Abstract

Background

   Low-grade gliomas (LGG) are a heterogeneous category of brain tumors
   characterized by a variable clinical course, frequently associated with
   unfavorable prognosis and therapeutic challenges. Understanding the
   molecular mechanisms underlying LGG progression is crucial for
   improving prognosis and therapeutic strategies. This study integrates
   single-cell RNA sequencing and bioinformatics to explore the role of
   METCGs (mitochondrial electron transport chain genes) in LGG and
   construct a predictive model for prognosis, and through in vitro
   experiments, the feasibility of this model was validated.

Methods

   We analyzed 5,691 cells and 22,947 genes from the [34]GSE117891
   dataset. Using cell marker genes from the CellMarker 2.0 database and
   classical markers, we identified four distinct cell types:
   oligodendrocytes, T cells, astrocytes, and microglial cells. The METCGs
   profiles were calculated using various algorithms, including AUCell,
   UCell, ssGSEA, and others. Differentially expressed genes (DEGs) were
   identified and enriched for relevant pathways. Machine learning
   algorithms were employed to construct a prognostic risk model based on
   five selected METCGs. The model was validated using independent LGG
   cohorts. Biological pathway analyses, immune infiltration profiles, and
   potential drug targets were also explored. To validate the reliability
   of this model through experiments, functional experiments, including
   Blue native Page (BN-Page), western blotting, immunofluorescence, and
   cell viability assays, were conducted to validate SDHB expression and
   its role in LGG progression.

Results

   Astrocytes exhibited the highest METCG scores, indicating their central
   role in mitochondrial energy regulation. The prognostic model,
   constructed using the StepCox[forward] + plsRcox approach, included
   five genes: SDHB, SDHC, SLC25A27, UQCRB, and NDUFA13. The model
   demonstrated high prognostic accuracy with an average C-index of 0.67
   and successfully stratified LGG patients into low- and high-risk
   groups. High-risk patients had worse survival outcomes, with
   significant differences observed in KEGG pathways, immune infiltration,
   and metabolic processes. The low-risk group exhibited higher immune
   cell infiltration, including follicular helper T and monocyte cells.
   AZD1208_1449 was identified as a potential drug targeting high-risk
   patients. Additionally, SDHB expression was significantly higher in LGG
   cells, and knockdown of SDHB inhibited cell proliferation and invasion,
   supporting its role in tumor progression.

Conclusion

   This study provides a comprehensive analysis of METCGs in LGG and
   develops a robust prognostic model for patient stratification. SDHB, a
   key subunit of Complex II, plays a crucial role in mitochondrial
   function and tumor progression. Our findings suggest that he high
   expression of SDHB in LGG contributes to maintaining elevated SDH and
   Complex II activity, ensuring the structural and functional integrity
   of mitochondrial ETC complexes. This supports the high ROS production
   and MMP required for the rapid growth of LGG, thereby promoting its
   proliferation and invasion. Thus, targeting SDHB and its associated
   pathways could offer new therapeutic avenues for LGG treatment.

   Keywords: Low-grade glioma, METCGs, SDHB, Prognostic model, Single-cell
   RNA sequencing

Introduction

   Low-grade gliomas (LGGs), originating from glial cells, represent a
   relatively rare subset of primary intracranial tumors, accounting for
   approximately 15% of central nervous system (CNS) neoplasms [[35]1].
   These tumors encompass diffuse oligodendrogliomas, astrocytomas, and
   oligoastrocytomas, as classified by the 2021 WHO Classification of
   Tumors of the Central Nervous System (5th Edition) [[36]2–[37]4]. While
   LGGs exhibit slower growth and less aggressive invasion compared to
   their high-grade counterparts, they carry a significant risk of
   malignant transformation over time, progressing to higher-grade gliomas
   with increased potential for neurological deterioration and mortality.
   The indolent nature of LGGs often results in asymptomatic early stages,
   delaying clinical detection until the disease has advanced to a more
   aggressive state [[38]5]. Although surgical resection, chemotherapy,
   and radiotherapy remain the mainstay of treatment for LGGs, current
   strategies primarily focus on managing the disease as a chronic
   condition and delaying its progression rather than achieving a
   definitive cure [[39]6]. The prognosis for patients with LGG is
   relatively unfavorable, with median survival typically ranging from 3
   to 15 years [[40]7]. With the advancement of research and the rapid
   progress in molecular biology, it has become evident that distinct
   molecular mechanisms underlie different types of LGGs. These molecular
   alterations not only contribute to more accurate diagnosis and
   classification but also serve as valuable prognostic indicators and
   hold potential as novel therapeutic targets in the future. Accordingly,
   this study aims to comprehensively investigate the mitochondrial
   mechanisms contributing to the pathogenesis of LGG, providing deeper
   insights into its molecular underpinnings.

   The mitochondrial electron transport chain (ETC) consists of five large
   protein complexes [[41]8]. These complexes function as electron donors
   and acceptors, playing a crucial role in oxidative phosphorylation
   (OXPHOS). During this process, ADP is phosphorylated to ATP utilizing
   the electrochemical gradient, and mitochondrial ROS (mtROS) is
   generated as a by-product [[42]9]. OXPHOS is also essential for
   maintaining mitochondrial membrane potential (MMP) and serves as an
   initiator of various metabolic pathways that are important for tumor
   progression and invasion [[43]10]. Recent studies have revealed the
   mutation or dysregulation of subunits, structure alteration and
   dysfunction of ETC protein complexes is correlated with the activation,
   progression and radioresistance of glioma [[44]11–[45]15].For example,
   NADH dehydrogenase [ubiquinone] 1 alpha subcomplex subunit 4-like
   2(NDUFA4L2) is a subunit of Complex I and significantly upregulated in
   GBM, knockdown of NDUFA4L2 GBM proliferation by protective mitophagy
   [[46]14]. Fractionated radiation enhances Complex IV activity through a
   switch in the cytochrome c oxidase subunit 4 (COX4), reducing ROS
   production and apoptosis while lowering labile iron and lipid
   peroxidation levels, ultimately promoting radioresistance [[47]13].
   Mutations in complexes III and IV impair mitochondrial respiratory
   chain activity, inhibiting glioblastoma proliferation [[48]12]. ATP
   synthase subunits alpha and beta (ATP5A1 and ATP5B), part of Complex V,
   are significantly upregulated in glioblastoma and correlate with
   microvascular proliferation [[49]16]. However, the role of Complex II
   and its subunits in the development, proliferation, and invasion of
   gliomas, particularly low-grade gliomas (LGG), remains elucidated. In
   this study, we first found Complex II and its subunit succinate
   dehydrogenase complex iron sulfur subunit B (SDHB) was upregulated in
   LGG.

