The FGD5-AS1/miR-142-5p/CDK5 axis promotes ESCC progression by regulating mitochondrial fission and mitophagy: insights from integrative bioinformatics and experimental analyses

Abstract

Esophageal squamous cell carcinoma (ESCC) continues to pose a significant health challenge worldwide because of its poor prognosis and limited therapeutic options. To identify key regulators, we performed integrative bioinformatic approaches, such as differential expression analysis, weighted gene co-expression network analysis (WGCNA), LASSO regression, support vector machine (SVM) modeling, and ceRNA network construction, followed by experimental validation. These analyses across three GEO cohorts consistently highlighted FGD5-AS1 and CDK5 as upregulated, while miR-142-5p was downregulated and emerged as a central regulator within the ceRNA network. Clinical samples and functional assays confirmed these findings, showing that FGD5-AS1 overexpression enhanced proliferation, invasion, and migration while suppressing apoptosis in ESCC cells. Moreover, FGD5-AS1 promoted mitochondrial fission and mitophagy through miR-142-5p sequestration, leading to increased CDK5 levels, thereby driving tumor progression. Animal experiments further demonstrated that modulation of FGD5-AS1 expression significantly altered tumor growth. Collectively, these integrative results reveal that the FGD5-AS1/miR-142-5p/CDK5 axis promotes ESCC malignancy by regulating mitochondrial dynamics, suggesting that the FGD5-AS1/miR-142-5p/CDK5 axis may represent a candidate molecular target and warrants further investigation for its potential clinical relevance in ESCC.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12957-026-04338-y.

Introduction

Although it exhibits significant geographical variation, esophageal cancer, including esophageal squamous cell carcinoma (ESCC) and esophageal adenocarcinoma (EAC) has become the seventh most common cancer and the sixth leading cause of cancer-related death worldwide [1]. Approximately 600,000 new cases are diagnosed annually, with deaths exceeding 500,000. The high mortality rate is largely attributed to the absence of obvious early symptoms, leading to delayed diagnosis and reduced treatment effectiveness [2]. In addition, ESCC is highly prone to metastasis, frequently spreading to lymph nodes, liver, lungs, and bones [3]. Given these characteristics, the limited effectiveness of current therapies is mainly associated with difficulties in early detection, high recurrence rates, and extensive metastasis [3]. Therefore, a deeper understanding of the molecular mechanisms underlying ESCC progression is essential for the development of more effective therapeutic strategies.

In recent years, studies have revealed that long non-coding RNAs (lncRNAs) are key regulators in ESCC development [4]. These non-coding RNAs exceeding 200 nt govern gene expression and cellular functions through various mechanisms despite not encoding proteins [5, 6]. LncRNAs can interact with DNA, RNA, or proteins to influence gene transcription, post-transcriptional modifications, and translation processes [7]. Through these regulatory activities, lncRNAs significantly affect tumor cell proliferation, apoptosis, invasion, and migration in ESCC. For example, lncRNA H19 has been reported to promote ESCC progression by modulating angiogenesis-related genes [8]. Notably, FGD5-AS1 has been identified as closely associated with cancer development [9]. Increased expression of FGD5-AS1 can accelerate cancer progression. In breast cancer, high expression of FGD5-AS1 correlates with enhanced cell proliferation [10], while in non-small cell lung cancer it activates FGFRL1 to promote tumor cell proliferation and survival [11]. These studies suggest that FGD5-AS1 may act as an oncogene in various cancers. However, its role and mechanisms in ESCC remain to be elucidated.

MicroRNAs (miRNAs), as one of the primary targets of lncRNAs, also serve as key regulators in ESCC development [12]. miR-34a has been found to induce apoptosis by targeting and inhibiting anti-apoptotic genes such as BCL-2 and SIRT1. Its reduced expression is often associated with the anti-apoptotic characteristics of ESCC [13]. These findings highlight the important involvement of miRNAs in ESCC progression. Among them, miR-142-5p has attracted considerable attention for its role in cancer biology [14]. In several cancer types, miR-142-5p functions as a tumor suppressor that inhibits tumor cell proliferation and invasion. For example, in pancreatic cancer it suppresses cell proliferation, migration, and invasion by targeting PIK3CA [15]. However, miR-142-5p in cervical cancer can exhibit oncogenic properties, promoting cancer progression by regulating the expression of key transcription factors [16]. Interestingly, in ESCC, miR-142-5p serves a tumor-suppressive function, supported by data showing significantly reduced expression levels in tumor tissues of ESCC patients [17]. This underscores the necessity to investigate the detailed functions of miR-142-5p in ESCC.

In addition to transcriptional regulation mediated by non-coding RNAs, mitochondrial dynamics have emerged as important regulators of tumor progression [18]. Mitochondrial fission and mitophagy are essential processes that maintain mitochondrial quality control and cellular metabolic homeostasis [19]. Increasing evidence indicates that dysregulated mitochondrial fission and mitophagy promote tumor cell proliferation, survival, and metabolic adaptation in multiple cancers [18, 20]. Moreover, recent studies suggest that non-coding RNAs can modulate mitochondrial dynamics by regulating key signaling pathways and mitochondrial regulatory proteins [21]. However, whether mitochondrial fission and mitophagy participate in the regulatory network of FGD5-AS1 in ESCC remains largely unclear, and the molecular mechanisms linking this lncRNA to mitochondrial regulation have not yet been fully elucidated. Notably, CDK5 has been implicated in the regulation of mitochondrial function and stress responses, suggesting that it may represent a potential molecular link between non-coding RNA regulatory networks and mitochondrial dynamics in tumor cells.

Therefore, this research sought to elucidate the role of the FGD5-AS1/miR-142-5p/CDK5 regulatory axis in ESCC progression by means of integrative bioinformatics analyses and experimental validation. Based on integrative bioinformatics analyses of multiple GEO datasets combined with experimental validation, we sought to identify key oncogenic drivers, uncover their functional impact on mitochondrial dynamics and mitophagy, and clarify the underlying molecular interactions. This work provides novel mechanistic insights into ESCC progression and identifies candidate molecular axes that warrant further investigation in independent cohorts from different geographic regions for their biological and potential translational relevance.

Materials and methods

Data acquisition and preprocessing

Publicly available transcriptomic datasets of ESCC were obtained from the Gene Expression Omnibus (GEO) database, including GSE130078 (23 tumor vs. 23 normal samples; South Korea), GSE213565 (10 vs. 10; China), and GSE111011 (7 vs. 7; United States). Raw data were downloaded using the R package GEOquery. Probe IDs were processed using quantile normalization to ensure comparability across samples.

Identification of differentially expressed genes (DEGs)

For each dataset, Tumor–normal differential expression was performed with the limma R package. Genes exhibiting |log₂ fold change| > 1 with adjusted P < 0.05 were identified as significant DEGs. To improve robustness, the overlapping DEGs among the three independent datasets were extracted and included in subsequent analyses.

