ITGBL1 and RPN1 Mark a Fibrotic NP Subpopulation with Coupled Integrin Signaling and N-Glycosylation Programs in IVDD Authors: Shuo Qiang, Yunke Liu, Yonghui Dong, Cheng Cheng, Yifan Huang, Jia Zheng, Chao Tang, Siyu Zhao Published: 2026 ABSTRACT BackgroundFibroNPC is a degeneration associated subtype in intervertebral disc degeneration (IVDD), yet its markers and immune crosstalk remain poorly characterized. ITGBL1 and RPN1 recur as IVDD associated genes in bulk studies, but where they are expressed within the heterogeneous disc has not been resolved.MethodsThree GEO datasets comprising 81 bulk profiles were integrated and screened by LASSO, SVM-RFE, and Random Forest, with external validation in an independent 15 sample cohort. Two human IVDD scRNA-seq cohorts were analyzed by subclustering, signature scoring, and within scRNA-seq per-subpopulation contribution analysis. CellChat, AUCell pathway scoring, and drug prediction were used as exploratory analyses; IL-1β stimulated HNPCs served as the in vitro model.ResultsMachine learning analysis prioritized ITGBL1 and RPN1 as consensus core genes (combined AUC = 0.925), with their differential expression further supported in an independent external cohort. Single cell mapping preferentially localized both genes to a FibroNPC subpopulation and associated this cell state with coupled integrin/focal adhesion and ER protein processing/N-glycosylation programs. Notably, although FibroNPCs comprised less than one-fifth of NP/AF cells, they contributed approximately 48% of total ITGBL1 expression and disproportionate shares of several canonical fibrotic transcripts. CellChat further inferred increased macrophage SPP1–integrin communication involving structural cell populations in IVDD. In IL-1β stimulated HNPCs, both ITGBL1 and RPN1 were increased at the mRNA and protein levels.ConclusionITGBL1 and RPN1 represent FibroNPC associated molecular markers linked to coupled integrin signaling and N-glycosylation programs in IVDD. The disproportionate contribution of FibroNPCs to ITGBL1 and fibrotic gene expression highlights the importance of cell state resolved localization beyond conventional bulk hub-gene identification. Further functional validation is required before therapeutic implications can be established. Introduction Intervertebral disc degeneration (IVDD) is recognized as a major structural contributor to chronic low back pain,1 a pervasive musculoskeletal disorder that imposes healthcare burden on a global scale.2–4 The healthy intervertebral disc is organized into three components: a central, highly hydrated gelatinous nucleus pulposus (NP); an annulus fibrosus (AF) that concentrically surrounds the NP; and cartilaginous endplates (EP) that cover the NP tissue.5–7 The onset of IVDD is ultimately facilitated by the NP tissue becoming more fibrous, which is accompanied by a decrease in type II collagen and aggrecan.8,9 IVDD is fundamentally driven by a decline in NP cellularity and disruption of extracellular matrix (ECM) homeostasis.10 In healthy discs, NP cells maintain ECM balance through integrin mediated mechanotransduction that coordinates matrix synthesis and catabolism. During degeneration, protein-degrading of the ECM generates fibronectin fragments that bind integrins (eg, α5β1) on NP cells,11 activating ERK and NF-κB signaling and establishing a positive feedback loop that accelerates matrix degradation.12–14 Yet which cell subpopulations drive integrin-mediated matrix degradation, and whether endogenous mechanisms exist to counterbalance this catabolic cascade, remain unknown. Recent single cell transcriptomic studies have revealed substantial cellular heterogeneity within the NP, including a fibrotic NP cell subpopulation (FibroNPC) linked to ECM remodeling and terminal differentiation during degeneration. However, the molecular heterogeneity within this fibrotic compartment and the transcriptional programs that distinguish FibroNPCs from other NP cell states remain incompletely characterized. In particular, two critical questions remain. First, molecular markers defining FibroNPC identity across independent cohorts have not been systematically established. Second, the cellular source of bulk-level biomarkers such as ITGBL1 and RPN1—whether they are expressed uniformly across NP cells or restricted to a particular subpopulation—has not been resolved. In the present study, integrated bioinformatics screening prioritized integrin subunit beta-like 1 (ITGBL1) and ribophorin I (RPN1) as candidate genes for further investigation. ITGBL1 is a secreted EGF-like domain containing protein that modulates integrin activity. Previous studies have implicated ITGBL1 in fibroblast/myofibroblast remodeling and fibrogenesis.15–17 Notably, however, ITGBL1 has also shown chondroprotective effects in osteoarthritis.18 Because NP fibrosis is a central pathological feature of disc degeneration,19 whether ITGBL1 is associated with a protective or pro-fibrotic state in IVDD may depend on its specific cellular context. Macrophage mediated inflammation is also an important component of the IVDD microenvironment,20 providing a rationale for examining the relationship between these candidate genes and local immune remodeling. RPN1 is a component of the oligosaccharyltransferase complex and is involved in N-linked glycosylation and endoplasmic reticulum (ER) homeostasis,21 while ER stress itself has been implicated in NP cell dysfunction during IVDD.22 Recent studies have further associated RPN1 with immunosuppressive microenvironments and immune cell infiltration in several cancers.23,24 Critically, it remains unknown whether ITGBL1 and RPN1 are expressed broadly across NP cells or preferentially localized to specific subpopulations, leaving a fundamental gap in understanding how their bulk level dysregulation relates to pathological cell states. To address these gaps, we designed a multi-layered analytical strategy structured around three aims. First, we integrated a bulk transcriptomic meta cohort with two independent human IVDD scRNA-seq cohorts that span complementary stages of the disease spectrum, allowing a cross-stage comparison that no single existing dataset provides. Second, instead of relying on a single hub gene, we used three machine learning algorithms with external validation to define a two gene signature, ITGBL1 and RPN1, of the degenerative state. Third, by combining subcluster level expression analysis with CellChat inferred communication networks, we localized both genes to a FibroNPC subpopulation, characterized the associated integrin/focal-adhesion and ER protein-processing/N-glycosylation programs, and examined their relationships with inferred intercellular signaling during degeneration. Materials and Methods Data Acquisition and Processing The discovery meta cohort comprised GSE23130 (23 human annulus-disc tissue profiles), GSE56081 (5 control and 5 degenerative human nucleus pulposus samples), and GSE70362 (48 human annulus fibrosus/nucleus pulposus tissue profiles spanning degeneration grades), for 81 bulk profiles in total. GSE15227 (10 IVDD and 5 control samples) was retained as an independent external expression validation cohort. The single-cell cohorts were GSE199866 (four paired NP/AF preparations from non-degenerating and degenerating discs) and GSE244889 (seven NP samples). Identification of Differentially Expressed Genes Prior to differential analysis, 13 hemoglobin and erythrocyte-related genes (HBA1, HBA2, HBB, HBD, HBE1, HBG1, HBG2, HBM, HBQ1, HBZ, ALAS2, GYPA, GYPB) were excluded to minimize blood contamination artifacts. Differential expression analysis was performed using the limma R package. DEGs were defined as genes with Benjamini–Hochberg adjusted P values (FDR) < 0.05 and |log2FC| > 0.585 (corresponding to 1.5-fold change). Results were visualized with a volcano plot (ggplot2) and a hierarchical clustering heatmap of the top 100 DEGs (50 upregulated and 50 downregulated, ranked by P-value; pheatmap). Machine Learning Based Screening of Hub Genes The top 150 DEGs ranked by a composite score of −log10(P-value) × |log2FC| were used as the candidate pool for machine learning feature selection. All procedures were initialized with set.seed(123) for reproducibility. Three complementary algorithms were applied: (1) LASSO logistic regression (glmnet, binomial family) with 10-fold cross-validation to determine the optimal λ (lambda.min); (2) SVM-RFE (e1071, kernlab, caret) with a radial basis function kernel, 5-fold cross-validation, and feature subset sizes of 1–30; and (3) random forest (randomForest, 500 trees), retaining the top 20 genes by Mean Decrease in Gini index. Consensus hub genes were defined as the intersection of all three algorithms and visualized using a Venn diagram (ggVennDiagram). Construction and Evaluation of the Diagnostic Model Based on the identified core genes, a logistic regression-based diagnostic nomogram was constructed using the regplot R package, with kernel density and box plots overlaid to visualize predictor distributions. Predictive accuracy was evaluated by calibration curves (bootstrap resampling, B = 1000; rms), the Hosmer–Lemeshow test, and the Brier Score. ROC analysis assessed diagnostic efficacy, with AUC and 95% CIs reported for individual and combined gene models.25 Decision curve analysis (dcurves) evaluated net clinical benefit across threshold probabilities of 0.01–0.99. ROC, calibration, and decision-curve analyses were used to characterize the two-gene model within the integrated discovery meta-cohort. GSE15227 was used for independent expression validation. Functional Enrichment and Immune Infiltration Analysis Gene Set Enrichment Analysis (GSEA) was performed on the integrated bulk transcriptomic profiles using KEGG pathway gene sets via the clusterProfiler