In this investigation, the Human Protein Atlas (HPA) repository was used to obtain gene expression datasets and subcellular localization images for MDH1 across various tissues. Data on MSI and TMB were sourced from The Cancer Genome Atlas (TCGA). The conservation of the MDH1 gene across vertebrate species was visualized using the University of California, Santa Cruz (UCSC) Cancer Genomics Browser. The UCSC XENA platform was used to acquire transcriptome RNA-seq data and clinical profiles from TCGA for a wide range of cancer types, while the Genotype-Tissue Expression (GTEx) project provided data for normal tissue samples. RNA-seq data, quantified as transcripts per million (TPM), were extracted for both cancerous and adjacent (para-cancer) samples from TCGA. The ComBat algorithm, which is part of the “sva” package in R, serves as an effective tool to alleviate the impact of batch effects stemming from non - biological technical factors in gene expression data across multiple datasets during data preprocessing. To explore the mutational and copy number variation (CNV) landscapes of MDH1 in various cancer types, the cBioPortal for Cancer Genomics was utilized (31). Clinical data related to immune therapies were obtained from the Tumor Immunotherapy Gene Expression Resource (TIGER) (32). Supplementary Table S1 summarizes the public data cohorts included in this study.

Recognizing the Cox proportional hazards model as a standard approach for survival analysis, this study applied it to a comprehensive cancer dataset sourced from the UCSC repository. The goal was to analyze the association between MDH1 expression and oncological outcomes. The diagnostic potential of MDH1 in cancer was assessed using Receiver Operating Characteristic (ROC) analysis. The Gene Expression Profiling Interactive Analysis 2 (GEPIA2) platform was used to examine correlations between MDH1 expression levels and key survival metrics, namely overall survival (OS) and disease-free survival (DFS). Additionally, survival data from multiple lung cancer cohorts were compiled from the Lung Cancer Explorer database (https://lce.biohpc.swmed.edu/lungcancer).

For this research, we accessed the GSE200972 dataset from the GEO repository, extracting 14 LUAD specimens for detailed analysis. We also integrated a diverse collection of single-cell datasets from various cancer types: GSE207422 (LUAD), GSE120575 [skin cutaneous melanoma (SKCM)], GSE145281 [bladder urothelial carcinoma (BLCA)], GSE169246 [triple-negative breast cancer (TNBC)], GSE212217 (UCEC), GSE229353 (lung carcinoma), and GSE235672 [glioblastoma multiforme(GBM)], all obtained from GEO. A subset of 54 samples from these datasets was selected for in-depth analysis. scRNA-seq analysis was performed using the “Seurat” R package. Cells were filtered based on the following criteria: expressed gene count per cell between 200 and 6000, with mitochondrial content less than 5%.

Batch effects were removed utilizing the harmony R package. Cluster-specific cell types were then identified using discrete cellular markers. Additionally, we obtained multiple scRNA-seq datasets from the Tumor Immunity Single Cell Center (TISCH) for supplementary analysis.

To assess cellular differentiation and progression, we used Monocle2 with its standard parameters to perform pseudotime trajectory analysis, organizing the cells along developmental pathways that were divided into distinct branches reflecting cellular progression or differentiation.

We randomly selected samples from each of the different cancer types across multiple GEO cohorts (GSE203612, GSE250636, GSE193460, GSE245704 and GSE194329) and Spatial Transcript Omics DataBase (https://db.cngb.org/stomics/) to create a spatial pan-cancer cohort. It consists of 9 different types of cancers, with at least three samples for each type of cancer. Spatial clustering was performed using a graph-based approach implemented in the Scanpy package (33). Principal component analysis (PCA)-based dimensionality reduction and subsequent spatial transcriptomic (ST) point grouping were executed via the RunPCA, FindNeighbors, and FindClusters algorithms. Leveraging histological data from hematoxylin and eosin (H&E)-stained tissue sections, clusters were annotated to clarify cell identities. Further analysis and visualization of the annotated clusters enabled the assessment of MDH1 and CD68 expression intensities as well as their spatial localization patterns within the tissue landscape.

Tumor Mutational Burden (TMB) was quantified by counting non-synonymous somatic mutations. Neoantigens from LUAD samples in TCGA were retrieved from The Cancer Immunome Atlas (TCIA) (34), and the tumor neoantigen burden (TNB) was calculated by aggregating the total predicted neoantigens.

DNA-damage-response (DDR) integrity was classified by scanning each tumor for non-silent variants within the seven core repair branches [Base Excision Repair (BER), Nucleotide Excision Repair (NER), Mismatch Repair (MMR), Homologous Recombination Repair (HRR), Non-Homologous End Joining (NHEJ), Fanconi Anemia (FA), or Translesion Synthesis (TLS)]; any lesion classified the sample as DDR-mutant (DDR-Mut), whereas absence of such events yielded DDR-wild-type (DDR-WT).