Weighted gene co-expression network analysis (WGCNA)

Gene co-expression was analyzed through WGCNA based on the dataset with the largest sample size (GSE130078). Outlier samples were removed after hierarchical clustering. A soft-thresholding power of 15 was selected to generate a scale-free topology. Genes were then clustered into separate modules using dynamic tree cutting, and module–trait relationships were calculated. Modules showing notable tumor correlations (|r| > 0.6, P < 0.05) were regarded as tumor-associated modules and selected for further analysis.

Identification of mitochondria-related genes

Mitochondria-associated genes were obtained via the Gene Ontology (GO) database using the term “mitochondrion localization” (GO:0051646). The overlap between these genes and the tumor-associated WGCNA modules was examined. Modules containing enriched mitochondrial genes with significant correlations to ESCC were prioritized for further functional analyses.

Machine learning-based gene selection

Machine learning approaches were employed as complementary feature selection tools. LASSO regression was performed using ten-fold cross-validation to reduce dimensionality and identify robust features with non-zero coefficients. In parallel, SVM-based feature ranking was applied to prioritize genes based on relative importance. To minimize the risk of overfitting associated with small sample sizes, only genes consistently identified across LASSO, SVM, WGCNA modules, and multiple independent GEO datasets were retained for downstream analyses.

Functional enrichment analysis

Functional annotation of the selected genes was executed with the R package clusterProfiler. GO enrichment for biological processes (BP), cellular components (CC), and molecular functions (MF), alongside Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses, were conducted. Enriched terms with P < 0.05 were defined as significant. The top-ranked GO terms and KEGG pathways were visualized to highlight the underlying biological roles of candidate genes.

Protein-protein interaction (PPI) network construction

To examine possible interactions among the candidate genes, a PPI network was constructed with STRING, using a minimum interaction score of 0.4. Network visualization was performed in Cytoscape, and hub genes were screened by the CytoHubba plug-in according to the maximal clique centrality (MCC) algorithm. Genes ranking highest in MCC were prioritized as central regulators in the ESCC network.

miRNA prediction and ceRNA network construction

Upstream regulatory microRNAs targeting the hub genes were predicted using TargetScan and miRDB. Only miRNAs consistently identified by both databases and showing differential expression in ESCC were considered. A lncRNA–miRNA–mRNA competing endogenous RNA (ceRNA) network was subsequently established in Cytoscape to depict the regulatory interactions.

Collection of clinical samples

Tumor tissues (ESCC group) and adjacent healthy tissues were collected from 10 patients with ESCC who underwent surgical treatment from January and April 2024 from the Affiliated Hospital of North Sichuan Medical College. The basic information of the patients is shown in Table 1. This study was approved by the Ethics Committee of Affiliated Hospital of North Sichuan Medical College (Approval number: 2024ER466-1), with informed consent obtained from all participants. During surgery, the surgeon excised tumor and adjacent tissues from the esophagus. These samples were segmented into appropriate sizes and stored at -80 °C. Inclusion criteria for ESCC patients: (1) Patients pathologically confirmed to have ESCC (2). Patients aged 18 and above with full legal capacity (3). Patients who signed written informed consent for study participation (4). Patients scheduled to undergo surgical tumor resection (5). Tumor tissue samples that could be obtained and processed immediately after surgery. Exclusion criteria: (1) Patients with a history of radiotherapy or chemotherapy (2). Patients with other malignancies or severe comorbidities (3). Patients with severe coagulation disorders or other surgical contraindications (4). Patients unable to provide sufficient tumor tissue samples for research purposes (5). Patients unwilling to participate or who failed to provide informed consent.

Table 1.

Clinic-pathological characteristic of all the studied patients

Characteristics	Total (n = 10)
Age (mean ± SD) (years)	60.5 ± 8.1
Gender
Male	7
Female	3
Tumor Stage
Stage I	2
Stage II	4
Stage III	3
Stage IV	1
Tumor Size (mean ± SD)(cm)	4.3 ± 1.2
Histological Type
Squamous Cell Carcinoma	9
Adenocarcinoma	1
Lymph Node Involvement
Positive	6
Negative	4
Distant Metastasis
Yes	3
No	7

Cell culture and treatment

Human esophageal epithelial cells Het-1 A (Catalog #CRL-2692, Supplier: ATCC) and esophageal squamous cell carcinoma cell lines TE-8 (Catalog #KYSE-1500, Supplier: JCRB), TE-1 (Catalog #CRL-2791, Supplier: ATCC), and KYSE150 (Catalog #ACC-375, Supplier: DSMZ) were obtained from our laboratory’s cell stock. After thawing, the cells were cultured to the third passage before experimentation. The culture medium for Het-1 A, KYSE150 cells was RPMI-1640, while TE-8 and TE-1 cells were grown in DMEM, with media containing 10% FBS and 1% penicillin-streptomycin. Cells were cultured under standard conditions (37 °C, 5% CO₂). Transfections were performed using Lipofectamine 2000. Cells were transfected with pcDNA3.1-CDK5 plasmid (OE-CDK5), the empty vector pcDNA3.1 (OE-NC), negative shRNA (shNC), FGD5-AS1 shRNA (sh-FGD5-AS1), negative mimic, miR-142-5p mimic, negative inhibitor, miR-142-5p inhibitor, and co-transfected with FGD5-AS1 shRNA and pcDNA3.1-CDK5, as well as FGD5-AS1 shRNA and miR-142-5p inhibitor.

The sense and antisense sequences used for shRNAs and miRNAs in this study are as follows: FGD5-AS1 shRNA: Sense: 5’-AGAACCGAGUCAUGUGGAA-3’, Antisense: 5’-UUCCACAUGACUCGGUUCU-3’. CDK5 shRNA: Sense: 5’-CCGGAAGGCAAAUCUCUCA-3’, Antisense: 5’-UGAGAGAUUUGCCUUCCGG-3’. shNC: Sense: 5’-UUCUCCGAACGUGUCACGU-3’, Antisense: 5’-ACGUGACACGUUCGGAGAA-3’. miR-142-5p mimics: Sense: 5’-CAUAAAGUAGAAAGCACUACU-3’, Antisense: 3’-UAGUGCUUUCUACUUUAUGUU-5’. Mimic-NC: Sense: 5’-UUCUCCGAACGUGUCACGUTT-3’, Antisense: 3’-ACGUGACACGUUCGGAGAATT-5’. miR-142-5p inhibitor: 5’-AGUAGUGCUUUCUACUUUAUG-3’. Inhibitor-NC: 5’-UUCUCCGAACGUGUCACGUUU-3’.

Cells were analyzed 48 h post-transfection.