R package. Targeted Gene Ontology (GO) enrichment analysis was additionally performed for the two core genes (ITGBL1 and RPN1) to characterize their associated biological processes. To evaluate the local immune landscape, the single sample GSEA algorithm was used to quantify the relative abundances of 25 immune cell subsets in each sample, based on curated marker gene signatures from established immunological literature; signatures with fewer than 3 detected genes were excluded (Supplementary Table S1). Differences in immune cell ssGSEA scores between IVDD and control groups were assessed using the Wilcoxon rank-sum test. P values were adjusted across the 25 immune-cell signatures using the Benjamini–Hochberg procedure, and FDR < 0.05 was considered statistically significant. Exploratory calculation of the Spearman rank correlation coefficient between the expression level of core genes and the immune cell enrichment score. Single-Cell RNA Sequencing Analysis The two scRNA-seq datasets described in Data acquisition and processing were processed with Seurat v5. Quality control excluded cells with <200 or >4000 (GSE199866) or >5000 (GSE244889) detected genes, or excessive mitochondrial content (>15% and >20%, respectively, consistent with original protocols). After normalization, variable feature selection (top 2000 by VST), scaling, and PCA, batch effects were corrected with Harmony, followed by SNN clustering (dims = 1:15, resolution = 0.5) and UMAP visualization. Cell types were annotated using canonical markers: ACAN/SOX9/COL2A1/KRT19 (NP/AF cells), COL1A1/COL1A2/DCN/LUM/POSTN/PDGFRA (fibroblasts), and CD68/CD163/C1QA/TYROBP (macrophages) (Supplementary Figure 1). Core gene expression across cell types was assessed with the Bonferroni-corrected Wilcoxon rank-sum test and visualized with violin and UMAP feature plots. After initial annotation, NP and AF cells were retained as distinct anatomical compartments. They were jointly subclustered only to identify a shared degeneration associated structural cell state across cohorts. This approach was adopted because GSE199866 contains both NP and inner AF cells, whereas GSE244889 contains NP cells only. Tissue origin was retained throughout downstream analyses. The combined NP/AF subset was re-normalized, Harmony corrected, and subclustered (resolution = 0.6). Subcluster markers were identified using FindAllMarkers (min.pct = 0.25, logfc.threshold = 0.25). Proportional shifts between control and IVDD groups were visualized with bar plots; changes exceeding 5 percentage points were considered biologically notable Each cell was scored against five curated NPC signatures—Homeostatic NPC, FibroNPC, Inflammatory NPC, ER-stressed NPC (Supplementary Table S2), and Progenitor NPC—and the results were visualized as a subcluster by signature heatmap. The joint NP/AF analysis was used to characterize the shared FibroNPC like state by subclustering, signature scoring, pseudotime analysis, and cell communication analysis. To quantify the relative contribution of individual cell subpopulations to the aggregate expression signal, we performed a within scRNA-seq per-subpopulation contribution analysis. For each gene and group, the reconstructed aggregate expression was calculated as: Σ[Proportion (cell type, group) × Mean expression (gene, cell type, group)]. This analysis estimates the relative contribution of each subpopulation to the aggregate expression signal reconstructed from the scRNA-seq data and does not represent deconvolution against an independent bulk transcriptomic cohort. Layer Integration JoinLayers was applied to consolidate per-sample layers in the merged Seurat v5 object into a unified expression matrix prior to CellChat analysis. Pseudotime Trajectory Analysis To order NP/AF cells along a degeneration-associated trajectory, the NP/AF subset—with Seurat v5 layers merged (JoinLayers)—was converted to a SingleCellExperiment object and analyzed with Slingshot. Lineage structure and per-cell pseudotime were inferred on the UMAP embedding using the NP/AF subclusters as cluster labels. To anchor the trajectory at the healthiest state, the starting cluster was defined data-drivenly as the subcluster containing the highest proportion of control (non-degenerated) cells. The resulting pseudotime was used to display the distribution of NPC subtypes along the trajectory and the expression of ITGBL1 and RPN1 along pseudotime. Cell-Cell Communication Analysis CellChat v1.6 was used to infer intercellular communication networks with the human ligand–receptor database. Communication probabilities were estimated with the trimean method and filtered to retain interactions involving ≥5 cells per type. AUCell was applied to score pathway activity at single-cell resolution for key IVDD associated signaling pathways using gene sets from KEGG and Reactome (Supplementary Table S3). Drug Gene Interaction Prediction Drug gene interaction prediction was performed by querying the two core genes (ITGBL1 and RPN1) against the Drug Signatures Database (DSigDB) via the Enrichr platform.26,27 Candidate compounds with significant associations (P < 0.05) were identified and classified by pharmacological category. A drug–gene interaction network and a three-layer drug pathway target regulatory network were constructed to visualize relationships among candidate drugs, enriched pathways, and core target genes. Cell Culture and qRT-PCR Validation SV40 immortalized human nucleus pulposus cells (HNPCs; Cat. No. SAc0646, Aoyinbio, China) were maintained in nucleus pulposus cell specific culture medium (SAPM0646, Aoyinbio) at 37°C in a humidified atmosphere of 95% air and 5% CO2. For inflammatory stimulation, cells were treated with 10 ng/mL recombinant human IL-1β for 24 h. Total RNA was extracted using the SteadyPure RNA Extraction Kit (AG21024, Accurate Biology, Hunan, China) and reverse-transcribed using the Evo M-MLV RT Kit for qPCR (AG11707, Accurate Biology). qRT-PCR was performed with the SYBR Green Pro Taq HS Premix qPCR Kit (AG11701, Accurate Biology) to quantify mRNA levels of ITGBL1, RPN1, MMP13, and MMP3, with GAPDH as the internal reference. Relative expression was calculated using the 2−ΔΔCt method. qRT-PCR experiments were independently performed three times using separately cultured and IL-1 treated HNPCs, with technical triplicate measurements within each experiment. Mean Ct values from the technical replicates were used for ΔCt and 2−ΔΔCt calculations. Statistical significance was assessed by Student’s t-test (Supplementary Table S5). For protein level validation, total protein was extracted from control and IL-1β treated HNPCs using RIPA lysis buffer. Equal amounts of protein were separated by SDS-PAGE, transferred onto PVDF membranes, and probed with primary antibodies against ITGBL1 (1:4000, 30010-1-AP, Proteintech), RPN1 (1:1500, HA722373, HUABIO), and GAPDH (1:80000, ET1601-4, HUABIO) as a loading control. After incubation with HRP conjugated secondary antibodies, bands were visualized using an ECL detection system and quantified by densitometry with Image J software. Western blotting was independently performed three times. For each experiment, ITGBL1 and RPN1 band intensities were normalized to the corresponding GAPDH signal from the same samples and expressed relative to the control group. Bands used for quantification were identified according to their expected molecular weight regions. Statistical Analysis All statistical analyses were performed using R software (version 4.5.1). Continuous variables between two groups were compared using Student’s t-test or Wilcoxon rank-sum test, as appropriate. Correlations were evaluated using Spearman’s rank correlation coefficient. P-values were adjusted for multiple comparisons using the Benjamini-Hochberg method where applicable. A two-sided P value < 0.05 was considered statistically significant. Subcluster proportions, per-subpopulation expression contributions, pseudotime distributions, and AUCell pathway scores were used as descriptive single-cell measures. Because the number of biological donors was limited, individual cells were not treated as independent biological replicates for confidence-interval estimation or confirmatory hypothesis testing. Results Study Design and Analytical Workflow The overall study design is illustrated in Figure 1. Briefly, three bulk RNA-seq datasets (GSE23130, GSE56081, GSE70362) and two scRNA-seq cohorts (GSE199866, GSE244889) were retrieved from GEO. Bulk data were used for differential expression analysis, machine learning based feature selection, diagnostic modeling, and functional enrichment. Single-cell data were analyzed for cell type annotation, per-subpopulation contribution analysis, and intercellular communication. Core gene findings were further validated by drug prediction analysis and IL-1β stimulated HNPCs experiments at the mRNA and protein levels.Figure 1Overview of the study design and analytical workflow. Schematic of the integrated bulk and single cell transcriptomic strategy used to characterize ITGBL1 and RPN1 in intervertebral disc degeneration (IVDD). Three bulk microarray datasets (GSE23130, GSE56081, GSE70362), an independent external validation cohort (GSE15227), and two human single-cell RNA-sequencing (scRNA-seq) cohorts (GSE199866, GSE244889) were retrieved from the GEO database. Bulk data were used for differential expression analysis, machine-learning–based feature selection, and diagnostic modeling (modules 1–2), as well as functional enrichment and immune-infiltration analysis (module 3). Single-cell data were used for cell-type annotation, per-subpopulation contribution analysis, and intercellular communication inference (modules 4–5). Core-gene findings were further explored by computational drug prediction (module 6) and validated experimentally in IL-1β–stimulated human nucleus pulposus cells (HNPCs; module 7).A schematic of intervertebral disc degeneration study design with seven modules.A schematic illustrating the study design for intervertebral disc degeneration (IVDD) with seven modules. Centered is ′Intervertebral Disc Degeneration (IVDD)′ with GEO datasets: GSE70362, GSE23130, GSE56081, GSE15227, GSE199866, GSE244889. Arrows point to modules: 1) Bulk transcriptomics, showing heatmaps and plots. 