Characterizing prognostic and immunological traits of kidney renal clear cell carcinoma via mitochondria-associated membranes and identifying DNM1L as a potential therapeutic target using machine learning

Abstract

Background

Mitochondrial-associated membranes (MAMs) participate in cellular metabolism, calcium signaling, and cancer reprogramming, but their role in kidney renal clear cell carcinoma (KIRC) remains unclear.

Methods

Clinical/transcriptomic data of KIRC and previously reported MAMs-related genes were obtained from TCGA, E-MTAB-1980, and GSE29609. After analyzing MAMs gene expression in KIRC, 101 models were generated via 10 machine learning algorithms to select the optimal MAMs-based scoring system. Survival (Kaplan-Meier) and Cox regression analyses evaluated its prognostic value; associations with immune cell infiltration, checkpoints, and drug sensitivity were explored, and key genes were validated in vitro and in vivo.

Results

Forty-two MAMs-related genes were identified, most highly expressed in KIRC tissues. A 9-gene MAMs scoring system was built using the Stepwise Cox model. High-score patients had worse prognosis, and the system was an independent prognostic factor for KIRC. High scores correlated with advanced TNM stage/grade, elevated CTLA4/PD1 (better immunotherapy response in CTLA4+/PD1+ subgroups), and sensitivity to temsirolimus/sunitinib (low scores sensitive to sorafenib). DNM1L was a key gene; its knockdown inhibited KIRC cell proliferation, invasion, and migration in vitro. In vivo experiments further confirmed that DNM1L knockdown suppressed tumor growth in BALB/c nude mice bearing KIRC xenografts.

Conclusion

This study identified high MAMs-related gene expression in KIRC, developed a 9-gene MAMs scoring system, and validated DNM1L as a key gene, providing new insights into KIRC prognostic biomarkers and therapeutic targets.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12885-026-15603-4.

Introduction

Renal cell carcinoma (RCC) is a common malignant tumor of the urinary system, with its global incidence steadily rising at an annual rate of approximately 2% [1]. The predominant pathological subtype is kidney renal clear cell carcinoma (KIRC), which accounts for more than 70% of cases [2]. While localized RCC is often amenable to surgical resection with favorable outcomes, metastatic RCC typically demonstrates resistance to conventional radiotherapy and chemotherapy, posing significant therapeutic challenges [3]. Notably, a substantial proportion of patients, approximately 20%–30%, are diagnosed with distant metastases at initial presentation. Furthermore, even after surgical treatment, around 30% of individuals with localized RCC may experience disease recurrence and progression to metastatic stages [4]. Although recent advances in immunotherapy and targeted therapies have expanded the therapeutic landscape and shown improved outcomes compared to traditional modalities, their efficacy remains variable among patients, and achieving long-term remission continues to be a challenge [5]. Consequently, the identification of reliable prognostic biomarkers and novel therapeutic targets remains a critical area of research in the ongoing effort to improve patient outcomes. Mitochondrial-associated membranes (MAMs) are specialized structures that serve as contact sites between the endoplasmic reticulum (ER) and mitochondria. As critical hubs facilitating communication between these organelles, MAMs play essential roles in various cellular processes, including calcium signaling, lipid transfer, and cell death regulation [6–9]. Cancer cells require substantial energy to support rapid proliferation and acquire malignant phenotypes. To meet these demands, they adopt multiple metabolic adaptations, such as increased glucose uptake and glycolytic activity, enhanced lipid synthesis and lipolysis, as well as the regulation of Ca²⁺ levels. Given their role in coordinating these pathways, MAMs are closely linked to cancer cell function and metabolism. Furthermore, alterations in MAM-associated proteins can disrupt these mechanisms, potentially leading to apoptosis inhibition and increased resistance to anticancer therapies [10]. However, the role of MAMs in KIRC remains unexplored. In the artificial intelligence era, machine learning has become a research focus in oncology. It has played a crucial role in cancer diagnosis, treatment and prognosis over the past decade [11]. Mounting evidence confirms its great potential in clinical tumor prognosis prediction [12]. Given the variety of machine learning algorithms, prognostic models exhibit distinct performance. Recently, cross-validation of 10 machine learning algorithms (101 combinations) has been widely adopted to select optimal algorithms for developing more accurate oncology prognostic models [13, 14].

In this study, we first curated 42 MAMs-related genes from recent literature and systematically characterized their expression patterns in KIRC. We then employed a comprehensive machine learning strategy, involving 10 distinct algorithms and 101 algorithmic combinations, to develop an optimal MAMs-based scoring model. Subsequently, we assessed its associations with clinical parameters and patient prognostic outcomes. Furthermore, we explored the impact of this MAMs score on the tumor immune microenvironment and potential therapeutic responses in KIRC. Finally, DNM1L was pinpointed as a key gene in the MAMs model, and experimental validation confirmed that DNM1L knockdown can suppress tumor growth both in vitro and in vivo.

