Molecular subtyping and prognostic evaluation in idiopathic pulmonary fibrosis: a focus on mechanical-related genes Authors: Zibin Chen, Yupeng Zhi, Bo Wu, Hongzhao Huang, Mingwei Zhang, Jinsheng Hong, Yansong Guo, Chun Chen Published: 2025 ABSTRACT ObjectiveIdiopathic pulmonary fibrosis (IPF) is a chronic lung disease characterized by fibroblast activation and extracellular matrix deposition. Although mechanical forces are known to influence critical processes such as tissue remodeling and cellular differentiation, the specific role of mechanosensitive genes in IPF pathogenesis remains poorly understood.MethodsThis study applied mechanical force–related scoring to IPF and control samples. Weighted Gene Co-expression Network Analysis (WGCNA) was used to identify IPF-specific genes associated with mechanical force–related scores, which were subsequently subjected to enrichment analysis. Core genes were screened through univariate Cox regression analysis and machine learning algorithms. Independent prognostic genes were determined using stepwise multivariate Cox regression analysis, which informed the construction of a nomogram integrating key clinicopathological variables. The core genes were further analyzed through single-cell RNA sequencing (scRNA-seq) and Connectivity Map (CMap) analyses to investigate molecular subtypes and identify potential therapeutic targets.ResultsIPF samples exhibited significantly higher scores related to mechanical-related genes (MRGs). Moreover, these samples demonstrated notable heterogeneity, with distinct patterns of high and low mechanical force–associated scores. Core gene analysis revealed two distinct molecular subtypes among idiopathic IPF samples. The C2 subtype was characterized by pronounced inflammatory responses, activation of mechanotransduction pathways, and more severe pathological features. CMap analysis identified nifekalant, which targets AGTR2, as a potential therapeutic agent for the C2 subtype. CD24, PPP1R14C, DSP, and CC2D2A were identified as diagnostic biomarkers, among which CD24, PPP1R14C, and CC2D2A also served as independent prognostic indicators. scRNA-seq demonstrated elevated expression of CD24 and CC2D2A in IPF samples, predominantly within ciliated cells. Pseudotime trajectory analysis revealed two distinct cell fate trajectories in IPF samples. Differentiation toward cell fate 1 was associated with enhanced protein synthesis and secretion, whereas cell fate 2 and the branching point origin region exhibited increased cell adhesion and activation of the p38 MAPK signaling pathway.ConclusionCD24, PPP1R14C, and CC2D2A were identified as prognostic genes associated with MRGs, functioning as markers for molecular subtyping in IPF. These genes may contribute to the pathogenesis of IPF by modulating mechanotransduction processes within epithelial cell subpopulations.Graphical Abstract Supplementary informationThe online version contains supplementary material available at 10.1186/s12967-025-07365-7. Introduction Idiopathic Pulmonary Fibrosis (IPF) is a chronic, progressive form of interstitial lung disease of unknown etiology, defined primarily by alveolar epithelial injury and aberrant repair mechanisms [1]. This pathological process culminates in abnormal fibroblast proliferation and excessive extracellular matrix (ECM) deposition, which in turn increases lung tissue stiffness and modifies its mechanical properties [2]. With a median survival of only 3–5 years, IPF profoundly impairs patients’ quality of life and is associated with high mortality rates [3]. Although anti-fibrotic agents such as pirfenidone and nintedanib are currently available to slow disease progression, their clinical utility is constrained by limited efficacy and significant adverse effects [4]. Physical forces are integral to the functioning of molecules, cells, and organs, exerting profound influences on mechanosensing, development, stem cell fate, and cancer [5, 6]. These forces, together with the cellular responses they elicit, play pivotal roles in both the initiation and resolution of tissue injury. Conversely, dysregulated repair mechanisms or recurrent injuries may culminate in pathological tissue remodeling and fibrosis. The respiratory dynamics and distinctive structural attributes of the lungs generate a mechanically complex environment that is susceptible to dysregulation. Research indicates that increased matrix stiffness promotes alveolar epithelial cell (AEC) epithelial–mesenchymal transition (EMT) [7], whereas abnormal mechanical stretching markedly alters AEC behavior [8]. The remodeled ECM, characterized by heightened stiffness, further activates fibroblasts, thereby perpetuating a detrimental cycle [9]. Despite these insights, the dynamic interplay between mechanical cues and cellular responses during disease progression remains systematically uncharacterized, constituting a substantial barrier to the development of targeted therapeutic strategies. To address this, the analysis of key genes that define the mechanical microenvironment during the progression of pulmonary fibrosis—using bulk transcriptome and single-cell sequencing data—offers a promising and potentially groundbreaking avenue for identifying novel therapeutic targets. Furthermore, investigation of the heterogeneity in mechanical signals and features across patients and disease stages can yield critical mechanistic insights. Precision medicine approaches targeting mechano-biochemical coupling signals in specific patient subtypes have the potential to overcome the limitations of current treatment modalities, which frequently lack specificity. This strategy constitutes a pivotal advance in elucidating the underlying mechanisms of pulmonary fibrosis and heralds a transformative shift in clinical management—from conventional antiinflammatory and anti-fibrotic therapies toward targeted remodeling of the mechanical microenvironment.Although IPF has been stratified into multidimensional subtypes based on PANoptosis signatures [10], clinical characteristics [11], and multi‑omics profiles encompassing miRNA and proteomic data [12], its onset and progression are also profoundly shaped by the lung’s biomechanical microenvironment; subtyping grounded in mechanobiological features holds promise for unveiling a novel dimension of disease heterogeneity and furnishing a refined framework for precision therapies targeting mechanical microenvironment remodeling. This study employed a comprehensive analytical framework to identify co-expression modules closely associated with mechanical-related genes (MRGs) in patients with IPF using Weighted Gene Co-expression Network Analysis (WGCNA). Using univariate Cox regression analysis in combination with machine learning techniques, we identified a core gene cluster exhibiting both disease relevance and prognostic significance. Notably, we developed a prognostic nomogram based on multivariate Cox stepwise regression, thereby identifying for the first time mechanical-related molecular subtypes exhibiting distinct diseasestate variations in patients with IPF. The identification of these MRGs introduces novel molecular biomarkers for IPF prognosis and offers new perspectives for elucidating disease mechanisms and designing targeted therapeutic strategies. Materials and methods Data acquisition For bulk RNA-seq analysis, lung tissue samples were obtained from GSE110147 (IPF = 22, Control = 11), GSE213001 (IPF = 62, Control = 41), and GSE32537 (IPF = 119, Control = 50). The GSE110147 and GSE213001 datasets were integrated for preprocessing, and batch effects between them were corrected using the ComBat function from the R package sva(Supplemental Fig. 8A, B). The GSE32537 dataset served as the validation set. Bronchoalveolar lavage fluid (BALF) transcriptomic data from 112 patients with IPF were obtained from GSE70866. Single-cell RNAs-equencing datasets from lung tissue included GSE128033 (IPF = 8, Control = 10) and GSE136831 (IPF = 31, Control = 26). GSE128033 contains singlecell sequencing data from both upper and lower lung lobes, enabling differential analyses between these regions and facilitating the identification of differentially expressed genes corresponding to early versus midtolate disease stages. GSE136831 was used to investigate in greater depth the associations of core genes across diverse cell types. The single-cell dataset GSE141259, derived from bleomycin (BLM)induced mouse models of IPF at multiple time points, was used for validation. Three MRG sets were compiled by retrieving entries from the Gene Ontology (GO) database (Supplemental Table S1). These sets comprised the Mechanical Microenvironment (MM), Cytoskeleton and Force Generation (CF), and Mechanosensation and Signaling (MS) categories. After removing redundant genes within each set, 1,252, 1,028, and 509 genes were retained for MM, CF, and MS, respectively. Processing of single-cell RNA sequencing datasets The raw unique molecular identifier (UMI) countbased