   Mitochondrial complex II, also known as succinate dehydrogenase (SDH),
   consists of four subunits encoded by the SDHx gene family: SDHA, SDHB,
   SDHC, and SDHD [[50]17]. Among these, SDHA and SDHB serve as catalytic
   subunits, whereas SDHC and SDHD function as anchoring subunits,
   stabilizing the complex within the mitochondrial inner membrane
   [[51]17, [52]18]. Complex II plays a pivotal role in cellular energy
   metabolism by catalyzing the oxidation of succinate to fumarate in the
   tricarboxylic acid (TCA) cycle, a key step in cellular respiration and
   ATP production [[53]18]. Dysfunction of complex II disrupts normal
   mitochondrial metabolism, leading to the accumulation of succinate—a
   metabolite that influences cellular signaling, gene expression, and
   oxidative stress [[54]19]. Among the causes of complex II dysfunction,
   mutations and dysregulation of SDHB subunit is the most prevalent.
   These alterations impair complex II assembly, disrupt mitochondrial
   oxidative phosphorylation, and rewire metabolic pathways. Consequently,
   they elevate mitochondrial reactive oxygen species (mtROS) production,
   enhance genetic instability, and contribute to tumorigenesis. SDHB
   mutation and dysregulation have been implicated in the initiation,
   progression, and invasion of several tumor types, including
   pheochromocytomas, paragangliomas (PPGLs), and gastrointestinal stromal
   tumors (GISTs) [[55]20–[56]22]. For instance, in PPGLs, SDHB mutations
   are associated with a higher risk of metastasis and shorter overall
   survival [[57]23]. Similarly, the presence of SDHB mutations in GISTs
   and PPGLs correlates with increased tumor aggressiveness and poor
   prognosis [[58]22]. Despite the established role of complex II and SDHB
   in these tumors, their contributions to glioma, particularly low-grade
   gliomas (LGG), remain poorly understood and warrant further
   investigation.

   In this study, we found SDHB and complex II was upregulated in LGG and
   correlated with the poor prognosis of patients with LGG. Moreover, we
   showed that the elevated expression of SDHB in LGG sustains high SDH
   and Complex II activity, preserving the structural and functional
   integrity of mitochondrial ETC complexes. This, in turn, facilitates
   increased ROS production and MMP, which are essential for the rapid
   growth of LGG, thereby driving its proliferation and invasion.

Materials and methods

Data collection

   W Single-cell RNA sequencing (scRNA-seq) and bulk RNA-seq data (counts
   and TPM), along with baseline clinical characteristics and prognostic
   outcomes, were retrospectively collected from publicly available
   databases for patients diagnosed with low-grade gliomas (LGG). The
   scRNA-seq dataset [59]GSE117891 included 73 tissue samples from 14 LGG
   patients. For model construction and validation, data from 1,490 LGG
   patients were sourced from TCGA (n = 498), CGGA325 (n = 172), CGGA693
   (n = 420), CGGA301 (n = 159), [60]GSE16011 (n = 98), and M-TAB-3982
   (n = 142).

Analysis of single-cell RNA sequencing data

   The scRNA-seq data were processed using the Seurat R package [[61]24].
   Quality control involved excluding genes expressed in fewer than three
   cells, cells with fewer than 300 detected genes, and cells with > 15%
   mitochondrial gene content or > 0.1% hemoglobin gene content.
   Mitochondrial genes were removed, and doublets were excluded.
   Normalization was performed using the NormalizeData function, followed
   by centering and scaling via the ScaleData function.

   Highly variable genes (n = 2000) were identified for principal
   component analysis (PCA). Sample integration was achieved using the
   Harmony R package, and the first 20 principal components were utilized
   for clustering analysis. Cell clusters were visualized using UMAP, with
   clustering performed using the FindClusters function
   (resolution = 0.5). Cell type annotation relied on marker genes from
   the CellMarker 2.0 database and prior literature [[62]25].
   Differentially expressed genes (DEGs) were identified using the
   FindAllMarkers function, with significance thresholds set at
   p-adjusted < 0.05 and absolute log2 fold change (log2FC) > 0.25.

Enrichment analysis of METCGs in cell clusters

   Enrichment scores for Mitochondrial Electron Transport Chain genes
   (METCGs) in each cell cluster were calculated using the irGSEA.score
   function from the irGSEA R package [[63]26]. The wiki pathway METCGs,
   “WP_ELECTRON_TRANSPORT_CHAIN_OXPHOS_SYSTEM_IN_MITOCHONDRIA.v2024.1 “,
   were retrieved from the MSigDB database
   ([64]https://www.gsea-msigdb.org/gsea/msigdb/human/geneset/WP_ELECTRON_
   TRANSPORT_CHAIN_OXPHOS_SYSTEM_IN_MITOCHONDRIA.html). Enrichment
   algorithms (AUCell, UCell, ssGSEA, Singscore, JASMINE, and viper) were
   applied to normalized RNA-seq data. The GSVA R package [[65]27] was
   employed to assess pathway activity in astrocyte clusters based on
   METCG-associated DEGs.

Identification of prognosis-associated genes

   Univariate Cox regression analysis within the TCGA-LGG cohort
   identified prognostic genes. Upregulated METCGs in the astrocyte
   cluster were extracted using the FindAllMarkers function. Prognostic
   METCGs were determined through intersection analysis.

Model construction and validation

   To develop the optimal prognostic model, 101 machine learning algorithm
   combinations were evaluated using the Mime package
   (https://github.com/l-magnificence/Mime) [[66]28]. These combinations
   included both single algorithms (e.g., StepCox[forward] and partial
   least squares regression for Cox [plsRcox]) and hybrid models that
   combined two distinct algorithms for feature selection and modeling
   (e.g., StepCox[forward] + plsRcox), as previously described [[67]29].
   First, univariate Cox regression (p < 0.05) identified prognostic genes
   in the the training set. Then, 101 machine learning algorithm
   combinations were trained using 10-fold cross-validation and tested in
   the other independent LGG cohorts. Final, the model with the highest
   average C-index was selected, identifying a five-gene signature (SDHB,
   SDHC, SLC25A27, UQCRB, NDUFA13). We compared our optimal model against
   95 published signatures (33 for glioma/22 for LGG) derived from Liu’s
   study [[68]28] and clinical characteristics using C-index. Kaplan-Meier
   curve analysis was used to evaluate the overall survival difference
   between high-risk and low-risk groups based on median cutoff of each of
   cohorts. Model-related plot was generated via the Mime package.