Xenograft experiment in nude mice

Female BALB/c nude mice (4–6 weeks) were selected, ensuring they were maintained in Specific Pathogen Free (SPF) conditions and approved by the Animal Ethics Committee (Approval number: 2024ER466-2). Prior to the experiment, mice were acclimatized for one week before tumor cell implantation. KYSE150 cells transfected with negative shRNA and FGD5-AS1 shRNA, as well as TE-8 cells transfected with negative pcDNA3.1 and pcDNA3.1-FGD5-AS1, were adjusted to 1 × 10⁷ cells/mL. A 200 µL aliquot was subcutaneously injected into the right axillary region of each mouse. Following injection, tumor volumes were measured weekly using calipers to assess the long diameter (L) and short diameter (W), and calculated using the formula V = (L × W²)/2. Tumor growth curves were recorded to monitor tumor progression. Mice were euthanized by CO₂ asphyxiation when tumor volumes reached approximately 1500 mm³ or if mice exhibited symptoms of distress (such as reduced activity or significant weight loss).

To ensure compliance with institutional and ethical guidelines (Directive 2010/63/EU), several measures were implemented to uphold animal welfare. The health and behavior of the mice were monitored daily for signs of distress, reduced activity, or significant weight loss. A gradual CO₂ fill rate of 20% of the chamber volume per minute was employed to ensure humane euthanasia, as recommended by AVMA guidelines. Death was confirmed by cessation of respiration and lack of a heartbeat. Additionally, the mice were kept in a controlled temperature and humidity environment to minimize stress throughout the experiment. No unexpected deaths were reported.

Tumor samples were collected, measured, and preserved at -80 °C for subsequent analyses.

qRT-PCR

TRIzol reagent was added to both tissues and cells for lysis. RNA was precipitated following chloroform extraction and isopropanol precipitation and subsequently dissolved in DEPC water, and its concentration and purity were quantified by a Nano-400 A ultra-micro nucleic acid analyzer (yph-bio China). cDNA was synthesized from 1 µg of total RNA using the Takara PrimeScript RT Reagent Kit, following the manufacturer’s protocol. The reaction system included gDNA Eraser treatment to remove genomic DNA contamination and then reverse transcription at 37 °C (15 min) and 85 °C (5 s). The qPCR reaction system was prepared using Takara TB Green Premix Ex Taq, cDNA template, specific primers (final concentration 0.2 µM), and SYBR Green mix. The qRT-PCR was performed on an ABI 7500 Fast Real-Time PCR platform. All samples were analyzed at least three times, normalized to the reference gene GAPDH, and relative expression was obtained via the 2^-ΔΔCt approach.

Primers used for qPCR are listed below: FGD5-AS1, F: 5’- CAGTGCCGACTTCAATGTGC − 3’, R: 5’- CCCAGTGCTTTGTGGTTGTG − 3’; miR-142-5p, F: 5’-CAATAAGTAGAAAGCACTACT-3’, R: 5’-GTGCAGGGTCCGAGGT-3’;CDK5, F: 5’-TGGGATTCTGTCATAGCCGC-3’, R: 5’-GAGTAACAGCGGACGGGAAT-3’; GAPDH, F: 5’- AGAAGGCTGGGGCTCATTTG − 3’, R: 5’- AGGGGCCATCCACAGTCTTC − 3’; U6, F: 5’- CTCGCTTCGGCAGCACATATACT − 3’, R: 5’- AACGCTTCACGAATTTGCGT − 3’.

MTT assay

Cells were seeded at a density of 5 × 10³ cells per well of a 96-well plate, with each well containing 100 µL of complete culture medium. The cells were incubated at 37 °C in a 5% CO₂ humidified incubator for 24 h to allow for attachment and entry into logarithmic growth phase. Subsequently, cells were transfected with shRNA or expression vectors, and further incubated for 48 h. Then, 10 µL of 5 mg/mL MTT solution (Sigma-Aldrich) was introduced into wells and incubated for 4 h. Supernatant was removed, followed by the addition of 150 µL DMSO per well. Optical density at 570 nm was assessed via a SpectraMax i3x plate reader (Molecular Devices, USA). Cell viability for each group was calculated as a percentage of the control.

TUNEL assay

KYSE150 and TE-8 cells were plated onto coverslips in 24-well plates (1 × 10⁵ cells/well) and maintained under standard culture conditions (37 °C, 5% CO₂) for 24 h. They were then transfected with shRNA or expression vectors and further incubated for 48 h. Afterward, cells were subjected to two PBS washes and fixation in 4% paraformaldehyde for 30 min, then rinsed again with PBS. Next, cells were permeabilized using 0.1% Triton X-100 and 0.1% sodium citrate for 2 min on ice. TUNEL assay was performed by incubating cells with 50 µL of reagent from the In Situ Cell Death Detection Kit (Roche) for 1 h. Cells were rinsed three times with PBS and stained with DAPI for 10 min to visualize nuclei. Cells were visualized and photographed with an Axio Observer 7 microscope (Zeiss, Germany) to record TUNEL-positive cells (apoptotic cells) and total cell counts stained with DAPI. The excitation/emission wavelengths used were 488 nm/520 nm for detecting TUNEL and 358 nm/461 nm for detecting DAPI.

Transwell assay

Pre-coated Transwell inserts (8 μm pore size) with diluted Matrigel matrix (1:8) were prepared by applying 50 µL per well and allowing them to incubate for 30 min at 37 °C. KYSE150 and TE-8 cells (1 × 10⁵ cells/mL, logarithmic phase) were loaded into the upper chambers of the Transwell inserts, transfected with shRNA or expression vectors, and further incubated for 48 h. Simultaneously, 600 µL of complete medium containing 10% FBS was added to the lower chambers. The Transwell inserts were then incubated at 37 °C in a 5% CO₂ humidified incubator for an additional 48 h. After incubation, the inserts were removed, and cells on the lower membrane were fixed with 4% paraformaldehyde at room temperature for 30 min. Post-fixation, cells underwent two PBS washes, crystal violet staining (0.1%, 20 min, room temperature), and microscopic observation. Stained cell numbers were quantified in five randomly chosen fields to assess invasion capability.

Scratch assay

KYSE150 and TE-8 cells were grown in 6-well plates (5 × 10⁵ cells/well) and kept for 24 h at 37 °C, 5% CO₂. Using a 200 µL sterile pipette tip, a straight scratch was produced in the center of the monolayer. Cells were then transfected with shRNA or expression vectors and further incubated for an additional 48 h. Scratch images were taken at 0 and 24 h via an inverted microscope to monitor cell migration. The scratch width was measured using ImageJ, and cell migration distance was calculated to assess migration capability.

