Machine learning identification of mitochondrial autophagy diagnostic biomarkers and immune infiltration analysis in steroid-induced osteonecrosis of the femoral head

doi:10.12307/2025.332

Abstract

Abstract: BACKGROUND: Mitochondrial autophagy is closely related to the occurrence and development of steroid-induced osteonecrosis of the femoral head (SONFH), but specific biomarkers and regulatory mechanisms remain unclear.
OBJECTIVE: To identify the key biomarkers of mitochondrial autophagy in steroid-induced osteonecrosis of the femoral head using machine learning algorithms and to conduct an immune infiltration analysis.
METHODS: The SONFH datasets GSE123568 and GSE74089 were downloaded from the GEO database, serving as the training and validation sets, respectively. Differentially expressed genes between SONFH and control groups were selected, and weighted gene co-expression network analysis was performed. Mitochondrial autophagy-related genes were obtained from MitoCarta3.0 and intersected with differentially expressed genes and module genes. Two machine learning algorithms were utilized to identify key genes of SONFH mitochondrial autophagy, and validated using an external validation set. CIBERSORT and immune infiltration analysis were employed to assess the proportion of immune cells, and ssGSEA was used to analyze the correlation between mitochondrial autophagy genes and immune cells.
RESULTS AND CONCLUSION: Differential analysis identified a total of 1 163 differentially expressed genes, including 663 upregulated genes and 500 downregulated genes. Weighted gene co-expression network analysis identified 4 key modules, comprising 1 412 module genes. Intersection with mitochondrial autophagy genes yielded 39 intersecting genes as disease-related mitochondrial autophagy genes. Gene ontology enrichment analysis showed that the biological processes were mainly related to heme metabolism, mitochondrial transport, nucleotide bisphosphate metabolism and thioester metabolism, and the cellular components were mainly related to mitochondrial matrix, mitochondrial outer membrane, organelle outer membrane and mitochondrial inner membrane, and the molecular functions were mainly related to fatty acid ligase activity, iron-sulfur cluster binding, and cofactor A ligase activity. Kyoto Encyclopedia of Genes and Genomes enrichment analysis mapped out a total of six pathways, which were mainly related to fatty acid degradation, mitochondrial autophagy, butyric acid metabolism, fatty acid biosynthesis and cofactor biosynthesis. Through LASSO regression and RFE-SVM algorithm analysis, four intersecting genes (ALDH5A1, FBXL4, MCL1, and STOM) were identified. The receiver operating characteristic curves of the four core genes and the diagnostic column chart validation set were all greater than 0.9. The occurrence and development of SONFH were related to immune cells such as dendritic cells, bone marrow-derived suppressor cells, regulatory T cells, and central memory CD8 T cells. To conclude, the four key mitochondrial autophagy genes ALDH5A1, FBXL4, MCL1, and STOM play a crucial role in the progression of SONFH through osteoclast differentiation and immune mechanisms. Additionally, all four genes have good disease prediction efficacy and can serve as biomarkers for the diagnosis and treatment of SONFH.

中国组织工程研究杂志出版内容重点：组织构建；骨细胞；软骨细胞；细胞培养；成纤维细胞；血管内皮细胞；骨质疏松；组织工程

Key words: steroid-induced osteonecrosis of the femoral head, mitophagy, machine learning algorithms, immune cell infiltration, key marker

CLC Number:

Huang Keqi, Chen Yueping, Chen Shangtong, Li Jiagen. Machine learning identification of mitochondrial autophagy diagnostic biomarkers and immune infiltration analysis in steroid-induced osteonecrosis of the femoral head[J]. Chinese Journal of Tissue Engineering Research, 2025, 29(11): 2402-2410.