Biological pattern evaluation

   Differential gene expression analysis was performed using the DESeq2
   algorithm [[69]30] to identify DEGs between high-risk and low-risk
   groups in the TCGA-LGG cohort, with thresholds of log2 fold change
   (log2FC) > 1 and adjusted p-value < 0.05.

   Pathway enrichment analysis utilized KEGG, HALLMARK, and GO gene sets
   from MSigDB ([70]https://www.gsea-msigdb.org/gsea/msigdb/index) via
   Gene Set Enrichment Analysis (GSEA), with significance set at adjusted
   p-value < 0.05 and absolute normalized enrichment score (NES) > 1. GO
   analysis focused on biological processes (BP), cellular components
   (CC), and molecular functions (MF). Gene mutation analysis identified
   somatic and chromosomal region-associated mutations to uncover
   potential therapeutic targets.

   Immune cell infiltration was assessed using the Cibersort algorithm and
   IOBR R package [[71]31], analyzing 22 immune cell types and comparing
   immune activation, suppression factors, and MHC-related molecule
   expression between risk groups. Results were visualized with the BEST
   tool ([72]https://rookieutopia.hiplot.com.cn/app_direct/BEST/).

Identification of potential therapeutic agents based on risk model
stratification

   Drug sensitivity was evaluated using IC50 values of 198 compounds
   across five LGG cohorts, predicted via the OncoPredict R package
   [[73]32] and GDSC2 database [[74]33]. IC50 comparisons between risk
   groups were complemented by ROC curve analysis to assess predictive
   accuracy. Potential therapeutic agents were identified through
   intersection analysis of drugs with significant distribution
   differences.

Molecular Docking analysis

   PDB files were obtained from the Protein Data Bank (RCSB PDB,
   [75]https://www.rcsb.org/structure/). Compound structures were
   downloaded in sdf format from the PubChem database
   ([76]https://pubchem.ncbi.nlm.nih.gov/). The cavity-detection guided
   blind docking algorithm [[77]34] (CB-Dock2) website
   ([78]https://cadd.labshare.cn/cb-dock2/index.php) was used to perform
   molecular docking analysis between SDHB and AZD1208.

Nomogram scoring system

   Significant factors from univariate Cox regression (p < 0.05) were
   incorporated into a multivariate model to construct a nomogram.
   Predictive performance was evaluated using AUC, time-dependent C-index,
   and calibration plots comparing predicted versus actual overall
   survival outcomes.

Cell culture

   HEB and LGG cell lines (HS683 and SHG-44) were cultured in DMEM (Gibco,
   C11330500BT) supplemented with 10% fetal bovine serum (FBS, Excell Bio,
   FSP500) and 1% penicillin-streptomycin (Gibco) at 37 °C in a humidified
   atmosphere with 5% CO₂. Upon reaching 70–80% confluence, cells were
   washed with DPBS (calcium- and magnesium-free) and detached with
   500–1000 µL of trypsin for 1–2 min at 37 °C. Trypsinization was halted
   with DMEM containing 10% FBS, followed by centrifugation at 1000 rpm
   for 5 min. The cell pellet was resuspended in fresh medium, transferred
   to new culture dishes, and incubated for further growth.

SiRNA transfection

   Cells were seeded in 6-well plates and grown to 40–50% confluence.
   siRNA (Sangon Biotech, China) was prepared in DEPC-treated water (20
   µM) and complexed with RNATransMate (Sangon Biotech, E607402) in serum-
   and antibiotic-free DMEM. After a 10-minute incubation at room
   temperature, the complexes were added to cells and distributed by
   gentle rocking. Following 8 h of incubation at 37 °C with 5% CO₂, the
   medium was replaced with complete medium. Transfected cells were
   cultured for 24–48 h before further analysis. siRNA sequences were as
   follows: siRNA#2: F-UCUUGAUCUCUGCAAUAGC, R-GCUAUUGCAGAGAUCAAGA;
   siRNA#3: F-UCAAAUCGGGAACAAGAUC, R-GAUCUUGUUCCCGAUUUGA.

Cell migration and proliferation assays

   For migration assays, cells were trypsinized, centrifuged at 1000 rpm
   for 5 min, and resuspended in serum-free medium. A total of 5 × 10⁴
   cells in 300 µL serum-free medium were seeded into the upper chamber of
   Transwell inserts (Corning, 3422), with 500 µL DMEM containing 20% FBS
   in the lower chamber. After 24–48 h, migrated cells were fixed with
   paraformaldehyde, stained with 0.1% crystal violet (Abiowell,
   AWI0364a), and counted in five random microscopic fields.

   For proliferation assays, 5 × 10³ cells were seeded into 96-well plates
   and treated with CCK-8 reagent (Beyotime, C0046) according to the
   manufacturer’s protocol. Absorbance was measured at 450 nm using a
   microplate reader 2 h post-treatment. Proliferation was assessed over
   time by repeating the CCK-8 assay at 24-hour intervals.

SDS-PAGE and Western blotting

   Cells were lysed with 4× SDS sample buffer containing 1 mM phosphatase
   inhibitor (Absin, abs9162) and 1 mM PMSF (Beyotime, ST507-10 ml).
   Protein lysates were separated by SDS-PAGE at 150–200 V for 60 min and
   transferred to methanol-activated PVDF membranes (Vazyme, E801-01).
   Membranes were blocked with 5% non-fat milk in TBST for 1 h and
   incubated overnight at 4 °C with primary antibodies diluted in 1% BSA.
   After washing with TBST, membranes were incubated with HRP-conjugated
   secondary antibodies (Sangon, D123504) for 3 h at room temperature.
   Protein bands were visualized using ECL substrate (Abbkine, BMU102-CN)
   and imaged using a chemiluminescence detection system. Primary
   antibodies included anti-SDHB (Proteintech, 10620-1-AP), SDHA (Promab,
   32308), anti-GAPDH (Proteintech, 60004-1-Ig).