Mitochondrial membrane potential (JC-1) Assay

JC-1 assay kit (C2006, Beyotime, China) was used to determine mitochondrial membrane potential. KYSE150 and TE-8 cells were plated in 6-well plates (5 × 10⁵ cells/well), followed by transfection with shRNA or expression vectors. Following a 48 h incubation, JC-1 staining solution (1 mL) was introduced, and cells were incubated for 20 min at 37 °C, 5% CO₂. After two PBS washes, JC-1 fluorescence was measured using a BD FACSCanto II cytometer (BD Biosciences, USA). For JC-1 polymers, excitation and emission occurred at 525 and 590 nm, respectively, whereas JC-1 monomers were measured at 488 nm (excitation) and 530 nm (emission).

ATP content assay

Cellular ATP content was determined using an ATP assay kit (S0026, Beyotime, China). KYSE150 and TE-8 cells were plated in 6-well plates (5 × 10⁵ cells/well) and transfected with shRNA or expression vectors. After 48 h of incubation, an equal volume of CellTiter-Glo^® reagent was added to each well. The 96-well plate was gently shaken for 2 min and then incubated in the dark at room temperature for 10 min. The fluorescence intensity of each well was measured using a microplate reader (SpectraMax i3x, Molecular Devices, USA).

Western blot

After two PBS washes on ice, cells were lysed using ice-cold RIPA buffer at the appropriate volume. The lysates were placed in 1.5 mL microcentrifuge tubes and kept on ice for 30 min, followed by centrifugation (12,000 rpm, 10 min, 4 °C). Total protein was obtained from the supernatants, and protein levels were measured with a BCA protein assay kit (P0009, Beyotime, China). Protein samples (30 µg each) were diluted in 5 × SDS buffer and boiled (95 °C, 5 min). A 10% SDS-PAGE gel was used to separate the proteins, which were then electrotransferred to PVDF membranes under constant current (300 mA, 1.5 h). After transfer, 5% non-fat milk was used to block the membranes at room temperature for 1 h. Overnight incubation at 4 °C was performed with primary antibodies against MFN1 (A21293), MFN2 (A19678), Drp1 (A2705), BECN1 (A21191), P62 (A19700), PINK1 (A25301), Parkin (A0968), CDK5 (A2689) (all 1:1000; ABclonal, China), p-Drp1 (3455), LC3-II (3868), LC3-I (4599) (all 1:1000; CST, USA), and β-actin (1:50000, AC026, ABclonal, China). Following primary antibody incubation, membranes were treated with HRP-labeled secondary antibodies — Goat anti-Rabbit IgG (1:5000, AS014, ABclonal, China) and Goat anti-Mouse IgG (1:5000, AS003, ABclonal, China) — for 1 h at room temperature. The membrane was then treated with ECL substrate (Clarity™ Western ECL Substrate, Bio-Rad, USA) for 2 min. Signals were captured using the ChemiDoc™ MP system (Bio-Rad, USA). Band intensities were quantified with ImageJ, and the target protein levels were normalized to the β-actin internal control.

Dual-luciferase reporter assay

The dual-luciferase reporter assay was employed to investigate the targeting relationship between FGD5-AS1 and miR-142-5p, as well as between miR-142-5p and CDK5. Synthetic wild-type sequences containing the potential miR-142-5p predicted miR-142-5p target sites in FGD5-AS1 (WT-FGD5-AS1) and CDK5 (WT-CDK5), along with their corresponding mutant sequences (MUT-FGD5-AS1 and MUT-CDK5), were inserted downstream of the luciferase gene in the pGL3-Promoter vector. This resulted in the construction of the pGL3-WT-FGD5-AS1, pGL3-MUT-FGD5-AS1, pGL3-WT-CDK5, and pGL3-MUT-CDK5 reporter vectors. Exponentially growing KYSE150 and TE-8 cells were seeded at appropriate densities in 24-well plates and grown until reaching 70–80% confluence. Using Lipofectamine 2000, the constructed reporter vectors were co-transfected with miR-142-5p mimics or negative controls (NC) into KYSE150 and TE-8 cells. Following a 48 h incubation at 37 °C with 5% CO₂, luciferase activity was analyzed using the Dual-Luciferase^® Reporter Assay System (E1330, Promega, USA). Firefly to Renilla luciferase ratios were calculated for each sample to standardize luciferase activity.

Data analysis

All results are reported as mean ± standard deviation (SD). Data analysis was carried out with GraphPad Prism 8.0 software. Two-group comparisons employed Student’s t-test, whereas one-way analysis of variance (ANOVA) with Tukey’s post hoc test analyzed differences among multiple groups. P < 0.05 was deemed significant.

Result

Identification of DEGs in ESCC

Differential expression analysis was conducted using three independent GEO datasets to identify dysregulated genes in ESCC. In GSE130078, 5,471 DEGs were detected, with 3,124 showing upregulation and 2,347 downregulation (Fig. 1A). Similarly, analysis of GSE213565 revealed 7,419 DEGs, of which 4,765 were upregulated and 2,654 were downregulated (Fig. 1B). For GSE111011, 6,913 DEGs were obtained, comprising 3,116 upregulated and 3,797 downregulated genes (Fig. 1C). Integration of the three datasets identified 1,795 overlapping DEGs that were consistently dysregulated across cohorts (Fig. 1D). These robust DEGs were selected for subsequent network and functional analyses, as their consistent dysregulation across independent GEO cohorts from different geographic regions supports their robustness.

Fig. 1.

Identification of differentially expressed genes in ESCC. A Volcano plot of DEGs in GSE130078. B Volcano plot of DEGs in GSE213565. C Volcano plot of DEGs in GSE111011.Red dots represent significantly upregulated genes, green dots represent significantly downregulated genes, and black dots indicate non-significant genes. D Venn diagram showing the overlap of DEGs among the three datasets, with 1,795 common DEGs identified

WGCNA identifies tumor-associated mitochondria-related modules in ESCC

To identify gene co-expression patterns associated with ESCC, WGCNA was performed on the GSE130078 dataset. After sample clustering, several outliers were removed to ensure data quality (Fig. 2A). The soft-thresholding power was set to 15 as it satisfied the scale-free topology criterion with high scale independence and acceptable mean connectivity (Fig. 2B). Genes were then clustered into multiple co-expression modules through hierarchical clustering and dynamic tree cutting (Fig. 2C). Module-trait correlation analysis revealed that several modules were significantly associated with the tumor phenotype, among which the blue and midnightblue modules exhibited the strongest positive correlations (r = 0.64, P < 0.01; r = 0.62, P < 0.01, respectively) (Fig. 2D). Further analysis confirmed that gene significance for tumor was highly correlated with module membership in the blue module (cor = 0.68, P < 0.001) (Fig. 2E) and the midnightblue module (cor = 0.69, P < 0.001) (Fig. 2F).

Fig. 2.