2) Diagnostic model, with graphs and box plots. 3) Functional enrichment, featuring various charts. 4) Dual-cohort scRNA-seq analysis, displaying bar graphs and plots. 5) Subpopulation decomposition, with bar charts and heatmaps. 6) Drug prediction, showing bar graphs and network diagrams. 7) In vitro experimental validation (HNPCs), with bar graphs and protein bands. Overview of the study design and analytical workflow. Schematic of the integrated bulk and single cell transcriptomic strategy used to characterize ITGBL1 and RPN1 in intervertebral disc degeneration (IVDD). Three bulk microarray datasets (GSE23130, GSE56081, GSE70362), an independent external validation cohort (GSE15227), and two human single-cell RNA-sequencing (scRNA-seq) cohorts (GSE199866, GSE244889) were retrieved from the GEO database. Bulk data were used for differential expression analysis, machine-learning–based feature selection, and diagnostic modeling (modules 1–2), as well as functional enrichment and immune-infiltration analysis (module 3). Single-cell data were used for cell-type annotation, per-subpopulation contribution analysis, and intercellular communication inference (modules 4–5). Core-gene findings were further explored by computational drug prediction (module 6) and validated experimentally in IL-1β–stimulated human nucleus pulposus cells (HNPCs; module 7). Machine Learning Based Feature Selection and Identification of Core Genes To investigate transcriptomic alterations in IVDD, three independent microarray datasets were integrated. Initial PCA manifested substantial batch effects, with samples clustering mainly by dataset of origin (Figure 2A). After ComBat correction, post correction PCA demonstrated that samples were clearly separated into Control and IVDD groups based on biological phenotype (Figure 2B), providing a reliable meta cohort for later analyses.Figure 2Machine-learning–based feature selection identifies ITGBL1 and RPN1 among nine consensus core genes. (A and B) Principal component analysis of the three integrated bulk datasets before (A) and after (B) ComBat batch correction. Samples clustered mainly by dataset of origin before correction and separated into Control and IVDD groups by biological phenotype afterward. (C) Hierarchical clustering heatmap of the top 100 differentially expressed genes (DEGs; 50 up- and 50 down-regulated), annotated by batch and group; ITGBL1 and RPN1 are indicated (arrows). (D) Volcano plot of DEGs (limma; adjusted P < 0.05, |log2FC| > 0.585), with representative up-regulated genes labeled. (E and F) SVM-RFE feature selection showing five-fold cross-validation accuracy ((E); maximum 0.917) and error ((F); minimum 0.083) across feature-subset sizes. (G) Co-expression correlation matrix among the nine consensus core genes. (H) LASSO logistic regression: binomial deviance versus −log10(λ), with the optimal penalty (λ = 0.0232) retaining 14 features. (I) Random forest out-of-bag error stabilizing as the number of trees approached 500. (J) Venn diagram of the features shared by LASSO, SVM-RFE, and random forest, defining nine consensus core genes (CHST10, CYP1B1, TMTC1, EMILIN1, CCND1, ITGBL1, ST6GALNAC2, RPN1, MLPH).Image with PCA, heatmap, volcano plot, SVM-RFE, correlation, LASSO, random forest, Venn diagram.Image A: PCA pre-ComBat, samples by dataset. Image B: PCA post-ComBat, samples in Control and IVD groups. Image C: Heatmap of top 100 DEGs, annotated by batch and group. Image D: Volcano plot of DEGs with labeled up-regulated genes. Image E: SVM-RFE feature selection, five-fold cross-validation accuracy max 0.917. Image F: SVM-RFE error min 0.083. Image G: Co-expression matrix of nine consensus core genes. Image H: LASSO logistic regression, binomial deviance vs negative log10 lambda, optimal lambda 0.0232, 14 features retained. Image I: Random forest out-of-bag error stabilizes as trees approach 500. Image J: Venn diagram of features shared by LASSO, SVM-RFE and random forest, defining nine consensus core genes. Machine-learning–based feature selection identifies ITGBL1 and RPN1 among nine consensus core genes. (A and B) Principal component analysis of the three integrated bulk datasets before (A) and after (B) ComBat batch correction. Samples clustered mainly by dataset of origin before correction and separated into Control and IVDD groups by biological phenotype afterward. (C) Hierarchical clustering heatmap of the top 100 differentially expressed genes (DEGs; 50 up- and 50 down-regulated), annotated by batch and group; ITGBL1 and RPN1 are indicated (arrows). (D) Volcano plot of DEGs (limma; adjusted P < 0.05, |log2FC| > 0.585), with representative up-regulated genes labeled. (E and F) SVM-RFE feature selection showing five-fold cross-validation accuracy ((E); maximum 0.917) and error ((F); minimum 0.083) across feature-subset sizes. (G) Co-expression correlation matrix among the nine consensus core genes. (H) LASSO logistic regression: binomial deviance versus −log10(λ), with the optimal penalty (λ = 0.0232) retaining 14 features. (I) Random forest out-of-bag error stabilizing as the number of trees approached 500. (J) Venn diagram of the features shared by LASSO, SVM-RFE, and random forest, defining nine consensus core genes (CHST10, CYP1B1, TMTC1, EMILIN1, CCND1, ITGBL1, ST6GALNAC2, RPN1, MLPH). Differential expression analysis between the IVDD and Control groups identified a set of DEGs. The volcano plot (Figure 2D) spotlights prominently upregulated genes in degenerated discs, such as CYP1B1, ITGBL1, CFD and CCND1. Graded clustering of the top 100 most significant DEGs revealed a distinct transcriptomic contrast between the IVDD and Control cohorts (Figure 2C), indicating deep molecular reprogramming during IVDD. Three machine learning algorithms were applied for feature selection. LASSO logistic regression, using the optimal penalty parameter (λ = 0.0232) determined via 10-fold cross validation, extracted 14 feature genes (Figure 2H). SVM-RFE achieved a maximum cross-validation accuracy of 0.917 (Figure 2E and F). Random Forest analysis showed that the out-of-bag error rate stabilized as the number of decision trees approached 500 (Figure 2I). By intersecting the results of all three algorithms, 9 consensus core genes (CHST10, CYP1B1, TMTC1, EMILIN1, CCND1, ITGBL1, ST6GALNAC2, RPN1, and MLPH) were obtained (Figure 2J). The co-expression matrix among the 9 core genes confirmed coordinated regulatory relationships (Figure 2G). External Cohort Validation and Evaluation of the Clinical Diagnostic Model To assess the generalizability of the 9 consensus genes, we evaluated their mRNA expression levels in an independent external validation cohort. As shown in Figure 3E, while most candidate genes showed no statistical significance, ITGBL1 and RPN1 maintained differential expression between the Control and IVDD groups (P < 0.05). Consequently, ITGBL1 and RPN1 were selected as the final candidate biomarkers.Figure 3External validation and model performance of ITGBL1 and RPN1. (A) Logistic regression–based diagnostic nomogram estimating individualized IVDD probability from ITGBL1 and RPN1 expression, with kernel-density and box overlays of predictor distributions. ** P < 0.01 for the corresponding regression coefficients. (B) Calibration curve (1000 bootstrap resamples) comparing bias-corrected and ideal predictions; Hosmer–Lemeshow P = 0.452, Brier score = 0.124. (C) Receiver operating characteristic curves for ITGBL1 alone (AUC = 0.885, 95% CI 0.796–0.974), RPN1 alone (AUC = 0.824, 95% CI 0.701–0.947), and the combined model (AUC = 0.925, 95% CI 0.854–0.997). (D) Decision curve analysis showing the net clinical benefit of each model across threshold probabilities relative to treat-all and treat-none strategies. (E) Expression of the nine candidate core genes between Control and IVDD groups in the independent external cohort (GSE15227); only ITGBL1 and RPN1 retained significant differential expression (Wilcoxon rank-sum test, P < 0.05), whereas the remaining genes were non-significant (ns).A multi graph figure evaluating an IVD model using RPN1 and ITGBL1 and gene expression boxplots.Image A presents a nomogram predicting IVD using RPN1 and ITGBL1, with scales for points, RPN1, ITGBL1, total points and disease probability. An example shows total points of 92 and disease probability of 0.706. Image B displays a calibration curve with predicted vs actual IVD probabilities, showing Hosmer Lemeshow p=0.452 and Brier Score=0.124. The curves align with the ideal diagonal at mid-range probabilities. Image C illustrates ROC curves for sensitivity vs false positive rate, with the combined model (red) outperforming ITGBL1 (blue) and RPN1 (orange) alone. AUCs are 0.925, 0.885 and 0.824 respectively. Image D shows decision curve analysis, where the combined model (red) offers higher net benefit across thresholds compared to ITGBL1 (blue) and RPN1 (orange). Image E compares Log2 Expression Levels between Control and IVD groups across nine genes, with ITGBL1, RPN1 and CCND1 showing significant differences, while others show non-significant changes. External validation and model performance of ITGBL1 and RPN1. (A) Logistic regression–based diagnostic nomogram estimating individualized IVDD probability from ITGBL1 and RPN1 expression, with kernel-density and box overlays of predictor