Methods

KIRC (kidney renal clear cell carcinoma) datasets processing

RNA sequencing data and clinical information for 537 KIRC patients and 71 normal tissue specimens were downloaded from The Cancer Genome Atlas (TCGA) database (https://portal.gdc.cancer.gov/). Additionally, the E-MTAB-1980 dataset, consisting of 101 ccRCC samples, was obtained from the ArrayExpress database (ebi.ac.uk/arrayexpress/). For external validation, the GSE29609 dataset, which includes transcriptomic data and clinical annotations of 39 ccRCC patients, was downloaded from the GEO repository (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE29609). Immunotherapy data for ccRCC patients were obtained from The Cancer Immunome Atlas (TCIA, https://www.tcia.at/home). Fifty MAMs (mitochondria-associated membranes)-related genes were obtained from previous studies [6–9], and the gene list is provided in Supplementary Table 1. The intersection of TCGA-KIRC, E-MTAB-1980, and MAMs genes was visualized using a Venn diagram.

Explore the expression patterns and characteristics of MAMs-related genes in KIRC

We performed differential expression analysis of MAMs-related genes between renal clear cell carcinoma (ccRCC) patients and normal tissue specimens using the limma package in R. The criteria for differential expression analysis were set at | log2(fold change) | > 1 and a false discovery rate (FDR) < 0.05. The results of the differential analysis were visualized using a heatmap. Additionally, we explored the differential expression of all MAMs genes between ccRCC patients and normal tissue specimens using boxplots, with p < 0.05 considered significant.

Prognostic scoring models for MAMs-Related genes in CcRCC using 10 models and 101 algorithms

We integrated 101 combinations of 10 machine learning algorithms, including Lasso, partial least squares regression for Cox (plsRcox), CoxBoost, random survival forest (RSF), elastic network (Enet), Ridge, stepwise Cox, supervised principal components (SuperPC), survival support vector machine (survival-SVM), and generalized boosted regression modeling (GBM), to develop scoring models for MAMs-related genes associated with the prognosis of renal clear cell carcinoma (ccRCC) in the TCGA and E-MTAB-1980 datasets. The optimal model for both datasets was selected based on the concordance index (C-index). The model score was calculated as: .  

Clinical correlation analysis and performance evaluation of the MAMs model

Based on the MAMs model, we calculated the score for each ccRCC patient and categorized the patients into high and low score groups according to the median score. Our model was mainly developed based on the TCGA database, and we validated the accuracy of the model in the E-MTAB-1980 and GSE29609 datasets. First, Kaplan-Meier (K-M) survival curves were used to compare potential differences in prognosis between the groups. Next, univariate and multivariate Cox regression analyses were conducted to further explore the prognostic value of the MAMs model in ccRCC patients. Additionally, box plots were employed to investigate the correlation between the MAMs model and clinical pathological variables. Finally, the performance of the MAMs model was assessed using ROC curves.

Evaluation of immune microenvironment and drug sensitivity

We employed the CIBERSORT algorithm to estimate immune infiltration scores for each patient and conducted Spearman correlation analysis to explore the relationships between genes in the MAMs model and immune infiltration scores. The results were visualized through a heatmap. Additionally, we examined the association between different MAMs scores and immune checkpoints. Furthermore, we analyzed the potential sensitivity of high and low score groups to immunotherapy, using data from renal clear cell carcinoma (ccRCC) patients in the Cancer Immunome Atlas (TCIA) database, which provides immune-related data and treatment responses. Lastly, we assessed the sensitivity of different score groups to common RCC targeted therapies, such as sunitinib, pazopanib, and sorafenib, by using IC50 values, which represent the concentration of a drug needed to inhibit 50% of the biological response. This analysis was conducted with the pRRophetic package in R to predict drug sensitivity. For the drug sensitivity assay, 5000 cells per well of 786-O and ACHN cells were seeded into 96-well plates, respectively. Sorafenib (catalog number # HY-10201 ) was added at a series of concentrations (0.01, 0.03, 0.1, 0.3, 1, 3, 10, 30, 100, and 300 µmol/mL) and incubated with the cells for 36 h. Each concentration was set up with 4 replicate wells. After incubation, cell viability was measured using the CCK-8 assay, and the half-maximal inhibitory concentration (IC50) value of sorafenib was calculated accordingly.

Identification of core genes in kidney renal clear cell carcinoma (KIRC)

The impact of genes in the MAMs model on the prognosis of KIRC patients was assessed using Kaplan–Meier survival curves. Core genes were identified based on hazard ratio (HR) values greater than 1, suggesting an increased risk of poor outcomes, and p-values less than 0.05.