For MDH1, we integrated four orthogonal layers of genomic regulation. Somatic mutation and copy-number landscapes were retrieved from cBioPortal; promoter and gene-body methylation were profiled with UALCAN and refined by MEXPRESS (https://mexpress.ugent.be/). Clinically relevant alternative-splicing events were extracted via OncoSplicing (ClinicalAS tool) (35) and visualized with PanPlot, comparing PSI values between TCGA tumors and GTEx normal tissues for events spanning ≥3 cancer types.

The responsiveness of cancer cell lines (CCLs) to various drugs was evaluated using data from three sources: the PRISM (36), CTRP (37), and GDSC (38) databases. Both CTRP and PRISM assess drug sensitivity using the AUC, while GDSC uses the IC50. It is important to note that lower IC50 or AUC values indicate higher responsiveness to the treatment compounds.

MDH1 structure was modelled with AlphaFold3 (39). Binding poses of candidate ligands from PubChem were predicted using AutoDockTools within an 80 × 80 × 80 Å grid centered at (65, 20, 180); complexes were inspected in PyMOL. To assess the stability of the BI-2536–MDH1 complex, 100-ns MDS was performed in GROMACS (40, 41).

The LUAD cell lines, A549 and PC-9, normal bronchial epithelial cells (BEAS), and human monocyte cell line (THP-1) were obtained from the Shanghai Cell Bank, affiliated with the Chinese Academy of Sciences. These cells were cultured in growth medium containing 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin mixture at 37°C. A549 and PC-9 cell lines were grown in DMEM, while THP-1 cells were cultured in RPMI 1640 medium.

A total of thirty LUAD tissue specimens were obtained from the First Affiliated Hospital of Wenzhou Medical University. Participants provided their written consent to be involved in the study. BI-2536 was acquired from MedChemExpress (#HY50698, USA).

RNA was isolated from both cell and tissue samples using Trizol reagent (Takara, Japan). Subsequently, cDNA synthesis was carried out with the PrimeScript RT enzyme kit. qRT-PCR was performed using the SYBR Green fluorescence detection system. Primer sequences utilized in this study are listed in Supplementary Table S2 .

In this study, custom-designed MDH1 small interfering RNA (siRNA) and its respective scrambled non-targeting control (SiNC) were used for investigation. The specific sequences of the MDH1 siRNA are provided in Supplementary Table S3 . LUAD cell lines A549 and PC-9 were transfected with the siRNA, and cells were collected 48 hours post-transfection for further analysis.

Following a 48-hour incubation, the transfected cells were harvested and plated into 96-well plates at a density of 5,000 cells per well. Cell proliferation was assessed every 24 hours using the CCK-8 assay, with absorbance measurements taken after a 1-hour incubation with the reagent. A549 and PC-9 cells were seeded into 96-well plates; 24 h later, they were exposed to a gradient of BI-2536 for an additional 24 h, after which viability was quantified with the CCK-8 assay.

Forty-eight hours post-transfection, 300 μl of serum-free medium containing 4 × 10^4 cells was added to the upper chamber, while the lower chamber was filled with 600 μl of medium supplemented with 20% FBS. After 24 hours, the medium was changed, and non-migratory cells were removed with cotton swabs. The migrated cells were then fixed and stained with 0.01% crystal violet. Finally, the cells that had migrated to different regions of the membrane were counted under a microscope.

Following transfection, cells were seeded in a 6-well plate and cultured until they reached 80%–90% confluence. A linear wound was then created by carefully scraping the cell monolayer using a 1000 μl pipette tip. Detached cells were removed by rinsing with phosphate-buffered saline (PBS). Wound closure was tracked at designated time points by capturing images using an inverted microscope.

After transfection, the cells were cultured in fresh medium for an additional 48 hours. The lactate concentration in the culture supernatant was subsequently measured using spectrophotometric measurements.

For the invasion assay, Matrigel was diluted at a ratio of 1:10 and applied at 70 μL per well to the upper chamber of the transwell system. In the migration assay, 600 μL of medium containing 20% FBS was added to the lower chamber, while 9 × 10^4 cells in 300 μL of serum-free medium were seeded in the upper chamber. For the M2 macrophage infiltration assay, THP-1 cells were differentiated into macrophages by treating with 100 ng/mL of phorbol 12-myristate 13-acetate (PMA; MCE, USA) for 24 hours. Polarization towards the M2 phenotype was induced by incubating the cells with 20 ng/mL of interleukin 4 (IL-4; MCE, USA) for 48 hours. The M2 macrophages were then cultured in the upper chamber with serum-free medium, while A549 and PC-9 cells were cultured in the lower chamber. After 48 hours of incubation, cells were fixed for further analysis.

