly decreased in the si-ALDH18A1-1 and si-ALDH18A1-2 groups compared with the si-NC group (p < 0.01) (Fig. 4C). Western blot showed that the expression levels of DRP1 and MFN2 were upregulated, whereas the expression of OPA1 was significantly decreased after transfection of si-ALDH18A1-1 and si-ALDH18A1-2 in A549 cells (p < 0.001) (Fig. 4D). Collectively, these findings demonstrate that ALDH18A1 silencing induces mitochondrial dysfunction in A549 cells.

Fig. 4.

ALDH18A1 silencing induces mitochondrial dysfunction and alters mitochondrial dynamics-related protein expression in A549 cells. (A) Intracellular reactive oxygen species (ROS) levels in A549 cells after ALDH18A1 silencing were detected using 2′,7′-dichlorodihydrofluorescein diacetate (DCFH-DA) staining (scale 100 μm; magnification 200 ×). (B) Mitochondrial membrane potential was assessed using JC-1 staining. Red fluorescence indicates JC-1 aggregates, green fluorescence indicates JC-1 monomers, and merged images show the overall mitochondrial membrane potential status. Quantitative analysis of the JC-1 red/green fluorescence ratio is shown on the right. The y-axis represents the JC-1 red/green ratio (scale 100 μm; magnification 200 ×). (C) The relative level of ATP in A549 cells after silencing ALDH18A1. The y-axis represents the relative ATP level compared with the si-NC group. (D) Western blot analysis was used to determine the protein levels of dynamin-related protein 1 (DRP1), mitofusin 2 (MFN2), and optic atrophy 1 (OPA1) in A549 cells after ALDH18A1 silencing. Data are presented as mean ± standard deviation. Statistical significance was determined by one-way ANOVA followed by Tukey’s post hoc test. N ≥ 3. ^**p < 0.01 and ^***p < 0.001 vs. si-NC group

The PPAR signaling pathway is a potential downstream pathway of ALDH18A1

To identify the potential downstream pathway of ALDH18A1, a total of 5962 ALDH18A1-related genes were screened, among which 37 overlapped with 44 mitoDEGs (Fig. 5A). KEGG enrichment analysis showed that these overlapping genes were mainly enriched in arginine and proline metabolism, glycine, serine and threonine metabolism, tryptophan metabolism, histidine metabolism, carbon metabolism, the citrate cycle (TCA) cycle, pyruvate metabolism, biosynthesis of cofactors, and the PPAR signaling pathway (Fig. 5B). In the GO-BP category, overlapped genes were abundant in carboxylic acid metabolic process, carboxylic acid biosynthetic process, and organic acid biosynthetic process. In the CC category, these genes were enriched in mitochondrial membrane, mitochondrial inner membrane, and organelle inner membrane. In the MF category, genes were enriched in heme binding, tetrapyrrole binding, and vitamin binding (Fig. 5C). Because the enrichment analysis was based on genes overlapping between ALDH18A1-related genes and mitoDEGs, broad mitochondrial terms such as the TCA cycle and mitochondrial membrane were expected and were considered background enrichment. The PPAR signaling pathway is closely associated with lipid metabolism, mitochondrial oxidative metabolism, ROS regulation, and tumor progression [26, 27]. Moreover, previous evidence has shown that the PPARγ/FABP4 axis suppresses lung cancer growth through ROS-related mechanisms [28], consistent with the mitochondrial changes observed after ALDH18A1 silencing. Therefore, PPARγ and its downstream effector FABP4 were selected for subsequent validation. RT-qPCR analysis showed that ALDH18A1 silencing significantly increased the mRNA expression levels of PPARγ and FABP4 compared with the si-NC group (p < 0.001) (Fig. 5D). Consistently, western blot analysis demonstrated that the protein levels of PPARγ and FABP4 were also significantly elevated after ALDH18A1 knockdown (Fig. 5E). These results suggest that ALDH18A1 silencing is associated with activation of the PPARγ/FABP4 axis in LUAD cells.

Fig. 5.

