Development of a prognostic prediction signature for idiopathic pulmonary fibrosis by integrating multiple programmed cell death-related genes and machine learning algorithms

Introduction

Idiopathic pulmonary fibrosis (IPF) is a chronic progressive interstitial lung disease characterized by pulmonary fibrosis and destruction of the alveolar structure, primarily manifested by symptoms such as dyspnea, cough, and worsening lung function (1). Currently, the main pharmacological treatments for early-stage IPF patients include pirfenidone (2) and nintedanib (3); although these drugs have been shown to slow the decline in lung function, their therapeutic effects remain limited and do not fundamentally alter the disease’s progression (4). For advanced IPF patients, lung transplantation is the only treatment option available. However, due to the scarcity of donor organs and the associated surgical risks, many patients are unable to access this treatment. Therefore, early detection and accurate prognostic predictive of IPF are crucial for improving patient outcomes and quality of life (5). Despite recent advances, existing prognostic tools for IPF remain limited in scope and accuracy. Current clinical models, such as the Gender-Age-Physiology (GAP) index (6), rely heavily on conventional parameters (e.g., pulmonary function decline, radiological staging) (7), or single circulating biomarkers (8), which often lack molecular specificity. Furthermore, gene expression–based prognostic models developed to date typically focus on isolated signaling pathways or individual gene sets, without considering the interconnected cellular processes driving alveolar epithelial cell (AEC) loss and fibroblast activation.

In IPF, abnormal programmed cell death (PCD) not only affects the survival and function of AECs but also significantly influences the activation and quantity of fibroblasts (9). Excessive apoptosis of AECs is considered a crucial trigger in the early stages of IPF, while autophagy plays a dual role in protecting and damaging cells. Additionally, necrosis and other novel forms of cell death, such as pyroptosis (10) and ferroptosis (11), are gradually being revealed as involved in the pathological processes of IPF. Modulating cell death pathways to protect AECs or inhibit fibroblast activation to slow down or reverse the pathological progression of IPF is currently a valuable research direction. However, few studies have incorporated these mechanisms into predictive modeling. This oversight may limit the biological relevance and clinical utility of existing signatures.

In light of this, there is growing interest in identifying mechanism-informed biomarkers that not only predict disease progression but also reflect key pathological processes (12). This study aimed to develop a risk score prognostic model based on core PCD-related genes involved in IPF, providing a new perspective for clinical individualized prognostic assessment and promoting more precise treatment strategies. We present this article in accordance with the ARRIVE and TRIPOD reporting checklists (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2024-2173/rc).

Methods

Workflow and data sources of the study

A protocol was prepared before the study without registration. This study incorporated multiple datasets related to IPF from the Gene Expression Omnibus (GEO) (13), including GSE24206, GSE47460, GSE53845, GSE93606, and GSE110147. These datasets, derived from various sample sources such as lung tissue and blood, were utilized during the gene discovery phase to identify robust PCD-related hub genes associated with IPF. For model construction and validation, we exclusively selected bronchoalveolar lavage (BAL) transcriptomic datasets with high-quality expression profiles and available survival data. Specifically, GSE70866, which includes 196 BAL-derived samples with longitudinal clinical follow-up, was designated as the training dataset, serving for both training and internal validation. GSE70867, an independent BAL dataset, was used as the external validation cohort. The consistent use of BAL samples across training and validation phases was intended to minimize tissue-specific bias, enhance transcriptomic comparability, and improve the biological validity of the prognostic model. The key regulatory genes of 16 PCD patterns were obtained from various sources, including the Kyoto Encyclopedia of Genes and Genomes (KEGG) Database (14), the GeneCards Database (15), the Molecular Signatures Database (16), the Reactome Database (17) and review articles. The final gene list of 16 different patterns of PCD are presented in available online: https://cdn.amegroups.cn/static/public/jtd-2024-2173-1.xlsx. The code used for machine learning analysis and model construction is available at https://github.com/jiangrx0711/code-of-JTD-2024-2173.