RNA extraction and sample collection from KIRC tissues

RNA was extracted from 10 pairs of kidney renal clear cell carcinoma (KIRC) cancerous and adjacent non-cancerous tissues collected from the First Affiliated Hospital of Nanchang University, China. Ethical approval for the study and the use of biological samples was granted by the hospital’s Ethics Committee (Approval: (2023) CDYFYYLK (03–013)), and all procedures followed informed consent from the participants.Total RNA was extracted using TRIZOL (Yeasen Biotechnology, Shanghai, China, #10606ES60), and cDNA synthesis was performed using FastKing cDNA (TIANGEN, Beijing, China, #KR116). RT-qPCR was carried out using SYBR Green qPCR Master Mix (Servicebio, Hubei, China, #G3322-05). The relative expression of the DNM1L gene was calculated using the 2 − ΔΔCt method, with ACTIN as the reference gene. The primer sequences used for RT-qPCR were as follows:

DNM1L_F: 5’- GATGCCATAGTTGAAGTGGTGAC-3’,

DNM1L_R: 5’-CCACAAGCATCAGCAAAGTCTGG-3’,

ACTIN_F: 5’-TCTCCCAAGTCCACACAGG-3’

ACTIN_R: 5’-GGCACGAAGGCTCATCA-3’.

Cell culture, SiRNA Transfection, and knockdown validation

Cell lines and culture conditions were described in our previous article [15]. The two designed interference RNA sequences are as follows: DNM1L-Homo-686 (S: 5’-GCAGAACCCUAGCUGUAAUTT-3’, AS: 5’-AUUACAGCUAGGGUUCUGCTT-3’) and DNM1L-Homo-803 (S: 5’- GGAGCCAGCUAGAUAUUAATT-3’, AS: 5’- UUAAUAUCUAGCUGGCUCCTT-3’). The negative control (NC) group sequence is S: 5’-UUCUCCGAACGUGUCACGUTT-3’, AS: 5’-ACGUGACACGUUCGGAGAATT-3’. SiRNA transfection was performed using Lipofectamine 3000 (Thermo Fisher Scientific, Catalog #L3000015). 786-O and ACHN cells were seeded in 6-well plates, and transfection was carried out the next day when the cell density reached approximately 70%. Knockdown efficiency was validated through quantitative real-time PCR (RT-qPCR).

CCK-8 and colony formation assays for cell proliferation

Cell viability and proliferation were assessed using the Cell Counting Kit-8 (CCK-8, Servicebio, Hubei, China, #G4103-5ML) and colony formation assays. After 48 h of siRNA transfection, cells were seeded at a density of 2000 cells per well in 96-well plates, with three replicate wells for each group. Cell viability was measured on days 0, 1, 2, 3, and 4 by adding 10 µL of CCK-8 reagent to each well, followed by a 1.5-hour incubation at 37 °C. The optical density (OD) at 450 nm was then measured using a microplate reader. For colony formation assays, 1000 cells were seeded in 6-well plates and cultured for 10 days. After the incubation period, colonies were fixed using 4% paraformaldehyde (Servicebio, Hubei, China, #G1101-500ML), stained with 0.1% crystal violet (Servicebio, Hubei, China, #G1014-50ML), and quantified using ImageJ software.

Scratch wound-healing and transwell assays for cell migration and invasion

For the wound healing (scratch) assay, after transfection with siRNA, cells were cultured until they reached confluence in a 6-well plate. A sterile pipette tip was then used to create a scratch in the cell monolayer. After washing to remove any detached cells, serum-free medium was added to the wells, and migration was observed at 0 h and 24 h. Images were captured at these time points to assess the migration ability of the cells. In the transwell assay, two separate experiments were performed. For both experiments, 2 × 10^4 cells were seeded into the upper chamber of the transwell. In the first experiment, to assess cell migration, the transwell chambers were not coated with Matrigel. Cells were allowed to migrate for 36 h, after which the non-migrating cells on the upper side of the membrane were removed. The cells that migrated to the lower surface of the membrane were fixed, stained with crystal violet, and counted under a microscope. In the second experiment, to assess cell invasion, the transwell chambers were pre-coated with Matrigel (MedChemExpress, USA, #HY-K6001). Cells were seeded in the upper chamber, and after 36 h of incubation, the non-invading cells on the upper side of the membrane were removed. The invading cells on the lower side were then fixed, stained with crystal violet, and counted under a microscope.

RNA interference and lentivirus transduction

For stable DNM1L-knockdown cell lines, we used lentiviral vectors pLV3-U6-DNM1L-shRNA1 (#P67808, Miaoling) and pLV3-U6-DNM1L-shRNA2 (#P67680, Miaoling). Lentiviruses were packaged by transfecting HEK 293T cells with these vectors plus packaging plasmids pMD2.G (Addgene #12259) and psPAX2 (Addgene #12260) using Lipofectamine 2000 (Invitrogen) per manufacturer’s instructions. To generate stable 786-O lines, cells were infected with lentiviruses for ≥ 24 h, then selected with complete medium containing 3 µg/ml puromycin for ≥ 1 week.

