Figure 1 illustrates the study’s workflow. This study was approved by the Ethics Committee of Second Affiliated Hospital of Jilin University (Ethical Approval No.: 2023-207). Between February and November 2023, nine patients diagnosed with non-traumatic ONFH who underwent hip replacement at the Department of Joint Surgery, The Second Affiliated Hospital of Jilin University, were enrolled in the ONFH group, alongside six healthy controls from routine physical exams. The diagnosis of ONFH was based on imaging, clinical history, and Ficat staging. Patients with ONFH secondary to trauma, tumors, autoimmune diseases, congenital hip anomalies, or genetic disorders were excluded from the study. The controls had no history of ONFH, hip disorders, malignancies, immune diseases, heavy alcohol consumption, or prolonged corticosteroid use. Fasting peripheral blood samples (5 mL) were collected between 6:00 and 7:00 AM on the second day after admission. Plasma was isolated and stored at −80 °C.

Total RNA was extracted from plasma using TRIzol™ reagent (Invitrogen, Thermo Fisher Scientific, China) according to the manufacturer’s protocol. rRNA was removed using the MGIEasy rRNA Depletion Kit (MGI Tech Co., Ltd., China) to enrich lncRNAs and mRNAs. The remaining RNA was fragmented and reverse-transcribed into double-stranded cDNA using the MGIEasy RNA Directional Library Preparation Kit (MGI Tech Co., Ltd., China). The double-stranded cDNA was adenylated and ligated with adapters, followed by amplifying the ligated product and circularizing the PCR product to generate a single-stranded circular DNA library. Sequencing was performed on the DNBSEQ platform (MGI Tech Co., Ltd., China).

This RNA extraction was performed on a total of 15 plasma samples (9 ONFH patients and 6 healthy controls), which were used exclusively for RNA sequencing to identify differentially expressed lncRNAs.

The raw sequencing data were strictly filtered using SOAPnuke (v1.5.2). Clean reads were aligned to the human reference genome using HISAT (v2.0.4), and Bowtie2 was further employed to align the clean reads to the reference genome, generating alignment results. Transcriptome sequencing was conducted on human plasma samples using the NCBI reference genome Homo_sapiens (GCF_000001405.39_GRCh38.p13). In the quantitative transcript expression analysis, transcript randomness, coverage, and sequencing saturation were evaluated to ensure the biological and statistical validity of the data. Pearson correlation coefficients of gene expression levels between samples were calculated to assess sample correlations. Principal component analysis (PCA) was applied to gene expression data for dimensionality reduction, identifying outliers and closely related sample clusters. Box plots of gene expression were generated for each sample to assess data dispersion and gene expression density plots were created to identify the primary distribution range of gene expression. Genes were classified based on expression levels (TPM ≤ 1, TPM: 1-10, TPM ≥ 10), and TPM values were log²-transformed before downstream analysis to stabilize variance. Their distribution across samples was visualized using stacked bar charts. Differentially expressed lncRNAs and mRNAs between the ONFH and control groups were identified using DESeq2, with thresholds of |log² (fold change) | ≥ 1 and Q-value ≤ 0.05. Volcano plots and heatmaps were generated using the R (Version 4.4.2) packages ggplot2 and pheatmap.

Seven lncRNAs were selected for ceRNA network analysis based on the following criteria: (1) |log²(fold change)| ≥ 1 and adjusted p-value < 0.05; (2) consistent and detectable expression across ONFH plasma samples; (3) annotation in both starBase v2.0 (53) (https://rnasysu.com/encori/index.php) and LncBase v3 databases (54) (https://diana.e-ce.uth.gr/lncbasev3/interactions); and (4) reliable primer design feasibility for Quantitative real-time Polymerase Chain Reaction (qRT-PCR).

MiRNAs predicted to interact with the selected lncRNAs were identified using both starBase v2.0 and LncBase v3, and only the overlapping miRNAs from the two databases were retained. Subsequently, miRNA-targeted mRNAs were predicted using the MiRWalk database (http://mirwalk.umm.uni-heidelberg.de/), and intersected with the DEmRNAs identified from ONFH plasma samples to generate the final target mRNA set. The ceRNA regulatory network was then constructed based on these filtered lncRNA–miRNA and miRNA–mRNA interaction pairs.

Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed on the final mRNAs using the Metascape database (55) (https://metascape.org). GO terms and KEGG pathways with p-values less than 0.05 were considered statistically significant, and the enrichment results were visualized using the ggplot2 R package.