Prognostic model based on multiple machine learning combinations

To construct a high-accuracy and stable predictive model based on PCD-related genes, we employed a combination of various machine learning techniques. We integrated four machine learning algorithms: random forest (RF), extreme gradient boosting (XGboost), generalized boosted regression modeling (GBM), and support vector machine-recursive feature elimination (SVM-RFE). The process of generating the prognostic model involved the following steps:

• Differential analysis was conducted between normal samples and IPF samples across the abovementioned five datasets, setting the criteria of log fold change (FC) >0.58 and P<0.05, and these genes were regarded as key hub genes in IPF samples.
• The previously identified hub genes were intersected with various PCD-related genes to obtain key PCD-related genes in IPF. Subsequently, we applied four types of machine learning algorithms to derive crucial genes.
• In the training cohort of GSE70866, we utilized the ‘limma’ and ‘unicox’ R packages to filter differentially expressed genes (DEGs) and prognostically relevant genes. We then intersected these genes and applied the least absolute shrinkage and selection operator (LASSO)-Cox algorithm to construct the predictive model.

Enrichment analysis

The R software and its affiliated packages “org.hs.eg.db”, “clusterprofiler”, and “ggplot2” were utilized to perform Gene Ontology (GO) and KEGG pathway analyses. Based on Homo sapiens, and screening criteria of P.adjust <0.1 and P value <0.2 were used to identify the primary enriched functions and pathways.

Development and evaluation of nomogram

To validate the value of the cell death score (CDS) as an independent prognostic indicator for patients with IPF, we developed a prognostic nomogram based on the GSE70866 cohort using the R packages “rms” and “replot”. The performance of these nomograms was evaluated through calibration curves and receiver operating characteristic (ROC) curves.

Immune cell enrichment analysis

Gene expression data from GSE70866 were combined with the CIBERSORT algorithm to assess the abundance of immune cells in individual samples.

Potential drug screening

Prognostic-related DEGs identified from the differential analysis in GSE70866 (logFC >0.585) were input into the Broad Institute’s Connectivity Map (cMAP, https://clue.io/command?q=/home) database (18) for enrichment analysis. The expression correlations of the targets of selected drugs with the genes in the prognostic model were analyzed (correlation coefficient =0.2).

Animals models

The 8-week-old Specific Pathogen-Free (SPF) grade male C57Bl/6 mice were raised in the Experimental Animal Center of Xi’an Jiaotong University Health Science (at 23–25 ℃, relative humidity of 50%, and a 12-h night/day cycle). Mice were randomly assigned to either the control group (saline) or the bleomycin-treated group (n=3 per group). Pulmonary fibrosis was induced by a single intratracheal administration of bleomycin (Selleck, NSC125066) at a dose of 5 mg/kg. On day 21 post-treatment, mice were euthanized, and lung tissues were harvested and stored at −80 ℃ for subsequent RNA extraction and quantitative polymerase chain reaction (PCR) analysis. All animal experiments were performed under a project license (No. XJTU1AF-CR-2023-006) granted by Ethics Committee of Xi’an Jiaotong University, in compliance with the national guidelines for the care and use of animals.

Quantitative real-time PCR (qRT-PCR)

We utilized the seven identified genes—TLR2, PDCD1LG2, AKT3, SLC1A4, ANTXR2, HTRA1, and TIMP1—to validate the mRNA expression of the prognostic genes. Primers used in this study are shown in Table 1.

Ethical approval

As the sequencing data generated by GEO are publicly available, no additional ethical committee approval was required. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Statistical analysis

Statistical analyses were performed using R software (version 4.1.0). Continuous variables are reported as means with standard errors and were compared using Student’s t-test or the Wilcoxon rank-sum test. Categorical data were assessed using the Chi-squared test. Statistical significance was defined as P<0.05.

Results

Identification of core genes related to PCD in IPF patients

The workflow of this study is shown in Figure 1.

Figure 1 Flow chart of this research. DEG, differentially expressed gene; IPF, idiopathic pulmonary fibrosis; LASSO, least absolute shrinkage and selection operator.