Mouse xenograft tumor assay

Four-week-old male BALB/c nude mice were purchased from Charles River Laboratories (Beijing, China). All animals used in this study received standardized humane care in accordance with relevant animal experiment regulations, policies, and guidelines. All animal experimental procedures were approved by the Institutional Animal Care and Use Committee (IACUC) of the First Affiliated Hospital of Nanchang University. For xenograft establishment, control 786-O cells, shDNM1L#1-transfected 786-O cells, and shDNM1L#2-transfected 786-O cells (1 × 10⁶ cells per cell line) were each mixed with Matrigel at a 1:1 ratio, followed by subcutaneous injection into the flanks of the nude mice. Tumor size was measured with calipers every 7 days, and tumor volume was calculated using the formula: Volume = length × width² × 0.5. For animal sacrifice, mice were first anesthetized with 2% isoflurane (inhalation, maintained until loss of consciousness was confirmed by the absence of pedal reflex) to ensure they were unconscious. Subsequently, euthanasia was performed via cervical vertebra dislocation. This method was chosen as it is a rapid, humane approach consistent with the guidelines of the Institutional Animal Care and Use Committee (IACUC), minimizing potential distress to the animals.

Statistical analysis

Bioinformatics analyses were conducted using R software, and the experimental data were visualized with GraphPad Prism 10.0. Student’s t-test was employed to compare the means between two groups, while one-way or two-way ANOVA was used for comparisons across multiple groups. A p-value of less than 0.05 was considered statistically significant.

Results

Differential expression analysis of MAMs-related genes in KIRC

The schematic flowchart of the entire study is presented in Fig. 1. The Venn diagram (Fig. 2A, Table S2) revealed a total of 42 overlapping genes among TCGA-KIRC, E-MTAB-1980, and MAMs-related genes. Differential expression analysis identified seven MAMs-related genes (Table S3). The heatmap (Fig. 2B) demonstrated a trend toward elevated expression of these genes in KIRC compared to normal tissues. Moreover, as depicted in Fig. 2C, the majority of MAMs-related genes showed a tendency for higher expression in KIRC tumor tissues relative to their normal counterparts.

Fig. 1.

Schematic flowchart of the entire study

Fig. 2.

Exploration of Mitochondrial-associated membranes (MAMs) gene expression patterns in kidney renal clear cell carcinoma (KIRC). A Venn diagram showing 42 overlapping genes among TCGA-KIRC, E-MTAB-1980, and MAMs-related genes. B The heatmap demonstrates the expression patterns of MAMS-related differential genes in kidney clear cell carcinoma tissues versus adjacent normal tissues (| log2(fold change) | > 1 and a false discovery rate (FDR) < 0.05). C The differential expression analysis of all MAMS genes in normal tissues versus tumor tissues (*, p < 0.05; **p < 0.01, ***p < 0.001)

Stepwise Cox model selection and survival validation

All model calculation results are shown in Supplementary Table 4. Based on the concordance index (C-index), the Stepwise Cox [both] model was identified as the optimal model, achieving a C-index of 0.701 in the TCGA dataset and 0.707 in the E-MTAB-1980 dataset (Fig. 3A). This model comprised nine genes, and the MAMs score was calculated using the following formula: MAMs SCORE = APP expression * -0.2377554 + DNM1L expression * 0.3034758 + PINK1 expression * -0.4826166 + REEP1 expression * -0.2032927 + BECN1 expression * 0.4848243 + PDK4 expression * -0.2896075 + RTN1 expression * -0.228418 + PDZD8 expression * -0.2687954 + PML expression * 0.2599408. Based on the MAMS scoring system, we obtained the score for each patient, as shown in Supplementary Table 5. The Kaplan-Meier (K-M) survival curves demonstrated that patients in the high MAMs score group had worse prognoses compared to those in the low MAMs score group, both in the TCGA-KIRC dataset (p < 0.001, Fig. 3B), the E-MTAB-1980 dataset (p = 0.043, Fig. 3C), and the external validation dataset GSE29609 (p = 0.021, Fig. 3D).

Fig. 3.