distributions. ** P < 0.01 for the corresponding regression coefficients. (B) Calibration curve (1000 bootstrap resamples) comparing bias-corrected and ideal predictions; Hosmer–Lemeshow P = 0.452, Brier score = 0.124. (C) Receiver operating characteristic curves for ITGBL1 alone (AUC = 0.885, 95% CI 0.796–0.974), RPN1 alone (AUC = 0.824, 95% CI 0.701–0.947), and the combined model (AUC = 0.925, 95% CI 0.854–0.997). (D) Decision curve analysis showing the net clinical benefit of each model across threshold probabilities relative to treat-all and treat-none strategies. (E) Expression of the nine candidate core genes between Control and IVDD groups in the independent external cohort (GSE15227); only ITGBL1 and RPN1 retained significant differential expression (Wilcoxon rank-sum test, P < 0.05), whereas the remaining genes were non-significant (ns). ROC curve analysis showed that ITGBL1 and RPN1 yielded individual AUC values of 0.885 (95% CI: 0.796–0.974) and 0.824 (95% CI: 0.701–0.947), respectively. The combined model achieved an AUC of 0.925 (95% CI: 0.854–0.997; Figure 3C). Given the limited size of the external validation cohort, these performance estimates should be interpreted cautiously and require confirmation in larger independent cohorts. A diagnostic nomogram was constructed to estimate individualized IVDD probability (Figure 3A). In the discovery meta-cohort, the calibration curve showed agreement between the apparent and bias-corrected predictions (Figure 3B), and the Brier Score was 0.124. Decision curve analysis indicated that the combined model provided net clinical benefit across a range of threshold probabilities (Figure 3D). Given the invasive nature of disc tissue sampling, these genes are therefore more appropriately framed as molecular correlates of disease state and candidates for further mechanistic investigation rather than established clinical diagnostic tools. Systemic Pathway Shifts and Immune Microenvironment Remodeling in IVDD To further elucidate the underlying biological mechanisms driving IVDD and the functional roles of the identified biomarkers, GSEA was performed on the integrated global transcriptomic profiles. As illustrated in Figure 4A, degenerated discs exhibited significant enrichment in pathways associated with ECM homeostasis, cellular stress, and immune regulation, notably Integrin signaling, N-Glycan biosynthesis, Protein processing in endoplasmic reticulum, protein export, and Antigen processing and presentation. The targeted functional enrichment analysis of the two core genes further corroborated these transcriptomic changes, with the enriched biological processes mainly involving integrin-mediated cell adhesion, collagen fibril organization, and glycoprotein metabolic and biosynthetic processes (Figure 4B). Additional enrichment terms related to protein glycosylation and glycoprotein metabolism are shown in Supplementary Figure 2. These results indicate that the progression of IVDD is tightly coupled with integrin regulated matrix remodeling.Figure 4Pathway-level reprogramming and immune-microenvironment remodeling in IVDD. (A) Gene set enrichment analysis (GSEA; KEGG) of the integrated bulk transcriptome. Degenerated discs were enriched for Integrin signaling, N-Glycan biosynthesis, Protein processing in the endoplasmic reticulum, Protein export, and Antigen processing and presentation (P and adjusted-P values inset). (B) Targeted Gene Ontology enrichment of the two core genes (ITGBL1 and RPN1), highlighting integrin-mediated cell adhesion, collagen fibril organization, and glycoprotein metabolic and biosynthetic processes (bar color, adjusted P). (C) Single-sample GSEA estimates of the relative abundance of 25 immune-cell subsets in Control versus IVDD discs; activated CD4 T cells were significantly increased in IVDD (Wilcoxon rank-sum test; P < 0.05). (D) Spearman correlations between ITGBL1/RPN1 expression and immune-cell abundance; both genes correlated positively with myeloid-derived suppressor cells and neutrophils and inversely with selected adaptive subsets, including effector-memory CD4 T cells. Asterisks indicate statistical significance: *P < 0.05, **P < 0.01, and ***P < 0.001.A multi plot figure with a line plot, a bar chart, grouped boxplots and a correlation heatmap.The image A showing a line graph with x axis label, Rank in Ordered Dataset, ranging from 0 to 10000 and y axis label, Running Enrichment Score, ranging from negative 0.4 to positive 0.6. Five curves are labeled, Antigen processing and presentation, Integrin signaling, N Glycan biosynthesis, Protein export and Protein processing in endoplasmic reticulum. A table lists pvalue and p adjust: Antigen processing and presentation pvalue 1.349e minus 03, p adjust 0.03900; Integrin signaling pvalue 3.006e minus 03, p adjust 0.02688; N Glycan biosynthesis pvalue 7.964e minus 04, p adjust 0.03515; Protein export pvalue 1.184e minus 03, p adjust 0.03900; Protein processing in endoplasmic reticulum pvalue 9.633e minus 05, p adjust 0.01114. The curves rise early then decline toward 0 by rank 10000. The image B showing a horizontal bar chart titled, Functional Enrichment of 2 Core Genes. The x axis label is Count, ranging from 0 to 4. The y axis lists terms: cell adhesion mediated by integrin, glycoprotein biosynthetic process, glycoprotein metabolic process, protein glycosylation, macromolecule glycosylation, estrogen metabolic process, glycosylation, collagen fibril organization, protein O linked glycosylation. Bar lengths are approximately 3.6, 4.0, 4.0, 3.0, 3.0, 2.0, 3.0, 2.0, 2.0 respectively. A legend label reads p adjust. The image C showing grouped horizontal boxplots with legend labels, Control and IVD. The x axis label is ssGSEA Enrichment Score, ranging from negative 1.0 to positive 0.5. The y axis label is Cell Type. Cell types listed include Macrophage, Effector memory CD4 T cell, Monocyte, Activated dendritic cell, Type 17 T helper cell, MDSC, Plasmacytoid dendritic cell, Regulatory T cell, Type 2 T helper cell, Effector memory CD8 T cell, Mast cell, Immature dendritic cell, Natural killer T cell, Activated B cell, Central memory CD4 T cell, Neutrophil, T follicular helper cell, Activated CD8 T cell, Activated CD4 T cell, Eosinophil, Central memory CD8 T cell, Natural killer cell. Boxplots for each cell type show paired distributions for Control and IVD, with an outlier point near positive 0.4 on the Activated CD4 T cell row. The image D showing a heatmap with x axis gene labels, ITGBL1 and RPN1 and y axis label, Immune Cell Types. The color bar is labeled Correlation, ranging from negative 1.0 to positive 1.0. Rows include Type 2 T helper cell, Type 17 T helper cell, T follicular helper cell, Regulatory T cell, Plasmacytoid dendritic cell, Neutrophil, Natural killer T cell, Natural killer cell, Monocyte, MDSC, Mast cell, Macrophage, Immature dendritic cell, Eosinophil, Effector memory CD8 T cell, Effector memory CD4 T cell, Central memory CD8 T cell, Central memory CD4 T cell, Activated dendritic cell, Activated CD8 T cell, Activated CD4 T cell, Activated B cell. Several cells contain asterisks, including single, double and triple asterisks. Pathway-level reprogramming and immune-microenvironment remodeling in IVDD. (A) Gene set enrichment analysis (GSEA; KEGG) of the integrated bulk transcriptome. Degenerated discs were enriched for Integrin signaling, N-Glycan biosynthesis, Protein processing in the endoplasmic reticulum, Protein export, and Antigen processing and presentation (P and adjusted-P values inset). (B) Targeted Gene Ontology enrichment of the two core genes (ITGBL1 and RPN1), highlighting integrin-mediated cell adhesion, collagen fibril organization, and glycoprotein metabolic and biosynthetic processes (bar color, adjusted P). (C) Single-sample GSEA estimates of the relative abundance of 25 immune-cell subsets in Control versus IVDD discs; activated CD4 T cells were significantly increased in IVDD (Wilcoxon rank-sum test; P < 0.05). (D) Spearman correlations between ITGBL1/RPN1 expression and immune-cell abundance; both genes correlated positively with myeloid-derived suppressor cells and neutrophils and inversely with selected adaptive subsets, including effector-memory CD4 T cells. Asterisks indicate statistical significance: *P < 0.05, **P < 0.01, and ***P < 0.001. Given the prominent enrichment of antigen presentation pathways accompanied by ECM degradation, we systematically analyzed the local immune infiltration profile utilizing the single sample GSEA algorithm. The overall immune profile showed distributional differences between the control and IVDD groups (Figure 4C). Nominal and Benjamini–Hochberg adjusted P values for all 25 immune cell signatures are provided in Supplementary Table S4. To explore the potential regulatory crosstalk between the structural/metabolic biomarkers and local inflammation, a correlation matrix was constructed between the expression of the core genes and infiltrating immune cell abundances (Figure 4D). Importantly, both ITGBL1 and RPN1 showed robust positive correlations with innate myeloid populations, particularly myeloid derived suppressor cells and neutrophils. In contrast, they demonstrated meaningful inverse associations with distinct adaptive immune subsets, such as effector memory CD4 T cells. In short, these findings show that in addition to their normative role in structural and metabolic regulation, the expression of ITGBL1 and RPN1 is closely related to the interaction between ECM degradation and local immune microenvironment remodeling during disc degeneration. Single-Cell Analysis Localizes ITGBL1 and RPN1 to a Fibrotic NPC Subpopulation To pinpoint the cellular origin of ITGBL1 and RPN1, we analyzed two independent human IVDD scRNA-seq cohorts spanning complementary disease stages. NP/AF cells from