Flow cytometry

   Cells were incubated with 5 µM MitoSox Red (Thermo, [79]M36008) at
   37 °C for 30 min, protected from light. After washing twice with PBS,
   cells were trypsinized, resuspended in PBS, and analyzed using a flow
   cytometer. Data were processed with FlowJo software to quantify
   mitochondrial ROS levels. Controls included untreated cells and cells
   treated with Tempo (KKLMED, KM17362).

Immunohistochemistry

   Tumor and adjacent tissues were formalin-fixed, paraffin-embedded, and
   sectioned (4–5 μm). Sections were deparaffinized, rehydrated, and
   subjected to antigen retrieval in citrate buffer. After blocking with
   5% goat serum, sections were incubated overnight at 4 °C with primary
   antibodies. Following washing, biotinylated secondary antibodies were
   applied, and sections were developed using DAB substrate (Sigma,
   D4418). Hematoxylin counterstaining was performed, and protein
   expression was assessed microscopically.

Immunofluorescence staining

   HEB and LGG cells were incubated with 50 nM MitoTracker Red
   (ThermoFisher, M7512) at 37 °C for 30 min. After fixation with 4%
   paraformaldehyde and permeabilization with 0.1% Triton X-100, cells
   were blocked with 1% BSA in PBST and incubated overnight at 4 °C with
   primary antibodies. Following washing, secondary antibodies were
   applied for 3–4 h at room temperature in the dark. Nuclei were
   counterstained with Hoechst 33,258, and slides were mounted for
   imaging.

ATP level detection

   ATP levels were measured using the Enhanced ATP Detection Kit
   (Beyotime, S0027). Lysates were prepared, centrifuged, and supernatants
   were mixed with ATP detection reagent. Luminescence was measured using
   a luminometer, and relative light units (RLU) were recorded to quantify
   ATP concentrations.

SDH activity assay

   SDH activity was assessed using the Succinate Dehydrogenase Activity
   Assay Kit (Boxbio, AKAC011M). Cell lysates were prepared, and enzyme
   activity was measured by recording absorbance changes at 600 nm over
   time. Final SDH activity was calculated after adjusting for background
   values.

Mitochondrial respiratory chain complex II activity assay

   Complex II activity was measured using the Mitochondrial Respiratory
   Chain Complex II Activity Detection Kit (Boxbio, AKOP006M).
   Mitochondria were isolated, and enzyme activity was determined by
   monitoring absorbance changes at 605 nm. Results were adjusted for
   background to calculate activity.

Blue-native PAGE

   Cells were harvested from 10 cm dishes, trypsinized, and centrifuged at
   room temperature for 5 min. The pellet was resuspended in 1 mL cold
   Digitonin buffer (1% Digitonin, 20 mM Tris-HCl pH 7.4, 0.1 mM EDTA, 50
   mM NaCl, 10% glycerol, 1 mM PMSF, 1 mM PI) and incubated on ice for
   30 min with gentle pipetting (10 repetitions). The lysate was
   centrifuged at maximum speed for 15 min at 4 °C, and the supernatant
   was collected, discarding the insoluble pellet. Blue-Native PAGE was
   performed at 600 V with the apparatus connected to a cooling pump to
   maintain low temperatures. Once the samples entered the separating gel,
   the buffer was replaced with 1× Cathode buffer (without Coomassie
   G-250), and electrophoresis continued for 3 h. Proteins were
   transferred to a PVDF membrane pre-activated with methanol for 1 min
   and rinsed in PBS. The transfer was carried out at 300 mA for 3 h using
   transfer buffer. The membrane was blocked with 5% non-fat milk in TBST
   for 60 min and incubated overnight at 4 °C with primary antibodies,
   including NDUFS1 (Sangon, D122747-0025), UQCRC1 (Sangon, D123504-0025),
   ATP5A1 (Proteintech, 14676-1-AP), MTCO2 (Proteintech, 55070-1-AP), and
   SDHB (Proteintech, 10620-1-AP). The following day, the membrane was
   incubated with the corresponding secondary antibody (Sangon, D123504),
   and protein bands were detected using chemiluminescence.

Results

Cell type annotation and METCG score calculation

   The workflow of this study is outlined in Fig. [80]1A. From the
   [81]GSE117891 dataset, we analyzed 5,691 cells and 22,947 genes. Using
   common brain tissue cell marker genes from the CellMarker 2.0 database
   and classical markers, we annotated four distinct cell types:
   oligodendrocytes (MBP, MOP, PLP1), T cells (CD3D, CD3E, CD8A),
   astrocytes (GFAP, AQP4, SOX9), and microglial cells (TMEM119, CX3CR1,
   P2RY12). The distribution of these cells was as follows:
   oligodendrocytes (858 cells, 15.1%), T cells (405 cells, 7.1%),
   astrocytes (2,876 cells, 50.5%), and microglial cells (1,552 cells,
   27.3%). The cell type distribution was visualized using UMAP and
   DotPlot analyses (Figs. [82]1B-C).

Fig. 1.

   [83]Fig. 1
   [84]Open in a new tab

   Analysis of METCGs in Glioma Single-Cell RNA Sequencing and Their
   Biological Significance.(A) Study flowchart. LGG = Low-grade glioma.
   (B) UMAP plot illustrating the distinct clustering patterns of four
   predominant cell lineages. (C) Dot plot depicting the expression levels
   of marker genes across the four major cell lineages. (D) Distribution
   of METCG activity scores across different clusters, calculated using
   six distinct algorithms. (E) Results of Gene Ontology (GO) analysis,
   with the horizontal axis representing -Log10 (adjusted p-value). (F)
   Bar plot showing KEGG pathway enrichment and the top five
   cluster-related differentially expressed genes

   To examine the METCG profiles of each cell type, glycosylation scores
   were calculated based on the expression of genes involved in the
   “WP_ELECTRON_TRANSPORT_CHAIN_OXPHOS_SYSTEM_IN_MITOCHONDRIA” pathway. By
   employing multiple algorithms (AUCell, UCell, ssGSEA, Singscore,
   JASMINE, and viper), we found that astrocytes exhibited the highest
   METCG scores compared to other cell types (Fig. [85]1D; Table 1, all
   adjust p value < 0.001). These findings suggest that METCGs play a
   critical role in astrocyte function. Furthermore, KEGG pathway
   enrichment analysis (Fig. [86]1E) revealed that these DEGs were
   significantly associated with nucleocytoplasmic transport and
   neurodegenerative disease pathways, emphasizing the role of astrocytes
   in neurodegenerative processes.