The PPAR signaling pathway is identified as a downstream pathway of ALDH18A1. (A) Venn diagram displays a total of 37 shared genes at the intersection of ALDH18A1-related genes obtained from the GeneCards database and 44 mitoDEGs. (B) Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis of the overlapping genes. The y-axis represents enriched KEGG pathways, and the x-axis represents the enrichment ratio. Dot size indicates the number of enriched genes, and dot color represents −log10(p value). The PPAR signaling pathway is highlighted with a red box and arrow. (C) Gene Ontology (GO) enrichment analysis of the overlapping genes, including biological process (BP), cellular component (CC), and molecular function (MF) categories. The y-axis represents enriched GO terms, and the x-axis represents the enrichment ratio. Dot size indicates the number of enriched genes, and dot color represents −log10(p value). (D-E) RT-qPCR and western blot was used to quantify the expression of peroxisome proliferator-activated receptor gamma (PPARγ) and fatty acid-binding protein 4 (FABP4) in A549 cells after ALDH18A1 silencing. Data are presented as mean ± standard deviation. Statistical significance was determined using Student’s t-test. N ≥ 3. ^**p < 0.01 and ^***p < 0.001 vs. si-NC group

ALDH18A1 silencing suppresses LUAD cell progression and alters mitochondrial function in association with PPAR signaling pathway activation

To investigate whether the PPAR signaling pathway was involved in ALDH18A1-mediated regulation of LUAD cells, A549 cells were treated with the PPAR pathway inhibitor GW9662 after ALDH18A1 knockdown. A si-NC + GW9662 group was included as a GW9662-only control. CCK-8 assays showed that ALDH18A1 silencing significantly reduced cell viability compared with the si-NC group. GW9662 treatment partially restored cell viability in ALDH18A1-silenced cells. However, compared with the si-NC + GW9662 group, the si-ALDH18A1 + GW9662 group still showed lower cell viability, indicating that GW9662 did not completely abolish the inhibitory effect of ALDH18A1 knockdown (Fig. 6A). Flow cytometry analysis showed that ALDH18A1 silencing markedly increased the apoptotic rate compared with the si-NC group. GW9662 treatment significantly attenuated apoptosis in ALDH18A1-silenced cells. Nevertheless, apoptosis remained higher in the si-ALDH18A1 + GW9662 group than in the si-NC + GW9662 group, suggesting that ALDH18A1 knockdown still promoted apoptosis under GW9662 treatment (Fig. 6B). Transwell and wound healing assays were used to evaluate cell invasion and migration. ALDH18A1 silencing significantly decreased the invasive ability and wound healing rate of A549 cells compared with the si-NC group. GW9662 treatment partially reversed these inhibitory effects. Compared with the si-NC + GW9662 group, cells in the si-ALDH18A1 + GW9662 group still exhibited reduced invasion and migration (Fig. 6C and D). Mitochondrial functional changes were then assessed. DCFH-DA staining showed that ALDH18A1 silencing significantly increased intracellular ROS levels compared with the si-NC group. GW9662 treatment reduced ROS accumulation in ALDH18A1-silenced cells, but ROS levels remained higher in the si-ALDH18A1 + GW9662 group than in the si-NC + GW9662 group (Fig. 6E). JC-1 staining further showed that ALDH18A1 knockdown decreased the JC-1 red/green fluorescence ratio, indicating a loss of mitochondrial membrane potential. GW9662 treatment partially restored the JC-1 red/green ratio, whereas the ratio remained lower in the si-ALDH18A1 + GW9662 group than in the si-NC + GW9662 group (Fig. 6F). Western blot analysis showed that ALDH18A1 expression was markedly reduced after si-ALDH18A1 transfection. ALDH18A1 silencing increased the protein expression levels of DRP1, MFN2, PPARγ, and FABP4 compared with the si-NC group. GW9662 treatment reduced these increases in ALDH18A1-silenced cells. Compared with the si-NC + GW9662 group, the si-ALDH18A1 + GW9662 group still showed lower ALDH18A1 expression and higher DRP1, MFN2, PPARγ, and FABP4 expression (Fig. 6G). Together, these findings indicate that GW9662 partially reverses the effects of ALDH18A1 knockdown, suggesting that the PPARγ/FABP4 axis contributes to, but does not fully account for, ALDH18A1-mediated regulation of LUAD cell malignant phenotypes and mitochondrial dysfunction.

Fig. 6.