each cohort were subclustered, and in Cohort I the recovered NPC states aligned along a continuous differentiation trajectory that separated endplate associated, fibrotic, homeostatic, regulatory, and reparative NPCs, while subpopulation composition shifted appreciably between control and degenerated discs, indicating that degeneration reorganizes the NPC landscape rather than uniformly altering a single transcriptional state (Figures 5A and B). Across the Cohort I NPC subtypes, both genes were concentrated in the fibrotic, homeostatic and regulatory NPCs and near absent from endplate-associated and reparative cells, marking the fibrotic-leaning compartment as their principal expression niche (Figure 5C and D). ITGBL1 in particular showed its fullest, most upward shifted distribution in the fibrotic NPCs of degenerated discs, while RPN1 was more broadly distributed across NPC states. Among the eight Cohort I subclusters, Sub_2 carried the strongest fibrotic ECM program—high COL1A1, COL1A2, COL3A1, FN1, POSTN and LOXL1/2—setting it apart from the homeostatic, COL2A1/ACAN/SOX9-leaning subclusters (Figure 5E).Figure 5Single-cell analysis localizes ITGBL1 and RPN1 to a fibrotic NP subpopulation (Cohort I, GSE199866). (A) Subpopulation composition (Sub_0–Sub_7) of NP/AF cells in Control versus IVDD discs. (B) Pseudotime density distribution across NPC states (endplate-associated, fibrotic, homeostatic, regulatory, and reparative NPCs). (C and D) Violin plots of ITGBL1 (C) and RPN1 (D) expression across NPC states in Control and IVDD; ITGBL1 showed its most upward-shifted distribution in fibrotic NPCs of degenerated discs, whereas RPN1 was more broadly distributed. (E) Dot plot of fibrotic versus homeostatic ECM markers across subclusters; Sub_2 carried the strongest fibrotic program (high COL1A1, COL1A2, COL3A1, FN1, POSTN, LOXL1/2), distinguishing it from the COL2A1/ACAN/SOX9-leaning homeostatic subclusters (dot size, percentage expressed; color, scaled mean expression). (F and G) UMAP feature plots of ITGBL1 (F) and RPN1 (G) over the same fibrotic territory. (H) Detection rate of ITGBL1 and RPN1 in FibroNPC (Sub_2) versus other NPCs (ITGBL1, 34.2% vs 10.7%; RPN1, 45.0% vs 19.7%). (I) Per-subpopulation contribution analysis of ECM-gene contribution in Control discs; Sub_2 (red) supplied a disproportionate share of ITGBL1 (48%), COL1A1 (65%), POSTN (77%), LOXL1 (69%), and RPN1 (29%). (J and K) ITGBL1 (J) and RPN1 (K) expression along pseudotime, showing higher expression toward the fibrotic end of the inferred trajectory.Infographic on NP/AF cell subpopulations and gene expression differences in Control vs IVDD discs.The infographic is a grid of panels A to K analyzing NP/AF cell subpopulations and gene expression in Control versus IVDD discs. A shows subpopulation composition, highlighting Sub2′s increase in IVDD. B depicts pseudotime density, with fibrotic NPCs peaking. C and D are violin plots showing ITGBL1 and RPN1 expression, higher in fibrotic NPCs. E is a dot plot comparing fibrotic vs homeostatic ECM markers, with Sub2 showing strong fibrotic markers. F and G are UMAP plots for ITGBL1 and RPN1. H shows detection rates: ITGBL1 at 34.2 percent and RPN1 at 45.0 percent in FibroNPCs. I details ECM gene contributions, with Sub_2 contributing 48 percent to 77 percent for various genes. J and K show ITGBL1 and RPN1 expression along pseudotime, increasing towards fibrotic states. The infographic highlights differences in cell behavior and gene expression between healthy and degenerated discs. Single-cell analysis localizes ITGBL1 and RPN1 to a fibrotic NP subpopulation (Cohort I, GSE199866). (A) Subpopulation composition (Sub_0–Sub_7) of NP/AF cells in Control versus IVDD discs. (B) Pseudotime density distribution across NPC states (endplate-associated, fibrotic, homeostatic, regulatory, and reparative NPCs). (C and D) Violin plots of ITGBL1 (C) and RPN1 (D) expression across NPC states in Control and IVDD; ITGBL1 showed its most upward-shifted distribution in fibrotic NPCs of degenerated discs, whereas RPN1 was more broadly distributed. (E) Dot plot of fibrotic versus homeostatic ECM markers across subclusters; Sub_2 carried the strongest fibrotic program (high COL1A1, COL1A2, COL3A1, FN1, POSTN, LOXL1/2), distinguishing it from the COL2A1/ACAN/SOX9-leaning homeostatic subclusters (dot size, percentage expressed; color, scaled mean expression). (F and G) UMAP feature plots of ITGBL1 (F) and RPN1 (G) over the same fibrotic territory. (H) Detection rate of ITGBL1 and RPN1 in FibroNPC (Sub_2) versus other NPCs (ITGBL1, 34.2% vs 10.7%; RPN1, 45.0% vs 19.7%). (I) Per-subpopulation contribution analysis of ECM-gene contribution in Control discs; Sub_2 (red) supplied a disproportionate share of ITGBL1 (48%), COL1A1 (65%), POSTN (77%), LOXL1 (69%), and RPN1 (29%). (J and K) ITGBL1 (J) and RPN1 (K) expression along pseudotime, showing higher expression toward the fibrotic end of the inferred trajectory. UMAP feature plots placed ITGBL1 and RPN1 expression over this same fibrotic territory (Figure 5F and G), and quantification confirmed the enrichment: the two genes were detected in 34.2% and 45.0% of Sub_2 cells versus 10.7% and 19.7% of the remaining NPCs (Figure 5H). A per-subpopulation contribution analysis showed how disproportionate this was—although the fibrotic subcluster made up under one-fifth of NP/AF cells in control discs, it supplied roughly half of total ITGBL1 expression (48%) and the dominant share of canonical fibrotic transcripts (COL1A1 65%, POSTN 77%, LOXL1 69%), along with 29% of RPN1 (Figure 5I). Plotted along pseudotime, both genes peaked as cells advanced into the fibrotic state (Figure 5J and K). These data identify a fibrotic NPC subpopulation (Sub_2) as a preferential expression niche for ITGBL1 and RPN1 within the analyzed cohort and motivated a focused dissection of its transcriptional program. ITGBL1 and RPN1 are Associated with Coupled Integrin-Signaling and N-Glycosylation Programs in FibroNPCs Scoring each Cohort I subcluster against canonical NPC signatures formally identified Sub_2 as the FibroNPC, reaching the maximum FibroNPC score (1.00) while scoring low for homeostatic, inflammatory, and progenitor states (Figure 6A). Sub_2 was also the only subcluster jointly elevated on two functional modules, scoring highest for both an integrin/focal-adhesion module (0.25, well above every other subcluster) and an ER protein-processing/N-glycosylation module (0.18), whereas the remaining states stayed near zero or negative on at least one axis (Figure 6B). ITGBL1 and RPN1 mirrored this pattern, expressed most strongly in Sub_2 (Figure 6C). ITGBL1 and RPN1 therefore map to the two major transcriptional programs of the FibroNPC, supporting their association with this cell state.Figure 6ITGBL1 and RPN1 are associated with coupled integrin-signaling and N-glycosylation programs in FibroNPCs. (A) Heatmap of canonical NPC signature scores across subclusters; Sub_2 reached the maximum FibroNPC score (1.00) while scoring low for homeostatic, inflammatory, and progenitor states, supporting the annotation of Sub_2 as the FibroNPC subpopulation. (B) Module scores for an integrin/focal-adhesion program and an ER protein-processing/N-glycosylation program; Sub_2 was the only subcluster jointly elevated on both axes (0.25 and 0.18, respectively). (C) Violin plots confirming the strongest ITGBL1 and RPN1 expression in Sub_2. (D and E) Gene Ontology biological-process (D) and KEGG (E) enrichment of the FibroNPC identity program, resolving into a matrix arm (collagen fibril organization, extracellular matrix organization, cell-substrate adhesion; Focal adhesion and Integrin signaling) and a secretory arm (protein N-linked glycosylation, response to ER stress; Protein processing in the endoplasmic reticulum and N-Glycan biosynthesis). (F) Inferred ligand–receptor interactions in FibroNPC versus other subclusters; ECM-to-integrin pairs including FN1→ITGA5, COL1A2→ITGB1, CTGF→TGFBR2, and ITGBL1→ITGA5 were strongest in the FibroNPC. (G) GSEA (biological process) of FibroNPC in IVDD versus Control, showing activation of collagen metabolism, ECM organization, and angiogenesis, with suppression of cytoplasmic translation and ribosome biogenesis. (H) Volcano plot of FibroNPC transcriptional reprogramming during degeneration (IVDD vs Control, within Sub_2); COL1A1, COL1A2, COL3A1, POSTN, FN1, TGFBI, CTHRC1, and SPP1 were markedly induced.Charts and plots analyze FibroNPC subclusters and degeneration.Image A is a heatmap. The x-axis shows subclusters Sub0 through Sub7. The y-axis lists NPC signature types: ProgenitorNPC, InflammatoryNPC, HomeostaticNPC, FibroNPC and ERstressedNPC. The mean score scale ranges from negative 0.5 to 1.0. Sub2 scores highest for FibroNPC at 1.00, while scoring low for homeostatic (negative 0.02), inflammatory (negative 0.11) and progenitor (0.16) states. Image B is a heatmap. The x-axis shows Sub0 through Sub7. Two module scores are shown: integrin slash focal adhesion and ER processing slash N-glycan. Sub2 scores highest on both at 0.25 and 0.18 respectively; all other subclusters score near zero or negative on at least one axis. Image C shows violin plots. The x-axis lists subclusters Sub0 through Sub7. The y-axis shows expression level from 0.0 to 2.0. ITGBL1 and RPN1 both peak in Sub_2. Image D is a horizontal bar chart. The x-axis shows count from 0 to 100 plus. The y-axis lists Gene Ontology biological process terms. Glycoprotein metabolic process and Golgi vesicle transport show the highest counts near 100. Image E is a horizontal bar chart. The x-axis shows count from 0 to 80. The y-axis lists KEGG pathways. Protein processing in endoplasmic reticulum and focal adhesion show the highest counts near 80. Image F is a heatmap of ligand-receptor interactions. Rows