single-cell RNA-sequencing dataset was preprocessed in strict accordance with the standard pipeline of the Seurat package (version 5.2.1). During quality control, cells with a mitochondrial gene proportion greater than 20% (indicative of cell damage) or fewer than 200 detected genes (likely empty droplets) were excluded. The expression matrix was subsequently lognormalized using the NormalizeData function. The FindVariableFeatures function, in combination with the meanvariance plotting (MVP) method, was applied to identify 2,000 highly variable genes (HVGs). The expression values of these HVGs were scaled using the ScaleData function to balance their contributions in downstream analyses. Principal component analysis (PCA) was performed on the HVGs. The optimal number of principal components (PCs) was determined as 20 by integrating results from the JackStraw test and the elbow method. Unsupervised clustering (FindClusters, resolution = 1) was performed on the PCA-reduced data, and the results were visualized using the UMAP algorithm. Cell-type-specific marker genes were identified using the FindAllMarkers function (selection criteria:\|average log₂ fold change\| > 2 and adjusted p < 0.01). Key gene expression patterns were further examined through multidimensional visualization using the DimPlot and VlnPlot functions. Differential gene analysis To identify differentially expressed genes (DEGs) in bulk transcriptome data, we used the limma R package to screen DEGs from the integrated GSE110147 and GSE213001 datasets, based on normalized count data (FPKM) [13]. Samples in these datasets were derived from human lung tissues, including those from patients with idiopathic pulmonary fibrosis (IPF) and from IPF patients who had undergone lung transplantation. The criteria for DEG selection were\|log₂ fold change\| > 0.585 and adjusted p < 0.05. For the singlecell RNAsequencing dataset GSE128033, we used the FindAllMarkers function in Seurat to identify DEGs between lowerlobe and upperlobe groups. The selection criteria were adjusted p < 0.05 and\|log₂ fold change\| > 0.585. Furthermore, we used the R package ggplot2 to generate volcano plots of the DEGs, providing a visual representation of their distribution to facilitate subsequent analyses. Screening of highly variable genes Genes obtained during this process were filtered using the getMostVarGenes function, with the following parameter settings for highly variable gene selection: fitThr = 0.8 and minMeanForFit = 1 (minimum mean expression for model fitting).[14] WGCNA analysis of MRG modules Based on the expression profile data of the IPF disease group from the integrated datasets GSE110147 and GSE213001, this study employed WGCNA for a systematic investigation. First, sample clustering was performed using a single-sample gene set enrichment analysis (ssGSEA) scoring matrix constructed from highly variable genes (HVGs) in IPF patient samples to identify and exclude potential outliers. Subsequently, the optimal soft-threshold power was determined by constructing a scalefree topological network, and a topological overlap matrix (TOM) was generated from the adjacency matrix. Hierarchical clustering of co-expression modules was performed using TOM-based dissimilarity metrics. Furthermore, the module most strongly associated with mechanical features was identified by calculating the correlation between module eigengenes and ssGSEA scores of mechanically related HVGs. Finally, the genes within this module were designated as HVGs associated with mechanical features in IPF samples. Screening of prognosis - related genes This study systematically investigated the clinical significance of MRGs in the prognostic evaluation of IPF using a multi‑stage bioinformatics analysis strategy. First, using the GSE70866 dataset containing complete survival information, we performed genome‑wide screening with univariate Cox proportional hazards regression models (significance threshold: p < 0.05), followed by multiple hypothesis testing correction using the Benjamini–Hochberg method. Applying stringent selection criteria—ranking genes by ascending adjusted p-values and excluding non‑significant genes whose hazard ratio (HR) confidence intervals included the null value of 1—we identified 40 MRGs significantly associated with overall survival (OS) in this dataset. To construct a precise prognostic model, we integrated least absolute shrinkage and selection operator (LASSO) regression—implemented with the glmnet package using 10‑fold cross‑validation to determine the optimal penalty parameter (lambda.min)—and the random forest algorithm for dual feature selection. This approach effectively addressed the challenge of dimensionality reduction in high‑dimensional data while further refining features by evaluating gene variable importance. Intersecting the screening results from both algorithms yielded a core set of MRGs with the strongest prognostic predictive value. To validate the diagnostic efficacy of these core genes, we performed receiver operating characteristic (ROC) curve analysis for each gene in the integrated dataset (GSE110147 and GSE213011) and in GSE32537 using the pROC package, defining genes with area under the curve (AUC) values > 0.7 as potential diagnostic biomarkers. Finally, using the survival and survminer packages for Kaplan–Meier survival curve analysis combined with log‑rank tests, we examined the dose–response relationship between MRG expression levels and patient OS. Establishment of the prognostic nomogram Based on the results of multivariate stepwise Cox regression analysis, a prognostic nomogram was constructed using the rms and regplot packages in R. The performance of the prognostic model was evaluated using time-dependent ROC curves, with the timeROC package employed to plot the curves and calculate the AUC values, thereby determining the predictive performance of each model. Additionally, calibration curves were generated using the rms package to assess the predictive accuracy of the nomogram and to compare the agreement between predicted and observed survival rates. Subtyping of IPF patient samples To subtype patients with idiopathic pulmonary fibrosis (IPF) using bulk RNAsequencing data, consensus clustering was performed with the ConsensusClusterPlus package, using hub mechanical-related genes (MRGs) as features. Strict parameters were applied, including 80% resampling of both samples and genes, a maximum cluster number of nine (k = 9), and 1,000 iterations to ensure robust results. Hierarchical clustering with Pearson correlation–based distance (1 – Pearson) was applied to assess gene expression patterns. Principal component analysis (PCA) was performed to validate clustering results and to visualize subtype separation based on mechanical core genes, revealing clear distinctions among molecular subtypes. For single-cell RNA-sequencing clustering, the AddModuleScore function in Seurat was used to score epithelial cell clusters based on prognostic core genes. Singlecell samples from patients with IPF were then divided into high-and lows-core groups using the median score as the cutoff. immune infiltration and pathway scoring ssGSEA was used to quantify the infiltration levels of 28 immune cell subsets. Gene sets from GeneCards and the Molecular Signatures Database (MSigDB) were integrated to assess immune phenotypes and mechanotransduction pathways implicated in IPF progression. Boxplots generated with the ggpubr R package illustrated differences in immune cell composition, phenotypes, and pathway activity among IPF molecular subtypes. To explore the immunoregulatory roles of key genes, differential expression analysis was conducted using the limma package, and ggplot2 was used to visualize correlations between three core genes and immune cell infiltration, thereby providing insights into the immune microenvironment in IPF. Protein-protein interaction network (PPI) The protein–protein interaction (PPI) network was constructed using the STRING database ( < https://cn.string-db.org>), with Homo sapiens (human) specified as the organism of interest and “medium confidence” set as the minimum interaction score threshold. Network visualization was performed using Cytoscape software (version 3.7.1). Cell trajectory analysis During the derivation of lung ciliated cell trajectories, the Monocle package (version 2.18.0) was used to infer cell developmental trajectories [15]. Raw read counts served as input data, with the expressionFamily parameter set to a negative binomial distribution. Size factors and dispersion were estimated using the estimateSizeFactors and estimateDispersions functions. The Monocle package identified highly variable genes to construct the cell trajectory, designating the root state as the predominant control cell state. The orderCells function was used to sort cells by their actual states. The BEAM algorithm identified key genes associated with cell fate decisions, selecting those expressed in at