GW9662 partly reverses the effects of ALDH18A1 knockdown on malignant phenotypes and mitochondrial dysregulation in LUAD cells in vitro. A549 cells were transfected with si-NC or si-ALDH18A1 and treated with or without the PPARγ antagonist GW9662. (A) Cell viability was detected by CCK-8 in each group after adding GW9662 in si-ALDH18A1 group. (B) Cell apoptotic rate was assessed using flow cytometry after adding GW9662 in si-ALDH18A1 group. (C) The transwell assay was used to detect the invasion ability in the si-NC, si-ALDH18A1, si-ALDH18A1 + GW9662, and si-NC + GW9662 groups (Scale bar = 100 μm). (D) Cell migration in the si-ALDH18A1 group following GW9662 treatment was assessed using a wound healing assay. (E) The level of intracellular ROS after GW9662 treatment in ALDH18A1-silenced cells was detected using DCFH-DA staining (scale 100 μm; magnification 200 ×). (F) JC-1 probe detects mitochondrial membrane potential after treatment of GW9662 in si-ALDH18A1 group (scale 100 μm; magnification 200 ×). (G) Western blot analysis determines the protein levels of DRP1, MFN2, PPARγ and FABP4 after treatment of GW9662 in si-ALDH18A1 group. Data are presented as mean ± standard deviation. Statistical significance was determined by one-way ANOVA followed by Tukey’s post hoc test. N ≥ 3. ^*p < 0.05, ^**p < 0.01, and ^***p < 0.001 vs. si-NC, si-ALDH18A1 or s-NC + GW9662 group

ALDH18A1 knockdown suppresses LUAD progression in vivo and is associated with PPAR signaling pathway activation and mitochondrial dysfunction

To validate the role of ALDH18A1 in vivo and evaluate whether the PPAR signaling pathway was involved in this process, a tumor-bearing mouse model was established. Four groups were included: sh-NC, sh-ALDH18A1, sh-ALDH18A1 + GW9662, and sh-NC + GW9662. The sh-NC + GW9662 group was used as a GW9662-only control. RT-qPCR analysis showed that ALDH18A1 expression was significantly decreased in the sh-ALDH18A1 group compared with the sh-NC group. GW9662 treatment significantly increased ALDH18A1 expression in sh-ALDH18A1 tumors, whereas ALDH18A1 expression remained lower in the sh-ALDH18A1 + GW9662 group than in the sh-NC + GW9662 group (Fig. 7A). Tumor growth analysis showed that ALDH18A1 knockdown markedly reduced tumor volume compared with the sh-NC group. GW9662 administration notably restored tumor growth in ALDH18A1-knockdown mice. However, tumor volume in the sh-ALDH18A1 + GW9662 group remained lower than that in the sh-NC + GW9662 group, indicating that GW9662 did not completely abolish the tumor-suppressive effect associated with ALDH18A1 knockdown (Fig. 7B and C). HE staining showed reduced tumor cell density and increased tissue damage in the sh-ALDH18A1 group compared with the sh-NC group. These changes were partially attenuated by GW9662 treatment. In contrast, tumors in the sh-NC + GW9662 group showed densely arranged tumor cells, whereas those in the sh-ALDH18A1 + GW9662 group displayed a relatively sparse tumor cell distribution. (Fig. 7D). IHC staining showed that Ki67-positive cells were significantly reduced after ALDH18A1 knockdown. GW9662 treatment increased Ki67 expression in ALDH18A1-knockdown tumors, whereas Ki67 expression remained lower in the sh-ALDH18A1 + GW9662 group than in the sh-NC + GW9662 group (Fig. 7E and F). Mitochondrial functional indicators were then examined. Compared with sh-NC tumors, ALDH18A1-deficient tumors exhibited significantly increased ROS levels, which were reduced following GW9662 administration. In addition, ROS levels remained higher in the sh-ALDH18A1 + GW9662 group than in the sh-NC + GW9662 group (Fig. 7G). Conversely, ALDH18A1 knockdown markedly decreased ATP levels relative to the sh-NC group. Although GW9662 treatment substantially restored ATP generation, ATP levels were still lower in sh-ALDH18A1 + GW9662 tumors than in sh-NC + GW9662 tumors (Fig. 7H). Western blot analysis showed that ALDH18A1 protein expression was decreased in the sh-ALDH18A1 group. Meanwhile, the expression levels of DRP1, MFN2, PPARγ, and FABP4 were increased after ALDH18A1 knockdown compared with the sh-NC group. GW9662 treatment partially reversed these protein changes in ALDH18A1-knockdown tumors. However, compared with the sh-NC + GW9662 group, the sh-ALDH18A1 + GW9662 group still showed lower ALDH18A1 expression and higher DRP1, MFN2, PPARγ, and FABP4 expression (Fig. 7I and J). Collectively, knockdown of ALDH18A1 inhibited LUAD progression in mice and may be associated with PPAR signaling pathway activation and mitochondrial dysfunction.