WGCNA identifies tumor-associated mitochondria-related modules in ESCC (GSE130078). A Sample clustering dendrogram for quality control and detection of outliers. B Analysis of scale independence and mean connectivity for soft-threshold power selection. C Gene clustering dendrogram with dynamic tree cutting to define co-expression modules. D Heatmap of module–trait correlations, showing that blue and midnightblue modules are significantly associated with ESCC. E Scatterplot of module membership versus gene significance for tumor in the blue module. F Scatterplot of module membership versus gene significance for tumor in the midnightblue module

Machine learning-based screening of candidate genes and functional enrichment analysis

To further refine hub genes associated with ESCC, machine learning approaches were employed. LASSO regression analysis demonstrated that the regression coefficients of candidate genes gradually shrank to zero with increasing penalty parameter λ, and 33 genes remained with non-zero coefficients (Fig. 3A). Cross-validation error plots identified the optimal λ value, yielding the most parsimonious model with minimal mean squared error (Fig. 3B). In parallel, SVM analysis ranked genes according to feature importance, and several key genes including FGD5-AS1 and CDK5 were among the top candidates (Fig. 3C). By intersecting the results of LASSO regression and SVM, 27 overlapping genes were identified as robust candidates (Fig. 3D). Further integration with DEGs revealed 14 overlapping genes that were both machine learning–selected and significantly dysregulated in ESCC (Fig. 3E). Functional enrichment analysis of these candidate genes showed significant enrichment in biological processes linked to cell cycle regulation, mitochondrial localization, and DNA repair (Fig. 3F).

Fig. 3.

Machine learning-based screening of candidate genes and enrichment analysis. A Coefficient profile plot of LASSO regression, showing the trajectory of gene coefficients with increasing penalty parameter λ. B Cross-validation error curve of LASSO regression, identifying the optimal λ value. C.SVM feature selection ranking, highlighting top candidate genes. D. Venn diagram showing the overlap between LASSO- and SVM-selected genes (27 common genes). E Venn diagram showing the overlap between machine learning–selected genes and DEGs, yielding 14 robust candidates. F. GO enrichment analysis of candidate genes, showing significant enrichment in mitochondrial localization, cell cycle regulation, and DNA repair pathways

Construction of the Candidate Gene–miRNA Regulatory Network

To explore upstream regulatory mechanisms of the identified candidate genes, miRNA–mRNA interactions were predicted using TargetScan and miRDB, and a ceRNA regulatory network was established. The resulting network incorporated lncRNA, mRNAs, and their potential regulatory miRNAs (Fig. 4). Among the network components, FGD5-AS1 was predicted to interact with multiple miRNAs, including miR-142-5p, which has been documented to exert tumor-suppressive functions in ESCC. In parallel, CDK5 was recognized as a direct target of miR-142-5p, suggesting that FGD5-AS1 may regulate CDK5 expression through a competitive binding mechanism. Other candidate genes, such as E2F1, HAP1, KAT2A, and INPP5E, were also integrated into the network, interacting with different miRNAs.

Fig. 4.

Construction of the candidate gene–miRNA regulatory network

FGD5-AS1 is Upregulated in ESCC Tissues and Cells

To investigate the function of FGD5-AS1 in ESCC, we initially examined its expression levels. The findings revealed that FGD5-AS1 expression was markedly higher in the EC tissues relative to the Adjacent tissue (Fig. 5). These findings indicate that FGD5-AS1 expression is elevated in ESCC tissues and cells.

Fig. 5.

Upregulation of FGD5-AS1 in ESCC tissues and cells. A qRT-PCR analysis of FGD5-AS1 expression levels in ESCC group tissues compared to Adjacent tissue group tissues. B qRT-PCR analysis of FGD5-AS1 expression levels in Het-1 A, TE-8, TE-1, and KYSE150 cell lines. ^*p < 0.05, ^**p < 0.01

FGD5-AS1 promotes cell viability, invasion, and migration, while inhibiting apoptosis in ESCC cells in vitro

Prior evidence indicates that FGD5-AS1 expression influences tumor biological behaviors [10]. To confirm this and examine its mechanisms, we knocked down FGD5-AS1 in KYSE150 cells (Fig. 6A) and overexpressed FGD5-AS1 in TE-8 cells (Fig. 6B). Subsequently, we assessed cell viability, apoptosis, invasion, and migration using MTT, TUNEL, Transwell, and scratch assays. The results demonstrated that knockdown of FGD5-AS1 in KYSE150 cells significantly reduced cell viability, increased apoptosis, and decreased invasion and migration abilities. Conversely, overexpression of FGD5-AS1 in TE-8 cells significantly enhanced cell viability, decreased apoptosis, and increased invasion and migration abilities (Fig. 6C-E). These findings suggest that FGD5-AS1 promotes tumorigenic behaviors in ESCC cells.

Fig. 6.

FGD5-AS1 promotes cell viability, invasion, and migration, while inhibiting apoptosis in ESCC cells in vitro. A qRT-PCR analysis of FGD5-AS1 expression levels in KYSE150 cells transfected with shNC or sh-FGD5-AS1. B qRT-PCR analysis of FGD5-AS1 expression levels in TE-8 cells transfected with OE-NC or OE-FGD5-AS1. C The apoptosis levels of KYSE150 cells transfected with shNC or sh-FGD5-AS1, and TE-8 cells transfected with OE-NC or OE-FGD5-AS1 was evaluated using TUNEL assay. D The invasion capability of KYSE150 cells transfected with shNC or sh-FGD5-AS1, and TE-8 cells transfected with OE-NC or OE-FGD5-AS1 was evaluated using a Transwell assay. E The migration ability of KYSE150 cells transfected with shNC or sh-FGD5-AS1, and TE-8 cells transfected with OE-NC or OE-FGD5-AS1 was evaluated using scratch assay measuring. ^**p < 0.01

FGD5-AS1 modulates mitochondrial fission and mitophagy in ESCC cells in vitro

In order to comprehensively assess the effects of FGD5-AS1 on ESCC, we investigated its influence on cellular energy metabolism in cells expressing varying levels of FGD5-AS1. Results demonstrated that FGD5-AS1 knockdown in KYSE150 cells significantly enhanced mitochondrial membrane potential and ATP synthesis. Conversely, overexpression of FGD5-AS1 in TE-8 cells significantly decreased mitochondrial membrane potential while increasing ATP synthesis (Fig. 7A-B). Next, we assessed changes in the expression of mitochondrial dynamics-related proteins via western blotting. Knockdown of FGD5-AS1 led to pronounced MFN1 and MFN2 expression, accompanied by suppressed p-Drp1 levels. In contrast, overexpression of FGD5-AS1 led to reduced MFN1 and MFN2 and elevated p-Drp1 (Fig. 7C). In addition, we examined changes in the expression of mitophagy-linked proteins. FGD5-AS1 knockdown markedly reduced LC3-II, BECN1, PINK1, and Parkin, whereas P62 levels were elevated and the LC3-II/LC3-I ratio declined. Conversely, overexpression produced the opposite effects (Fig. 7D). These results demonstrate that enhanced FGD5-AS1 expression facilitates mitochondrial fission alongside mitophagy.