least 100 cells. The plot_genes_branched_heatmap function visualized developmentally relevant genes, whereas the plot_genes_branched_pseudotime function depicted dynamic gene expression patterns over pseudotime. Intercellular ligand-receptor communication analysis The CellChat package (version 1.5.0) was used to analyze intercellular communication among all lung epithelial cell subpopulations [16], including aberrant basaloid, alveolar type I (ATI), alveolar type II (ATII), basal, ciliated, club, goblet, ionocyte, mesothelial, and pulmonary neuroendocrine cells (PNECs). By calculating the number of ligands and the weighting values of cell–cell interactions, a weighted, directed network was constructed, incorporating significantly interacting ligands between cell populations. This analysis integrated 2,021 experimentally validated ligand–receptor molecular interaction signals, from which autocrine/paracrine-related signaling pathways were selected to identify conserved and condition-specific signaling information flows between high- and low-scoring groups of patients with IPF. Subsequently, the signal strengths of the identified conserved and condition-specific pathways were quantified across different cell types. Drug screening and molecular docking The Connectivity Map (CMap; < https://portals.broadinstitute.org/cmap/>) is an online analysis platform based on genomewide transcriptional expression data, designed to investigate mechanisms of action and facilitate drug repurposing of small molecules [17]. According to its pattern-matching algorithm, positive scores indicate that a small molecule induces the query signature, whereas negative scores suggest an inhibitory effect. In this study, CMap was used to screen for potential small-molecule drugs targeting the upregulated gene set of the C2 subtype, with the five compounds exhibiting the lowest scores selected as candidate drugs. The chemical structures of these drugs were retrieved from PubChem ( < https://pubchem.ncbi.nlm.nih.gov/>) [18] The crystal structure of the key target encoded by the hub C2 gene was obtained from the RCSB Protein Data Bank (PDB; < http://www.rcsb.org/>) at the highest available resolution [19]. Subsequently, to examine the interaction profiles and binding affinities of the putative targets with AGTR2 (e.g., PDB ID: 5UNF), molecular docking was performed using SybylX 2.0, which generated precise ligand–protein complex binding geometries. The results were visually analyzed using PyMOL software (version 4.6.0). BLM mouse model of pulmonary fibrosis The mice were anesthetized via intraperitoneal injection of 1% sodium pentobarbital (50 mg/kg), after which 150 μL of bleomycin (BLM; 5 mg/mL, BIO000001) was administered to the lungs by tracheal instillation. This animal experiment was approved by the Experimental Animal Welfare Ethics Committee of Fujian Medical University (No. IACUC FJMU 2023-0129) and conducted in accordance with the National Research Council’s Guide for the Care and Use of Laboratory Animals. Quantitative real-time PCR (qPCR) Total RNA (TRIzol-extracted) from lung tissue (20 mg) was reverse-transcribed using Hifair® III cDNA Synthesis SuperMix (YEASEN, China). qPCR (Hieff® SYBR Green Master Mix, YEASEN) was performed with primers (Supplemental Table S3; Shangya Biotechnology, China). Relative mRNA levels were calculated via the 2−ΔΔCt method. Results Mechanical factors and genetic correlates in IPF During fibrogenesis, the ECM undergoes substantial compositional and structural alterations, thereby reshaping the mechanical microenvironment of encapsulated cells. These cells detect mechanical stimuli via specialized mechanosensors, which transduce these cues into intracellular biochemical signals that subsequently elicit specific cellular responses. Additionally, cells actively generate mechanical forces, including those arising from cytoskeletal contractile activities [20]. Importantly, aberrant mechanical signals originating from local cells can propagate via intercellular mediators, thereby triggering cascading effects in neighboring cells and amplifying fibrotic responses at the tissue level [20]. Thus, guided by these three core aspects—mechanosensation and signaling (MS), mechanical microenvironment (MM), and cytoskeleton and force generation (CF)—we curated MRGs from the Gene Ontology (GO) database, eliminated redundant entries, and categorized them accordingly (Supplemental Fig. S1A; Supplemental Table S1). IPF, pathological manifestations exhibit pronounced regional heterogeneity between the upper and lower lobes of the lung. Pathological changes in the upper lobe predominantly reflect early-stage IPF, whereas those in the lower lobe are indicative of advanced disease [21]. This regional disparity results from multifactorial mechanisms. The lower lobe, influenced by greater gravitational forces and increased blood perfusion, facilitates the accumulation and deposition of inflammatory mediators [22]. Furthermore, anatomical differences highlight the importance of mechanical factors, as the lower lobe experiences greater tensile stress and higher alveolar deformation rates than the upper lobe. To elucidate the molecular basis, differential gene expression analysis (criteria:\|log2FC\| > 0.585, p < 0.05) was performed on upper and lower lobe tissues using the publicly available single-cell dataset GSE128033, identifying 3,369 differentially expressed genes (DEGs) (Fig. 1A). Gene set enrichment analysis (GSEA) of these DEGs revealed significant upregulation of biological processes associated with small GTPase-mediated signaling and ECM assembly in the lower lobe relative to the upper lobe (Fig. 1B). Notably, Rho family small GTPases (e.g., RhoA), which are pivotal mechanotransduction molecules, become activated to promote fibrogenesis by regulating downstream effectors such as Rho-associated kinase (ROCK). This activation modulates cytoskeletal remodeling and cellular contractility, ultimately driving fibroblast activation and collagen secretion [23].Fig. 1Mechanical factors and genetic correlates in IPF. (A) The volcano plot illustrates the differential gene analysis between upper and lower lung tissues during IPF progression. From the public single-cell dataset GSE128033(Up). Integration with bulk transcriptome datasets GSE2014 and GSE213001 yielded 2783 differentially expressed genes significantly associated with the IPF disease state(down). (B) Gene set enrichment analysis (GSEA) of differentially expressed genes in upper and lower lung tissues.(C) A venn diagram demonstrates the intersection between gene sets characterizing IPF progression, the core IPF disease gene set, and gene sets associated with three biomechanical features (mechanosensation & signaling, mechanical microenvironment, cytoskeleton & force generation). (D) UPSET plot visualizes the overlapping relationships among different gene sets Mechanical factors and genetic correlates in IPF. (A) The volcano plot illustrates the differential gene analysis between upper and lower lung tissues during IPF progression. From the public single-cell dataset GSE128033(Up). Integration with bulk transcriptome datasets GSE2014 and GSE213001 yielded 2783 differentially expressed genes significantly associated with the IPF disease state(down). (B) Gene set enrichment analysis (GSEA) of differentially expressed genes in upper and lower lung tissues.(C) A venn diagram demonstrates the intersection between gene sets characterizing IPF progression, the core IPF disease gene set, and gene sets associated with three biomechanical features (mechanosensation & signaling, mechanical microenvironment, cytoskeleton & force generation). (D) UPSET plot visualizes the overlapping relationships among different gene sets However, DEGs identified from single-cell sequencing datasets may reflect not only disease progression and mechanical influences but also intrinsic structural differences between the upper and lower lobes. To prioritize disease-associated changes, we integrated the bulk transcriptomic datasets GSE2014 and GSE213001 for differential expression analysis using the same criteria, which yielded 2,783 DEGs significantly associated with IPF disease status (Fig. 1A). Intersection analyses were subsequently conducted among gene sets representing IPF progression, core IPF disease genes, and the three MRG sets (Fig. 1C, D). The intersecting genes were visualized using a heatmap (Supplemental Fig. S2), which revealed distinct clustering patterns across samples from IPF patients. These findings suggest potential heterogeneity in mechanically related molecular features among individual IPF patients. We hypothesize that inter-individual variations in mechanically related molecular signatures may influence disease progression rates and clinical outcomes in patients with IPF. Screening and functional analysis of hypervariable genes based on MRGs To more precisely classify IPF patient samples into subtypes based on MRGs, this study