Five datasets from GEO were included for differential analysis, specifically using the “DESeq2” package for identification. A total of 14,218 genes were identified as key hub genes in IPF (Figure 2A), with the specific differential analysis results for each dataset presented in available online: https://cdn.amegroups.cn/static/public/jtd-2024-2173-2.xls. Intersecting the hub genes with PCD-related genes, we obtained 1,491 hub genes associated with PCD in IPF (Figure 2B). Subsequently, differential analysis and univariate Cox regression analysis were conducted (Figure 2C,2D), resulting in 241 prognosis-related DEGs (Figure 2E), which will be used for machine learning screening and constructing the prognostic model. The 241 DEGs were classified based on PCD, and without excluding the possibility that a single gene may participate in multiple forms of PCD, apoptosis was found to potentially have the greatest impact on IPF prognosis. Specifically, 68 genes associated with intrinsic apoptosis and 84 genes associated with extrinsic apoptosis were identified as prognosis-related DEGs. The next analyses focused on immunogenic cell death, necrosis, necroptosis, and ferroptosis (Figure 2F).

Figure 2 Identification of the DEGs for machine learning. (A) Differential analyses were performed from GSE53845, GSE93606, GSE110147, GSE47460 and GSE24206, respectively, to screen for key hub genes in IPF. (B) The key hub genes were taken to intersect with PCD related genes. (C) Prognostic analyses were performed in GSE70866 using the uniCox algorithm. (D) The volcano plot of DEGs between IPF patients and control group from GSE70866. Black dots represent genes with P<0.05. (E) The Venn diagram of the DEGs and prognostic genes in GSE70866. (F) Types of PCD involved in 241 prognostically relevant DEGs. CI, confidence interval; DEG, differentially expressed gene; IPF, idiopathic pulmonary fibrosis; PCD, programmed cell death.

Enrichment analysis

To further investigate the potential biological functions of the DEGs in IPF, GO and KEGG enrichment analyses were constructed. The results of the GO enrichment analysis indicated that the DEGs were mainly enriched in cell death, PCD, apoptosis, and the regulation of PCD (Figure 3A). The KEGG pathway enrichment analysis showed that the DEGs were primarily enriched in pathways such as apoptosis, the TNF signaling pathway, and the NF-κB signaling pathway (Figure 3B).

Figure 3 Functional enrichment analysis. The GO (A) and KEGG (B) pathways of 241 prognostic DEGs. DEG, differentially expressed gene; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Construction of prognostic gene features for IPF patients using multiple machine learning algorithms

Four validated machine learning algorithms [XGB, RF, SVM-RFE, generalized linear model (GLM)] were applied to identify key feature genes associated with IPF (Figure 4A,4B). Among the four machine learning methods, the GLM selection yielded the lowest predictive performance, with an area under the curve (AUC) of only 0.628, while the other three methods had AUC exceeding 0.95 (Figure 4C). Each of the four machine learning methods was used for screening, and the top ten most important genes were identified (Figure 4D). Additionally, seven feature genes (TLR2, PDCD1LG2, AKT3, SLC1A4, ANTXR2, HTRA1, and TIMP1) were identified using the LASSO algorithm and were ultimately selected as biomarkers for IPF to construct the prognostic prediction model (Figure 4E,4F). The following formula was used to calculate the CDS for each patient: CDS = (0.4002 × TLR2 exp) + (−0.0751 × PDCD1LG2 exp) + (−0.058 × AKT3 exp) + (0.0179 × SLC1A4 exp) + (0.1027 × ANTXR2 exp) + (0.1530 × HTRA1 exp) + (0.1570 × TIMP1 exp).

Figure 4 Machine learning: identification of crucial genes by using four machine-learning algorithms. (A) The feature importance created by four machine learning models. (B) Reverse cumulative distribution of residuals in four machine learning models. (C) ROC analysis of four machine learning models. (D) Top 10 genes for importance of four machine learning outputs. LASSO Cox analysis identified seven-gene signature in GSE70866 (E), and the (F) optimal λ was selected by cross-validation. GLM, generalized linear model; LASSO, least absolute shrinkage and selection operator; RF, random forest; ROC, receiver operator characteristic; SVM, support vector machine; XGB, extreme gradient boosting.