Fig. 7.

GW9662 partially attenuates the effects of ALDH18A1 knockdown on tumor growth, mitochondrial dysfunction, and PPARγ/FABP4 signaling in vivo. (A) RT-qPCR quantifies the expression levels of ALDH18A1 in tumor tissues. (B) Tumor volume was measured in each group. (C) Representative gross images of excised tumors from each group. (D) Hematoxylin and eosin (HE) staining analyzes the changes of the tumor tissues (scale 50 μm; magnification 400 ×). (E-F) Immunohistochemistry staining (IHC) was used to detect Ki67 expression in tumor tissues, and the percentage of Ki67-positive cells was quantified. (scale 50 μm; magnification 400 ×). (G) The production of ROS in the tumor tissues after treatment of GW9662 in ALDH18A1 knockdown mice. (H) The relative level of ATP in the tumor tissues after treatment of GW9662 in ALDH18A1 knockdown mice. (I-J) Western blot analysis was used to detect the protein levels of DRP1, MFN2, PPARγ and FABP4 after treatment of GW9662 and ALDH18A1 knockdown. Data are presented as mean ± standard deviation. Statistical significance was determined by one-way ANOVA followed by Tukey’s post hoc test. N ≥ 6. ^*p < 0.05, ^**p < 0.01, and ^***p < 0.001 vs. sh-NC, sh-ALDH18A1, or sh-NC + GW9662 group

Discussion

We identified eight hub mitoDEGs, among which ACACB, ACADL, and ALDH2 were downregulated in LUAD samples, whereas SHMT2, ALDH18A1, PC, IDH2, and PAICS were upregulated in LUAD. These hub mitoDEGs showed potential discriminatory ability in LUAD. Silencing ALDH18A1 enhanced apoptosis and suppressed invasion and migration of LUAD cells. Additionally, silencing ALDH18A1 induces mitochondrial dysfunction via increasing DRP1 and MFN2 but decreasing OPA1 in LUAD cells. The PPAR signaling pathway was identified as a potential downstream effector of ALDH18A1. ALDH18A1 silencing inhibited LUAD progression and induced mitochondrial dysfunction, at least partly through activation of the PPAR signaling pathway.

DRP1 is a major mediator of mitochondrial fission, whereas MFN2 and OPA1 regulate the fusion of the outer and inner mitochondrial membranes, respectively [29]. Recent studies have shown that mitochondrial dynamics-related genes, including MFN1, MFN2, and DRP1, are associated with the risk of lung cancer [30]. In addition, regulation of mitochondrial dynamics mediated by the p32/OPA1 axis has been shown to be associated with cisplatin resistance in NSCLC [31]. These studies indicate that mitochondrial dynamics are closely associated with lung cancer development, progression, and therapeutic response. Regarding DRP1, while it is broadly considered a pro-tumorigenic factor in many solid tumors, its function in lung cancer appears highly context-dependent. In lung cancer cells, sustained or moderate DRP1 activation, particularly mitochondrial fission driven by Ser616 phosphorylation, is frequently associated with proliferation, invasion, migration, recurrence, and therapeutic resistance. For example, extracellular signal-regulated kinase (ERK)/protein kinase B (AKT)- and cyclin-dependent kinase 2 (CDK2)-mediated phosphorylation of DRP1 at serine 616 activates mitochondrial fission signaling and promotes the proliferation and invasion of LUAD cells [32]. Leflunomide/teriflunomide reduces DRP1 Ser616 phosphorylation and mitochondrial fragmentation, and synergizes with carboplatin or lurbinectedin to exert antitumor effect [33]. However, tissue array analysis of human specimens reveals that DRP1 expression is significantly reduced in malignant lung cancer tissues compared with adjacent normal tissues, with decreased expression observed in approximately 78% of lung cancer patients. Moreover, the loss of DRP1 expression is more pronounced in tumors with advanced grades [34]. In addition, DRP1-mediated mitochondrial fission contributes to baicalein-induced apoptosis and autophagy in lung cancer via activation of adenosine monophosphate-activated protein kinase (AMPK) signaling pathway [35]. QiDongNing triggers NSCLC cell apoptosis by enhancing mitochondrial fission mediated by the p53/DRP1 axis [36]. These findings position DRP1-driven fission as a pro-death rather than pro-survival mechanism under specific therapeutic or stress-related conditions. Moreover, upregulated DRP1 facilitates T-cell activation, potentiates the inhibitory effects of T cells on lung cancer cells, increases CD8 + T-cell infiltration in tumors and the spleen, and significantly reinforces pembrolizumab-mediated antitumor immunity [37]. Therefore, the paradoxical role of DRP1 in lung cancer may essentially reflect the difference between tumor cell-intrinsic pro-survival fission and immune cell-mediated antitumor fission or stress-induced apoptotic fission.