A total of thirty LUAD tissue sections, embedded in paraffin, were prepared for IF staining. The sections were deparaffinized, treated with 3% hydrogen peroxide to block endogenous peroxidase activity, and underwent antigen retrieval. They were then incubated with 2% bovine serum albumin for 30 minutes to prevent non-specific binding. After blocking, the sections were incubated with primary antibodies against MDH1 (rabbit polyclonal, 1:100 dilution, Proteintech) and CD68 (mouse monoclonal, 1:3000 dilution, Proteintech). Following this, secondary antibodies were applied: CoraLite488-labeled Goat Anti-Mouse IgG and CoraLite594-labeled Goat Anti-Rabbit IgG. Finally, cell nuclei were visualized by counterstaining with 4’,6-diamidino-2-phenylindole (DAPI).

Cell samples were collected 48 hours after transfection, and WB analysis was performed according to established protocols (42). Total protein was extracted using RIPA buffer, supplemented with phosphatase and protease inhibitors. The proteins were then denatured and prepared for WB analysis. Primary antibodies targeting MDH1 (1:5000 dilution, Proteintech), HK2(1:5000 dilution, Proteintech), LDHA (1:2000 dilution, Proteintech) and Tubulin (1:10000 dilution, Proteintech) were used in this experiment.

Medium was collected in chilled sterile tubes, cleared by 4°C centrifugation (1000 × g, 10 min), and aliquoted into microtubes. Cytokine quantification (IL-10 and TNF-α) was performed with the commercial ELISA reagents (Solarbio, Beijing) according to the vendor’s specifications.

CETSA was carried out following a published protocol (43) with minor modifications. Briefly, A549 cell were exposed to BI-2536 (500 nM) or vehicle for 24 h, harvested, and resuspended in PBS supplemented with protease inhibitors. The suspension was split into seven aliquots that were individually heated (55, 60, 65, 70, 75, 80, 85 °C) for 3 min. Thermal‐treated samples were snap-frozen in liquid nitrogen for 8 h, thawed at 26°C for 6 min, and the freeze–thaw cycle was repeated twice to ensure complete lysis. After centrifugal clarification, MDH1 levels in the supernatants were determined by immunoblotting.

The mean plus the standard error of the mean was shown after performing at least three independent experiments for data analysis. Statistical analyses were carried out using R version 4.1.1 and GraphPad Prism 10.0. Prior to statistical testing, data were assessed for normality using the Shapiro-Wilk test and for homogeneity of variances using Levene’s test. For group comparisons, if data met the assumptions of normality and equal variances, one-way ANOVA with Tukey’s post-hoc correction for multiple comparisons was used. If assumptions were violated, the non-parametric Kruskal-Wallis test with Dunn’s post-hoc correction was applied instead of Student’s t-tests. For correlations between variables, Pearson’s correlation coefficient was used for parametric data, and Spearman’s rank correlation coefficient was used for non-parametric data. To control for multiple hypothesis testing, the Benjamini-Hochberg procedure was applied to adjust p-values, and a corrected p-value threshold of less than 0.05 was considered statistically significant.

We analyzed MDH1 gene expression in a cohort of healthy subjects using data from the GTEx project. Our results showed that MDH1 was widely transcribed across various organs, with particularly high expression levels observed in the cerebral cortex and basal ganglia, as shown in Figure 2A . Additionally, we examined MDH1 protein expression in a diverse group of healthy individuals, using data from the HPA. This analysis identified a significant upregulation of MDH1 in the cerebral cortex and stomach, as illustrated in Figure 2B .

In the subsequent phase of our study, we explored MDH1 mRNA expression in cancerous tissues involved in 33 cancer types, utilizing data from TCGA. Our results identified significant upregulation of MDH1 mRNA in several cancers, such as cholangiocarcinoma (CHOL), cervical squamous cell carcinoma and endocervical adenocarcinoma (CESC), hepatocellular carcinoma (LIHC), head and neck squamous cell carcinoma (HNSC), LUAD, lung squamous cell carcinoma (LUSC), uterine corpus endometrial carcinoma (UCEC) and stomach adenocarcinoma (STAD). Conversely, MDH1 mRNA expression was found to be reduced in breast invasive carcinoma (BRCA), colon adenocarcinoma (COAD), glioblastoma multiforme (GBM), kidney chromophobe (KICH), kidney renal clear cell carcinoma (KIRC), and thyroid carcinoma (THCA), as illustrated in Figure 2C .