Model predictive performance validation

Based on the calculated CDS values, we divided the IPF patients in the GSE70866 and GSE70867 datasets into CDS-high and CDS-low groups. In both cohorts, the CDS effectively distinguished between the CDS-high and CDS-low groups (Figure 5A,5B). Furthermore, the selected biomarkers demonstrated good predictive performance in both the training and validation cohorts, with the AUC for 1–3 years in the training cohort of 0.819, 0.805, and 0.790, respectively (Figure 5C); in the validation cohort, the AUC for 1–3 years were 0.797, 0.768, and 0.779, respectively (Figure 5D). Both groups of samples showed better prognostic predictions in the low-risk group (Figure 5E,5F).

Figure 5 Construction and validation of CDS related prognostic model. Distribution of risk scores according to the (A,B) overall survival, survival status and heatmap of seven PCD-related genes’ expression in high/low-risk groups. (C,D) ROC curve analysis and (E,F) Kaplan-Meier survival analysis of patients in GSE70866. AUC, area under the curve; CDS, cell death score; CI, confidence interval; HR, hazard ratio; PCD, programmed cell death; ROC, receiver operator characteristic.

Development and evaluation of nomogram

To evaluate the independent prognostic significance of the CDS, we constructed a prognostic nomogram model in the training cohort to predict the overall survival (OS) of IPF patients at 1, 3, and 5 years, with risk stratification related to the prognostic model shown in (Figure 6A). Calibration curves demonstrated the accurate predictive ability of the nomogram model for 1-, 3-, and 5-year survival rates (Figure 6B). The AUC for the CDS-related risk model is 0.907, better than nomogram and age (Figure 6C). In the gene interaction network, there were 20 nodes surrounding the 7 central nodes, with significant co-expression patterns observed among HLA-DQB2, IL3RA, IFNGR1, and FPR3, and these four nodes were identified as participating in immune receptor activity (Figure 6D).

Figure 6 Construction of nomogram and GeneMANIA analysis. (A) The nomogram based on the risk-related prognostic signature and clinicopathological factors, including gender and age. ***, P<0.001. (B) Calibration curves for the nomogram for predicting 1-, 3-, and 5-year survival in IPF patients from GSE70866. (C) ROC curve of nomogram, risk-score and age. (D) GeneMANIA analysis, red indicates the immune receptor activity. AUC, area under the curve; IPF, idiopathic pulmonary fibrosis; OS, overall survival; ROC, receiver operator characteristic.

CDS correlates with immune characteristics in IPF patients

To further elucidate the potential association between risk scores and immune responses, we quantified immune cell infiltration using gene expression data analyzed by CIBERSORT. In the GSE70866 samples, the content of B cells was relatively high (Figure 7A). Figure 7B explores the interactions among immune cells, showing a strong positive correlation between dendritic cells and naïve B cells (r=0.56). In contrast, activated and resting natural killer (NK) cells exhibited the most pronounced negative correlation (r=−0.75).

Figure 7 The immune infiltration analysis in GSE70866. (A) Stacking diagram by CIBERSORT algorithm. (B) Correlation matrix of ratios of immune cells. (C) Box-plot showing immune infiltration levels differ in risk groups. *, P<0.05; **, P<0.01; ***, P<0.001. (D) Analysis of the correlation between seven CDS-risk related genes’ expression and immune cell infiltration. (E) Kaplan-Meier survival curve of patients grouped by the degree of monocyte infiltration. (F) Kaplan-Meier survival curve of patients grouped by the degree of infiltration of activated NK cells. CDS, cell death score; NK, natural killer.

When assessing the differences in immune cell infiltration between risk groups, we found significant differences in naïve and memory B cells, memory CD4⁺ T cells (resting and activated), resting and activated NK cells, monocytes, M1/M2 macrophages, dendritic cells, and neutrophils (Figure 7C). Among these immune cell infiltrates, monocytes and activated NK cells were more prevalent in the high-risk group. We then performed correlation analysis between the expression levels of the seven signature genes and the relative abundance of 22 immune cell subsets estimated by CIBERSORT. We found that monocytes and activated NK cells exhibited the strongest positive correlations with several signature genes, including TLR2, TIMP1, and HTRA1. These results suggest that the poor prognosis associated with high-CDS patients may be mediated by an immunosuppressive and pro-fibrotic microenvironment involving innate immune activation and matrix remodeling (Figure 7D). Based on CIBERSORT-derived immune profiles in the GSE70866 dataset, we stratified IPF patients into high- and low-infiltration groups for monocytes and activated NK cells, respectively. Kaplan-Meier survival curves demonstrated that patients with higher infiltration levels of either monocytes or activated NK cells had significantly worse OS (log-rank P<0.05 for both comparisons), suggesting their adverse prognostic roles. (Figure 7E,7F).