The role of MFN2 in lung cancer is equally complex. MFN2 expression is found to be significantly elevated in LUAD tissues compared to adjacent normal tissues, with knockdown of MFN2 disrupting cell proliferation, cell cycle, and invasion [38]. Other evidence supports a tumor-suppressive role of MFN2. In LUAD, MFN2 deficiency reduces uncoupling protein 4 (UCP4) expression, impairs ATP production and calcium homeostasis, and induces mitochondrial dysfunction. Low MFN2/UCP4 expression is also associated with poor prognosis [39]. Consistently, NSCLC patients with low MFN2 expression have shorter mean survival times, and MFN2 protein levels are lower in tumor tissues than in adjacent normal tissues. Moreover, Shenmai injection enhances cisplatin-induced apoptosis by upregulating MFN2 expression [40]. Mitochondria-shaping protein OPA1 is a mitochondrial fusion protein, and counteracts cellular senescence induced by extensive mitochondrial elongation [41]. Highly expressed OPA1 considerably promotes mitochondrial fusion and stemness of cancer stem-like cells (CSCs) in lung cancer [42]. In LUAD, OPA1 is upregulated in tumor tissues, and its high expression is correlated with worse prognosis [43]. Mitochondrial fission disrupts cristae architecture, and tumor cells with excessive fission activity depend on OPA1 to preserve electron transport chain function [44]. Importantly, loss of OPA1 induces the imbalance of mitochondrial dynamics, and thereby impairing respiratory function and increasing tumor epithelial sensitivity to CD8 + T cells in NSCLC [45], emphasizing the role of OPA1 in the mitochondria respiratory function and immune response to CD8 + T cells of LUAD. In this study, the expression of OPA1 was upregulated whereas silencing ALDH18A1 decreased the expression of OPA1. The findings suggest that silencing ALDH18A1 could maintain OPA1-mediated mitochondrial homeostasis through the PPAR signaling pathway, in addition to cellular senescence, cell stemness, and immune response. In addition, a clinical study examining mitochondrial dynamics across disease stages shows that fission protein expression is increased only in the early stages of NSCLC. In locally advanced and metastatic stages, fusion protein expression increases, suggesting that the balance between mitochondrial fission and fusion is dynamic and changes with disease progression [46]. Collectively, these discrepancies likely reflect the influence of tumor histological subtype, stage, experimental model system, and the metabolic state of the tumor microenvironment. These observations underscore the complexity of mitochondrial dynamics in lung cancer and emphasize that the functional significance of individual regulators cannot be generalized without careful consideration of biological context. In the present study, ALDH18A1 knockdown increased the expression of DRP1 and MFN2 but decreased the expression of OPA1. This pattern may reflect a mitochondrial stress response, in which excessive fission and impaired inner membrane fusion disrupt cristae integrity, reduce mitochondrial membrane potential, and promote apoptosis. The increase in MFN2 may represent a compensatory response to mitochondrial damage or a shift toward mitochondrial quality control. Therefore, ALDH18A1 may support malignant growth by maintaining mitochondrial dynamics and bioenergetic stability in LUAD.