Additionally, by integrating data from GTEx and TCGA, we observed significant overexpression of MDH1 in 19 out of the 33 cancer types analyzed ( Figure 2D ). We further examined IHC results for MDH1 in the HPA database, which showed strong expression levels in LUAD, LUSC, PAAD, and LIHC, while expression was lower in COAD and BRCA. Collectively, these results underscore the significant overexpression of MDH1 in various cancer types, highlighting its essential involvement in tumorigenesis, as depicted in Figure 2E .

To explore the clinical implications of MDH1 expression, we conducted a comprehensive analysis to assess its relationship with patient survival metrics, including OS, DSS, DFS, and PFS, across 33 cancer types in the TCGA database. Our result unraveled that MDH1 expression was an adverse factor for OS, DSS, and PFI in LUAD, Uveal Melanoma (UVM), and KICH ( Figure 3A ). Furthermore, Kaplan-Meier survival analysis showed that MDH1 expression significantly impacted OS. Specifically, it acted as a risk factor for OS in LUAD, KICH, acute myeloid leukemia (LAML), and UVM, while it was associated with improved OS in cervical squamous cell carcinoma (CESC), KIRP, LUSC, and sarcoma (SARC), as shown in Figure 3B ( Supplementary Table S4 ). Additionally, analysis of TCGA data identified a positive correlation between MDH1 expression and tumor stage, as well as the metastatic (M) and nodal (N) classifications in LUAD ( Figure 4A , Supplementary Figures S2A, B ). These findings indicate that MDH1 expression is most strongly associated with the aggressiveness of LUAD. To further validate these results, we extracted data from 40 LUAD patient cohorts in the TIGER database and Lung Cancer Explorer (https://lce.biohpc.swmed.edu/lungcancer/). Survival analysis in these cohorts showed that high MDH1 expression was linked to poorer survival outcomes in most LUAD patient groups, as detailed in Supplementary Figures S2C-E .

Additionally, the diagnostic significance of MDH1 across various tumors was assessed utilizing ROC curves. The results identified that MDH1 exhibited high diagnostic accuracy for CESC, CHOL, BRCA, LUSC, LIHC, KIRC, KICH, and READ, as indicated by the area under the curve (AUC) presented in Figure 4B . Furthermore, MDH1 protein expression patterns were analyzed using the HPA database, revealing that MDH1 was mainly localized in the cytoplasm and co-localized with the endoplasmic reticulum. These findings indicate that MDH1 may be involved in lipid and protein biosynthesis and maturation, as shown in Figure 4C .

Genomic techniques provide a solid framework for understanding oncogenic processes (44). MDH1 gene amplification was most commonly observed in non-small cell lung carcinoma (NSCLC), UCEC, BLCA, and ovarian cancers, while mutations were predominantly found in COAD. Additionally, a significant frequency of SNVs was detected in SKCM, UCEC, and KICH ( Supplementary Figure S3A ). A detailed analysis of the mutational landscape of MDH1 is presented in Supplementary Figure S3B . On the epigenetic level, the methylation status of various sites within the MDH1 promoter and gene body showed a strong positive correlation with mRNA expression levels in most tumor samples ( Supplementary Figure S3C ).

Through computational analysis, we successfully identified the m6A modification sites within the MDH1 gene sequence, as shown in Supplementary Figure S3D . Our analysis revealed four m6A sites with a high degree of confidence, located at nucleotide positions 2287, 2350, 6534, and 16326. Dysregulated gene expression in cancer is often linked to abnormal DNA methylation patterns, as highlighted by previous studies (45). To explore the potential connection between aberrant transcriptional profiles of MDH1 and its methylation status, we utilized the UALCAN database (46) to examine methylation patterns in both cancerous and non-cancerous tissues. By comparing the differential methylation between cancerous and non-cancerous samples in the UALCAN database ( Supplementary Figure S4A ), we observed a correlation between higher MDH1 expression levels in LUAD, BLCA, thyroid carcinoma (THCA), and testicular germ cell tumors (TGCT) and reduced methylation levels ( Supplementary Figure S4A ). Furthermore, in the MEXPRESS database, methylation changes at sites cg14736172 (R = -0.098) and cg10164987 (R = 0.107) in LUAD showed an inverse correlation with MDH1 expression levels ( Supplementary Figure S3E ). Additionally, changes in methylation at sites cg02914501 (R = -0.093), cg17972846 (R = -0.136), cg24055349 (R = -0.188), cg09303977 (R = -0.124), cg19484848 (R = 0.094), and cg14458626 (R = 0.092) in SKCM were found to correlate with MDH1 expression ( Supplementary Figure S3F ).