screened for highly variable genes (HVGs) within the intersecting gene sets, ultimately identifying 25, 46, and 31 HVGs, respectively (Fig. 2A). These HVGs exhibited significant intercorrelations, particularly at the cytoskeleton and force generation (C&F) level, suggesting strong biological connectivity (Supplemental Fig. S3C). Gene Ontology (GO) enrichment analysis of these HVGs produced results highly consistent with the functions of the previously defined three categories of MRGs (Fig. 2B), thereby reinforcing that the selected HVGs are associated with specific mechanical features and may contribute to the aberrant mechanical properties observed in IPF.Fig. 2Screening and functional analysis of hypervariable genes based on mechanobiology - related genes. (A)Numbers of hypervariable genes (HVGs) obtained from screening. (B) Consistency between the gene ontology (GO) enrichment analysis results of HVGs and the functions of three characteristic mechanobiology - related gene categories. (C) Significant difference between the normal control group and the IPF patient group after scoring with the custom gene sets of HVGs by single - sample gene set enrichment analysis (ssGSEA). (D) Correlation between the ssGSEA scores of up - regulated and down - regulated gene sets shown by heatmap. (E) Protein - protein interaction (PPI) network diagram showing that HVGs cluster into two main modules. ***p < 0.001 Screening and functional analysis of hypervariable genes based on mechanobiology - related genes. (A)Numbers of hypervariable genes (HVGs) obtained from screening. (B) Consistency between the gene ontology (GO) enrichment analysis results of HVGs and the functions of three characteristic mechanobiology - related gene categories. (C) Significant difference between the normal control group and the IPF patient group after scoring with the custom gene sets of HVGs by single - sample gene set enrichment analysis (ssGSEA). (D) Correlation between the ssGSEA scores of up - regulated and down - regulated gene sets shown by heatmap. (E) Protein - protein interaction (PPI) network diagram showing that HVGs cluster into two main modules. ***p < 0.001 Subsequently, the HVGs were categorized into upregulated and downregulated gene sets according to their expression changes in the disease state. Using these custom gene sets, single-sample gene set enrichment analysis (ssGSEA) was performed to score both the normal control and IPF patient groups. The scoring effectively distinguished between the normal and IPF groups, revealing statistically significant differences (Fig. 2C). Furthermore, the HVGs demonstrated robust functional associations. Heatmap analysis revealed that ssGSEA scores of the upregulated gene set were positively correlated with one another but negatively correlated with those of the downregulated gene set (Fig. 2D). Additionally, the protein–protein interaction (PPI) network revealed dense interconnections among the HVGs, which were distinctly organized into two major modules (Fig. 2E). Screening of MGR modules in IPF patients samples This study employed WGCNA to identify gene modules correlated with a mechanically derived HVG score, herein designated the “mechanics-related score.” First, the mechanics-related score was calculated from mechanically derived HVGs for patients with IPF and used to stratify the cohort. Hierarchical clustering analysis revealed distinct and consistent grouping of high-score samples, with no detectable outliers (Fig. 3A).Fig. 3Screening of MGR modules in IPF patients. (A)Demonstration of the clustering of patient samples in the high mechanobiology - related score group. (B) Illustration of determining the soft - threshold value as 4 by evaluating scale independence and average connectivity. (C) Presentation of clustering genes into 14 co - expression modules using the dynamic tree - cutting algorithm. (D) Results of the correlation between the score based on 6 up - regulated high - variable genes related to mechanobiology and turquoise and green modules. (E) Correlation analysis between gene significance of C&F - type mechanobiology - related scores and module membership of the turquoise module. (F) A venn diagram is used to visualize the process of selecting 465 core genes. (G) GO enrichment analysis of core genes to reveal the potential functional characteristics of mechanobiology - related genes in IPF Screening of MGR modules in IPF patients. (A)Demonstration of the clustering of patient samples in the high mechanobiology - related score group. (B) Illustration of determining the soft - threshold value as 4 by evaluating scale independence and average connectivity. (C) Presentation of clustering genes into 14 co - expression modules using the dynamic tree - cutting algorithm. (D) Results of the correlation between the score based on 6 up - regulated high - variable genes related to mechanobiology and turquoise and green modules. (E) Correlation analysis between gene significance of C&F - type mechanobiology - related scores and module membership of the turquoise module. (F) A venn diagram is used to visualize the process of selecting 465 core genes. (G) GO enrichment analysis of core genes to reveal the potential functional characteristics of mechanobiology - related genes in IPF The selection of an appropriate soft-thresholding power is a critical step in constructing a robust co-expression network. Based on analyses of scale independence and mean connectivity, a soft-thresholding power of β = 4 was selected (Fig. 3B), which successfully established a scale-free network topology (R2 > 0.85). Subsequently, the dynamic tree cut algorithm was applied, which clustered genes into 14 distinct co-expression modules (Fig. 3C). We then assessed the correlation between each module eigengene and the mechanics-related score. This analysis revealed that the mechanics-related score exhibited the strongest positive correlation with the eigengenes of both the Turquoise and Green modules (Fig. 3D). Furthermore, a significant positive correlation was identified between gene significance for the mechanics-related score and module membership within the Turquoise module (Fig. 3E). By applying stringent selection criteria (p.Gene Significance < 0.01 for the target module, p.Module Membership < 0.01 for the Turquoise module, and\|MM.Turquoise\| ≥ 0.7 with\|GS.target module\| ≥ 0.7), we identified 577 genes within the Turquoise module as candidate mechanotransduction-responsive genes strongly associated with the mechanics-related high-variance gene score. Intersecting these genes with single-cell differentially expressed genes (DEGs_sc) associated with IPF progression yielded 465 core genes (Fig. 3F). GO enrichment analysis of the 465 core genes identified putative functional characteristics of mechanics-related genes in IPF (Fig. 3G). The analysis showed significant enrichment across multiple biological processes (BPs), cellular components (CCs), and molecular functions (MFs) directly implicated in mechanotransduction. At the BP level, notably enriched terms (RichFactor = 0.4–0.8) included axonemal central apparatus assembly, mucociliary clearance, and epithelial cilium movement associated with left–right asymmetry. Moreover, concurrent enrichment of microtubule bundle formation and other cross-system pathway components further substantiated the presence of widespread dysregulation in mechanotransduction networks in IPF. Multi-stage bioinformatics analysis identifies prognostic MRGs in IPF We implemented a multi-stage bioinformatics analysis framework to systematically evaluate the clinical significance of mechanotransduction-related genes in the prognostic assessment of idiopathic pulmonary fibrosis (IPF). Using the GSE70866 expression profile dataset, we identified 40 mechanotransduction-signature genes that were significantly associated with overall survival (OS) (Fig. 4A). An initial LASSO regression analysis was conducted to select 13 key genes from the statistically significant univariate candidates identified in IPF patients. Subsequent refinement using random forest analysis identified 15 diagnostic markers that were strongly correlated with pulmonary fibrosis (Fig. 4B).Fig. 4Multi-stage bioinformatics analysis identifies prognostic MRGs in IPF. (A) 40 mechanobiology-related signature genes significantly associated with overall patient survival were screened from the GSE70866 dataset. (B) Biomarkers strongly correlated with pulmonary fibrosis diagnosis were identified through LASSO regression and random forest analysis. (C) Seven prognostic core genes with the highest predictive efficiency were selected via venn diagram analysis. (D) A gene expression heatmap illustrates the relationships among the genes. (E) A correlation heatmap demonstrates the associations between the genes. (F) Single-cell transcriptome analysis revealed that the seven core genes are specifically highly expressed in lung epithelial cells. (G) The diagnostic performance of the core genes was evaluated using ROC curves in both the integrated dataset and independent validation set, determining genes with diagnostic value and their corresponding AUC values. (H) Kaplan-Meier survival analysis and log-rank tests revealed a dose-response relationship between core gene expression levels and overall patient survival, along with p-values.