Fig. 7.

FGD5-AS modulates mitochondrial fission and mitophagy in ESCC cells in vitro. A Measurement of mitochondrial membrane potential (JC-1) in KYSE150 cells (shNC and sh-FGD5-AS1 groups) and TE-8 cells (OE-NC and OE-FGD5-AS1 groups) using a commercial assay kit. B Measurement of ATP levels in KYSE150 cells (shNC and sh-FGD5-AS1 groups) and TE-8 cells (OE-NC and OE-FGD5-AS1 groups) using a commercial assay kit. C Western blot analysis of MFN1, MFN2, Drp1, and p-Drp1 protein levels in KYSE150 cells (shNC and sh-FGD5-AS1 groups) and TE-8 cells (OE-NC and OE-FGD5-AS1 groups). D Western blot analysis of LC3-II, LC3-I, BECN1, P62, PINK1, and Parkin protein expression levels in KYSE150 cells (shNC and sh-FGD5-AS1 groups) and TE-8 cells (OE-NC and OE-FGD5-AS1 groups). ^**p < 0.01

FGD5-AS1 knockdown inhibits tumor growth in vivo

Earlier research identified FGD5-AS1 expression as a factor that FGD5-AS1 facilitates mitochondrial autophagy and mitochondrial fission in ESCC cells, thereby promoting tumorigenic behaviors. We further investigated the impact of FGD5-AS1 knockdown or overexpression on tumor development in nude mice. The results indicated that knockdown of FGD5-AS1 in KYSE150 cells led to a significant reduction in tumor volume and weight in nude mice, whereas overexpression of FGD5-AS1 in TE-8 cells led to marked enlargement of tumor volume and weight (Fig. 8A-C). Therefore, knockdown of FGD5-AS1 inhibits tumor progression in vivo.

Fig. 8.

Knockdown of FGD5-AS1 inhibits tumor growth in vivo. A Representative images of tumors from mice in the shNC, sh-FGD5-AS1, OE-NC, and OE-FGD5-AS1 groups. B Tumor volume measurements of mice in the shNC, sh-FGD5-AS1, OE-NC, and OE-FGD5-AS1 groups, taken every four days. C Tumor weights from mice in the shNC, sh-FGD5-AS1, OE-NC, and OE-FGD5-AS1 groups, measured four weeks’ post-inoculation. ^**p < 0.01

FGD5-AS1 Acts as a miR-142-5p Sponge

Prior evidence has demonstrated that miR-142-5p has tumor-suppressive functions and is downregulated in ESCC tissues [16]. Interestingly, FGD5-AS1 was identified as a regulator of miR-142-5p [22]. However, the relationship between these two during ESCC progression is not fully understood. To explore this relationship, we initially examined miR-142-5p expression in ESCC. The outcomes revealed that tumor tissues exhibited significantly lower miR-142-5p levels compared to adjacent tissues (Fig. 9A), and a clear negative association with FGD5-AS1 was observed (Fig. 9B). Notably, knockdown of FGD5-AS1 expression significantly increased miR-142-5p levels, whereas its overexpression inhibited miR-142-5p expression (Fig. 9C). Predicted targeting relationships suggest a direct interaction, which was validated through dual-luciferase reporter analysis showing that miR-142-5p mimics co-transfection significantly suppressed the activity of FGD5-AS1-WT luciferase reporters (Fig. 9E). These findings indicate that FGD5-AS1 serves as a miR-142-5p sponge, regulating its levels.

Fig. 9.

FGD5-AS1 Functions as a sponge for miR-142-5p. A qRT-PCR analysis of miR-142-5p expression levels in ESCC tissues and Adjacent tissue samples. B Pearson correlation analysis of the expression levels between FGD5-AS1 and miR-142-5p. C qRT-PCR analysis of miR-142-5p expression levels in KYSE150 cells (shNC and sh-FGD5-AS1 groups) and TE-8 cells (OE-NC and OE-FGD5-AS1 groups). D Predicted target sequences of FGD5-AS1 and miR-142-5p. E Dual-luciferase reporter assay to verify the interaction between FGD5-AS1 and miR-142-5p. ^**p < 0.01

CDK5 Identified as a miR-142-5p target and positively linked to FGD5-AS1 expression

Prior research has demonstrated that CDK5 is typically overexpressed in ESCC tissues, contributing to the progression and malignancy of the disease. Additionally, research indicates that miR-142-5p is capable of modulating CDK5 expression levels [23]. It is hypothesized that FGD5-AS1 drives ESCC development through miR-142-5p–mediated regulation of CDK5 expression. To investigate the relationships among these factors in ESCC, we first examined CDK5 expression levels. The findings demonstrated that CDK5 expression was significantly increased in EC tissues compared to Adjacent tissue samples (Fig. 10A). Moreover, there was a negative correlation between CDK5 and miR-142-5p expression levels (Fig. 10B). Overexpression of miR-142-5p led to a significant decrease in CDK5 mRNA and protein levels, while knockdown of miR-142-5p resulted in increased CDK5 mRNA and protein levels (Fig. 10C-D). The predicted miR-142-5p–CDK5 interaction was validated through a dual-luciferase reporter assay, demonstrating that miR-142-5p mimics significantly inhibited the luciferase activity of the CDK5-WT construct (Fig. 10E-F), indicating that miR-142-5p directly regulates CDK5 expression. Additionally, we found a positive linkage between FGD5-AS1 and CDK5 levels (Fig. 10G). Overexpression of FGD5-AS1 upregulated CDK5, while FGD5-AS1 knockdown reduced it (Fig. 10H-I). These results collectively demonstrate that CDK5 is a downstream miR-142-5p target positively associated with FGD5-AS1.

Fig. 10.