A pie chart representation of CNVs revealed the widespread occurrence of MDH1 CNVs across all 33 cancer categories in the TCGA dataset, as shown in Supplementary Figure S3G . Moreover, an increased proportion of tumors exhibited copy number gains for the MDH1 gene, while copy number deletions were observed in only a small fraction of the samples ( Supplementary Figure S3H ). A higher incidence of MDH1 CNVs was particularly noted in BRCA, LUSC, BLCA, ESCA, LUAD, and CESC ( Supplementary Figure S3I ).

Distinct alterations in gene splicing patterns within tumor cells can drive the progression of malignancy. Detecting alternative splicing (AS) events may therefore enhance prognostic and diagnostic assessments for patients, as suggested by previous studies (47). A pan-cancer analysis showing the percent spliced-in (PSI) values for the MDH1_alt_3prime_143356 event is presented in Supplementary Figure S5A . Compared to normal tissue samples, elevated PSI levels were observed in kidney renal KIRP, LIHC, LUSC, PCPG, and SKCM, whereas a contrasting pattern was noted in KICH and THYM. Supplementary Figure S5B summarizes the statistical differences in PSI between tumor and adjacent normal tissues, along with their prognostic significance. Kaplan-Meier survival curves revealed a relationship between PSI of MDH1 AS events and survival outcomes in select cancers, as shown in Supplementary Figure S5C . Overall, these findings suggest that AS events in MDH1 may take a significant role in the progression of various cancers.

As shown in Supplementary Figure S6A , both TMB and TNB scores exhibited a progressive and significant increase with higher MDH1 expression, suggesting a potential association between MDH1 and enhanced immunogenicity as well as tumor heterogeneity in LUAD. Additionally, MDH1 showed strong positive associations with both synonymous (r = 0.19, p < 0.001) and non-synonymous (r = 0.19, p < 0.001) mutations in LUAD samples, as illustrated in Supplementary Figure S6C . Using the OncodriveCLUST algorithm, PPP3CA was identified as a shared driver gene in both the MDH1-L and MDH1-M cohorts, while no driver gene was detected in the MDH1-H group, as presented in Supplementary Figure S6B . A bar plot demonstrated a decrease in PPP3CA mutation frequency with increasing MDH1 expression, as shown in Supplementary Figure S6D . Significantly mutated genes were identified in the MDH1-L+M groups, visualized in the forest plot ( Supplementary Figure S6E ). In the MDH1-L+M group, the most frequently mutated genes were ACACB, PTPRC, and ZNF676. A ternary plot was used to depict the distribution of all non-silent mutations across the three MDH1 groups, including the 10 most common non-silent mutations (TP53, AMT, PRKDC, FANCM, BRCA2, POLE, HFM1, MSH4, ATR, and BRIP1; Supplementary Figure S7A ). It was found that the top 10 mutated genes were more likely to occur in the MDH1-L group ( Supplementary Figure S6F ). Additionally, a higher frequency of co-occurring and mutually exclusive mutations was observed with decreasing MDH1 levels, indicating that MDH1 is associated with increased somatic mutation activity in LUAD ( Supplementary Figure S6G ). DNA damage response (DDR) genes are frequently mutated in LUAD, and these alterations are associated with favorable immune checkpoint inhibitors (ICIs) outcomes in LUAD patients, as reported in the literature (48). The mutation rates of eight DDR pathways (CPF, HRR, NER, FA, MMR, BER, NHEJ, and TLS) across different MDH1 groups are summarized in Supplementary Figure S6H .

The integrity of the cancer genome critically depends on the efficacy of DNA repair processes, including DNA methyltransferase (DNMT), MMR, and HRR. Moreover, the heatmap revealed a positive correlation between MDH1 expression and DNMT, MMR, and HRR-related genes in the majority of tumors ( Supplementary Figures S6I , S7B ).

To explore the functions of MDH1 in the immune microenvironment, we analyzed TMB and MSI. As shown in Supplementary Figure S8A , a positive correlation was observed between MDH1 expression levels and TMB across various cancer types, including BLCA, BRCA, COAD, DLBC, HNSC, LUAD, READ, SKCM, STAD, and UCEC. Additionally, MDH1 expression was positively associated with MSI in COAD, STAD, and UCEC, whereas it showed an adverse relationship with MSI in LUAD and PRAD, as illustrated in Supplementary Figure S8C .

Next, we investigated the correlation between MDH1 expression and various immune cells. We revealed that MDH1 expression was positively association with both M2 and M1 macrophages. In contrast, MDH1 expression was negatively associated with regulatory T cells, memory B cells, and plasma cells ( Supplementary Figure S8B ).