show pairs including FN1 to ITGA5, COL1A2 to ITGB1, POSTN to ITGAV, COL3A1 to ITGA2, CTGF to TGFBR2 and ITGBL1 to ITGA5. FibroNPC column shows the highest interaction scores, with CTGF to TGFBR2 at 0.580 and FN1 to ITGA5 at 1.200 being the strongest. Image G is a dot plot titled GSEA biological process: FibroNPC activated versus suppressed. The x-axis shows gene ratio from 0.4 to 0.8. The y-axis lists biological process terms. Activated processes include cytoplasmic translation, ribosomal biogenesis and collagen metabolic process. Suppressed processes include cytokine production and necrosis factor production. Dot size encodes count up to 150 and shading encodes adjusted p-value. Image H is a volcano plot titled FibroNPC transcriptional reprogramming during degeneration. The x-axis shows log subscript 2 fold change from negative 4 to positive 4. The y-axis shows negative log subscript 10 adjusted p-value from 0 to above 200. Markedly upregulated genes include TGFBI, CTHRC1, SPP1, FN1, COL1A2, COL3A1, COL1A1 and POSTN. Three categories are shown: down-regulated, not significant and up-regulated. ITGBL1 and RPN1 are associated with coupled integrin-signaling and N-glycosylation programs in FibroNPCs. (A) Heatmap of canonical NPC signature scores across subclusters; Sub_2 reached the maximum FibroNPC score (1.00) while scoring low for homeostatic, inflammatory, and progenitor states, supporting the annotation of Sub_2 as the FibroNPC subpopulation. (B) Module scores for an integrin/focal-adhesion program and an ER protein-processing/N-glycosylation program; Sub_2 was the only subcluster jointly elevated on both axes (0.25 and 0.18, respectively). (C) Violin plots confirming the strongest ITGBL1 and RPN1 expression in Sub_2. (D and E) Gene Ontology biological-process (D) and KEGG (E) enrichment of the FibroNPC identity program, resolving into a matrix arm (collagen fibril organization, extracellular matrix organization, cell-substrate adhesion; Focal adhesion and Integrin signaling) and a secretory arm (protein N-linked glycosylation, response to ER stress; Protein processing in the endoplasmic reticulum and N-Glycan biosynthesis). (F) Inferred ligand–receptor interactions in FibroNPC versus other subclusters; ECM-to-integrin pairs including FN1→ITGA5, COL1A2→ITGB1, CTGF→TGFBR2, and ITGBL1→ITGA5 were strongest in the FibroNPC. (G) GSEA (biological process) of FibroNPC in IVDD versus Control, showing activation of collagen metabolism, ECM organization, and angiogenesis, with suppression of cytoplasmic translation and ribosome biogenesis. (H) Volcano plot of FibroNPC transcriptional reprogramming during degeneration (IVDD vs Control, within Sub_2); COL1A1, COL1A2, COL3A1, POSTN, FN1, TGFBI, CTHRC1, and SPP1 were markedly induced. Functional enrichment of the FibroNPC identity program resolved into two coordinated arms (Figure 6D and E). One arm centered on matrix biology—collagen fibril organization, extracellular matrix organization, and cell-substrate adhesion at the GO level, and Focal adhesion and Integrin signaling at the KEGG level—the signaling context in which ITGBL1, an integrin-modulating ECM protein, operates. The other centered on secretory protein maturation—glycoprotein metabolic process, protein N-linked glycosylation, and response to ER stress, with Protein processing in the endoplasmic reticulum as the leading KEGG term and N-Glycan biosynthesis close behind—the machinery in which RPN1, a core OST subunit, functions. Ligand-receptor analysis reinforced the matrix arm: ECM-to-integrin pairs including COL1A2→ITGB1, FN1→ITGA5, CTGF→TGFBR2, and ITGBL1→ITGA5 were strongest in the FibroNPC relative to other subclusters (Figure 6F). During degeneration the FibroNPC sharply intensified its fibrotic program. Differential testing of FibroNPC between degenerated and control discs showed pronounced induction of COL1A1, COL1A2, COL3A1, POSTN, FN1, TGFBI, CTHRC1, and SPP1 (Figure 6H), and GSEA over the same comparison activated extracellular matrix organization, collagen metabolism, and angiogenesis while suppressing cytoplasmic translation and ribosome-biogenesis programs (Figure 6G). Given its disproportionate contribution to ITGBL1 and several fibrotic transcripts within the analyzed cohort, the FibroNPC exhibits an increasingly fibrotic, integrin engaged transcriptional profile during degeneration, supporting an association of this subpopulation with active disc degeneration and providing the context for the intercellular signaling examined next. Intercellular Communication Rewiring During IVDD To examine the immune microenvironment around ITGBL1/RPN1, and because macrophages were captured only in Cohort I, we applied CellChat to this cohort. Given the low abundance of macrophages (~0.3% of recovered cells), the macrophage related communication patterns described below should be interpreted as exploratory. The total number of inferred interactions increased from 125 (Control) to 180 (IVDD), and the overall interaction strength rose from 3.735 to 4.625 (Figure 7A), reflecting an amplified signaling complexity in degenerating discs.Figure 7CellChat inferred intercellular communication changes during IVDD (Cohort I). (A) Total number of inferred interactions (125 → 180) and overall interaction strength (3.735 → 4.625) in Control versus IVDD. (B) Inferred ligand–receptor pairs with altered communication probabilities from macrophages to structural cells in IVDD, including SPP1–CD44, multiple integrin-mediated SPP1 pairs [SPP1–(ITGAV+ITGB5), SPP1–(ITGAV+ITGB1), SPP1–(ITGA5+ITGB1)], TNFSF12–TNFRSF12A, and NAMPT–(ITGA5+ITGB1) (dot size, P value; color, communication probability). (C) Relative information flow per signaling pathway in Control versus IVDD; FN1, TENASCIN, NCAM, and HSPG remained dominant in both conditions, whereas SPP1, MHC-II, THY1, and others were preferentially active in IVDD. (D) Differential interaction-strength network (IVDD vs Control) among fibroblasts, macrophages, and NP/AF cells; macrophage-directed signaling was the most enhanced component (red, increased; blue, decreased). (E and F) Two-dimensional signaling-role plots (incoming versus outgoing interaction strength) in Control (E) and IVDD (F); macrophages shifted from minimal to substantial engagement upon degeneration, while NP/AF cells showed contracted involvement.CellChat multi plot: interaction counts, strengths, ligand-receptor dots, flows, networks, roles.The image A showing two bar charts. Left bar chart title, Total Number of Interactions. X axis labels, Control and IVD. Y axis label, Number of inferred interactions, with range 0 to 180. Bar values: Control 125, IVD 180. Right bar chart title, Total Interaction Strength. X axis labels, Control and IVD. Y axis label, Interaction strength, with range 0 to 4.625. Bar values: Control 3.735, IVD 4.625. The image B showing a dot plot titled, Altered L R Pairs from Macrophages to Structural Cells. X axis categories: Macrophages to Fibroblasts in IVD and Macrophages to NP AF Cells in IVD. Y axis categories: TNFSF12 minus TNFRSF12A; SPP1 minus CD44; SPP1 minus left parenthesis ITGAV plus ITGB5 right parenthesis; SPP1 minus left parenthesis ITGAV plus ITGB1 right parenthesis; SPP1 minus left parenthesis ITGA5 plus ITGB1 right parenthesis; PPIA minus BSG; PGE2 minus PTGES3 minus PTGER2; NAMPT minus left parenthesis ITGA5 plus ITGB1 right parenthesis. Dots appear in both x axis columns for TNFSF12 minus TNFRSF12A and SPP1 minus CD44 and single dots appear for the remaining rows. A size legend labeled, p value and a vertical scale labeled, Communication prob, with endpoints, min and max. The image C showing a horizontal stacked bar chart titled, Signaling Pathway Information Flow Shift. X axis label, Relative information flow, with range 0.00 to 1.00. Y axis categories listed top to bottom: FN1, TENASCIN, NCAM, HSPG, GAS, PTN, TGFb, PDGF, CypA, COLLAGEN, ANGPTL, VISFATIN, FLRT, PERIOSTIN, CD46, CADM, MPZ, COMPLEMENT, NOTCH, EGF, ICAM, RANKL, APP, THY1, GAP, LAMININ, MHC II, Prostaglandin, TWEAK, ADGRE, THBS, IGFBP, CD99, MIF, SPP1. Legend labels: Control and IVD. Bars show varying splits between Control and IVD across pathways, with many lower pathways dominated by IVD. The image D showing a directed network diagram titled, Differential Interaction Strength, with subtitle, IVD vs Control. Three nodes labeled, Fibroblasts, Macrophages and NP AF Cells. Curved connections link nodes in both directions, including self loops on Fibroblasts and NP AF Cells. Multiple arcs connect Fibroblasts to Macrophages and NP AF Cells to Macrophages and arcs connect Fibroblasts and NP AF Cells. The image E showing a scatter plot titled, Control. X axis label, Outgoing interaction strength, with range 0.0 to 2.5. Y axis label, Incoming interaction strength, with range 0 to 3. A bubble size legend labeled, Count, with values 0, 25, 50, 75, 100, 125. Three labeled points: Macrophages near left bottom around x 0.0 and y 0.0; NP AF Cells around x 1.2 and y 0.6; Fibroblast around x 2.3 and y 3.2. The image F showing a scatter plot titled, IVD. X axis label, Outgoing interaction strength, with range 1.0 to 2.5. Y axis label, Incoming interaction strength, with range 1.1 to 2.1. A bubble size legend labeled, Count, with values 80, 100, 120, 140. Three labeled points: NP AF Cells near x 1.0 and y 1.15; Macrophages around x 1.2 and y 1.35; Fibroblast around x 2.4 and y 2.1. CellChat inferred intercellular communication changes during IVDD (Cohort I). (A) Total number of inferred interactions (125 → 180) and overall interaction strength (3.735 → 4.625) in Control versus IVDD. (B) Inferred ligand–receptor pairs with altered communication probabilities from macrophages to structural cells in IVDD, including SPP1–CD44, multiple integrin-mediated SPP1 pairs [SPP1–(ITGAV+ITGB5), SPP1–(ITGAV+ITGB1), SPP1–(ITGA5+ITGB1)], TNFSF12–TNFRSF12A, and NAMPT–(ITGA5+ITGB1) (dot size, P value; color, communication probability). (C) Relative information flow per signaling pathway in Control versus IVDD; FN1, TENASCIN, NCAM, and HSPG remained dominant in both conditions, whereas SPP1, MHC-II, THY1, and others were preferentially active in IVDD. (D) Differential interaction-strength network (IVDD vs Control) among fibroblasts, macrophages, and NP/AF cells; macrophage-directed signaling was the most enhanced component (red, increased; blue, decreased). (E and F) Two-dimensional signaling-role plots (incoming versus outgoing interaction strength) in Control (E) and IVDD (F); macrophages shifted from minimal to substantial engagement upon degeneration, while NP/AF cells showed contracted involvement. Comparative analysis of the relative information flow across signaling pathways revealed that FN1, TENASCIN, NCAM, HSPG, and COLLAGEN remained the most dominant signaling axes in both conditions (Figure 7C). Notably, PTN, PDGF, FN1, and ANGPTL pathways exhibited increased information flow in the IVDD group, while pathways such as THY1, MHC-II, VISFATIN, and SPP1 were preferentially active in degenerated discs, consistent with an inflammatory and pro-fibrotic microenvironmental shift. The differential interaction strength analysis revealed that macrophage directed signaling was the most dramatically enhanced component in the IVDD group (Figure 7D). At the ligand-receptor level, the most prominent altered L-R pairs from macrophages to structural cells included TNFSF12-TNFRSF12A and SPP1-CD44 (macrophage-to-fibroblast), as well as multiple integrin-mediated SPP1 pairs (SPP1-ITGAV+ITGB5, SPP1-ITGAV+ITGB1, SPP1-ITGA5+ITGB1) and NAMPT-(ITGA5+ITGB1) (macrophage-to-NP/AF cells) (Figure 7B). These integrin dependent interactions are particularly noteworthy given the molecular function of ITGBL1 as a known modulator of integrin signaling.18 Taken together, these CellChat derived findings suggest a potential association between macrophage SPP1–integrin communication and the ITGBL1 enriched FibroNPC compartment. Macrophage-derived SPP1 was predicted to communicate with NP/AF cells through multiple integrin receptor pairs, while ITGBL1 expression was enriched in FibroNPCs. This overlap suggests that the ITGBL1 enriched FibroNPC compartment may be involved in macrophage-associated integrin signaling, although this relationship remains computationally inferred. The two dimensional signaling-role analysis further showed changes in inferred communication patterns during degeneration (Figure 7E and F): macrophages shifted from minimal to greater signaling involvement (incoming strength ~1.5, outgoing ~1.0), whereas NP/AF cells showed reduced overall signaling involvement. Cell Type Resolved Pathway Activity, Computational Drug Prediction, and in vitro Validation To bridge the single-cell transcriptomic findings with potential therapeutic implications, we first assessed the activity of IVDD associated signaling pathways at single-cell resolution using the AUCell algorithm. Violin plots revealed cell type and condition-dependent activity patterns (Figure 8A): ECM receptor interaction and collagen fibril organization were most active in NP/AF cells, integrin signaling and TGF-β signaling predominated in fibroblasts, and protein processing in the endoplasmic reticulum and unfolded protein response pathways were enriched in macrophages. The aggregated AUCell heatmap further confirmed that, upon degeneration, NP/AF cells exhibited upregulated integrin signaling and ECM receptor interaction activity, while macrophages displayed enhanced NF-κB signaling and inflammatory response activity (Figure 8B).Figure 8Cell-type-resolved pathway activity, computational drug prediction, and in vitro validation. (A) AUCell violin plots of IVDD-associated pathways across cell types and conditions; ECM-receptor interaction and collagen fibril organization were most active in NP/AF cells, integrin and TGF-β signaling in fibroblasts, and ER protein processing and the unfolded protein response in macrophages. (B) Aggregated AUCell heatmap of pathway activity by cell type and condition; upon degeneration, NP/AF cells gained integrin signaling and ECM-receptor interaction activity, while macrophages gained NF-κB signaling and inflammatory-response activity. (C) DSigDB drug–gene interaction prediction for ITGBL1 and RPN1; eight candidate compounds passed the significance threshold (dashed line, P = 0.05; dot size, target-gene count; color, −log10 P). (D) Functional classification of the candidate compounds by pharmacological category. (E) Three-layer drug–pathway–target regulatory network linking candidate drugs to enriched pathways and the two core genes (Orange, ITGBL1-targeting; blue, RPN1 targeting; purple, dual-targeting). (F) Drug–gene interaction network for ITGBL1 and RPN1. (G) Representative qRT-PCR results for ITGBL1 and RPN1 in IL-1β stimulated versus control HNPCs (10 ng/mL, 24 h). The experiment was independently repeated three times, with technical triplicate measurements within each experiment. (H) Densitometric quantification of ITGBL1 and RPN1 protein normalized to GAPDH from three independent experiments. (I) Representative Western blots of ITGBL1, RPN1, and GAPDH in control and IL-1β treated HNPCs. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.Infographic on AUCell pathways, drug candidates and IL-1beta validation of ITGBL1 and RPN1.A multi-part infographic arranged in a grid and read A through I, summarizing AUCell pathway activity by cell type and condition, drug candidate selection and validation of ITGBL1 and RPN1 under IL-1beta. The image A showing six AUCell violin plots titled, ECM-receptor interaction, Integrin signaling, TGF-beta signaling, Collagen fibril organization, Protein processing in ER and Unfolded protein response. Each plot compares NP/AF Cells, Fibroblasts and Macrophages under Control and IVDD. The y-axis in each plot is AUCell score with tick labels from 0.00 up to about 0.40 depending on the pathway. Higher distributions appear for ECM-receptor interaction and Collagen fibril organization in NP/AF Cells, Integrin signaling and TGF-beta signaling in Fibroblasts and Protein processing in ER and Unfolded protein response in Macrophages. The image B showing a clustered heatmap titled, Pathway Activity (AUCell) by Cell Type and Condition. Columns are NP/AF Cells Control, NP/AF Cells IVDD, Fibroblasts Control, Fibroblasts IVDD, Macrophages Control and Macrophages IVDD. Rows list, TGF-beta signaling, Unfolded protein response, N-glycan biosynthesis, Inflammatory response, NF-kB signaling, Collagen fibril organization, Protein processing in ER, ECM-receptor interaction and Integrin signaling. The color scale bar is labeled from minus 1.5 to 1.5. NP/AF Cells IVDD shows higher Integrin signaling and ECM-receptor interaction than NP/AF Cells Control. Macrophages IVDD shows higher NF-kB signaling and Inflammatory response than Macrophages Control. The image C showing a scatter plot of candidate compounds with a note, Dashed line indicates P = 0.05 significance threshold. The x-axis label is minus log10(P-value) with ticks 0.0, 0.5, 1.0, 1.5, 2.0. The y-axis lists, Rofecoxib, Ibuprofen, IL-6-(Butanoylamino)P., Lycopene, Simvastatin, Mg-262, Medroxyprogesterone, Estradiol. A size key labeled Target Gene Count shows 1 and 2. A color key labeled minus log10(P-value) shows 1.4, 1.6, 1.8, 2.0. Points for Rofecoxib and Ibuprofen are near 2.0 on the x-axis. The image D showing a bar chart titled, Functional Classification of Candidate Drugs. The y-axis label is minus log10(P-value) with values up to 2.0. Bars are labeled with values: Lycopene 1.62, Rofecoxib 2.00, Medroxyprogesterone 1.41, Estradiol 1.33, Ibuprofen 1.83, IL-6-(Butanoylamino)P. 1.65, Mg-262 1.45, Simvastatin 1.62. A legend titled Drug Category lists, Alkaloid, COX-2 inhibitor, Hormone, NSAIDs, Nucleotide analogue, Proteasome inhibitor, Statin. The image E showing a three-column network labeled, Drug Candidates, Enriched Pathways and Core Genes. Drug Candidates list includes, Rofecoxib, Ibuprofen, IL-6-(Butanoylamino)P., Lycopene, Simvastatin, Mg-262, Medroxyprogesterone, Estradiol. Enriched Pathways list includes, NF-kB signaling, TGF-beta signaling, N-glycan biosynthesis, Protein processing in ER, Unfolded protein response, Inflammatory response, ECM-receptor interaction, Integrin signaling. Core Genes are ITGBL1 and RPN1. Lines connect drugs to pathways and pathways to ITGBL1 and RPN1. A small key labels, ITGBL1-targeting, RPN1-targeting and Dual-targeting. The image F showing a drug and gene interaction network with two labeled gene nodes, ITGBL1 and RPN1, connected to drug nodes labeled, Rofecoxib, Ibuprofen, Mg-262, Simvastatin, Lycopene, Estradiol, Medroxyprogesterone and Butanoylamino-purine. A legend labels node types, RPN1, ITGBL1, Shared Drug and Specific Drug. The image G showing two bar charts. Left chart title, ITGBL1, with y-axis label Relative ITGBL1 level and ticks 0 to 5; bars for Control and IL-1 with IL-1 higher and a significance mark, double asterisk. Right chart title, RPN1, with y-axis label Relative RPN1 level and ticks 0 to 8; bars for Control and IL-1 with IL-1 higher and a significance mark, four asterisks. The image H showing two bar charts. Left chart title, RPN1, with y-axis label Relative RPN1 expression level and ticks 0.0 to 2.0; Control is about 1.0 and IL-1 is higher, with a single asterisk. Right chart title, ITGBL1, with y-axis label Relative ITGBL1 expression level and ticks 0.0 to 1.5; Control is about 1.0 and IL-1 is higher, with three asterisks. The image I showing Western blot bands labeled at top, Control and IL-1beta 24h. Three rows are labeled RPN1, ITGBL1 and GAPDH, with molecular weight labels 70KD, 