Identification of prognosis-associated genes

   To construct an effective prognostic model, we evaluated 101
   combinations of machine learning algorithms using the Mime package
   ([87]https://github.com/l-magnificence/Mime). These combinations
   included single algorithms (e.g., StepCox[forward] and partial least
   squares regression for Cox [plsRcox]) and hybrid models combining two
   distinct algorithms for feature selection and modeling (e.g.,
   StepCox[forward] + plsRcox) [[88]29]. The TCGA cohort served as the
   training set, while additional independent cohorts were used for
   validation. For each cohort, the C-index was computed, and the model
   with the highest average C-index was selected. Ultimately, the
   StepCox[forward] + plsRcox model emerged as the optimal prognostic risk
   score model, identifying five key genes: SDHB, SDHC, SLC25A27, UQCRB,
   and NDUFA13.

   Kaplan-Meier survival analysis, univariate and multivariate Cox
   regression analyses, time-dependent receiver operating characteristic
   (ROC) curves, and the C-index were employed to evaluate the model’s
   prognostic accuracy. Comparative analysis with previously published
   glioma prognostic models across six independent LGG cohorts confirmed
   the model’s robustness and generalizability, as evidenced by superior
   hazard ratios (HRs) and C-index values.

Construction of the prognostic risk model

   The StepCox[forward] + plsRcox model demonstrated strong predictive
   power and generalization across six cohorts, with an average C-index of
   0.67 (Figure S1, Fig. [89]2B). Meta-analysis confirmed the model’s
   significance as a prognostic factor (all p < 0.05, Fig. [90]2 C).
   Patients were categorized into low- and high-risk groups based on
   median risk scores. Low-risk patients exhibited significantly better
   survival outcomes compared to high-risk patients (all p < 0.05, Fig.
   [91]2D). ROC analysis demonstrated the model’s predictive performance,
   achieving AUC values of 0.77 at one year, 0.74 at three years, and 0.72
   at five years (Fig. [92]2E).

Fig. 2.

   [93]Fig. 2
   [94]Open in a new tab

   Prognostic Gene Selection and Model Construction.(A) Venn diagram
   demonstrating the intersection of astrocyte subtype-specific
   METCG-related differentially expressed genes, the METCG set, and common
   genes identified across six LGG cohorts. (B) C-index comparison of
   different machine learning models across cohorts, with the
   StepCox[forward] + plsRcox model achieving the highest mean C-index.
   (C) Meta-analysis illustrating the hazard ratio distribution across six
   LGG cohorts. (D) Kaplan-Meier survival analysis comparing high- and
   low-risk groups, based on the StepCox[forward]+ plsRcox model, with
   median survival times shown. P-values were calculated using the
   Log-rank test. (E) ROC curve analysis demonstrating the predictive
   performance of the StepCox[forward] + plsRcox model at 1, 3, and 5
   years

Evaluation of model generalization ability

   To assess robustness, the StepCox[forward] + plsRcox model was compared
   to existing prognostic models. Univariate Cox regression analysis
   confirmed its reliability (Fig. [95]3 A), while consistent C-index
   values across six cohorts (Fig. [96]3B) indicated stable performance.
   The model outperformed existing clinical and molecular markers in
   predictive accuracy (Fig. [97]3 C), validating its generalizability and
   prognostic value.

Fig. 3.

   [98]Fig. 3
   [99]Open in a new tab

   Evaluation of Model Generalization.(A) Univariate Cox regression
   analysis comparing the prognostic power of the proposed model to
   previously established LGG models. (B) Generalized C-index assessment
   across six cohorts, highlighting the stable performance of the proposed
   model in predicting overall survival compared to most other prognostic
   models. (C) Bar plot comparing the C-index values of the proposed model
   with various clinical and molecular markers. PQ = 1p19q co-deletion;
   IDH = IDH mutation

Exploration of biological pathways based on risk model

   Pathway enrichment analyses revealed significant biological alterations
   associated with risk stratification. KEGG analysis indicated that
   high-risk patients were enriched in pathways related to oxidative
   phosphorylation, neurodegenerative diseases, and pyruvate metabolism.
   Hallmark analysis identified associations with oxidative
   phosphorylation, adipogenesis, and fatty acid metabolism. GO pathway
   analysis highlighted processes including ATP metabolism, the electron
   transport chain, and mitochondrial respiratory chain complex assembly
   (Fig. [100]4 A, p < 0.01). Subpopulation analyses corroborated these
   findings (Fig. [101]4B, adjusted p < 0.01). Additionally, high-risk
   patients exhibited significant mutations in CTNNB1 and APOB7, as well
   as chromosomal aberrations such as 5q35.3 gain and 8p23.2 loss (Fig.
   [102]4 C, p < 0.05).

Fig. 4.

   [103]Fig. 4
   [104]Open in a new tab

   Biological Function Analysis of the Model.(A) GSEA pathway enrichment
   analysis comparing high- and low-risk groups using KEGG, Hallmark, and
   GO gene sets. (B) GO term enrichment analysis showing biological
   processes, cellular components, and molecular functions enriched in the
   high-risk group (red) and low-risk group (blue). (C) Mutation analysis
   identifying key somatic mutations and chromosomal regions associated
   with high-risk patients. (D) Box plots illustrating the relative
   distribution of immune cells across six LGG cohorts, analyzed using the
   CIBERSORT algorithm.(E) Correlation analysis between immune modulators
   and the model-derived risk scores across six LGG cohorts

   Immune infiltration analysis revealed distinct microenvironmental
   characteristics between risk groups. The low-risk group demonstrated
   higher infiltration of follicular helper T cells and monocytes, while
   the high-risk group was associated with M2-type macrophages (Fig.
   [105]4D). Furthermore, the low-risk group exhibited a “hot” immune
   microenvironment, characterized by increased expression of immune
   stimulators, MHC pathways, and immune inhibitors (Fig. [106]4E).