Data from Spatial DB, obtained through spatial transcriptomic analysis, unraveled significant overlap in the expression profiles of MDH1 and the macrophage marker CD68, suggesting potential co-localization of these genes ( Supplementary Figure S8D ). Furthermore, single-cell RNA sequencing data from TISCH and HPA confirmed the presence of MDH1 transcripts in both macrophages and malignant tumor cells across most of the cancer types studied ( Supplementary Figures S8E, F ).

To elucidate the immunological divergence under MDH1-based stratification, we next examined how intrinsic immune phenotypes aligned with MDH1 abundance. Tumors falling into the MDH1-low set were strongly skewed toward a C3 (inflammation-centric) milieu, whereas those with elevated MDH1 expression were preferentially classified as C1 (wound-healing) or C2 (IFN-γ–polarized) immune subtypes ( Supplementary Figure S9A ).

To further investigate the function of MDH1 in LUAD and validate the previously obtained analytical results, we utilized scRNA-seq data from the GSE200972 repository, which includes gene expression profiles of 93,201 cells derived from 14 LUAD samples. We have implemented rigorous quality control procedures ( Supplementary Figure S10A ). Dimensionality reduction of the gene expression data was performed using principal component analysis (PCA), revealing 33 distinct cell clusters. The cellular identity of each cluster was then annotated based on immunological markers described in the literature (49, 50) ( Figure 5A ). As expected, MDH1 was predominantly expressed in macrophages and epithelial cells ( Figures 5B, D ). Moreover, the expression profile of MDH1 showed a significant correlation with the macrophage-specific marker CD68 ( Figure 5C ).

To further explore the relationship between MDH1 and the dynamic evolution of macrophages in LUAD, we purified monocyte- and macrophage-enriched subpopulations and assigned them to M1, M0, or M2 phenotypes using marker genes previously established in the literature (51, 52) ( Figure 5E ). Using the Monocle2 toolkit, we ordered single cells along a pseudo-temporal developmental trajectory, which revealed a distinct and uniform maturation pathway for macrophages ( Figure 5F ). Notably, the expression patterns of MDH1 and CD68 closely matched the macrophage’s dynamic evolutionary trajectory ( Figure 5G ).

Following this, we categorized macrophages into high-MDH1 and low-MDH1 groups based on their MDH1 expression levels. A differential gene expression analysis was then conducted, with a log2 fold change > 0.5 and a p-value < 0.05 as the cutoff criteria ( Figures 5H, I ). In the biological process (BP) category, the differentially expressed genes (DEGs) were enriched in the positive regulation of cytokine production, cell adhesion, and leukocyte-related functions. KEGG pathway analysis revealed that these DEGs were predominantly involved in cytokine-cytokine receptor interactions, cell cycle regulation, and T cell receptor signaling, among other pathways (p< 0.05) ( Figures 5J, K ).

Given the potential role of MDH1 in epithelial cells, we also stratified epithelial cells into high-MDH1 and low-MDH1 groups, followed by differential gene expression analysis using the same criteria ( Figures 5L, M ). The DEGs in this category were associated with cellular migration, apoptosis, and oncogenic processes. KEGG pathway analysis indicated that these DEGs were primarily involved in oxidative stress responses and immune reactions ( Figures 5O, P ).

Initially, the influence of MDH1 on immune cell infiltration was assessed. The results revealed a positive correlation between MDH1 expression and the infiltration of myeloid-derived suppressor cells (MDSCs) across various tumor types, as illustrated in Supplementary Figure S11A . Next, cytotoxic T lymphocytes (CTLs), which are an essential functional subset of CD8+ T cells, were examined. Impaired CTL function is a critical factor in tumor immune evasion and resistance to immunotherapy, as highlighted in the literature (53). As depicted in Figure 6A , MDH1 expression showed an inverse correlation with CTL activity in UCEC, metastatic melanoma, neuroblastoma, leukemia, and TNBC.

Furthermore, the correlation between MDH1 and immunomodulatory genes was assessed. MDH1 showed a significant positive association with most chemokines and chemokine receptors in UCEC, LUAD, STAD, TGCT, KIRP, KIRC, PCPG, BLCA, COAD, BRCA, PAAD, READ, KICH, and UVM ( Figures 6B, C ). Additionally, we found that MDH1 expression was positively correlated with immune checkpoint genes in UCEC, STAD, PCPG, LUAD, LIHC, LGG, LAML, KIRC, BRCA, and BLCA.

Next, we explored the predictive value of MDH1 for immunotherapy response in the TCGA-LUAD cohort. As expected, the MDH1-high group exhibited higher TIS scores and IFNG expression compared to the MDH1-low group ( Figures 6D, E ). Moreover, we evaluated the sensitivity to anti-PD1 and anti-CTLA4 immunotherapy utilizing the SubMap algorithm. As anticipated, higher MDH1 expression in LUAD patients increases the likelihood of benefiting from immunotherapy ( Figure 6F ).