CDK5 as a target of miR-142-5p and its positive correlation with FGD5-AS1 expression. A qRT-PCR analysis of CDK5 expression levels in ESCC tissues and Adjacent tissue samples. ^**p < 0.01; B Pearson correlation analysis of CDK5 and miR-142-5p expression levels. C-D qRT-PCR and Western blot analysis of CDK5 mRNA and protein expression levels in cells overexpressing or knocking down miR-142-5p. ^**p < 0.01 vs. NC mimics; ^# p < 0.05, ^## p < 0.01; E Predicted target sequences of CDK5 and miR-142-5p. F Dual-luciferase reporter assay validating the interaction between CDK5 and miR-142-5p. ^**p < 0.01; G Pearson correlation analysis of FGD5-AS1 and CDK5 expression levels. H-I qRT-PCR and Western blot analysis of CDK5 expression levels in cells overexpressing or knocking down FGD5-AS1. ^*p < 0.05, ^**p < 0.01

The FGD5-AS1/miR-142-5p/CDK5 Axis Influences ESCC Progression

To gain deeper insight into the association between FGD5-AS1, miR-142-5p, and CDK5, we conducted experiments where FGD5-AS1 and miR-142-5p were simultaneously knocked down in KYSE150 cells, or CDK5 was overexpressed while FGD5-AS1 was knocked down. Compared to the sh-FGD5-AS1 group, CDK5 protein expression was markedly elevated in in both the sh-FGD5-AS1 + miR-142-5p inhibitor and sh-FGD5-AS1 + OE-CDK5 groups (Fig. 11A). MTT, TUNEL, Transwell, and wound healing assays demonstrated that, relative to the sh-FGD5-AS1 group, both the sh-FGD5-AS1 + miR-142-5p inhibitor and sh-FGD5-AS1 + OE-CDK5 groups exhibited significantly increased cell viability, decreased apoptosis levels, and enhanced invasion and migration capabilities (Fig. 11B-E). Using commercial assay kits, we found that, relative to the sh-FGD5-AS1 group, JC-1 and ATP levels in the sh-FGD5-AS1 + miR-142-5p inhibitor and sh-FGD5-AS1 + OE-CDK5 groups were significantly reduced (Fig. 11F-G). Western blotting revealed that, relative to the sh-FGD5-AS1 group, MFN1 and MFN2 protein expression was significantly decreased, and p-Drp1 was significantly increased in both the sh-FGD5-AS1 + miR-142-5p inhibitor and sh-FGD5-AS1 + OE-CDK5 groups (Fig. 11H). Moreover, compared to the sh-FGD5-AS1 group, the protein expression levels of LC3-II, BECN1, PINK1, and Parkin were significantly increased, the protein level of P62 was significantly decreased, and the LC3-II/LC3-I ratio was significantly elevated in the sh-FGD5-AS1 + miR-142-5p inhibitor group and the sh-FGD5-AS1 + OE-CDK5 group (Fig. 11I). These findings suggest that FGD5-AS1/miR-142-5p/CDK5 signaling axis influences ESCC progression.

Fig. 11.

The FGD5-AS1/miR-142-5p/CDK5 axis influences ESCC progression. A Western blot analysis of CDK5 protein levels in shNC, sh-FGD5-AS1, sh-FGD5-AS1 + miR-142-5p inhibitor, and sh-FGD5-AS1 + OE-CDK5 groups of KYSE150 cells. B MTT assay of cell viability in the four groups. C TUNEL assay of apoptosis levels in the four groups. D Transwell assay of invasion capabilities in the four groups. E Scratch assay of migration capabilities in the four groups. F JC-1 assay of mitochondrial membrane potential in the four groups. G ATP assay of ATP levels in the four groups. H Western blot analysis of MFN1, MFN2, Drp1, and p-Drp1 protein levels in the four groups. I Western blot analysis of LC3-II, LC3-I, BECN1, P62, PINK1, and Parkin protein levels in the four groups. ^**p < 0.01 vs. Control, ;^##p < 0.01 vs. sh-FGD5-AS1

Discussion

The threat posed by ESCC to human health is increasingly severe. Due to its subtle early symptoms, most patients present at an advanced stage, leading to poor overall survival. Thus, the dentification of molecular regulators involved in ESCC progression is particularly important, as it may provide a foundation for future diagnostic and therapeutic research. In this study, we first performed integrative bioinformatics analyses across multiple independent GEO cohorts derived from distinct institutions in South Korea, China, and the United States, including differential expression analysis, WGCNA, and machine learning, which consistently prioritized FGD5-AS1 and CDK5 as tumor-associated genes and identified miR-142-5p as their key regulatory miRNA. Based on these computational findings, we subsequently carried out experimental validation. We focused on FGD5-AS1, discovering its role and mechanism in ESCC, thereby offering new insights for the treatment of this disease.

In ESCC, lncRNAs regulate gene expression and cellular processes through various mechanisms, influencing tumor growth, apoptosis, migration, and invasiveness [24]. Consistent with our bioinformatics analyses, which identified FGD5-AS1 as significantly upregulated across multiple GEO datasets and enriched in mitochondria-related co-expression modules. We found a marked elevation of FGD5-AS1 in ESCC tissues. Moreover, overexpression of FGD5-AS1 in vitro promotes malignant characteristics of ESCC cells, such as viability, invasion, and migration. This indicates that FGD5-AS1 exerts oncogenic functions in ESCC, positively facilitating its progression. Notably, we provide the novel evidence that FGD5-AS1 is associated with enhanced mitochondrial fission and mitophagy in ESCC. This finding was also supported by enrichment analysis of machine learning–selected candidate genes, which revealed significant involvement in biological processes such as mitochondrion localization and negative regulation of the cell cycle.The expression levels of LC3-II, BECN1, PINK1, and Parkin were analyzed to evaluate their roles in mitochondrial fission and mitophagy in ESCC cells. These proteins are essential regulators of autophagy and mitophagy, processes that help maintain cellular homeostasis by eliminating damaged organelles, particularly mitochondria [25]. This analysis aimed to demonstrate that the FGD5-AS1/miR-142-5p/CDK5 axis promotes mitophagy, thereby enhancing the survival and proliferation of ESCC cells. Specifically, LC3-II and BECN1 are markers of autophagosome formation, while PINK1 and Parkin are involved in the clearance of dysfunctional mitochondria through mitophagy [26].

Previous research has demonstrated that lncRNAs can function as homologous or competitive binding sites for transcription factors, directly affecting gene transcription. For example, FGD5-AS1 has been reported to function as a competing endogenous RNA (ceRNA) in breast cancer by sponging miR-497-5p and regulating its downstream target gene MACC1, thereby promoting tumor progression [10]. Lin et al. reported that FGD5-AS1 inhibits miR-520a-3p function to promote KIAA1522 expression, accelerating pancreatic cancer cell proliferation and migration [27]. Additionally, FGD5-AS1 can activate oral cancer tumor growth by targeting miR-153-3p [28]. This suggests that FGD5-AS1 can regulate cancer progression through multiple miRNAs.In our study, ceRNA network construction further revealed that miR-142-5p serves as the central microRNA linking FGD5-AS1 and CDK5. This computational prediction not only complements previous experimental findings but also suggests that the FGD5-AS1/miR-142-5p/CDK5 axis represents one of the regulatory pathways through which FGD5-AS1 may contribute to ESCC progression. However, the miRNA network regulated by FGD5-AS1 in ESCC remains to be fully characterized, making it a highly promising area for further exploration. It should be noted that the ceRNA hypothesis primarily describes a functional mode of post-transcriptional regulation mediated by shared microRNA response elements, rather than obligatorily requiring the demonstration of stable physical RNA–RNA complexes. As discussed in previous critical evaluations of endogenous microRNA sponges, functional ceRNA interactions can be supported by sequence-specific dual-luciferase assays, reciprocal expression regulation, and phenotypic rescue experiments, even in the absence of detectable RNA immunoprecipitation or pull-down evidence [29]. Therefore, in the present study, we define the FGD5-AS1/miR-142-5p/CDK5 axis as a functional regulatory pathway.