Potential drug screening

To identify potential therapeutic drugs for IPF associated with CDS, we screened for small molecules on the basis of prognosis-related DEGs related to PCD, resulting in the identification of several small molecules. Combining these findings with existing literature, we identified a total of 25 small molecules. These drugs mainly include mineralocorticoid receptor antagonists, histone deacetylase (HDAC) inhibitors, acetylcholine receptor antagonists, 5-alpha reductase inhibitors, and thyrotropin-releasing hormone receptor agonists (Figure 8A). Vorinostat has been reported as a HDAC inhibitor that promotes fibroblast apoptosis and improves pulmonary fibrosis in mice (19). In order to explore the potential mechanism of abovementioned anti-IPF drugs, we started a KEGG enrichment analysis by 106 drug targets from five mentioned drug types. These drug targets enriched in several signal pathways, such as: MAPK signal pathway, pathways in cancer, and oxytocin signal pathway (Figure 8B). Therefore, we performed co-expression analysis of the HDAC inhibitor-related targets of vorinostat with the genes associated with the CDS prognostic model and constructed a correlation network (Figure 8C).

Figure 8 Related drug screening and potential downstream drug targets. (A) Sankey diagram of 25 IPF potential drugs; left: drug name, middle: mechanism of action, right: primary target. (B) KEGG enrichment analysis of drug targets. (C) Co-expression network of CDS-risk related genes with vorinostat’s targets. CDS, cell death score; IPF, idiopathic pulmonary fibrosis; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Expression validation of the seven prognostic genes in IPF models

Differential expression analysis was first conducted using 20 normal samples and 176 IPF samples from the GSE70866 dataset, which includes transcriptomic data derived from human BAL cells. The results showed that all seven genes included in the prognostic model (TLR2, PDCD1LG2, AKT3, SLC1A4, ANTXR2, HTRA1, and TIMP1) were significantly dysregulated in IPF samples compared to controls (P<0.05 for all; Figure 9A).

Figure 9 Differential expression of the seven prognostic genes in human and murine IPF samples. (A) Expression levels of TLR2, PDCD1LG2, AKT3, SLC1A4, ANTXR2, HTRA1, and TIMP1 in BAL samples from IPF patients and healthy controls in the GSE70866 dataset. (B) Relative mRNA expression of the same genes in lung tissues from bleomycin-induced pulmonary fibrosis mice versus control mice (n=3), as assessed by qRT-PCR. *, P<0.05; **, P<0.01; ***, P<0.001; ****, P<0.0001. BAL, bronchoalveolar lavage; IPF, idiopathic pulmonary fibrosis; qRT-PCR, quantitative real-time polymerase chain reaction.

To validate these findings in vivo, we established a bleomycin-induced mouse model of pulmonary fibrosis and performed qRT-PCR on lung tissues from IPF mice and matched controls. Consistent with the human BAL data, all seven genes showed significantly altered expression in fibrotic lung tissues (P<0.05; Figure 9B). While species-specific differences may exist, this experimental validation supports the biological relevance and cross-tissue consistency of the identified prognostic genes in pulmonary fibrosis. In addition, we analyzed gene expression patterns of the seven signature genes in several independent IPF-related transcriptomic datasets from the GEO database, including GSE24206, GSE47460, GSE53845, GSE93606 and GSE110147 (Figure S1). Consistent with our findings in GSE70866, the majority of genes exhibited similar dysregulation trends between IPF and control samples across datasets. However, we also observed partial variability in gene expression levels, likely due to differences in sample sources, cohort characteristics, or technical platforms. These results further underscore the robustness and partial context-dependence of the model genes across heterogeneous IPF datasets.