*p < 0.05, ***p < 0.001 Multi-stage bioinformatics analysis identifies prognostic MRGs in IPF. (A) 40 mechanobiology-related signature genes significantly associated with overall patient survival were screened from the GSE70866 dataset. (B) Biomarkers strongly correlated with pulmonary fibrosis diagnosis were identified through LASSO regression and random forest analysis. (C) Seven prognostic core genes with the highest predictive efficiency were selected via venn diagram analysis. (D) A gene expression heatmap illustrates the relationships among the genes. (E) A correlation heatmap demonstrates the associations between the genes. (F) Single-cell transcriptome analysis revealed that the seven core genes are specifically highly expressed in lung epithelial cells. (G) The diagnostic performance of the core genes was evaluated using ROC curves in both the integrated dataset and independent validation set, determining genes with diagnostic value and their corresponding AUC values. (H) Kaplan-Meier survival analysis and log-rank tests revealed a dose-response relationship between core gene expression levels and overall patient survival, along with p-values.*p < 0.05, ***p < 0.001 Venn diagram analysis (Fig. 4C) further narrowed the selection to seven core genes with the highest prognostic predictive performance. Single-cell transcriptomic analysis revealed that these core genes were predominantly and specifically expressed in lung epithelial cells (Fig. 4F). Gene expression heatmap clustering, combined with correlation analysis (Fig. 4D,E), demonstrated significant interactions among CD24, PPP1R14C, DSP, and CC2D2A, with DSP and CC2D2A exhibiting the strongest correlation (r = 0.81). To assess diagnostic performance, receiver operating characteristic (ROC) curve analysis was performed on both the integrated dataset and the independent validation set GSE43292 (Fig. 4G). Applying an area under the curve (AUC) threshold greater than 0.7 as the selection criterion, CD24, PPP1R14C, DSP, and CC2D2A were validated as diagnostic markers. Notably, CC2D2A exhibited outstanding diagnostic performance in both datasets (AUC > 0.9), whereas the other genes achieved AUC values of 0.996 for CD24, 0.836 for PPP1R14C, and 0.798 for DSP. Finally, Kaplan–Meier survival analysis combined with log-rank tests (Fig. 4H) revealed a dose–response relationship between core gene expression levels and patient overall survival (OS). High-expression groups of PPP1R14C and CD24 were associated with significantly poorer prognosis (p < 0.001), whereas the low-expression group of CC2D2A showed a significant trend toward reduced survival (p < 0.001). Construction and calibration of the nomogram To evaluate the predictive performance of the final set of mechanistically relevant core prognostic genes, systematic validation was conducted using univariate and multivariate Cox proportional hazards regression analyses (Fig. 5A). Multivariate Cox regression analysis revealed that CD24, PPP1R14C, and CC2D2A retained significant independent prognostic value in the patient cohort after adjustment for potential confounding factors. Furthermore, patient stratification based on Cox model–predicted risk scores demonstrated that the high-risk group was significantly associated with poorer survival outcomes (Fig. 5B). To assess the classification model’s performance, ROC curve analysis was conducted (Fig. 5C), yielding AUC values of 0.82, 0.84, and 0.76 for 1-, 3-, and 5-year predictions, respectively. For individualized survival prediction, a prognostic nomogram was developed based on multivariate Cox stepwise regression analysis (Fig. 5D). This predictive model demonstrated excellent calibration in estimating 1-, 3-, and 5-year survival probabilities, with validation curves indicating strong concordance between predicted and observed survival rates (Fig. 5E), thereby confirming its clinical utility.Fig. 5Construction and calibration of the nomogram. (A) Systematically validated Mechanical-related core prognostic genes through univariate and multivariate cox regression analyses. (B) Grouped patients based on cox model risk scores, revealing poorer survival outcomes in the high-risk group. (C) Evaluated classifier performance using ROC curves, presenting AUC values for 1-, 3-, and 5-year survival. (D) Constructed a predictive nomogram via multivariate cox stepwise regression for individualized survival prediction. (E) Calibration curves demonstrated strong agreement between predicted and observed survival probabilities at 1-, 3-, and 5-year time points Construction and calibration of the nomogram. (A) Systematically validated Mechanical-related core prognostic genes through univariate and multivariate cox regression analyses. (B) Grouped patients based on cox model risk scores, revealing poorer survival outcomes in the high-risk group. (C) Evaluated classifier performance using ROC curves, presenting AUC values for 1-, 3-, and 5-year survival. (D) Constructed a predictive nomogram via multivariate cox stepwise regression for individualized survival prediction. (E) Calibration curves demonstrated strong agreement between predicted and observed survival probabilities at 1-, 3-, and 5-year time points Molecular subtyping of IPF patients based on mechanical - related prognostic genes Based on the identified mechanobiology-related prognostic genes, consensus clustering was applied to achieve molecular stratification of IPF patients. The proportion of ambiguous clustering (PAC) remained low, indicating high stability of clustering solutions and minimal variation in sample allocation across iterations. The optimal cluster number (K = 2) was identified using the minimum PAC criterion. The consensus matrix heatmap (Fig. 6A) for K = 2 distinctly separated patient samples into two subtypes, C1 and C2. Cumulative distribution function (CDF) analysis indicated near-saturation at K = 2 (Fig. 6A), whereas the area under the curve exhibited minimal increase at K = 3 (Fig. 6A), supporting the selection of K = 2. The sample classification tracking plot further corroborated the superior clustering stability at K = 2 (Fig. 6A). Principal component analysis (PCA) visually demonstrated clear separation between the two subtypes (Fig. 6B), validating the stratification accuracy. Notably, core gene expression analysis (Fig. 6C) revealed significant differences among C1, C2, and the control group. Relative to C1, all core genes were downregulated in C2, suggesting a distinct mechanotransduction regulatory mechanism in this subtype.Fig. 6Molecular subtyping of IPF patients based on mechanics - related prognostic genes. (A)Cluster analysis of the original dataset based on mechanistic prognostic genes (including consensus matrix heatmap, CDF curve, and sample classification tracking plot) identified the optimal number of clusters K = 2 and revealed subtype characteristics; (B) Principal component analysis of the original dataset demonstrated a separation trend between the two subtypes; (C) Expression differences of core genes between subtypes and the control group in the original dataset; (D-E) Consensus clustering analysis of validation set GSE32537 using mechanistic prognostic genes; (F) Expression differences of core genes between subtypes and the control group in the validation set. **p < 0.01, ***p < 0.001 Molecular subtyping of IPF patients based on mechanics - related prognostic genes. (A)Cluster analysis of the original dataset based on mechanistic prognostic genes (including consensus matrix heatmap, CDF curve, and sample classification tracking plot) identified the optimal number of clusters K = 2 and revealed subtype characteristics; (B) Principal component analysis of the original dataset demonstrated a separation trend between the two subtypes; (C) Expression differences of core genes between subtypes and the control group in the original dataset; (D-E) Consensus clustering analysis of validation set GSE32537 using mechanistic prognostic genes; (F) Expression differences of core genes between subtypes and the control group in the validation set. **p < 0.01, ***p < 0.001 To validate the pivotal role of mechanosensitive prognostic genes in subtype stratification, the independent dataset GSE32537 was employed. As shown in Figures 6D and 6E, consensus clustering of the validation set yielded an identical subtype classification (K = 2), and the gene expression profiles of both subtypes were highly concordant with those in the original dataset (Figs. 6C,6F). These findings confirm that the two molecular subtypes defined by mechanosensitive core genes exhibit cross-cohort generalizability, underscoring