Selection of the optimal model. A We utilized 10 distinct machine learning techniques to build a total of 101 models, and ultimately chose the Stepwise Cox [both] model, from which a scoring system for 9 MAMS was derived. B-D K-M survival curve analyses in TCGA-KIRC (B), E-MTAB-1980 (C), and the external validation set GSE29609 (D) consistently showed that patients in the high-score group had poorer prognosis (p < 0.05 for all analyses)

Prognostic significance of the MAMs model

In the TCGA dataset, univariate and multivariate Cox regression analyses indicated that age, T stage, M stage, and MAMs score were associated with poorer prognosis (Hazard ratio > 1, p < 0.05, Fig. 4A, B). Similarly, in the external validation dataset E-MTAB-1980, T stage, N stage, and MAMs score were identified as prognostic risk factors (Hazard ratio > 1, p < 0.05, Fig. 4C, D). Boxplot analysis demonstrated significant associations between MAMs score and clinicopathological features. In the TCGA dataset (Fig. 4E), higher tumor grade corresponded to an increased MAMs score (G1-G2 vs. G3-G4, p = 1.3e-07). Moreover, MAMs score was positively correlated with increasing T stage (T1 vs. T2 vs. T3 vs. T4, p < 0.05). Similarly, patients with higher N stage and M stage exhibited elevated MAMs scores (N0 vs. N1, p = 0.059; M0 vs. M1, p = 2.5e-06). In the external validation dataset (Fig. 4F), the results were consistent, showing that higher clinical stage and grade were associated with a higher MAMs score, which might contribute to the worse prognosis observed in the high-score group.

Fig. 4.

Clinical correlation analysis and performance evaluation of the MAMs Model. Univariate and multivariate Cox analyses both reveal that the MAMS scoring system is associated with prognosis (P < 0.05) in the TCGA (A, B) and the E-MTAB-1980 dataset (C, D). Clinical correlation analysis shows that higher TNM stage and Grade are associated with higher MAMS scores in both the TCGA (E) and the E-MTAB-1980 dataset (F). ROC curves for predicting overall survival at 1, 3 and 5 years in the TCGA (G) and E-MTAB-1980 cohort (H). ROC curves for different clinical characteristics in the TCGA and the E-MTAB-1980 dataset

To assess the prognostic performance of the MAMs model, ROC curves were plotted. In the TCGA dataset, the AUC values for 1-, 3-, and 5-year survival prediction were 0.768, 0.728, and 0.718, respectively (Fig. 4G). In the external validation dataset, the corresponding AUC values were 0.896, 0.730, and 0.720, respectively (Fig. 4H). Additionally, a comparative analysis with other clinicopathological features suggested that, in both the TCGA and E-MTAB-1980 datasets, the MAMs score may serve as a relatively strong prognostic predictor (TCGA: AUC = 0.896, Fig. 4I; E-MTAB-1980: AUC = 0.768, Fig. 4J).

Immune microenvironment characteristics and drug sensitivity analysis

As shown in the heatmap (Fig. 5A), most genes in the MAMs scoring system exhibited negative correlations with Tregs, gamma delta T cells, follicular helper T cells, CD8 + T cells, activated NK cells, and M0 macrophages. In contrast, they showed positive correlations with CD8 + T cells, monocytes, resting mast cells, M2 macrophages, and M1 macrophages. The immune checkpoint analysis (Fig. 5B) revealed that, except for HHLA2, KIR3DL1, and NRP1, the majority of immune checkpoints tended to have higher expression levels in the high MAMs score subgroup. Moreover, TCIA analysis indicated that the high MAMs score group appeared to be more responsive to immunotherapy only when both CTLA4 and PD-1 were positive (p < 0.05), whereas no notable differences were observed under other conditions (Fig. 5C). Furthermore, drug sensitivity analysis suggested that patients in the low-score group had relatively lower half-maximal inhibitory concentration (IC50) values for Sorafenib and Erlotinib, indicating a potential for increased sensitivity to these drugs. In contrast, the high-score group exhibited relatively lower IC50 values for Sunitinib and Temsirolimus, implying a possible tendency toward higher sensitivity to these targeted therapies (Fig. 5D). As shown in Supplementary Fig. 2A and 2B, knockdown of DNM1L in 786-O cells significantly decreased the IC50 value of sorafenib from 11.04 µmol/mL to 4.10 µmol/mL. Consistent results were observed in ACHN cells: the IC50 value of sorafenib was reduced from 18.32 µmol/mL to 10.47 µmol/mL after DNM1L knockdown (Supplementary Fig. 2C and 2D). These experimental findings were consistent with the results of our bioinformatic prediction of drug sensitivity.

Fig. 5.