Enrichment analysis showed that amino acid metabolism, such as arginine and proline metabolism, pyruvate metabolism, carbon metabolism, the TCA cycle, cofactor biosynthesis, and PPAR signaling pathway may participate in ALDH18A1-associated metabolic regulation. Studies have shown that ALDH18A1 variants can affect amino acid metabolism and antioxidant metabolism [47]. ALDH18A1/P5CS acts as a rate-limiting enzyme in proline biosynthesis, and its inhibition unexpectedly enables cancer cells to maintain proliferation under glutamine-restricted conditions by reducing glutamine consumption through proline synthesis [23]. PPARs are nuclear receptor transcription factors that play important roles in lipid metabolism, energy balance, and inflammatory responses [48, 49]. Previous studies have shown that PPARs are involved in the regulation of mitochondrial oxidative metabolism [50], and PPARγ appears to be closely associated with tumorigenesis [51]. Hua et al. reported that inhibition of the oncogenic driver Src activates PPARγ-dependent lipolysis and induces FABP4 expression, which subsequently increases endogenous ROS production and suppresses tumor growth [52]. In addition, high FABP4 expression is associated with favorable prognosis in lung cancer [52]. These findings suggest that activation of the PPARγ/FABP4 axis may represent a potential therapeutic strategy for lung cancer. Recent evidence also showed that FABP4 expression is markedly downregulated in LUAD tumor tissues, and reduced FABP4 expression may impair the cytotoxic function of natural killer cells through the glycerophospholipid metabolism pathway [53].Therefore, decreased FABP4 expression may impair the immune function in LUAD. This study demonstrated decreased PPARγ and FABP4 expression in both LUAD cells and tumor tissues, while silencing ALDH18A1 induced higher expression levels of PPARγ and FABP4. Meanwhile, silencing ALDH18A1 promoted the production of ROS. The findings indicated that silencing ALDH18A1 exerted inhibitory effects on LUAD through activating PPARγ/FABP4-mediated mitochondrial oxidative metabolism. Whether silencing ALDH18A1 affects immune function in lung cancer by activating PPARγ/FABP4 requires further investigation. Furthermore, GW9662 partially attenuated the effects of ALDH18A1 silencing on PPARγ/FABP4 expression and mitochondrial dysfunction-related changes, suggesting that the PPARγ/FABP4 axis may be involved in ALDH18A1 knockdown-associated mitochondrial dysfunction and tumor suppression. Notably, GW9662 partially increased ALDH18A1 expression in ALDH18A1-knockdown tumors, suggesting that GW9662-related effects may involve not only PPARγ inhibition but also ALDH18A1 expression modulation. Therefore, the rescue effects of GW9662 should be interpreted as supportive rather than definitive evidence for a linear ALDH18A1-PPARγ/FABP4 pathway.

Despite these findings, several limitations should be acknowledged. First, LASSO regression and SVM-RFE analyses were performed after PPI-based prioritization rather than using all 44 mitoDEGs as direct input. Although this two-step approach reduced feature redundancy and focused on topologically important genes, it may have overlooked biologically relevant mitoDEGs with lower network connectivity. Future studies should apply machine-learning algorithms directly to the full mitoDEG set and validate additional candidate genes in independent datasets and experimental models. In addition, the ALDH18A1-associated changes in the PPARγ/FABP4 axis and mitochondrial dynamics were primarily validated in LUAD cell lines and xenograft tumor models. Although public datasets consistently demonstrated ALDH18A1 upregulation in LUAD, direct validation of ALDH18A1, PPARγ/FABP4, and mitochondrial dynamics-related proteins in clinical LUAD tissues remains limited. Future studies using paired tumor and adjacent non-tumor tissues are needed to determine whether ALDH18A1 expression is associated with PPAR signaling activity and mitochondrial dysfunction in patients. Moreover, the present study focused mainly on PPARγ, whereas the potential roles of other PPAR family members, including PPARα and PPARδ, as well as additional PPAR ligands and downstream effectors, were not investigated. Further studies are needed to clarify the precise contribution of PPAR signaling to ALDH18A1-associated mitochondrial dysfunction and LUAD progression.

Conclusion

This study identifies eight hub mitoDEGs with potential diagnostic performance in LUAD, including SHMT2, ALDH18A1, PC, ALDH2, IDH2, ACACB, PAICS, and ACADL. ALDH18A1 silencing inhibits proliferation, invasion, and migration in LUAD cells and reduces tumor growth in tumor-bearing mice. These effects may be partly associated with PPAR signaling pathway activation and altered mitochondrial function. This study provides ALDH18A1 as a potential biomarker and the PPAR signaling pathway as a regulatory target for LUAD.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Taixu Sun: Conceptualization; Formal analysis; Methodology; Writing - original draft; Validation; Resources; Jianming Deng and Menglin Yang: Formal analysis; Methodology; Validation; Editing; Wenjie Wang and Shubin Guan: Data curation; Investigation; Software; Review & editing; All authors have read and approved the manuscript.

Funding

Internal hospital fund of the affiliated Guangdong Second Provincial General Hospital of Jinan University (No. YN2024-005).

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

The experiments conformed to the Guide for the Care and Use of Laboratory Animals. Animal study has been approved by the Animal Ethics Committee of the Affiliated Guangdong Second Provincial General Hospital of Jinan University.(No. 2025-DW-KZ-142-01). All methods are reported in accordance with ARRIVE guidelines.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.