their biological relevance in IPF stratification. Immune infiltration analysis reveals characteristics and related molecular mechanisms of different subtypes in IPF ssGSEA of immune infiltration revealed significant differences between C1 and C2 subtypes of IPF patients (Fig. 7A). Statistical analysis indicated significant disparities in the enrichment levels of multiple immune cell populations between the two subtypes (p < 0.05). The C2 subtype displayed elevated overall immune activity (as evidenced by significantly higher median immune infiltration scores in boxplots) and a pro-inflammatory microenvironment, characterized by markedly greater infiltration of effector memory CD8+ T cells, macrophages, and NK cells relative to C1. In contrast, C1 exhibited immunosuppressive tendencies. Furthermore, correlation analysis among immune cell types (Fig. 7B) revealed widespread positive associations across diverse immune cell populations.Fig. 7Immune infiltration analysis reveals characteristics and related molecular mechanisms of different subtypes in IPF. (A)Differences in immune infiltration between C1 and C2 subtypes in IPF patients. (B)Correlation analysis of immune cell types. (C)Correlation between specific genes and immune cell infiltration. (D)Differences in gene - set scores related to EMT, ECM assembly and fibroblast activation among C1, C2 subtypes and normal control group. (E)Analysis of mechanobiology - related processes in C1 and C2 subtypes. (F)Analysis of genes specifically highly expressed in C2 subtype.*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 Immune infiltration analysis reveals characteristics and related molecular mechanisms of different subtypes in IPF. (A)Differences in immune infiltration between C1 and C2 subtypes in IPF patients. (B)Correlation analysis of immune cell types. (C)Correlation between specific genes and immune cell infiltration. (D)Differences in gene - set scores related to EMT, ECM assembly and fibroblast activation among C1, C2 subtypes and normal control group. (E)Analysis of mechanobiology - related processes in C1 and C2 subtypes. (F)Analysis of genes specifically highly expressed in C2 subtype.*p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 Notably, expression levels of CD24, PPP1R14C, and CC2D2A exhibited strong negative correlations with infiltration of multiple immune cell types in the fibrotic microenvironment (correlation coefficient < −0.6, p < 0.001), most prominently for NK cells (Fig. 7C). These findings suggest that these genes may suppress immune cell activity by modulating the immune microenvironment, encompassing both innate immunity (dendritic cells, NK cells, monocytes/macrophages) and adaptive immunity (T-cell subsets and B cells). This mechanism was especially evident for CD24, whose immunosuppressive function has been previously reported. Scoring of gene sets associated with EMT, ECM assembly, and fibroblast activation revealed significant differences between C1 and C2 subtypes and the normal control group (Fig. 7D). Beyond EMT scores, C2 consistently exhibited higher values than both C1 and controls across all metrics, a trend validated in an independent cohort (Fig. 7D). Analysis of mechanotransduction-related processes revealed significant differences between C1 and C2 in focal adhesion formation and regulation of integrin activation. Pathway scores for MAPK, RhoA/ROCK, TGF-β, and Wnt—each closely linked to mechanical signaling—indicated pronounced activation in C2 (Fig. 7E). In-depth analysis of the most significantly differentially regulated integrin activation gene set identified high-expression patterns of SELP, FERMT1, CXCL13, MZB1, COL16A1, CXCL12, and FBLIM1 in C2 (Fig. 7F). Notably, the chemokines CXCL13 and CXCL12 are pivotal in IPF pathogenesis. CXCL13 expression correlates with disease prognosis and may serve as a prognostic biomarker [24], whereas CXCL12 promotes fibrosis by mediating directional migration of fibrocytes to injury sites [25]. Single-cell and pseudotime analysis to determine the expression patterns of CD24, PPP1R14C, and CC2D2A in IPF samples This study identifies CD24, PPP1R14C, and CC2D2A as prognostic biomarkers in IPF. Preliminary single-cell RNA sequencing analysis indicated that these genes were predominantly localized within lung epithelial cell clusters (Fig. 4F). Accordingly, we reclustered the lung epithelial cell population (Figs. 8A,8B) to comprehensively characterize the expression distribution patterns of CD24, PPP1R14C, and CC2D2A across epithelial subpopulations. To assess changes in cell-type proportions, we generated stacked bar plots (Fig. 8C), which revealed a significant reduction in alveolar epithelial cells and an increase in ciliated cells in IPF lung tissues relative to healthy controls—findings consistent with alveolar epithelial barrier dysfunction and bronchial metaplasia in IPF pathogenesis. UMAP and violin plots illustrated gene expression distributions and levels (Fig. 8D), showing that CD24 and CC2D2A were predominantly enriched in ciliated cells, with CD24 also highly expressed in alveolar epithelial cells, whereas PPP1R14C was broadly expressed across all epithelial subtypes. Notably, CD24 and CC2D2A expression levels were significantly upregulated in IPF samples relative to controls.Fig. 8Single-cell and Pseudotime analysis to determine the expression patterns of CD24, PPP1R14C, and CC2D2A in IPF samples. (A) UMAP visualization of lung epithelial cell type distribution derived from the single-cell RNA sequencing dataset.(B) Bubble plot of marker genes in lung epithelial cell types.(C) Stacked bar plot showing the relative proportions of cellular clusters in IPF and control groups.(D) Distribution and expression patterns of CD24, PPP1R14C, and CC2D2A in epithelial cell populations.(E) UMAP visualization of ciliated cell type distribution in the lung.(F) Cell number distribution of four ciliated cell subtypes along the pseudotime trajectory.(G) Pseudotime analysis of the differentiation trajectory of lung ciliated cells.(H) BEAM heatmap analysis identifying genes associated with cell fate 1 and cell fate 2 differentiation. A total of three clusters were identified.(I) Relative pseudotime expression of CC2D2A, CD24, and PPP1R14C, as well as top genes of Cluster 1 (EGR1 and SOCS3), Cluster 2 (TMC5 and LMO7), and Cluster 3 (AGR2 and RPL21P75) in the BEAM heatmap. ***p < 0.001 Single-cell and Pseudotime analysis to determine the expression patterns of CD24, PPP1R14C, and CC2D2A in IPF samples. (A) UMAP visualization of lung epithelial cell type distribution derived from the single-cell RNA sequencing dataset.(B) Bubble plot of marker genes in lung epithelial cell types.(C) Stacked bar plot showing the relative proportions of cellular clusters in IPF and control groups.(D) Distribution and expression patterns of CD24, PPP1R14C, and CC2D2A in epithelial cell populations.(E) UMAP visualization of ciliated cell type distribution in the lung.(F) Cell number distribution of four ciliated cell subtypes along the pseudotime trajectory.(G) Pseudotime analysis of the differentiation trajectory of lung ciliated cells.(H) BEAM heatmap analysis identifying genes associated with cell fate 1 and cell fate 2 differentiation. A total of three clusters were identified.(I) Relative pseudotime expression of CC2D2A, CD24, and PPP1R14C, as well as top genes of Cluster 1 (EGR1 and SOCS3), Cluster 2 (TMC5 and LMO7), and Cluster 3 (AGR2 and RPL21P75) in the BEAM heatmap. ***p < 0.001 To clarify the differentiation status of ciliated cells in IPF, we performed pseudotime analysis on ciliated cells derived from IPF samples. Based on the inferred differentiation trajectory, we reclustered these cells into four subpopulations (clusters 0–3) (Fig. 8E). Monocle analysis (Fig. 8F, G) revealed that clusters 0 and 3 occupied the terminal ends of the trajectory, with cluster 3 constituting a distinct differentiation branch (cellfate2). Cluster 0 cells were distributed across both terminal branches (cellfate1 and cellfate2), with prominent localization near the bifurcation point and trajectory origin (Fig. 8F). We subsequently performed branch-dependent gene expression analysis (BEAM) at the bifurcation point. The resulting BEAM heatmap grouped genes into three clusters (Fig. 8H), corresponding to cellfate1, cellfate2, and the bifurcation-origin region. GO enrichment analysis indicated that, during dynamic differentiation, ciliated cells from IPF samples exhibited enhanced protein synthesis and secretion toward cellfate1, whereas cellfate2 and the bifurcation-origin region displayed elevated cell adhesion/junction functions and activation of p38 MAPK signaling (Fig. 8H). Visualization of CD24, PPP1R14C, CC2D2A, and the three BEAM-identified high-expression gene sets along the pseudotime axis (Fig. 8I) showed upregulation of CD24 and CC2D2A during differentiation, while other genes (AGR2, EGR1, LMO7, RPL21P75, SOCS3) exhibited elevated expression at terminal