The raw dataset GSE98895 (56), which contains samples from 20 normal individuals and 20 patients with MetS, was downloaded from the Gene Expression Omnibus (GEO) database (57) (https://www.ncbi.nlm.nih.gov/geo/). Weighted Gene Co-expression Network Analysis (WGCNA) (58) was executed to identify modules associated with MetS. A weighted adjacency matrix was constructed using a power function, and the optimal soft threshold (β) was determined using the “pickSoftThreshold” function. The adjacency matrix was then transformed into a Topological Overlap Matrix (TOM). The network connectivity of each gene was defined as the sum of its adjacency relationships with all other genes, and dissimilarity (1-TOM) was computed. Hierarchical clustering was applied to group genes with similar expression profiles into gene modules, with a minimum module size of n = 30. A gene dendrogram was constructed, and module eigengene dissimilarities were calculated. Finally, the feature gene network was visualized.

The mRNAs from the ONFH lncRNA ceRNA network were intersected with genes from the key modules in the MetS dataset to identify ONFH-MetS-mRNAs, which were subsequently subjected to GO and KEGG enrichment analysis using the method described in Section 2.4.

A protein-protein interaction (PPI) network was constructed using the STRING database (59) (www.string-db.org) with a minimum interaction score threshold of 0.400 and subsequently visualized using Cytoscape (60). Key nodes in the network were identified using the CytoHubba-MCC and MCODE plugins, which were applied to select the intersection of key ONFH-MetS mRNAs. The lncRNA ceRNA network for these mRNAs was then identified based on the constructed ceRNA network. To further prioritize candidate lncRNAs, Random Forest (RF) analysis (61) was performed using the “randomForest” R package (62) (version 4.7.1.2), with the number of trees (ntree) set to 500. A fixed random seed was applied before model training using set.seed(123) to ensure reproducibility. Variable importance was evaluated using the Mean Decrease in Gini index.

A nomogram was developed to provide clinical value for the diagnosis of ONFH. Based on the selected candidate lncRNAs, the nomogram was constructed using the “rms” R package. “Points” represent the score for each candidate lncRNA, and “Total Points” refers to the sum of the scores of all selected lncRNAs. We used the ‘pROC’ R package to evaluate the accuracy of our model by plotting the receiver operating characteristic (ROC) curve and calculating the area under the curve (AUC), with an AUC > 0.7 considered an ideal diagnostic threshold.

To mitigate overfitting and assess generalizability, we further implemented 3-fold cross-validation using the “caret” package. Cross-validated predictions were pooled to construct an overall ROC curve, which was used to evaluate model discrimination. These predictions were also used to generate Decision Curve Analysis (DCA) curves to assess clinical utility. Finally, to simulate real-world deployment, a final logistic regression model was retrained on the full dataset, and the corresponding nomogram and calibration curve were plotted. Calibration was performed using bootstrap resampling (B = 1000), demonstrating strong agreement between predicted and observed risks.

To validate the reliability of the results, RNA was extracted from plasma samples of 45 ONFH patients and 45 controls using the Starvio cfRNA Easy Kit (Shanghai Starvio Biotechnology Co., Ltd., China). Reverse transcription was conducted using the SureScript™ First-Strand cDNA Synthesis Kit (Cat. No. QP056, GeneCopoeia, China). qRT-PCR was used according to the manufacturer’s instructions using BlazeTaq™ SYBR^® Green qPCR Mix 2.0 (Cat. No. QP031, GeneCopoeia, China). The results were analyzed using the 2^−ΔΔCt method, with ACTB as the internal reference gene for normalizing lncRNA expression data. Five candidate lncRNAs (MRPS30-DT, LINC01106, MIR100HG, PELATON, and WDR11-AS1) were selected for validation. Primer sequences used for each lncRNA are provided in Table 1 at the end of the manuscript. ROC curves for each lncRNA were plotted, and AUC values were calculated to construct the diagnostic model.

This validation step involved an independent cohort of 90 plasma samples (45 ONFH patients and 45 healthy controls) and aimed to confirm the differential expression of key lncRNAs revealed by RNA sequencing.