It is noteworthy that in studies on ovarian cancer, FGD5-AS1 was found to promote cancer cell malignancy via modulation of miR-142-5p [30]. Current research suggests that miR-142-5p functions as a tumor suppressor gene in ESCC, though other reports indicate that it has complex functions. miR-142-5p can suppress cell proliferation by targeting proteins such as c-Myc and CDK6, thereby halting the cell cycle progression [31]. Additionally, miR-142-5p can suppress cancer cell migration and invasion by targeting epithelial-mesenchymal transition-related genes like ZEB1 and ZEB2 [32]. Interestingly, miR-142-5p can also target anti-apoptotic genes, increasing the apoptosis rate of cancer cells and thus limiting their survival [33]. Particularly, we observed a substantial reduction of miR-142-5p in ESCC tissues, negatively linked to FGD5-AS1 expression. The regulation of miR-142-5p by lncRNAs is well-documented. For instance, lncRNA TUG1 modulates ulcerative colitis via miR-142-5p [34], and lncRNA PTTG3P influences the progression of tongue cancer through miR-142-5p [35]. Notably, ceRNA network construction in our bioinformatics analysis identified miR-142-5p as the central microRNA linking FGD5-AS1 and CDK5, providing computational support for our experimental validation. This integrative evidence strengthens the notion that FGD5-AS1 promotes ESCC development through competitive binding miR-142-5p and thereby relieving its inhibitory effect on CDK5.Therefore, we hypothesized that miR-142-5p is an essential downstream effector of FGD5-AS1, and our findings confirmed this hypothesis. Moreover, research by Zhu et al. observed the involvement of the FGD5-AS1/miR-142-5p signaling axis in cisplatin resistance in lung adenocarcinoma cells [22], indicating the complexity of this regulatory network.

CDK5 (Cyclin-dependent kinase 5) is a protein kinase extensively studied in the nervous system, primarily involved in neuronal development and function [36]. However, recent years have revealed its significant role in various cancers, including esophageal carcinoma [37]. Studies indicate that CDK5 expression levels in ESCC tissues are markedly higher compared to normal esophageal tissues, and CDK5 upregulation is closely linked to cancer progression and unfavorable outcomes in esophageal carcinoma [37]. This suggests that increased CDK5 expression promotes the development of ESCC. Consistently, our bioinformatics analysis identified CDK5 as one of the significantly upregulated genes across multiple GEO cohorts, and it was further prioritized as a robust feature gene by both LASSO regression and SVM machine learning algorithms.The results from our experiments further validated the specificity of CDK5 overexpression in the OE-CDK5 group, ensuring the reliability of subsequent functional analyses. We observed a significant increase in CDK5 mRNA levels. This supports the hypothesis that CDK5 serves as a critical downstream effector of the FGD5-AS1/miR-142-5p axis, amplifying the oncogenic properties of FGD5-AS1.Together, these computational and experimental findings support the hypothesis that CDK5 serves as a critical downstream target in the FGD5-AS1/miR-142-5p signaling pathway, thereby amplifying the oncogenic properties of FGD5-AS1 in ESCC.

Moreover, research by Jia et al. demonstrated CDK5 is a direct miR-142-5p target [38], while we observed a significant correlation between CDK5 and FGD5-AS1 expression levels. Therefore, guided by the FGD5-AS1/miR-142-5p/CDK5 signaling axis, we investigated the interactions among these three factors. Results revealed that FGD5-AS1 promotes CDK5 expression by inhibiting miR-142-5p, thereby accelerating cancer progression. Co-knockdown of FGD5-AS1 and miR-142-5p markedly promoted the malignant characteristics of esophageal cancer cells. Importantly, CDK5 may also influence mitochondrial membrane potential and autophagy by acting as a phosphorylating enzyme for Drp1, Mfn1, Mfn2, and Parkin proteins [39, 40]. In summary, we have identified a novel regulatory network operating in ESCC that may contribute to disease progression.

However, this study has several limitations that should be acknowledged. First, the clinical sample size was relatively small and was primarily used for expression validation, and no follow-up information, such as survival outcomes or treatment responses, was available. Therefore, the prognostic and predictive value of the FGD5-AS1/miR-142-5p/CDK5 axis could not be fully evaluated. In addition, the lack of large multi-center cohorts with comprehensive clinical annotation, such as TCGA-ESCC, further limited the ability to perform robust prognostic analyses. From a mechanistic perspective, although our data support a regulatory role of CDK5 in mitochondrial dynamics, the precise molecular mechanisms underlying CDK5-mediated mitochondrial regulation require further investigation. Moreover, while LASSO and SVM were employed as feature selection approaches, their application to relatively small GEO cohorts may introduce a risk of overfitting. To mitigate this concern, these methods were integrated with WGCNA, multi-cohort overlap analysis, and extensive experimental validation. Nevertheless, future studies using larger and more diverse cohorts are warranted to further confirm the stability and generalizability of the identified features. Taken together, future research should focus on validating the diagnostic and prognostic utility of the FGD5-AS1/miR-142-5p/CDK5 pathway in larger, multi-center populations and further exploring its therapeutic potential in ESCC.

Conclusion

In summary, integrative bioinformatics and experimental analyses consistently identified the FGD5-AS1/miR-142-5p/CDK5 pathway as a critical driver of ESCC. This axis promotes tumor cell proliferation, migration, invasion, and mitochondrial fission while suppressing apoptosis, thereby facilitating ESCC progression. These results not only provide novel mechanistic insights into the mitochondrial regulation of ESCC but also highlight the FGD5-AS1/miR-142-5p/CDK5 axis as a potential biomarker and therapeutic target.

Authors’ contributions

Study concept and design: MH; Analysis and interpretation of data: SC, HDY, QMH; Drafting of the manuscript: MH; Critical revision of the manuscript for important intellectual content: MH; Statistical analysis: BLZ, HMZ, YG; Study supervision: all authors; all authors have read and approved the manuscript.

Funding

This study did not receive any funding in any form.

Data availability

The data used to support the findings of this study are available from the corresponding author upon request.

Declarations

Ethics approval and consent to participate

The animal experiments described in this study were authorized by the Committee of the Affiliated Hospital of North Sichuan Medical College and conducted in compliance with the institutional guidelines (Directive 2010/63/EU in Europe) for the care and use of animals (Approval number: 2024ER466-2).

This study was approved by the Medical Ethics Committee of the Affiliated Hospital of North Sichuan Medical College (Approval number: 2024ER466-1).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.