stages. Notably, these BEAM-identified gene sets encompassed key fibrosis-associated genes—including AGR2, EGR1, LMO7, RPL21P75, SOCS3, and TMC5—closely linked to pulmonary fibrosis pathogenesis. Intercellular communication between high- and low-scoring groups Analysis of single-cell transcriptomic data from 31 IPF patients revealed a consistent differential expression pattern of core prognosis-associated genes within the lung epithelial cell population (Supplemental Fig. S4). Using these genes, we performed AddModuleScore analysis and stratified samples into high- and low-score groups according to the median score (Fig. 9A, B). Comparative analysis showed a significantly higher proportion of pulmonary ciliated cells and a corresponding reduction in type II alveolar epithelial cells (AT II) in the high-score group relative to the low-score group (Fig. 10C).Fig. 9Intercellular communication between high- and low-scoring groups. (A) UMAP distribution of lung epithelial cell populations in high- and low-scoring groups (B) Violin plots comparing high-, low-scoring, and normal groups (C) Stacked bar chart of epithelial cell subpopulations in high-, low-scoring, and normal groups (D) Number and weight of intercellular communication connections in high- and low-scoring groups (E) Intercellular communication network among epithelial cell subpopulations in single-cell sequencing datasets (red: high-scoring group; blue: low-scoring group) (G) Heatmap showing differences in the number of interactions (left) and interaction strength (right) between epithelial cell types (F) Differences in overall information flow of 36 selected secretory ligand-receptor signaling pathways between high- and low-scoring groups. The information flow was calculated as the sum of communication probabilities between all cell population pairs in the inferred network (i.e., total network weight)Fig. 10C2 subtype-focused Drug discovery and hub gene validation in IPF models. (A) Protein-protein interaction network diagram. The nodes within the circle represent key targets of the C2 subtype identified via the degree algorithm. (B) Expression levels of AGTR2 in the C1 and C2 subtypes in the integrated datasets (GSE2014 and GSE213001). (C) Results of molecular docking performed using SYBYL software. (D) Structures and orthogonal views of the nifekalant-binding pocket in AGTR2. (E) HE staining images of the BLM model group and the control group. (F) Masson staining images of the BLM model group and the control group. (G) Relative mRNA levels of PPP1R14C, CC2D2A, and CD24. *p < 0.05, **p < 0.01,****p < 0.0001 Intercellular communication between high- and low-scoring groups. (A) UMAP distribution of lung epithelial cell populations in high- and low-scoring groups (B) Violin plots comparing high-, low-scoring, and normal groups (C) Stacked bar chart of epithelial cell subpopulations in high-, low-scoring, and normal groups (D) Number and weight of intercellular communication connections in high- and low-scoring groups (E) Intercellular communication network among epithelial cell subpopulations in single-cell sequencing datasets (red: high-scoring group; blue: low-scoring group) (G) Heatmap showing differences in the number of interactions (left) and interaction strength (right) between epithelial cell types (F) Differences in overall information flow of 36 selected secretory ligand-receptor signaling pathways between high- and low-scoring groups. The information flow was calculated as the sum of communication probabilities between all cell population pairs in the inferred network (i.e., total network weight) C2 subtype-focused Drug discovery and hub gene validation in IPF models. (A) Protein-protein interaction network diagram. The nodes within the circle represent key targets of the C2 subtype identified via the degree algorithm. (B) Expression levels of AGTR2 in the C1 and C2 subtypes in the integrated datasets (GSE2014 and GSE213001). (C) Results of molecular docking performed using SYBYL software. (D) Structures and orthogonal views of the nifekalant-binding pocket in AGTR2. (E) HE staining images of the BLM model group and the control group. (F) Masson staining images of the BLM model group and the control group. (G) Relative mRNA levels of PPP1R14C, CC2D2A, and CD24. *p < 0.05, **p < 0.01,****p < 0.0001 Further analysis of intercellular communication revealed that, although overall communication strength and interaction counts did not differ significantly between the high- and low-score groups (Fig. 9D), their respective communication patterns were altered (Fig. 9E). Regarding interaction counts, aberrant basaloid cells in the low-score group formed more active connections with other epithelial subpopulations, whereas in terms of interaction strength, pulmonary neuroendocrine cells (PNECs) in the high-score group exhibited stronger interactions with other epithelial cells (Fig. 9E). To elucidate the molecular mechanisms of cellular communication, we analyzed alterations in information flow across 36 secretory signaling pathways between high- and low-score groups. Results (Fig. 9F) showed that lung epithelial cells in the two groups displayed distinct ligand secretion profiles: the low-score group predominantly secreted molecules driving pro-inflammatory responses and pro-proliferation/migration/remodeling (e.g., IL6, IGF, FGF, UGRP1, SLITRK) [26], whereas the high-score group preferentially secreted factors promoting tissue repair, maintaining homeostasis, and modulating partial inflammatory responses (e.g., HGF, IL1, BMP). Notably, macrophages and their subpopulations emerged as principal responder targets of these epithelium-derived signals, collectively constituting a central interaction hub driving phenotypic divergence between groups. Cellular communication heatmap analysis (Fig. 9G) revealed substantial remodeling of specific subpopulation interactions: Club cells, the most active signal senders, showed the greatest increases in interaction counts with ATI, basal cells, and ionocytes. PNECs exhibited higher signal intensity when transmitting to other lung epithelial subpopulations, with a markedly enhanced bidirectional interaction strength between PNECs and aberrant basaloid cells. These findings suggest that Club cells act as key effectors in maintaining homeostasis, whereas the aberrantly enhanced bidirectional interaction between PNECs and aberrant basaloid cells indicates a potential pathological signaling axis. Such dysregulated communication patterns likely represent a critical mechanism driving pulmonary tissue progression toward fibrosis or precancerous lesions. Drug screening and molecular docking targeting the C2 subtype To identify potential therapeutic compounds targeting the characteristic upregulated genes of the C2 subtype (Supplemental Table S2), we conducted a CMap analysis, which yielded five lowest-scoring candidates: amiloride, NS-3694, zileuton, AC-55649, and necrostatin-1 (Supplemental Fig. 7A). Subsequently, to investigate key molecular targets in the C2 subtype, we constructed a protein–protein interaction (PPI) network of 10 interacting proteins (Fig. 10A). Degree-based network analysis revealed AGTR2 as the central node in the topological structure. In human lung fibroblasts, AGTR2 expression is inducible by certain cytokines, and its activation modulates ERK1/2 signaling, intracellular calcium release, and cytoskeletal reorganization, thereby influencing proliferation and migration and contributing to pulmonary fibrosis initiation and progression [27]. Transcriptomic data indicated significantly higher AGTR2 expression in the C2 subtype than in C1 (Fig. 10B). To identify promising AGTR2-targeting compounds, we used SYBYL software to perform molecular docking on low CMap-scoring candidates. This method evaluates compound–target binding stability via a Total Score metric, enabling precise selection of candidates with strong affinity. Docking results showed seven compounds with Total Scores > 4, indicating favorable binding stability with AGTR2. Among them, nifekalant achieved the highest Total Score (8.81) (Fig. 10C). To elucidate the molecular interaction mechanism, we visualized the three-dimensional binding pattern between nifekalant and AGTR2. Active-site interaction analysis (Fig. 10D) showed nifekalant forming a stable hydrogen-bond network with key residues THR-178, SER-174, and THR-125. Hub gene upregulation in BLM-Induced pulmonary fibrosis To examine hub gene expression changes, we employed a bleomycin (BLM)-induced mouse pulmonary fibrosis model. Hematoxylin–eosin (HE) staining (Fig. 10E) revealed disrupted alveolar architecture, inflammatory cell infiltration, and fibrosis in BLM-treated mice relative to controls. Masson’s trichrome staining (Fig. 10F) demonstrated marked collagen deposition. Total RNA was extracted from lung tissues of saline-treated controls and BLM-treated mice, and CD24, PPP1R14C, and CC2D2A expression levels were quantified by qPCR. qPCR