Immune cell infiltration analysis was conducted using the “CIBERSORT” R package (63) to estimate the relative abundance of 22 lymphocyte subtypes in both normal and ONFH samples. The LM22 signature matrix was used as a reference, with 100 permutations (perm = 100). A fixed random seed (set.seed(123)) was applied prior to model execution to ensure reproducibility. Cell types with zero abundance across all samples were removed from the results. To evaluate differences in immune cell composition between groups, Wilcoxon rank-sum tests were performed. Additionally, Spearman correlation analysis was executed between the expression levels of lncRNA-regulated key mRNAs and immune cell proportions, using ONFH samples only.

Data processing and analyses were performed using SPSS version 26.0 and R software (Version 4.4.2). Comparisons between samples were carried out using t-tests or chi-square tests. Patient baseline characteristics are presented as the mean ± standard deviation, and a p-value < 0.05 was considered statistically significant. Spearman correlation analysis was applied to examine the relationship between mRNAs regulated by key lncRNAs and immune cells. A p-value < 0.05 was considered statistically significant.

The average age of patients in the ONFH group was 58.67 ± 9.46 years, whereas the average age of healthy participants in the control group was 63.33 ± 9.22 years ( Supplementary Material, Table 1 ). Statistical analysis revealed no significant differences between the two groups in terms of age (P > 0.05). The results of data quality control were carefully evaluated to ensure the reliability of downstream transcriptomic analysis and are provided in the supplementary material ( Supplementary Material, Figure 1 - 11 , Tables 2 - 4 ).

In the ONFH patient group, 424 DElncRNAs (43 upregulated and 381 downregulated) ( Figures 2A–C ) and 1431 DEmRNAs (147 upregulated and 1284 downregulated) ( Figures 2D–F ) were identified, compared to the control group. These results indicate a substantial transcriptomic alteration associated with ONFH. Notably, among the differentially expressed mRNAs, several immune-related genes, including SPOP, TNF, CD22, CD1D, and others, were known to play key roles in immune cell activation and inflammatory regulation. Furthermore, ceRNA network analysis revealed that these immune genes are regulated by lncRNAs including MRPS30-DT, LINC01106, MIR100HG, PELATON, and WDR11-AS1 ( Figure 3C ). Detailed differential expression statistics of these lncRNA–mRNA pairs are provided in Supplementary Material, Table 5 .

Seven lncRNAs were selected for further analysis based on their FC value, Q-value, expression levels in the samples, and inclusion in starBase v2.0 and LncBase v3. Annotation details for the lncRNAs are summarized in Table 2 . These lncRNAs overlapped with 24 miRNAs predicted by starBase v2.0 and LncBase v3 ( Figure 4A ). 19,877 mRNAs were predicted as targets of the 24 miRNAs using miRWalk. By intersecting the lncRNA-predicted target mRNAs with 1,431 DEmRNAs from the plasma of ONFH patients, 683 target mRNAs were identified ( Figure 4B ). A ceRNA regulatory network was constructed, consisting of 7 lncRNAs, 24 miRNAs, and 683 mRNAs ( Supplementary Material, Table 6 ). This network reflects the potential post-transcriptional regulatory landscape mediated by lncRNAs in ONFH. KEGG analysis further revealed significant enrichment in fatty acid degradation, Wnt signaling pathways, NF-kappa B signaling pathways, Hedgehog signaling pathways, mTOR signaling pathways, PPAR signaling pathways, and fatty acid metabolism ( Figure 4C ). These pathways are associated with inflammation, lipid metabolism, and cell proliferation, suggesting their involvement in ONFH pathogenesis. GO analysis of these 683 mRNAs revealed significant enrichment in the biological process of protein modification by small protein conjugation or removal. The most enriched molecular functions included transcription coregulator activity, protein kinase binding, and kinase binding. The cellular component analysis revealed that the most enriched category was the external side of the plasma membrane ( Figure 4D ).

WGCNA was performed to identify gene modules significantly associated with MetS. A scale-free network was constructed using a soft threshold of β = 1, achieving a scale independence of R² = 0.9 ( Figures 5A, B ). A hierarchical clustering tree was generated, followed by dynamic tree cutting ( Figure 5C ), which resulted in eight distinct gene modules visualized as a heatmap ( Figure 5D ). The MEturquoise module (cor = 0.66, P = 6 × 10^-6) and the MEpurple module (cor = −0.64, P = 1 × 10^-5) showed the strongest positive and negative correlations with MetS, respectively. The MEturquoise module comprised 15,840 genes, indicating a potentially critical gene set for metabolic regulation.