55KD and 40KD. Cell-type-resolved pathway activity, computational drug prediction, and in vitro validation. (A) AUCell violin plots of IVDD-associated pathways across cell types and conditions; ECM-receptor interaction and collagen fibril organization were most active in NP/AF cells, integrin and TGF-β signaling in fibroblasts, and ER protein processing and the unfolded protein response in macrophages. (B) Aggregated AUCell heatmap of pathway activity by cell type and condition; upon degeneration, NP/AF cells gained integrin signaling and ECM-receptor interaction activity, while macrophages gained NF-κB signaling and inflammatory-response activity. (C) DSigDB drug–gene interaction prediction for ITGBL1 and RPN1; eight candidate compounds passed the significance threshold (dashed line, P = 0.05; dot size, target-gene count; color, −log10 P). (D) Functional classification of the candidate compounds by pharmacological category. (E) Three-layer drug–pathway–target regulatory network linking candidate drugs to enriched pathways and the two core genes (Orange, ITGBL1-targeting; blue, RPN1 targeting; purple, dual-targeting). (F) Drug–gene interaction network for ITGBL1 and RPN1. (G) Representative qRT-PCR results for ITGBL1 and RPN1 in IL-1β stimulated versus control HNPCs (10 ng/mL, 24 h). The experiment was independently repeated three times, with technical triplicate measurements within each experiment. (H) Densitometric quantification of ITGBL1 and RPN1 protein normalized to GAPDH from three independent experiments. (I) Representative Western blots of ITGBL1, RPN1, and GAPDH in control and IL-1β treated HNPCs. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Based on the two core genes (ITGBL1 and RPN1), drug–gene interaction prediction using DSigDB identified eight candidate compounds (P < 0.05; Figure 8C), predominantly enriched in anti inflammatory and immune‑modulatory categories (Figure 8D), consistent with immune‑mediated disc degeneration. After filtering, simvastatin and lycorine were prioritized as candidate compounds, both supported by previous preclinical evidence in IVDD models28,29 —simvastatin inhibits IL‑1β‑induced apoptosis and ECM degradation via NF‑κB/MAPK pathways,30 while lycorine suppresses NF‑κB signaling and catabolic enzymes (MMP3, MMP13, ADAMTS4/5).31 Other candidates (eg, MG-262, estradiol) represent exploratory leads or symptomatic agents requiring further validation. To contextualize these interactions, a drug–gene interaction network (Figure 8F) and a three layer drug pathway target regulatory network (Figure 8E) were constructed. The integrated network suggested two potentially relevant drug–pathway associations: the ITGBL1 associated branch was mainly linked to NF-κB signaling and integrin ECM interactions, whereas the RPN1 associated branch, represented by Simvastatin and MG-262, was linked to N-glycan biosynthesis, ER protein processing, and the UPR. Notably, NF-κB signaling emerged as a shared network node connecting these associations, suggesting a potential convergence between inflammation-related signaling and ECM remodeling in IVDD. These computational findings support further evaluation of Simvastatin and Lycorine as candidate compounds rather than established therapeutic agents. To provide experimental support for the bioinformatic predictions, we examined core gene expression in IL-1β stimulated HNPCs. qRT-PCR confirmed significant upregulation of both ITGBL1 (P < 0.01) and RPN1 (P < 0.0001) alongside the catabolic markers MMP13 (P < 0.0001) and MMP3 (P < 0.001) compared to untreated controls (Figure 8G). To further validate these findings at the protein level, Western blot analysis demonstrated that IL-1β treatment markedly increased the protein expression of both ITGBL1 and RPN1, with densitometric quantification showing approximately 1.35-fold and 1.26-fold upregulation, respectively, after normalization to GAPDH (Figure 8H and I, Supplementary Figure 3). The concordance between mRNA and protein level changes provides robust evidence that ITGBL1 and RPN1 are responsive to the pro-inflammatory catabolic milieu characteristic of IVDD progression. Discussion Intervertebral disc degeneration is driven in part by progressive NP fibrosis, but the cellular localization and transcriptional context of IVDD associated genes remain incompletely understood. In the present study, the main contribution extends beyond the identification of ITGBL1 and RPN1 as candidate IVDD associated markers. By integrating bulk and single-cell transcriptomic analyses, we preferentially localized ITGBL1 and RPN1 to a FibroNPC subpopulation and associated this cell state with integrin/focal-adhesion and ER protein processing/N-glycosylation programs. ITGBL1 is a secreted, integrin binding extracellular matrix protein that regulates integrin activity. It promotes fibrosis in the heart and liver,16,17 whereas in articular cartilage it protects chondrocytes and limits osteoarthritis.18 In our data, ITGBL1 was concentrated in the FibroNPC, co-expressed with COL1A1, POSTN, and LOXL1, and upregulated together with MMP13 and MMP3 in IL-1β-stimulated NP cells at both the mRNA and protein levels. This pattern indicates that, in the degenerating disc, ITGBL1 participates in fibrotic and catabolic remodeling rather than protecting the tissue, in line with its pro-fibrotic role in other organs and consistent with the FibroNPC moving away from a chondrogenic phenotype.19 CellChat inferred that macrophage derived SPP1 may communicate with NP/AF cells mainly through integrin receptors, consistent with a recent report implicating the SPP1–integrin axis in NP degeneration,32 while AUCell indicated increased integrin pathway activity with degeneration. Based on these convergent observations, we propose a hypothetical feed forward integrin model in which macrophage derived SPP1 may engage integrin receptors on FibroNPCs, while ITGBL1 may modulate integrin associated matrix remodeling. The resulting matrix changes could, in turn, favor further macrophage recruitment or activation. This model integrates the observed associations but is not directly demonstrated by the present cross-sectional data and requires further functional and temporal validation. RPN1 is a catalytic subunit of the oligosaccharyltransferase complex and is required for N-glycosylation of membrane and secreted proteins.21 It has been linked to macrophage polarization and an immunosuppressive microenvironment in several cancers.23 In our study, RPN1 marked the glycosylation arm of the FibroNPC, correlated positively with myeloid cells and negatively with effector memory CD4 T cells, and was induced by IL-1β. Because N-glycosylation modifies integrin receptors and antigen presentation, the increase in RPN1 may reflect the higher biosynthetic and ER load of a matrix secreting cell and a possible link to the disc immune environment. These results are preliminary and identify RPN1 as a candidate for further study. The two arms are probably connected, because a cell producing abundant matrix also carries a high N-glycosylation and ER load, which ties matrix output to the secretory machinery. Although the FibroNPC made up less than one fifth of NP/AF cells, it generated most fibrotic transcripts and, with degeneration, strongly induced COL1A1, COL1A2, COL3A1, POSTN, FN1, TGFBI, CTHRC1, and SPP1 without increasing in number; the heavy biosynthetic and ER load may restrain proliferation while inflammation raises per-cell output. This pattern was reproduced in the progression stage Cohort II. A recent multi-omics study of the same data instead highlighted NOTCH3 in NP fibrosis,33 suggesting that several programs converge on this state. Drug prediction identified simvastatin and lycorine, both acting on NF-κB and supported by earlier IVDD studies;28–30 targeting the FibroNPC state, for example by combining integrin-signaling modulation with relief of ER stress, may be more focused than general anti-inflammatory treatment. This study has several limitations. First, the relatively limited sample sizes, particularly the small external validation cohort, may affect model stability and generalizability and increase the risk of overfitting. Second, the included bulk and single cell datasets differed in platforms, tissue composition, and disease grading criteria, and some uncertainty in cell type annotation and subpopulation estimates therefore remains. Third, macrophages represented only a small fraction of the single cell data (~0.3%); accordingly, the CellChat derived communication patterns should be regarded as exploratory rather than evidence of established signaling mechanisms. Finally, the in vitro validation was limited to a single IL-1β condition in an immortalized HNPC line, without primary human tissue validation or direct loss or gain of function experiments. In addition, no in vivo experiments were performed to confirm the causal roles of ITGBL1 and RPN1 in IVDD. Further studies using primary human samples, functional manipulation, and in vivo models are needed to confirm these findings. Conclusion In summary, we identified ITGBL1 and RPN1 as candidate molecular markers associated with IVDD and found both genes to be preferentially localized to a FibroNPC subpopulation characterized by integrin/focal-adhesion and ER protein-processing/N-glycosylation programs. ITGBL1 was associated with a pro-fibrotic, catabolic phenotype, whereas RPN1 was associated with N-glycosylation and features of the local immune environment. CellChat further suggested potential macrophage SPP1–integrin communication involving structural cell populations, although these relationships remain computationally inferred. Together, these findings highlight ITGBL1, RPN1, and the FibroNPC state as priorities for future functional and mechanistic validation.