results showed significantly elevated mRNA levels of CD24 (p < 0.05), PPP1R14C (p < 0.01), and CC2D2A (p < 0.01) in BLM-treated mice relative to controls (Fig. 10G). Discussion Mechanical forces are intimately linked to the pathogenesis of multiple systemic diseases, such as cardiovascular disease [28 ], p]ulmonary fibrosis [29], and osteoporosis [30]. Extant literature highlights the critical involvement of mechanical forces in IPF pathogenesis [31]. Although these forces substantially modulate fibrotic disease progression, their precise mechanistic contributions in IPF remain poorly elucidated. Here, we systematically examine the prognostic implications of mechanical forces in IPF and characterize their mechanistic subtypes. Mechanical force–feature scoring demonstrated consistently elevated scores in IPF patient samples relative to controls, revealing distinct identified 465 IPF signature genes correlated with mechanical scores. Gene set enrichment analysis indicated that dysregulated cilia-associated gene expression may underlie the observed mechanical heterogeneity in IPF samples. Structural ciliary defects could compromise epithelial extracellular fluid transport and perturb mechanosensory function. Collectively, this study provides a comprehensive molecular perspective on the dysregulation of the mechanical microenvironment in IPF. Analysis of the core genes identified two distinct molecular subtypes in IPF patient samples. The C2 subtype was characterized by pronounced inflammatory responses, robust activation of mechanotransduction pathways, and more severe pathological manifestations, providing novel molecular insights into IPF heterogeneity. Based on the patient's clinical information, it is inferred that the phenotypic features of the C2 subtype may be mediated by a specific mechanotransduction or fibrosis-driving pathway that is independent of tobacco exposure but more readily activated in male individuals(Supplemental Fig. S9). Furthermore, scRNA-seq of BLM mouse model (GSE141259; Supplemental Fig. S5A) revealed a temporal expression pattern of CD24 in lung epithelial cells, peaking at day 21 during fibrosis development and declining by day 28. This temporal profile corresponded to reduced CD24 expression and severe pathological features in the C2 subtype. CMap analysis identified potential therapeutic agents for the C2 subtype, with nifekalant—an AGTR2-targeting drug—emerging as a promising candidate. Previous studies indicate that AGTR2 antagonism attenuates pulmonary fibrosis in bleomycin-induced mouse models and also suppresses the migration and proliferation of fibrotic fibroblasts [32]. Using univariate Cox regression and advanced machine learning algorithms (LASSO and random forest), we identified seven core genes. Analysis of internal training and external validation cohorts identified CD24, PPP1R14C, DSP, and CC2D2A as diagnostic biomarkers for IPF. Stepwise multivariate Cox regression further refined the model, identifying three independent prognostic genes: CD24, PPP1R14C, and CC2D2A. Integration of these gene markers with key clinicopathological variables enabled the development of a predictive nomogram, providing a novel quantitative tool for precise, individualized survival prediction. CD24 encodes a glycosylphosphatidylinositol (GPI)-anchored cell-surface protein mapped to human chromosome 6. It plays a pivotal role in cell adhesion, signal transduction, and immune response regulation [33] and may contribute to pulmonary fibrosis by modulating fibroblast phenotypic transformation. Additionally, adjacent alveolar septa may signal via CD24-mediated pathways, while immune infiltration may also communicate with fibroblast foci through CD24 [34]. CD24, enriched in lipid rafts, can assemble with EGFR and integrin β1 to form a “mechano–chemical co-receptor” complex, thereby amplifying ERK/JNK signaling [35]. We propose that elevated matrix stiffness in fibrotic foci promotes integrin β1 clustering, activating the FAK–Src–ERK cascade [20]. ERK-driven AP-1 then binds the CD24 promoter (~–1.2 kb), upregulating CD24, which may further stabilize EGFR signaling and drive alveolar type II epithelial cells toward a pro-fibrotic phenotype. PPP1R14C belongs to the protein phosphatase 1 regulatory subunit family, which modulates intracellular signaling pathways essential for cell proliferation and migration [36]. Notably, the family member PPP1R14D is overexpressed in lung adenocarcinoma and correlates with poor prognosis [37]. Previous studies have demonstrated that PPP1R14C, phosphorylated at Thr38 by RhoA–ROCK, inhibits MLCP and acts as a key “Ca2 + sensitization” switch in smooth muscle and myofibroblasts [38, 39]. Sustained tissue tension may thus activate RhoA–ROCK, phosphorylating PPP1R14C, suppressing MLCP, and maintaining myofibroblast contractility; its phosphorylated form may resist proteasomal degradation, creating a “mechanical memory” that promotes fibrosis. CC2D2A, located on human chromosome 4, encodes a protein containing coiled-coil and C2 domains. It is expressed in multiple tissues, including the lungs [40]. The CC2D2A protein, a constituent of the ciliary basal body, is essential for ciliogenesis. It interacts with CEP290, a protein implicated in multiple ciliopathies [41]. Similarly, CC2D2A, a ciliary transition zone scaffold, interacts with vesicular trafficking proteins such as NINL and RAB8 [42]. Although direct CaM binding evidence is limited, Ca2 + –CaM complexes regulate vesicular exocytosis [43]. Mechanical stretch or matrix stiffening can activate TRPV4 [44] and Piezo1, inducing Ca2 + influx that engages CaM-dependent pathways [43], Potentially enhancing CC2D2A-mediated trafficking. In secretory cells, Ca2 + signaling promotes TGF-β1 release, activating Smad2/3 in fibroblasts. Collectively, CC2D2A may integrate Ca2 + signals, mechanical cues, and pro-fibrotic cascades. To elucidate the potential roles and molecular mechanisms underlying IPF, we conducted single-gene GSEA using Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway gene sets for each of the three prognostic biomarkers. The analysis revealed significant downregulation of all three prognostic genes in the TGF-β signaling and ECM–receptor interaction pathways in the high-expression groups (Supplemental Fig. S6). In IPF, TGF-β signaling drives fibroblast activation and ECM deposition, culminating in progressive lung scarring [45]. Similarly, aberrant ECM–receptor interactions disrupt tissue architecture and cellular communication, fostering fibroblast proliferation and migration via altered ECM composition and receptor signaling, thereby exacerbating fibrosis [46]. This study has several limitations: (1) classification of mechanical subtypes is based solely on transcriptomic features, necessitating validation of spatial mechanical microenvironment heterogeneity through spatial transcriptomics; 2) the single-cell dataset has a limited sample size, requiring larger cohorts to confirm the stability of gene expression; 3) The potential therapeutic agent nifekalant, identified for the IPF C2 subtype, has not yet been validated in vivo, and its efficacy remains to be confirmed through animal models or clinical studies; 4) The prognostic utility of the proposed biomarker panel requires further validation in prospective, multicenter studies involving multi-ethnic populations. To further elucidate the causal roles and therapeutic potential of CD24, PPP1R14C, and CC2D2A in idiopathic pulmonary fibrosis (IPF), we plan to conduct systematic mechanistic and interventional studies. At the in vitro level, CRISPR-Cas9 technology will be employed to generate knockout models of alveolar epithelial cells and fibroblasts for the aforementioned genes. Fibrotic phenotypes will be induced using bleomycin and TGF-β1 to assess their effects on the TGF-β/Smad and Hippo/YAP/TAZ signaling pathways, as well as extracellular matrix remodeling, with validation via qPCR and Western Blot. For in vivo experiments, cell type-specific conditional knockout mouse models (e.g., Col1a2-Cre/ER, Sftpc-Cre/ER) will be constructed to analyze lung tissue phenotypes and transcriptomic changes at multiple time points, aiming to reveal the cell-specific mechanisms of these genes. In terms of drug evaluation, chemically induced and mechanically induced models will be combined to simulate the IPF microenvironment in vitro. Preliminary assessments of the anti-fibrotic effects of compounds such as nifekalant will be conducted in bleomycin-induced mouse models, with comparisons to AGTR2 inhibitors. If results are promising, further validation techniques, including gene rescue, overexpression/knockdown, and interaction assays such as SPR/ITC, will be employed to elucidate drug targets and mechanisms of action, thereby providing a foundation for clinical translation. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2 Supplementary Material 3 Supplementary Material 3 Supplementary Material 4 Supplementary Material 4