Discussion

This study incorporated 16 types of PCD related genes to construct a prognostic prediction model for patients with IPF from a comprehensive perspective of PCD. Through this model, we calculated a risk score based on PCD for each IPF patient and identified apoptosis that significantly influences prognosis. Additionally, we explored the relationships between the prognostic model and immune cell infiltration, functional enrichment, and potential drug screening, revealing the critical role of PCD in IPF and providing new directions for future personalized treatment and drug development.

Both apoptosis and apoptosis resistance are associated with fibrosis in multiple organ systems (20). Increased damage and apoptosis of AECs are considered initiating events in pulmonary fibrosis. Signals released by apoptotic cells, such as inflammatory mediators and cytokines, promote the activation and proliferation of fibroblasts, enhancing extracellular matrix (ECM) deposition, which leads to the progression of pulmonary fibrosis (5). Enrichment analysis in this study revealed that these DEGs are enriched primarily in apoptosis, the tumor necrosis factor (TNF) pathway, and the NF-κB pathway. TNF activates downstream signaling pathways through its receptors, particularly TNFR1, inducing apoptosis. Geranylgeranylacetone (GGA), as an inducer of HSP70, enhances the transition of BEAS-2B cells from epithelial to mesenchymal cells via the NF-κB/NOX4/ROS signaling pathway, significantly reducing apoptosis in BEAS-2B cells. Additionally, tetraspanin 1 inhibits TNF-α-induced cell apoptosis in AECs by activating the NF-κB signaling pathway (21). These findings reveal the critical role of apoptosis in pulmonary fibrosis and the complex interactions of the associated signaling pathways.

In this study, immune-related analyses indicated differences in monocytes, M2 macrophages, and NK cells between high-risk and low-risk groups. The observed positive correlation between monocytes/NK cells and key prognostic genes such as TLR2 and TIMP1 is biologically plausible. TLR2 is a well-known activator of innate immune responses and monocyte/macrophage recruitment (22), while TIMP1 has been implicated in matrix accumulation and immune modulation in fibrotic lungs (23). These findings further support the idea that the CDS score reflects not only molecular alterations, but also changes in the immune cell landscape that may contribute to disease progression. Kreuter et al.’s research suggests that the number of monocytes serves as a prognostic biomarker for IPF, with elevated monocyte counts associated with disease progression, increased hospitalization, and mortality risk (24). The high-risk group showed a higher infiltration rate of monocytes, and survival analysis revealed that higher monocyte infiltration correlated with lower survival rates, consistent with previously reported findings (25). NK cells are thought to mitigate bleomycin-induced pulmonary fibrosis (BIPF) progression by providing antifibrotic mediators and cytokines like IFN-γ; however, Monnier’s research (26) shows that depleting NK cells with anti-asialo-GM1 does not change the progression of fibrosis or affect levels of pro-inflammatory/fibrotic cytokines. Activating NK cells and promoting the apoptosis of hepatic stellate cells can inhibit liver fibrosis progression (27).

Macrophages (28), which differentiate from monocytes, are a crucial component of the immune system, involved in phagocytosis, pathogen clearance, inflammation regulation, and tissue repair. In pulmonary fibrosis, the degree of macrophage infiltration is closely related to the extent of fibrosis (29). The observed increase in macrophages in the low-risk group may reflect enhanced anti-inflammatory and repair capacities, aiding in the mitigation of fibrosis. In mice treated with a combination of bleomycin and lipopolysaccharide, the expression levels of the necroptosis marker mixed lineage kinase domain-like (MLKL) and CD68-positive macrophages were increased compared to those treated with bleomycin alone (30). When tissues are damaged or stimulated by inflammation, M0 macrophages differentiate into M2 macrophages under the influence of specific stimuli like cytokines interleukin (IL)-4 and IL-13. M2 macrophages have the capacity to promote fibroblast proliferation and collagen deposition, leading to the progression of fibrosis (31). Further studies are needed to clarify the roles of immune cells and PCD in IPF, which may help identify relevant potential drug targets.