Immune Microenvironment Characteristics and Drug Sensitivity Analysis. A Heatmap of Spearman Correlation between MAMs Model Genes and Immune Infiltration Scores. B Association between Different MAMs Scores and Immune Checkpoints. C Sensitivity of High and Low Score Groups to Immunotherapy in KIRC Patients from the TCIA Database. isp_ctla4_pos_pd1_pos indicates patients with both CTLA4 and PD1 expression being positive, isp_ctla4_pos_pd1_neg refers to patients with positive CTLA4 and negative PD1 expression, isp_ctla4_neg_pd1_pos denotes patients with negative CTLA4 and positive PD1 expression.isp_ctla4_neg_pd1_neg represents patients with neither CTLA4 nor PD1 expression (both negative). On the vertical axis, higher scores indicate a greater sensitivity to immunotherapy. D Drug-sensitive analysis. (*, p<0.05; **p<0.01, ***p<0.001)

Selecting DNM1L as the core gene for experimental validation and its functional analysis

Kaplan–Meier survival analysis indicated that patients with higher expression of DNM1L and PML (HR > 1 and p < 0.05) tended to have poorer prognoses (Supplementary Fig. 1). Recent studies have suggested that PML may inhibit p53 activity and cellular senescence in KIRC [16]. Based on these findings, we selected DNM1L for further experimental validation.

RT-qPCR analysis of 10 paired KIRC tumor and adjacent normal tissues showed that DNM1L expression was higher in eight tumor samples (Fig. 6A). Similarly, qPCR results from cell lines indicated that DNM1L expression was increased in most renal cancer cells compared to normal renal tubular epithelial cells (HK-2), with relatively higher levels observed in 786-O and ACHN cells (Fig. 6B). We verified the knockdown efficiency of siRNA sequences S1 and S2 in 786-O and ACHN cells (Fig. 6C). All PCR data, after normalization, are available in Table S6. CCK-8 and colony formation assays showed that DNM1L knockdown was associated with reduced proliferation in 786-O and ACHN cells (Fig. 6D–F). Additionally, scratch wound-healing and Transwell assays suggested that silencing DNM1L was linked to decreased migration and invasion in 786-O and ACHN cells (Fig. 6G–J). These observations suggest that DNM1L may contribute to cell proliferation, invasion, and migration in kidney renal clear cell carcinoma. To further investigate the role of DNM1L in tumorigenesis, we established a mouse subcutaneous tumorigenesis model by subcutaneously injecting 786-O cells into nude mice. In control mice, injection of 786-O cells resulted in the formation of large subcutaneous tumors; in contrast, mice injected with 786-O cells transduced with either of the two distinct DNM1L shRNAs showed reduced tumor formation (Fig. 6K-M). Collectively, these data indicate that DNM1L knockdown impairs the long-term proliferation and survival of kidney renal clear cell carcinoma cells both in vitro and in vivo.

Fig. 6.

Selecting DNM1L as the Core Gene for Experimental Validation. A qRT-PCR analysis illustrating the expression levels of DNM1L in 10 pairs of normal kidney tissue versus kidney cancer tissue. B Comparative qRT-PCR analysis showcasing DNM1L expression in normal renal cells and various kidney cancer cell lines. C Validation of DNM1L expression levels by qRT-PCR in 786-O and ACHN cell lines following transfection with S1-targeting and S2-targeting DNM1L siRNAs. D and E Proliferation curves of 786-O and ACHN cell lines were measured using the Cell Counting Kit-8 (CCK-8) assay after transfection with the indicated vectors. F A colony formation assay was conducted to assess the proliferative capacity of the cells following transfection with the specified vectors. G-J Cell migration capability was evaluated through wound healing (G, H) and transwell migration (I, J) assays. K 786-O shNC cells or shDNM1L cells were subcutaneously injected into Balb/c nude mice. Tumour volume growth curves from the indicated days and tumor growth was measured every 7 days. L After 28 days, mice were sacrificed and representative tumour images at the end of the experiment are presented. M Tumour weights were examined in the two groups. Statistical significance was indicated by *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001