Figures/Tables 8

2.1 激素性股骨头坏死差异基因使用“limma”包分析激素性股骨头坏死组和对照组的RNA-seq数据，设计校正后P < 0.05和|log2FC|使用> 0.58，总共筛选出1 163个差异基因，其中有663个上调基因和500个下调基因(图1A)。然后用火山图和热图可视化，火山图中显示上调的基因标注为橙点，下调的基因标注为蓝点。将排行前20个和后20个差异基因通过热图展示(图1B)。 2.2 WGCNA结果 WGCNA是发现基因与临床表型的好方法，通过结合差异表达基因挖掘出与临床疾病表型相关的关键基因。GSE123568没有异常值，数据集中未删除任何样本，对本数据集中位绝对偏差的前75%基因进行了WGCNA筛选，当软阈值功率=12时，评估参数R2大于0.5，构建了β=12软阈值能力的无标度网络(图2A)，并根据表达谱将基因分类为不同的模块。通过平均链接聚类(图2B)，模块的颜色越深，提示该模块相关性越强。在分析了正相关系数后选择了相关系数最强的模块，结果显示black模块(r=±0.52，P=6×10-4)、blue模块(r=±0.67，P=3×10-6)、brown模块(r=±0.67，P=4×10-6)和slmon模块(r=±0.7，P=4×10-7)的基因具有显著意义，其余无显著的模块(图2C)。最后，分别提取了上述4个模块中的基因，包括black模块43个基因、blue模块837个基因、brown模块335个基因和slmon模块中的197个基因，这1 412个基因被认为是与临床表型相关的关键基因。 2.3 整理激素性股骨头坏死的线粒体相关基因及GO、KEGG功能分析从“MitoCarta 3.0”数据库中收集和整理线粒体相关基因，共有1 136个基因，使用1 163个差异基因、WGCNA相关模块的1 412个基因和线粒体相关1 136个基因取交集得到激素性股骨头坏死线粒体相关基因，一共得到39个线粒体相关激素性股骨头坏死基因(图1C)。进行GO和KEGG功能富集分析(图3)，GO富集分析结果显示，生物过程主要涉及血红素代谢过程、线粒体运输、核苷酸双磷酸的代谢过程和硫酯代谢过程，细胞组分主要涉及线粒体基质、线粒体外膜、细胞器外膜和线粒体内膜，分子功能主要涉及脂肪酸连接酶活性、铁-硫簇结合和辅酶A连接酶活性，将每项功能的前5项分析结果可视化；KEGG富集分析结果显示共映射出6条通路，主要涉及脂肪酸降解、线粒体自噬、丁酸代谢、脂肪酸生物合成和辅因子生物合成。 2.4 两种机器学习筛选诊断标志物利用R软件的“random Forest”包对39个差异基因进行随机森林算法，并绘制分组误差图和基因重要性图谱，共筛出16个基因，以重要性分数0.5为基准共筛出9个特征基因(图4A，B)。利用LASSO模型进行内部验证，共获得5个特征基因(图4C)。将2种机器学习的结果取交集得到4个诊断标志物(图4D)，分别是ALDH5A1、FBL4、MCL1和STOM。 2.5 风险模型构建与验证将上述结果中的4个线粒体基因纳入风险研究，并基于二分类逻辑回归分析，构建列线图模型，以计算总评分来分析激素性股骨头坏死患者的风险概率(图5)。结果显示，线粒体4个基因的受试者工作特征曲线结果显示曲"

线下面积值均> 0.9(图6)，说明线粒体4个基因对激素性股骨头坏死具有良好的诊断价值。列线图的受试者工作特征曲线结果显示曲线下面积为0.98，说明列线图模型对股骨头坏死患者风险因素的评估具有较好的预测能力(图6A)。 2.6 免疫浸润分析通过CIBERSORT反卷积算法以P < 0.05样本进行筛选。热图左侧10个样本(GSM3507251-GSM3507260)为对照组，右侧30个样本(GSM3507261-GSM3507290)为激素性股骨头坏死组，免疫细胞比例箱图中展示了22种免疫细胞的浸润比例(图7)。创建了一个热图来描述免疫细胞之间的差异相关模式(图8A)，红色越深代表二者正相关性越高，蓝色越深代表二者负相关性越高，高正相关性的组合有活化的CD8T细胞与效应记忆CD8T细胞、骨髓来源的抑制性细胞与中心记忆CD4T细胞，其中发现活化的树突状细胞与T滤泡辅助细胞、嗜酸性粒细胞、中性粒细胞、浆细胞样树突状细胞和自然杀伤细胞等免疫细胞均具有高正相关性。负相关性较高的组合自然杀伤细胞与未成熟树突状细胞、2型T辅助细胞与巨噬细胞，其中发现活化的CD4T细胞与多种免疫细胞均呈高负相关，包括活化的树突状细胞、骨髓来源的抑制性细胞、调节性T细胞和中心记忆CD8T细胞等免疫细胞。免疫浸润热图结果显示，激素性股骨头坏死样本的免疫浸润丰度远高于对照组，其中包括中心记忆CD4T细胞、活化的CD4T细胞、骨髓来源的抑制性细胞和T滤泡辅助细胞等多数免疫浸润细胞(图8B)。免疫差异细胞箱向图结果显示，免疫细胞表达在激素性股骨头坏死和对照组有显著差异，中心记忆CD4T细胞、效应记忆CD8T细胞、骨髓来源的抑制性细胞和记忆B细胞在激素性股骨头坏死组中的表达显著高于对照组。在诊断标志物与免疫细胞相关性热图中，MCL1与骨髓来源的抑制性细胞、中性粒细胞、效应记忆CD8T细胞、T滤泡辅助细胞和活化的树突状细胞呈正相关，而ALDH5A1、FBXL4和STOM与上述免疫细胞呈负相关(图8C，D)。"