Vorinostat is an HDAC inhibitor that induces cell differentiation, blocks the cell cycle, and regulates cell functions. It promotes fibroblast apoptosis and improves pulmonary fibrosis in mice (32). The pan-HDAC inhibitor suberoylanilide hydroxamic acid (SAHA) has potential antifibrotic activity by preventing the downregulation of cyclooxygenase-2 (COX-2) in human lung fibroblasts induced by transforming growth factor beta-1 (TGF-β1) (19). Screening from the cMAP database suggests that vorinostat may be effective in treating IPF. The co-expression network of genes related to this prognostic model and vorinostat’s reported drug targets indicates a potential connection. Vorinostat also disrupts autophagy by inhibiting HDACs, promoting cancer cell survival (33). This study focuses on PCD and IPF, suggesting vorinostat may influence IPF through autophagy, though more research is needed to elucidate the mechanisms (34). Beyond vorinostat, we identified several other small-molecule compounds with reported anti-fibrotic or anti-inflammatory effects through literature mining. For instance, mineralocorticoid receptor antagonists (e.g., spironolactone) have been shown to alleviate cardiac and renal fibrosis by modulating inflammatory and fibrotic signaling pathways (35,36). Acetylcholine receptor antagonists have demonstrated efficacy in attenuating airway remodeling and fibrosis in chronic lung disease models by suppressing cholinergic-driven inflammation (37,38). Other compounds, including insulin-like growth factor 1 (IGF-1) inhibitors (39) and natural molecules targeting GSK-3β/β-catenin signaling (40), also show promise in regulating fibroblast activation and epithelial-mesenchymal crosstalk—key pathological features shared with IPF. Collectively, these drug classes represent potential therapeutic candidates that align mechanistically with the PCD and fibrosis-related pathways captured in our prognostic model. Although experimental validation was beyond the scope of this study, these findings offer a rational basis for prioritizing certain compounds—such as HDAC inhibitors or anti-cholinergic agents—for further investigation in preclinical and clinical IPF settings.