Discussion

Mitochondria-associated membranes (MAMs) are specialized contact sites between the endoplasmic reticulum and mitochondria, playing essential roles in calcium signaling, lipid metabolism, and cellular homeostasis [17–19]. Emerging evidence suggests that metabolic reprogramming is a hallmark of kidney renal clear cell carcinoma (KIRC), with fatty acid metabolism being a crucial driver of tumor progression [20]. However, despite the well-documented role of lipid metabolism in KIRC, the involvement of MAMs in this malignancy remains unexplored. Therefore, we conducted a comprehensive analysis to elucidate the role of MAMs-related genes in KIRC and to establish a prognostic model based on their expression profiles. We further demonstrated through in vitro experiments that DNM1L may represent a promising therapeutic target for KIRC treatment. Based on previous literature, we identified 42 MAMs-related genes and analyzed their expression profiles in KIRC. Heatmap and box plot results revealed that the majority of these MAMs-related genes were significantly upregulated in KIRC tumor tissues compared to normal tissues. To enhance the reliability of our findings, we incorporated three independent KIRC datasets from different sources for subsequent analyses. To further investigate the potential role of MAMS-related genes in KIRC, we employed machine learning algorithms and identified the Stepwise Cox [both] model as the optimal MAMs scoring model. This model demonstrated robust performance, with C-index values of 0.701 in the TCGA dataset and 0.707 in the E-MTAB-1980 dataset. Using this model, we calculated individual MAMs scores for each patient and stratified KIRC patients into high- and low-score groups based on the median score. Kaplan-Meier (K-M) survival analysis indicated that patients in the high MAMs score group had significantly worse prognoses compared to those in the low-score group, consistent across the TCGA dataset, the E-MTAB-1980 dataset, and the external validation dataset GSE29609 (p < 0.05). Univariate and multivariate Cox regression analyses further confirmed that the MAMs scoring system was an independent prognostic risk factor in both the TCGA-KIRC dataset and the E-MTAB-1980 validation set, with hazard ratios (HR) greater than 1 and p-values less than 0.05. Why do patients with high MAMs scores exhibit poorer prognoses? To address this question, we further investigated the correlation between MAMs scores and clinicopathological factors. The results demonstrated that patients with higher T, N, and M stages, as well as higher Grade classifications, tended to have higher MAMs scores. This trend was consistently observed in both the TCGA dataset and the external validation dataset. These findings suggest that the MAMs score may reflect more aggressive tumor biology and advanced disease progression, which could explain the association with poorer outcomes. Additionally, the predictive performance of the MAMs scoring model was evaluated using ROC curves. The AUC values exceeded 0.7 in both the TCGA and E-MTAB-1980 datasets, indicating good discriminative ability for prognostic assessment. The standard treatment for KIRC typically involves radical or partial nephrectomy with curative intent. However, metastatic recurrence occurs in 20–30% of patients after surgery. In these cases, systemic treatment mainly relies on targeted therapy and immunotherapy as the key therapeutic strategies [21]. We analyzed changes in the tumor immune microenvironment and immune checkpoints, as these factors may influence the effectiveness of immunotherapy [22]. Using the CIBERSORT algorithm, we assessed the correlation between MAMs scoring system genes and 22 immune cell types. The results showed that most genes in the MAMs scoring system exhibited negative correlations with Tregs, gamma delta T cells, follicular helper T cells, CD8 + T cells, activated NK cells, and M0 macrophages. In contrast, they showed positive correlations with CD8 + T cells, monocytes, resting mast cells, M2 macrophages, and M1 macrophages. Immune checkpoint analysis revealed that patients in the high MAMs score group had higher expression of CTLA4 and PDCD1. The TCIA immunotherapy cohort further indicated that patients with high MAMs scores were more responsive to immunotherapy in the CTLA4 + and PD1 + subgroups. The enrichment of CD8 + T cells and M1 macrophages, along with the elevated expression of immune checkpoints, may contribute to the increased sensitivity of the high-score group to immunotherapy [23–25]. Drug sensitivity analysis revealed that patients in the high MAMs score group had lower IC50 values for Temsirolimus and Sunitinib, indicating more sensitivity to these drugs. In contrast, patients in the low-score group exhibited higher sensitivity to Sorafenib. Based on these findings, it is suggested that patients in the high-score group may benefit from a combination of immunotherapy with Temsirolimus or Sunitinib, whereas those in the low-score group could be treated with immunotherapy plus Sorafenib or Sorafenib alone.

The MAMs scoring model consists of nine genes: APP, PINK1, REEP1, BECN1, PDK4, RTN1, PDZD8, PML, and DNM1L. APP regulates cell survival, adhesion, and migration [26] and is highly expressed in various cancers [27]. Its silencing inhibits proliferation and enhances treatment sensitivity by promoting apoptosis [28]. PINK1, a serine/threonine kinase, regulates mitophagy to maintain mitochondrial and redox homeostasis [29]. It contributes to drug tolerance and pazopanib resistance in clear cell renal cell carcinoma by remodeling lipid metabolism [30]. Receptor expression enhancing protein 1 (REEP1) interacts with NDUFA4 to maintain mitochondrial complex IV integrity in amyotrophic lateral sclerosis [31]. Reports on REEP1 in tumors are rare. BECN1 binds to proline-hydroxylated Von Hippel-Lindau (VHL) protein, which inhibits the interaction between BECN1 and VPS34, thereby suppressing the initiation of autophagy in clear-cell renal cell carcinoma [32]. Pyruvate dehydrogenase kinase 4 (PDK4) is upregulated in senescent cells, promoting aerobic glycolysis and lactate production while preserving mitochondrial activity. Its inhibition reduces tumor growth, DNA damage, and the senescence-associated secretory phenotype. PDK4 suppression also alleviates age-related frailty and dysfunction [33]. RTN1 expression is elevated in chronic kidney disease (CKD) and correlates with disease severity. It induces endoplasmic reticulum (ER) stress and apoptosis in renal cells, while its knockdown alleviates ER stress and fibrosis in various CKD models [34]. DZD8, phosphorylated by AMPK, promotes glutaminolysis by activating glutaminase 1 (GLS1) under low glucose conditions, preceding fatty acid utilization in skeletal muscle and macrophages [35]. Additionally, Hojo et al. reported that PDZD8 could serve as a drug target for sunitinib in gastric cancer [36], and further research is needed to explore whether it could also be a target for sunitinib in KIRC. Recent studies have suggested that PML may inhibit p53 activity and cellular senescence in KIRC [16]. DNM1L (DRP1) regulates mitochondrial dynamics and cancer progression. In glioma, inhibiting the DNM1L/DRP1-FIS1 axis disrupts mitochondrial remodeling and limits aerobic respiration, suppressing tumor growth [37]. In esophageal squamous cell carcinoma, DRP1 overexpression triggers mitochondrial dysfunction, activating the cGAS-STING pathway and autophagy [38]. In oral squamous cell carcinoma, DRP1 maintains mitochondrial morphology and stemness, while its suppression alters glutaminolysis, affects histone modification, and enhances ferroptosis-based anticancer effects [39]. The role of DNM1L in KIRC has not yet been reported.