Identification of potential targeted drugs based on the model

   Drug sensitivity analysis using IC50 values from the GDSC2 database
   identified AZD1208_1449 as a potential targeted therapy, consistently
   validated across five LGG cohorts with an AUC > 0.65 (Fig. [107]5A-B).
   ROC curve analysis demonstrated the predictive efficacy of the model
   for drug sensitivity (AUC range: 0.68–0.83, Fig. [108]5 C). Low-risk
   patients were more sensitive to AZD1208_1449, as indicated by lower
   IC50 values compared to high-risk patients. Molecular docking predicts
   that AZD1208 targets SDHB by binding its catalytic pocket with high
   affinity (Vina score: −8.9 kcal/mol), forming interactions with key
   electron transfer residues (HIS270/HIS363) and substrate-binding
   residue ASP74 (Table 2), potentially inhibiting succinate binding or
   disrupting [2Fe-2 S] cluster function to block SDHB activity (Figure
   S2).

Fig. 5.

   [109]Fig. 5
   [110]Open in a new tab

   Drug Sensitivity Analysis Based on METCGs. (A) Flowchart of the drug
   selection process. GDSC2 = Genomic Data Sharing for Cancer Research,
   version 2; AUC = area under the curve; IC50 = half-maximal inhibitory
   concentration. (B) ROC curves evaluating the predictive performance of
   the model in identifying AZD1208_1449 sensitivity across six cohorts.
   (C) Bar plot comparing the calculated IC50 values between high- and
   low-risk groups

Construction of the prognostic nomogram

   To enhance clinical utility, we performed univariate and multivariate
   Cox regression analyses to identify significant predictors of overall
   survival (OS) (Fig. [111]6 A, all p < 0.001). A nomogram incorporating
   model risk group, age, IDH mutation status, 1p19q co-deletion status,
   and WHO grade was developed to predict OS (Fig. [112]6 C). The nomogram
   outperformed other clinical predictors for 1-, 3-, and 5-year survival
   rates (Fig. [113]6D, DeLong test p < 0.05). Time-dependent C-index
   curves (Fig. [114]6E) and calibration plots (Fig. [115]6 F) further
   validated the nomogram’s accuracy and predictive performance.

Fig. 6.

   [116]Fig. 6
   [117]Open in a new tab

   METCG-Based Nomogram for Prognostic Evaluation in LGG Patients.(A-B)
   Univariate and multivariate Cox regression analyses comparing the
   model-derived risk groups with other clinical variables across LGG
   cohorts. (C) Nomogram plot integrating the model-derived risk groups
   with other clinical factors to generate predictive scores. (D) ROC
   curve analysis comparing the predictive performance of the nomogram and
   other variables for overall survival at 1, 3, and 5 years. (E)
   Time-dependent concordance index demonstrating the predictive
   performance of the nomogram compared to other clinical variables. (F)
   Calibration curves evaluating the accuracy of the nomogram in
   predicting 1-, 3-, and 5-year overall survival

SDHB expression and its functional role in LGG progression and invasion

   To robustly assess the reliability of this prognostic model, we
   selected SDHB, one of the five candidate genes, for further single-cell
   analysis and in vitro experimental validation. The UMAP plot (Fig.
   [118]7 A) revealed the heterogeneous distribution of SDHB expression
   across cell types of LGG, with astrocytes showing the highest
   expression levels and proportions, as indicated by the accompanying dot
   plot (Fig. [119]7B). Kaplan-Meier survival analyses using CGGA_693 and
   TCGA_LGG datasets demonstrated that higher SDHB expression correlates
   with poorer overall survival in glioma patients (Figs. [120]7 C, D;
   log-rank p < 0.0001). Immunohistochemistry (IHC) of LGG tissue and
   adjacent peritumor tissue confirmed a significant elevated expression
   of SDHB in tumor regions (Fig. [121]7E). Immunofluorescence staining
   further demonstrated SDHB localized mitochondria of LGG cells, with its
   expression being significantly higher than normal astrocytes (HEB)
   cells (HS683, SHG-44; Figs. [122]7 F, G). BN-PAGE also confirmed that
   SDHB forms protein complexes in LGG cells, with the expression levels
   of these protein complexes significantly higher than those observed in
   HEB cells (Fig. [123]7H). Western blot analysis validated this
   differential expression at the protein level across HEB and LGG cells
   (Fig. [124]7I). The above results suggest that SDHB and Complex II may
   play a crucial role in the growth and invasion of LGG.

Fig. 7.

   [125]Fig. 7
   [126]Open in a new tab

   SDHB was upregulated in LGG and correlated with invasion and
   proliferation. (A) UMAP visualization of SDHB expression across
   different clusters. Each dot represents an individual sample, with the
   color intensity corresponding to SDHB expression levels (scale on the
   left). (B) Dot plot depicting the average expression of SDHB across
   various cell types, including astrocytes, microglia, oligodendrocytes,
   and T cells. The size of each dot reflects the percentage of cells
   expressing SDHB, while the color indicates the average expression level
   (scale at the bottom right). (C-D) Kaplan-Meier survival curves for
   patients with high (blue) and low (yellow) SDHB expression in the CGGA
   693 and TCGA LGG cohorts. The log-rank test p-value is indicated,
   showing a significant difference in overall survival between the two
   groups. (E) Immunohistochemical staining of tumor and peritumoral
   tissues from three LGG patients revealed a marked upregulation of SDHB
   in the tumor tissues. Scale Bar=50μm. (F) Immunofluorescence analysis
   showed that the fluorescence intensity of SDHB in LGG cells was
   significantly higher than that in HEB cells.(G) Quantitative analysis
   of the green fluorescence intensity of SDH was performed on 30 cells in
   Figure 7F. ****P < 0.0001, n=30. (H-I) The results of BN-PAGE and
   SDS-PAGE indicated elevated expression levels of Complex II(H) and
   SDHB(I) in LGG and GBM cells. (J-K) Three siRNAs were used to knock
   down SDHB in HS683 and SHG44 cells, and SDS-PAGE confirmed that siRNA#2
   and siRNA#3 achieved higher knockdown efficiency. (L-M) After knocking
   down SDHB in SHG-44 cells with siRNA#2 and siRNA#3 for 48 hours,
   Transwell (L-N) and CCK-8(O-P) assays confirmed that the knockdown of
   SDHB significantly inhibited the invasion and growth of LGG cells.
   ****P < 0.0001

   Next, we used three different siRNAs to knock down SDHB in LGG cells
   and verified the knockdown efficiency through SDS-PAGE. siRNA#2 and
   siRNA#3 demonstrated higher knockdown efficiency. Therefore, we
   proceeded with further knockdown of SDHB in LGG cells using siRNA#2 and
   siRNA#3, and assessed the impact of SDHB knockdown on LGG cell growth
   and invasion through CCK-8 and Transwell assays. These results indicate
   that knockdown of SDHB in LGG cells (HS683 and SHG-44) leading to
   decreased cell invasion as shown by transwell assays (Fig. [127]7L, M,
   N) and significantly decreased proliferation rates (Figs. [128]7O, P).
   These findings suggest that SDHB can acts as a critical biomarker of
   glioma progression and invasion, potentially through its role in
   mitochondrial function.