References

[1] TAKASHIMA K, IWASA M, ANDO W, et al. MRI screening for osteonecrosis of the femoral head after COVID-19. Mod Rheumatol. 2023:road095. doi: 10.1093/mr/road095.
[2] IM GI. Regenerative medicine for osteonecrosis of the femoral head : present and future. Bone Joint Res. 2023;12(1):5-8.
[3] WEINSTEIN RS. Clinical practice. Glucocorticoid-induced bone disease. N Engl J Med. 2011;365(1):62-70.
[4] KURODA T, TANABE N, WAKAMATSU A, et al. High triglyceride is a risk factor for silent osteonecrosis of the femoral head in systemic lupus erythematosus. Clin Rheumatol. 2015;34(12):2071-2077.
[5] HOUDEK MT, WYLES CC, PACKARD BD, et al. Decreased Osteogenic Activity of Mesenchymal Stem Cells in Patients With Corticosteroid-Induced Osteonecrosis of the Femoral Head. J Arthroplasty. 2016; 31(4):893-898.
[6] ZHANG Q, SUN W, LI T, et al. Polarization Behavior of Bone Macrophage as Well as Associated Osteoimmunity in Glucocorticoid-Induced Osteonecrosis of the Femoral Head. J Inflamm Res. 2023;16: 879-894.
[7] CHENG Y, CHEN H, DUAN P, et al. Early depletion of M1 macrophages retards the progression of glucocorticoid-associated osteonecrosis of the femoral head. Int Immunopharmacol. 2023;122:110639.
[8] OSAWA Y, TAKEGAMI Y, KATO D, et al. Hip function in patients undergoing conservative treatment for osteonecrosis of the femoral head. Int Orthop. 2023;47(1):89-94.
[9] JENKINS BC, NEIKIRK K, KATTI P, et al. Mitochondria in disease: changes in shapes and dynamics. Trends Biochem Sci. 2024; 49(4):346-360.
[10] HARRINGTON JS, RYTER SW, PLATAKI M, et al. Mitochondria in health, disease, and aging. Physiol Rev. 2023;103(4):2349-2422.
[11] VICTORELLI S, SALMONOWICZ H, CHAPMAN J, et al. Apoptotic stress causes mtDNA release during senescence and drives the SASP. Nature. 2023;622(7983):627-636.
[12] TENG Y. Remodeling of Mitochondria in Cancer and Other Diseases. Int J Mol Sci. 2023;24(9):7693.
[13] VAFAI SB, MOOTHA VK. Mitochondrial disorders as windows into an ancient organelle. Nature. 2012;491(7424):374-383.
[14] DANG R, HOU X, HUANG X, et al. Effects of the Glucocorticoid-Mediated Mitochondrial Translocation of Glucocorticoid Receptors on Oxidative Stress and Pyroptosis in BV-2 Microglia. J Mol Neurosci. 2024;74(1):30.
[15] DU F, YU Q, SWERDLOW RH, et al. Glucocorticoid-driven mitochondrial damage stimulates Tau pathology. Brain. 2023;146(10):4378-4394.
[16] RATH S, SHARMA R, GUPTA R, et al. MitoCarta3.0: an updated mitochondrial proteome now with sub-organelle localization and pathway annotations. Nucleic Acids Res. 2021;49(D1): D1541-D1547.
[17] BARDOU P, MARIETTE J, ESCUDIE F, et al. jvenn: an interactive Venn diagram viewer. BMC Bioinformatics. 2014;15(1):293.
[18] TSUBOSAKA M, MARUYAMA M, LUI E, et al. Preclinical models for studying corticosteroid-induced osteonecrosis of the femoral head[. J Biomed Mater Res B Appl Biomater. 2024;112(1):e35360.
[19] TANG B, CHEN Y, ZHAO P, et al. MiR-601-induced BMSCs senescence accelerates steroid-induced osteonecrosis of the femoral head progression by targeting SIRT1. Cell Mol Life Sci. 2023;80(9):261.
[20] JIANG LY, YU X, PANG QJ. Research in the precaution of recombinant human erythropoietin to steroid-induced osteonecrosis of the rat femoral head. J Int Med Res. 2017;45(4):1324-1331.
[21] ANASTASILAKI E, PACCOU J, GKASTARIS K, et al. Glucocorticoid-induced osteoporosis: an overview with focus on its prevention and management. Hormones (Athens). 2023;22(4):611-622.
[22] ZHANG Y, WENG J, HUAN L, et al. Mitophagy in atherosclerosis: from mechanism to therapy. Front Immunol. 2023;14:1165507.
[23] CASTRO-SEPULVEDA M, FERNANDEZ-VERDEJO R, ZBINDEN-FONCEA H, et al. Mitochondria-SR interaction and mitochondrial fusion/fission in the regulation of skeletal muscle metabolism. Metabolism. 2023;144:155578.