Our study demonstrated that DNM1L expression was elevated in kidney renal clear cell carcinoma (KIRC) tumor tissues and renal cancer cell lines compared to adjacent normal tissues and HK-2 cells. Notably, 786-O and ACHN cells showed relatively higher DNM1L levels. Functional assays indicated that DNM1L knockdown led to a significant reduction in cell proliferation, as evidenced by CCK-8 and colony formation assays, while also impairing migration and invasion capabilities in 786-O and ACHN cells. Consistently, in vivo experiments using a mouse subcutaneous xenograft model further supported these observations: mice injected with DNM1L-knockdown 786-O cells (transduced with two distinct shRNAs) developed significantly smaller tumors with reduced volume and weight compared to controls (Fig. 6K-M). Collectively, these in vitro and in vivo data suggest that DNM1L plays a prominent role in promoting tumor progression in KIRC. However, our study has several limitations. First, although we incorporated kidney renal clear cell carcinoma (KIRC) datasets from different sources, they were obtained from public databases, which may introduce inherent biases. Second, the retrospective and indirect nature of predicting immunotherapy response underscores the necessity for prospective trials involving larger patient cohorts to enhance the reliability of our findings; additionally, the correlation between MAMs score and immune cell infiltration lacks functional validation. Third, the scope of in vitro cell models was restricted to two KIRC cell lines, and rescue experiments were not conducted to rule out potential off-target effects of the applied siRNAs/shRNAs; additionally, further mechanistic studies are needed to comprehensively elucidate the role of DNM1L in KIRC progression.

Conclusion

In this study, we identified the high expression of mitochondrial-associated membranes (MAMs)-related genes in kidney renal clear cell carcinoma (KIRC). Utilizing machine learning algorithms, we developed a scoring system comprising nine MAMs-related genes, which can be employed for prognostic evaluation and therapeutic response assessment. Furthermore, we identified DNM1L as the key gene within the MAMs scoring model and validated its functional role through experimental studies. These findings provide a certain insight into potential prognostic biomarkers and therapeutic targets for KIRC.

Abbreviations

MAMs

Mitochondrial-associated membranes

KIRC

Kidney renal clear cell carcinoma

TCGA

The Cancer Genome Atlas

plsRcox

Partial least squares regression for Cox

RSF

Random survival forest

Enet

Elastic network

SuperPC

Supervised principal components

survival-SVM

Survival support vector machine

GBM

Generalized boosted regression modeling

ccRCC

Renal clear cell carcinoma

TCIA

the Cancer Immunome Atlas

OD

Optical density

C-index

Concordance index

Authors’ contributions

XL, ZJ and BF contributed to the conception of the study. SL, JL, FZ and SX contributed significantly to the analysis and manuscript preparation.

Funding

This study was supported by Jiangxi Provincial Department of Urology Key Laboratory (Grant No. 2024SSY06111). National Natural Science Foundation of China (grant NO.82203365). the National Natural Science Foundation of Jiangxi Province (20232BAB206090). Jiangxi Provincial Health Commission Technology Plan Project (No. 202510025).

Data availability

The transcriptomic data, clinical data, and immunotherapy data used in this study were all accessible from public databases.

Declarations

Ethics approval and consent to participate

RNA was extracted from KIRC cancerous and adjacent non-cancerous tissues from the First Affiliated Hospital of Nanchang University. Human tissue procedures were approved by the hospital’s Ethics Committee (Approval ID: (2023) CDYFYYLK (03–013)) with patient informed consent. Animal protocols were approved by the hospital’s IACUC (Approval ID: CDYFY-IACUC-202505GR082). This research was conducted in strict adherence to the principles of the World Medical Association’s Declaration of Helsinki (revised 2013).

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.