Knockdown of SDHB impairs mitochondrial function and alters metabolic
activity in LGG cells

   To investigate the role of SDHB in mitochondrial function, we knocked
   down SDHB in HS-683 and SHG-44 cells using two siRNAs.
   Immunofluorescence analysis revealed a marked reduction in SDHB protein
   levels upon siRNA transfection, as evidenced by diminished SDHB
   staining (green) and disrupted mitochondrial structure, as visualized
   by Mitotracker staining (red) (Fig. [129]8 A and B). Quantification of
   Mitotracker fluorescence intensity demonstrated a significant increase
   in mitochondrial membrane potential (MMP) in siRNA-transfected cells
   compared to control siRNA (siCtrl)-treated cells (Fig. [130]8 C and D).
   Flow cytometry analysis of mitochondrial membrane potential, using
   Mitotracker fluorescence, further confirmed these findings, showing
   Increased MMP in SDHB-knockdown cells (Fig. [131]8E and F). To assess
   mitochondrial reactive oxygen species (ROS) levels, we also utilized
   MitoSOX staining. SDHB-knockdown cells exhibited significantly elevated
   mitochondrial ROS levels compared to siCtrl-treated cells (Fig. [132]8G
   and H), highlighting increased oxidative stress upon SDHB depletion.

Fig. 8.

   [133]Fig. 8
   [134]Open in a new tab

   Knockdown of SDHB impairs mitochondrial ETC complexes and alters
   metabolic activity in LGG cells. (A-D) siRNA#2 and siRNA#3 were
   utilized to knock down SDHB in HS683 and SHG-44 cells.
   Immunofluorescence staining was conducted to assess Mitotracker (red
   fluorescence), SDHB (green fluorescence), and mitochondrial morphology
   (A-B). Quantitative analysis of fluorescence intensity indicated that
   knockdown of SDHB led to a significant increase in fluorescence
   intensity in both HS683 (C) and SHG-44(D) cells. ****P < 0.0001, n=30.
   (E-H) Flow cytometric analysis was performed to evaluate the changes in
   fluorescence intensity of Mitotracker and MitoSOX in HS683 and SHG-44
   cells following SDHB knockdown. (I-K) The Luciferase ATP Assay Kit was
   used to measure changes in ATP levels in HEB cells, LGG cells, and SDHB
   knockdown LGG cells. ** P <0.01, **** P < 0.0001, n=8. (L-N) The
   Succinate Dehydrogenase (SDH) Activity Assay Kit was used to measure
   changes in SDH activity in HEB cells, LGG cells, and SDHB knockdown LGG
   cells. ** P <0.01, **** P < 0.0001, n=8. (O-Q) The Mitochondrial
   Respiratory Complex II Activity Assay Kit was used to measure changes
   in Complex II activity in HEB cells, LGG cells, and SDHB knockdown LGG
   cells. **** P < 0.0001, n=8. (R-S) SDS-PAGE(R) and BN-PAGE(S) were
   performed to detect the expression of the five mitochondrial
   respiratory chain complexes in HS683 and SHG-44 cells following SDHB
   knockdown

   Next, we aimed to investigate whether the expression level of SDHB in
   LGG affects ATP production, SDH activity, and Complex II (succinate
   dehydrogenase, SDH) activity. We found that ATP levels were
   significantly higher in LGG cells (Fig. [135]6I). Additionally, SDHB
   knockdown inhibited ATP production in LGG cells (Figs. [136]6 J, K). We
   also assessed the enzymatic activities of SDH, which were markedly
   reduced in SDHB-low expressed cells, as demonstrated in both the HS-683
   and SHG-44 cell models (Figs. [137]8L-N). In contrast, SDHA activity
   remained unaffected (Figure S3). As one of the most important subunits
   of Complex II, we hypothesized that SDHB knockdown would also inhibit
   Complex II (SDH) activity. Indeed, Complex II activity was
   significantly increased in SDHB-high expressed LGG cells (Fig.
   [138]8O), while SDHB knockdown reduced Complex II activity in both
   HS-683 and SHG-44 cell models (Figs. [139]8P-Q). Thus, we can conclude
   that high expression of SDHB promotes ATP production, SDH and Complex
   II activity in LGG cells.

   In summary, we conclude that SDHB knockdown increases oxidative stress
   and significantly impairs Complex II in LGG cells (Figs. [140]8A-Q),
   leading to widespread disruption of the electron transport chain (ETC).
   To further investigate the underlying mechanisms, we used SDS-PAGE and
   BN-PAGE to examine the expression and assembly of the five ETC
   complexes. Western blot analysis confirmed the efficient knockdown of
   SDHB and revealed concomitant reductions in the protein levels of
   several ETC components, including NDUFS1 (Complex I), SDHA (Complex
   II), UQCRC1 (Complex III), MTCO2 (Complex IV), and ATP5A1 (Complex V)
   in siRNA-transfected cells (Fig. [141]8R). Additionally, BN-PAGE
   further substantiated that SDHB knockdown led to a decrease in all five
   complexes, indicating compromised assembly and activity of the ETC
   complexes in SDHB-deficient cells (Fig. [142]8S).

   These results demonstrate that SDHB knockdown leads to mitochondrial
   dysfunction, reduced ATP production, elevated ROS levels, and impaired
   activities of respiratory complexes in glioma cells. These findings
   highlight the critical role of SDHB in maintaining mitochondrial
   integrity of ETC and metabolic homeostasis.