[24] SU E, VILLARD C, MANNEVILLE JB. Mitochondria: At the crossroads between mechanobiology and cell metabolism. Biol Cell.2023;115(9):e2300010.
[25] LU Y, LI Z, ZHANG S, et al. Cellular mitophagy: Mechanism, roles in diseases and small molecule pharmacological regulation. Theranostics. 2023;13(2):736-766.
[26] WANG S, LONG H, HOU L, et al. The mitophagy pathway and its implications in human diseases. Signal Transduct Target Ther. 2023;8(1):304.
[27] YANG N, SUN H, XUE Y, et al. Inhibition of MAGL activates the Keap1/Nrf2 pathway to attenuate glucocorticoid-induced osteonecrosis of the femoral head. Clin Transl Med. 2021;11(6):e447.
[28] ZHANG F, PENG W, ZHANG J, et al. P53 and Parkin co-regulate mitophagy in bone marrow mesenchymal stem cells to promote the repair of early steroid-induced osteonecrosis of the femoral head. Cell Death Dis. 2020;11(1):42.
[29] TIAN Z, CHEN Y, YAO N, et al. Role of mitophagy regulation by ROS in hepatic stellate cells during acute liver failure. Am J Physiol Gastrointest Liver Physiol. 2018;315(3):G374-G384.
[30] KOROLCHUK VI, MIWA S, CARROLL B, et al. Mitochondria in Cell Senescence: Is Mitophagy the Weakest Link? EBioMedicine. 2017;21:7-13.
[31] LIU J, HAN X, QU L, et al. Identification of key ferroptosis-related biomarkers in steroid-induced osteonecrosis of the femoral head based on machine learning. J Orthop Surg Res. 2023;18(1):327.
[32] LIANG XZ, LUO D, CHEN YR, et al. Identification of potential autophagy-related genes in steroid-induced osteonecrosis of the femoral head via bioinformatics analysis and experimental verification. J Orthop Surg Res. 2022;17(1):86.
[33] LENTING K, KHURSHED M, PEETERS TH, et al. Isocitrate dehydrogenase 1-mutated human gliomas depend on lactate and glutamate to alleviate metabolic stress. FASEB J. 2019;33(1):557-571.
[34] DENG XY, GAN XX, FENG JH, et al. ALDH5A1 acts as a tumour promoter and has a prognostic impact in papillary thyroid carcinoma. Cell Biochem Funct. 2021;39(2):317-325.
[35] BRENNENSTUHL H, DIDIASOVA M, ASSMANN B, et al. Succinic Semialdehyde Dehydrogenase Deficiency: In Vitro and In Silico Characterization of a Novel Pathogenic Missense Variant and Analysis of the Mutational Spectrum of ALDH5A1. Int J Mol Sci. 2020;21(22):8578.
[36] ZHENG Z, HAN L, LI Y, et al. Phospholipase A2-activating protein induces mitophagy through anti-apoptotic MCL1-mediated NLRX1 oligomerization. Biochim Biophys Acta Mol Cell Res. 2023;1870(6):119487.
[37] WIDDEN H, PLACZEK WJ. The multiple mechanisms of MCL1 in the regulation of cell fate. Commun Biol. 2021;4(1):1029.
[38] HOLLVILLE E, CARROLL RG, CULLEN SP, et al. Bcl-2 family proteins participate in mitochondrial quality control by regulating Parkin/PINK1-dependent mitophagy. Mol Cell. 2014;55(3):451-466.
[39] FELTON JM, DORWARD DA, CARTWRIGHT JA, et al. Mcl-1 protects eosinophils from apoptosis and exacerbates allergic airway inflammation. Thorax. 2020;75(7):600-605.
[40] JAMIL S, MOJTABAVI S, HOJABRPOUR P, et al. An essential role for MCL-1 in ATR-mediated CHK1 phosphorylation. Mol Biol Cell. 2008;19(8):3212-3220.
[41] HAKKI SS, BATOON L, KOH AJ, et al. The effects of preosteoblast-derived exosomes on macrophages and bone in mice. J Cell Mol Med. 2024;28(1):e18029.
[42] KIKYO N. Circadian Regulation of Macrophages and Osteoclasts in Rheumatoid Arthritis. Int J Mol Sci. 2023; 24(15):12307.
[43] SCHINDELER A, MCDONALD MM, BOKKO P, et al. Bone remodeling during fracture repair: The cellular picture. Semin Cell Dev Biol. 2008;19(5):459-466.
[44] WU L, LUO Z, LIU Y, et al. Aspirin inhibits RANKL-induced osteoclast differentiation in dendritic cells by suppressing NF-kappaB and NFATc1 activation. Stem Cell Res Ther. 2019;10(1):375.