A total of 506 overlapping mRNAs were identified by intersecting the mRNAs from the lncRNA ceRNA network with genes in the MetS-associated MEturquoise module ( Figure 6A ). KEGG pathway analysis indicated significant enrichment in the Hedgehog signaling pathway, fatty acid degradation, and Wnt signaling pathway ( Figure 6B ). These pathways are known to be involved in tissue development, lipid metabolism, and inflammatory signaling, and may contribute toxONFH pathogenesis through metabolic–immune crosstalk. GOxenrichment analysis revealed that the most significantly enriched biological process was protein modification by small protein conjugation or removal. The top molecular functions included acyltransferase activity and transcription coregulator activity. The most enriched cellular components included the ubiquitin ligase complex, cullin-RING ubiquitin ligase complex, and microtubule ( Figure 6C ).

A PPI network analysis was first conducted on 506 ONFH-MetS-related mRNAs ( Supplementary Material, Table 7 ), and hub genes with a degree ≥ 5 were visualized using Cytoscape v3.10.3 ( Figure 7A ). A key cluster of 15 mRNAs was further identified using MCODE ( Figure 7B ). The top 15 hub mRNAs were selected using the CytoHubba-MCC plugin ( Figure 7C ), and their intersection yielded 11 candidate mRNAs ( Figure 7D ), including BIRC5, KIF14, SEM1, SPOP, BTRC, CD1D, CD69, CDC6, TNF, CD22, and SDC1. These 11 mRNAs were regulated by six lncRNAs in the ceRNA network, namely LINC00630, MRPS30-DT, LINC01106, MIR100HG, PELATON, and WDR11-AS1. To construct a more robust diagnostic model, Random Forest was used to select the top five most important lncRNAs ( Figures 7E, F ), MRPS30-DT, LINC01106, MIR100HG, PELATON, and WDR11-AS1.

A nomogram incorporating five key lncRNAs was constructed ( Figure 8A ). Calibration curve analysis demonstrated minimal discrepancies between the predicted and observed risks ( Figure 8B ). The diagnostic performance of the model and each individual lncRNA was assessed by ROC analysis. The combined model demonstrated excellent discriminatory power, achieving an AUC of 1.00 on the training data ( Figure 8C ). To reduce overfitting and assess generalization, 3-fold cross-validation was performed, and the pooled cross-validated predictions were used to draw an overall ROC curve, yielding an AUC of 0.796 ( Figure 8D ). DCA analysis based on the same predictions showed a net clinical benefit over default strategies in the 0.4–0.8 threshold range ( Figure 8E ). The AUC values for the five lncRNAs were as follows: MRPS30_DT = 0.731, LINC01106 = 0.815, MIR100HG = 0.796, PELATON = 0.889, and WDR11_AS1 = 0.648 ( Figure 8F ). Although the combined model’s AUC was lower than that of some individual lncRNAs, this is likely due to the use of cross-validation, which provides a more realistic and conservative estimate. These findings support the robustness and clinical relevance of the multivariable model compared with individual biomarkers.

To further validate the accuracy of the integrated bioinformatics analysis, qRT-PCR was conducted to measure the expression levels of five candidate lncRNA diagnostic biomarkers in plasma samples from 45 patients with ONFH and 45 healthy controls. The results revealed that MRPS30-DT and LINC01106 were significantly upregulated in patients with ONFH, while MIR100HG exhibited an increasing trend. In comparison, PELATON and WDR11-AS1 show a downward trend ( Figure 9A ). These expression trends are consistent with those observed in the RNA sequencing data, confirming the robustness of the findings. The combined lncRNA model achieved an AUC greater than 0.7 ( Figure 9B ), surpassing the AUCs of individual lncRNAs ( Figure 9C ), indicating the potential clinical utility and translational value of this diagnostic model.

The CIBERSORT algorithm was employed to estimate the relative proportions of 22 immune cell types in each sample. Compared to the control group, the ONFH group exhibited elevated proportions of resting dendritic cells and monocytes, and a decreased proportion of M2 macrophages, indicating a substantial alteration in the immune microenvironment ( Figure 3A ). Further analysis revealed associations between the 11 hub mRNAs and immune cell infiltration. Specifically, BIRC5, BTRC, CD1D, CD22, KIF14, SEM1, SPOP, and TNF were correlated with various immune cell types, suggesting potential roles in immune regulation ( Figure 3B ). A ceRNA subnetwork comprising five lncRNAs and eight immune-related mRNAs was constructed ( Figure 3C ), highlighting possible lncRNA-mediated regulation of immune signaling pathways in ONFH.