To validate our hypothesis, we analyzed 16 immunotherapy cohorts from TIGER. As shown in Figures 6G-I , elevated MDH1 expression was associated with better sensitivity to immunotherapy in the majority of the cohorts.

A total of seven scRNA-seq datasets were utilized to explore the association between MDH1 expression levels and responses to immunotherapy, with the highest representation from UCEC samples (n = 17), LUAD (n = 11), and BLCA (n = 10) ( Supplementary Table S5 ). Firstly, we have implemented rigorous quality control procedures ( Supplementary Figure S10B ). To identify unique cellular populations, we applied uniform manifold approximation and projection (UMAP) for dimensionality reduction and clustering of cells from the seven scRNA-seq cohorts, which included patients treated with ICIs across seven different cancer types ( Figures 7A, B , Supplementary Figure S12A ). Figure 7C shows the UMAP distribution of immunotherapy responsiveness. Among these cells, LUAD (GSE207422) had the largest proportion of macrophages, while GBM (GSE235672) had the largest proportion of epithelial cells ( Supplementary Figure S13A ).

Consistent with our previous hypothesis, the group that was sensitive to immunotherapy exhibited elevated MDH1 expression levels ( Figure 7D ). However, the distribution of cellular composition between the two groups was found to be nearly identical ( Supplementary Figure S13B ). We then conducted a comparative analysis of MDH1 expression levels across various cell types in the two cohorts. Interestingly, we observed a notable increase in MDH1 expression in the immunotherapy-sensitive set across nearly all cell types, which was an unexpected finding ( Figure 7E ). Furthermore, as previously noted, MDH1 expression remained strongly correlated with macrophages and epithelial cells ( Supplementary Figure S13C ).

Figure 7F shows the stratification of macrophages into distinct clusters, with clear separation observed between patients who responded to immunotherapy (R) and those who did not (NR). This highlights the presence of macrophage subset-specific features that may influence responses to immunotherapeutic interventions. Trajectory analysis revealed that the differentiation trajectories of macrophages in the high MDH1 group were similar to those in the R group ( Figure 7G, H ).

Next, we extracted the macrophage subgroup ( Supplementary Figure S13D ) and performed differential gene analysis between responders and non-responders, as well as between the MDH1-high and MDH1-low groups. The DEGs in the BP category were enriched in the positive regulation of cytokine production, cell adhesion, and leukocyte-related functions ( Figure 7I ). KEGG pathway analysis indicated that these DEGs were primarily enriched in the T cell receptor signaling pathway, NOD-like receptor signaling pathway, and antigen processing and presentation ( Figure 7J ). The KEGG and GO analyses of the DEGs between the R and NR groups showed trends that were largely consistent with the previous findings ( Figures 7K, L ).

Furthermore, we extracted the epithelial cell subgroup ( Supplementary Figure S13E ) and conducted differential gene analysis between the R and NR groups, as well as between the MDH1-high and MDH1-low groups. The DEGs in the BP category were involved in cell migration and epidermis development ( Supplementary Figure S13F ). KEGG analysis revealed that these DEGs were mainly involved in apoptosis and carcinogenic pathways ( Supplementary Figure S13G ). KEGG and GO analyses of the DEGs between the R and NR groups exhibited trends that were largely consistent with the previous outcomes ( Supplementary Figures S13H, I ).

In summary, the data strongly suggest a temporal and spatial concordance between cells with high MDH1 expression and those that respond to immunotherapy.

Across 1,837 compounds curated from PRISM, CTRP and GDSC, high MDH1 expression consistently predicted enhanced sensitivity to BI-2536 (Spearman r < –0.30, p < 0.05); oxaliplatin and leflunomide emerged as the strongest DNA-replication inhibitors in this panel ( Figures 8A–C ). To examine the underlying binding mode, we docked BI-2536 into the AlphaFold3-derived MDH1 structure, revealing ΔG = –9.02 kcal mol⁻¹ and key hydrogen bonds with Val87, Ser89, Val129 and Ser242 ( Figure 8D ). Subsequent 100-ns molecular-dynamics simulations at pH 7.2 confirmed complex stability (RMSD ≈ 0.1–0.25 nm), reduced RMSF in the binding pocket, persistent H-bonding and compact SASA ( Supplementary Figures 14A–E ). Thus, computational and pharmacogenomic data converge on BI-2536 as a high-affinity MDH1-directed agent.