Discussion

   This study represents a pioneering integration of single-cell
   sequencing and advanced bioinformatics to elucidate the distribution,
   functionality, and prognostic significance of METCGs in LGG. Leveraging
   single-cell RNA sequencing from the [143]GSE117891 dataset, we
   accurately annotated four distinct cell types in brain tissue,
   including astrocytes, oligodendrocytes, T cells, and microglial cells,
   using a combination of CellMarker 2.0 and classical markers. The
   application of UMAP and DotPlot visualizations further enabled a
   comprehensive understanding of cell-specific characteristics,
   particularly highlighting astrocytes as the dominant cell type with
   unique metabolic profiles. Innovatively, we employed multiple scoring
   algorithms, including AUCell, UCell, and ssGSEA, to calculate METCGs
   scores, revealing astrocytes’ metabolic dominance in oxidative
   phosphorylation pathways. To validate whether this dominance was
   statistically significant, we conducted comprehensive inter-group
   comparisons across all scoring algorithms. As shown in Table 1,
   astrocytes exhibited significantly higher METCG scores compared to
   oligodendrocytes, T cells, and microglia (all adjusted
   p-values < 0.001). These results remained consistent across AUCell,
   UCell, ssGSEA, Singscore, JASMINE, and viper, confirming the robustness
   of our scoring framework. This significant enrichment suggests that
   astrocytes may serve as a key metabolic hub within the LGG
   microenvironment, with elevated oxidative phosphorylation activity
   potentially contributing to tumor-supportive functions. These findings
   underscore the power of single-cell analysis in revealing intratumoral
   heterogeneity. In parallel, we applied advanced machine learning
   (StepCox[forward] + plsRcox) to build a robust prognostic model, which
   demonstrated superior accuracy and generalizability across six LGG
   cohorts. Pathway and immune profiling further revealed distinct
   metabolic and immunological features between risk groups. Together,
   this work establishes a novel framework for LGG stratification and
   highlights the clinical potential of integrating single-cell and
   bioinformatics approaches.

   To further validate the reliability of the METCGs prognostic model in
   predicting potential diagnostic and therapeutic molecular targets in
   LGG, we conducted additional confirmation through cellular experiments.
   The model is composed of five genes, including SDHB, SDHC, SLC25A27,
   UQCRB, and NDUFA13, each contributing to various mitochondrial
   functions. SDHB and SDHC are subunits of Complex II, while the others
   belong to Complexes I and III or are involved in mitochondrial energy
   regulation [[144]17, [145]35, [146]36]. The role of Complex II in
   tumors is complex and context-dependent: it can promote tumor
   progression through succinate accumulation and activation of HIF-1α
   [[147]37], while also inhibiting tumor growth by maintaining
   mitochondrial function and inducing apoptosis [[148]17]. Therefore, a
   deeper understanding of the specific mechanisms through which Complex
   II regulates tumorigenesis in LGG is crucial for the development of
   novel metabolism-targeted therapeutic strategies. Thus, we selected
   SDHB for further study. SDHB is a key subunit of Complex II in the
   electron transport chain (ETC). It contributes to the proper
   functioning of mitochondrial oxidative phosphorylation, which is
   essential for ATP production and maintaining cellular energy
   homeostasis [[149]38]. In this study, we first observed that SDHB and
   Complex II were upregulated in LGG. Additionally, knockdown of SDHB
   inhibited the proliferation and invasion of LGG cells. Notably, SDHB
   exhibited the highest expression levels and proportions in astrocytes.
   In conclusion, our findings suggest that the elevated expression of
   SDHB and Complex II may be correlated with tumorigenesis and could
   serve as potential therapeutic targets for LGG. However, the underlying
   mechanisms by which SDHB modulates Complex II assembly, thereby
   promoting LGG proliferation and invasion, remain to be elucidated.

   To further investigate the mechanism by which SDHB affects
   mitochondrial function and inhibits LGG growth and invasion through
   disrupt the structure of the mitochondrial respiratory chain, we
   knocked down SDHB in two LGG cell lines and explored the effects using
   immunofluorescence, flow cytometry and BN-PAGE techniques. Our findings
   reveal that knockdown of SDHB impairs mitochondrial integrity, leading
   to a significant increase in oxidative stress, disrupted MMP, and
   elevated ROS levels. Notably, we also demonstrate that SDHB depletion
   reduces the activity of SDH and Complex II, indicating a global
   impairment of mitochondrial respiration chain. To further validate this
   hypothesis, we utilized BN-PAGE and SDS-PAGE to examine the assembly of
   the five mitochondrial ETC complexes and the expression levels of their
   key subunits. The results revealed widespread dysfunction across the
   ETC complexes, with a pronounced impact on Complex II. In summary, we
   propose the scientific hypothesis that the high expression of SDHB in
   LGG contributes to maintaining elevated SDH and Complex II activity,
   ensuring the structural and functional integrity of mitochondrial ETC
   complexes and ATP production. This supports the high ROS production and
   MMP required for the rapid growth of LGG, thereby promoting its
   proliferation and invasion.

   The novelty of this study lies in the comprehensive assessment of
   SDHB’s impact on both mitochondrial structure and function in LGG
   cells. While previous studies have linked SDHB to tumorigenesis and
   metabolic reprogramming [[150]38–[151]40], our work specifically
   addresses how SDHB knockdown affects the assembly and activity of all
   five mitochondrial ETC complexes, a mechanism not yet fully elucidated
   in LGG research. Additionally, our findings that SDHB depletion
   compromises the assembly of the ETC and impairs mitochondrial energy
   production provide new insights into the metabolic vulnerabilities of
   glioma cells. This underscores the potential of targeting SDHB or its
   associated pathways as a therapeutic strategy to disrupt mitochondrial
   function and limit tumor progression in LGG. These results contribute
   to the growing body of evidence linking mitochondrial ETC structure and
   functions to cancer biology, highlighting the potential for targeting
   METCGs in LGG therapy. From a clinical perspective, our findings hold
   significant translational implications. The METCG-based risk model
   demonstrates reliable prognostic value and may serve as a powerful tool
   for stratifying LGG patients and guiding individualized treatment
   planning. Moreover, SDHB emerges not only as a biomarker of poor
   prognosis but also as a promising therapeutic target. Our data suggest
   that pharmacological inhibition of SDHB, or disruption of Complex II
   function, may impair the metabolic fitness of LGG cells and suppress
   tumor growth. The identification of AZD1208_1449 as a candidate small
   molecule with selective efficacy in low-risk groups further supports
   the feasibility of developing mitochondria-targeted therapies based on
   the metabolic status of individual tumors. Further investigation into
   the downstream signaling pathways affected by SDHB depletion and their
   role in glioma progression could pave the way for novel therapeutic
   interventions targeting mitochondrial metabolism.

Acknowledgements