To accelerate the clinical translation of MDH1 as a predictive biomarker, we next interrogated whether MDH1 expression levels modulate drug sensitivity by comparing the IC50 values of standard-of-care antineoplastic agents across lung-cancer cell lines stratified into MDH1-high and MDH1-low cohorts. Elevated MDH1 was associated with heightened vulnerability to cisplatin, docetaxel, paclitaxel, vinblastine and vinorelbine, whereas lower MDH1 predicted superior sensitivity to erlotinib, gefitinib and gemcitabine ( Supplementary Figure S15A ).

To gain a deeper understanding of the pan-cancer implications of MDH1, a comprehensive analysis was conducted using the CancerSEA database. In this analysis, MDH1 expression was found to be positively associated with processes such as cell cycle progression, DNA damage response, DNA repair, epithelial-mesenchymal transition (EMT), invasiveness, metastatic potential, and proliferation. In contrast, MDH1 expression showed an inverse correlation with apoptotic pathways, hypoxic conditions, and inflammatory responses in lung adenocarcinoma, as shown in Supplementary Figure S16A .

To investigate the role of MDH1 in LUAD cell lines, we first demonstrated through PCR analysis that the expression levels of MDH1 were significantly higher in A549 and PC-9 cell lines compared to normal bronchial epithelial cells ( Figure 9A ). We then established MDH1 knockdown models by transfecting specific siRNAs. Following transfection with siRNAs targeting MDH1, WB analysis was conducted to assess knockdown efficiency. The results showed a significant reduction in MDH1 mRNA expression in A549 and PC-9 cells transfected with siRNA-MDH1–1 and siRNA-MDH1-4, compared to the control group ( Figure 9B ). These siRNA-transfected A549 and PC-9 cells were then used for further experiments.

The marked reduction in cell viability in both A549 and PC-9 cells transfected with si-MDH1, compared to the control group, between 48 and 96 hours post-transfection was demonstrated by CCK8 assays ( Figure 9C ). In order to assess the effect of MDH1 knockdown on cell proliferation, colony formation assays were conducted. These assays revealed a significant reduction in the number of colonies formed by A549 and PC-9 cells after MDH1 silencing ( Figures 9E, G ).

In addition, wound healing and transwell assays were utilized to evaluate the impact of MDH1 knockdown on cell migration. These experiments showed that silencing MDH1 significantly impaired the migratory capacity of LUAD cells ( Figures 9D, F, H, I ). Recent studies have indicated a potential role for MDH1 in hypoxia and glycolysis, but its function in LUAD remains unclear (54). Functional enrichment analysis has revealed that MDH1 is positively correlated with lactate metabolism in the vast majority of tumors ( Supplementary Figure S16B ). As expected, MDH1 knockdown increased lactate accumulation in culture supernatants ( Figure 9J ). Moreover, WB was utilized to examine the expression of glycolysis-related genes, and we found that MDH1 knockdown promoted glycolytic activity in LUAD cells ( Figure 9K ).

To further investigate the spatial relationship between MDH1 and the macrophage marker CD68, we used a pan-cancer spatial transcriptomics dataset. MDH1 exhibited a spatial distribution pattern that closely matched CD68, particularly in LUAD tissues (10A, Supplementary Figures S8D and S17A ). Additionally, IF staining revealed elevated CD68 protein expression in regions with high MDH1 expression, while areas with low MDH1 expression showed reduced CD68 levels ( Figure 10B ). The coexpression of MDH1 and CD68 in LUAD tissues supported our previous findings. Based on this, we further evaluated the impact of MDH1 on M2 macrophages, observing a significant reduction in M2 macrophage infiltration in LUAD cells following MDH1 knockdown ( Figure 10C ).

To clarify whether MDH1-regulated macrophage function depends on lactate production, we co-cultured THP-1-derived macrophages with LUAD cells in which MDH1 had been knocked down or left intact. ELISA quantification of IL-10 and TNF-α in the conditioned medium revealed that MDH1 depletion significantly impaired M2 polarization ( Supplementary Figure S18A ). Importantly, pharmacologic inhibition of lactate (Fx11 (55)) failed to reverse this effect, indicating that MDH1 deficiency drives macrophages toward an M1 phenotype through a lactate-independent mechanism.

Our research revealed a positive correlation between MDH1 expression and the IC50 of BI-2536, indicating that elevated levels of MDH1 might be involved in fostering BI-2536 resistance in cancer cells. Additionally, the CCK8 assay reveal that diminishing the expression of MDH1 can enhance the susceptibility of LUAD cells to BI-2536 treatment ( Supplementary Figures S19A, B ).

To confirm the molecular interaction between MDH1 and BI-2536, we performed CETSA. The ligand-induced thermal stabilization of MDH1 in A549 cells verified their binding within the cellular context ( Supplementary Figure S19C ).