The low survival rate and poor treatment response in IPF are partly due to cancer-like pathological events, including epigenetic and genetic changes, abnormal signaling responses, and microRNA dysregulation. These lead to the activation of pathways critical to cancer progression. For example, tyrosine kinase receptors play essential roles in both IPF and cancer (41). Growth factors like platelet-derived growth factor (PDGF) have significant profibrotic activity in IPF, with nintedanib being a standard treatment. Inhibition of PDGF signaling reduces pulmonary fibrosis (42), reflecting similarities to cell proliferation in tumor microenvironments. This understanding, combined with research on PCD, supports the development of new therapeutic strategies for IPF. TLR2, a member of the Toll-like receptor family, is an indispensable component of the immune system. Its involvement in IPF may be significant, particularly in inflammation and immune regulation. Activation of TLR2 can facilitate the recruitment and activation of inflammatory cells, thereby contributing to the development of IPF (43). TLR2-HIF1α-mediated glycolysis contributes to pyroptosis and oxidative stress in allergic airway inflammation (44). Reports indicate that SREBP2 regulates the TLR2/NF-kappa B/NFATc1/ABCA1 signaling network, promoting the viability, proliferation, and migration of TGF-β1-induced airway smooth muscle cells while inhibiting apoptosis (45). This suggests that TLR2 may play a role in various forms of cell death. AKT3, as part of the PI3K/AKT signaling pathway, is known to be crucial for cell survival and proliferation. miR-29a-3p regulates autophagy in SiO2-induced pulmonary fibrosis by targeting Akt3-mediated mTOR (46). Antibodies against AKT3 are associated with skin and lung fibrosis in systemic sclerosis patients. SLC1A4 is a sodium-dependent amino acid transporter that plays a significant role in tumor progression. Limited studies suggest its involvement in ferroptosis within tumor cells. It has been reported that SLC1A4 promotes epithelial-mesenchymal transition (EMT) in hepatocellular carcinoma cells by regulating the PI3K/AKT signaling pathway. Therefore, investigating the expression and function of SLC1A4 in IPF pulmonary fibroblasts, as well as its potential regulatory effect on the PI3K/AKT pathway, is crucial for understanding the pathogenesis of IPF and identifying novel therapeutic targets (47). A study has shown that ANTXR2 promotes the conversion of matrix metalloproteinase 2 (MMP2) from its inactive to active form. Active MMP2 degrades collagen, thereby facilitating ECM remodeling and fibrosis resolution (48). HTRA1, a member of the HTRA protein family, is a serine protease with multiple substrates. HTRA1 can degrade various ECM proteins, such as fibronectin. The resulting fibronectin fragments can induce synovial cells to upregulate the production of MMP1 and MMP3. Furthermore, it inhibits signaling mediated by members of the TGF-β family. Research has confirmed that by acting on TGF-β signaling, HTRA1 interferes with the disease process of fibrosis. A previous study indicates that overexpression of HTRA1 in BAL cells from IPF patients is associated with a significantly worse prognosis (49). Elevated levels of HTRA1 have been shown to suppress cell proliferation and induce apoptosis in esophageal cancer both in vitro and in vivo (50). Additionally, silencing HTRA1 affects the proliferation, invasion, and apoptosis of retinal epithelial cells (51). Upregulation of HTRA1 activates the NLRP3 inflammasome, promoting pyroptosis in human endometrial stromal cells derived from neutrophils (52). TIMP1, an apoptosis inhibitor (53), regulates the MMP9/TIMP1 signaling pathway, effectively reducing inflammation and fibrosis induced by cigarette smoke in vitro in alveolar cells (54). In the context of fibrotic processes such as pulmonary fibrosis, TIMP levels increase. The involvement of pro-inflammatory cytokines can lead to the overexpression of MMPs, thereby increasing their activity and contributing to airway remodeling. Consequently, TIMP1 can exert regulatory effects on IPF by modulating MMPs (55). Together, although these seven genes are involved in distinct molecular processes, they appear to function synergistically within a common fibrotic framework that is highly relevant to IPF pathogenesis. Several genes, including TLR2, PDCD1LG2, and HTRA1, are linked to immune-inflammatory regulation, particularly through pyroptosis, immune checkpoint signaling, and inflammasome activation, which may collectively shape a pro-fibrotic immune microenvironment. Meanwhile, AKT3 and SLC1A4 are associated with autophagy, oxidative stress responses, and cell survival, implicating them in epithelial cell injury and impaired regenerative capacity—two key features of progressive fibrosis. TIMP1 and ANTXR2 contribute to ECM remodeling and fibrosis maintenance by enhancing cell–matrix interactions and suppressing matrix degradation. Notably, TIMP1 has been shown to define a macrophage subset with strong pro-fibrotic potential, and ANTXR2 is implicated in collagen-mediated signaling. Although these genes participate in distinct pathways, their expression patterns suggest convergence within fibrotic niches characterized by epithelial dysfunction, immune activation, and fibroblast accumulation. This functional integration supports the prognostic power of our signature and highlights its potential to reflect the multifactorial cellular stress environment underlying IPF.

To explore the potential pathological relevance of the seven prognostic genes, we re-analyzed the transcriptomic dataset used in this study. TGF-β1 and alpha-smooth muscle actin (α-SMA; encoded by ACTA2) are well-established markers of fibrotic activity and are commonly used to assess IPF severity, particularly in relation to fibroblastic foci and ECM accumulation. Although not yet standardized in clinical diagnostics, accumulating evidence supports their close association with key pathological features of IPF. In our analysis, we found that the expression levels of several model genes were significantly correlated with TGF-β1 and ACTA2 (Figure S2), suggesting that these genes may be mechanistically linked to fibrosis progression in IPF.

There are several limitations in this study. First, although we integrated multiple modes of PCD and employed various machine learning algorithms to construct the prognostic model, we did not perform a comparative assessment of the individual predictive power of each PCD subtype or its constituent genes. Future studies could investigate the distinct contributions of ferroptosis, pyroptosis, apoptosis, and other PCD pathways to prognostic stratification. Second, although we used multiple GEO datasets for gene discovery, the transcriptomic data were derived from diverse sample types (BAL, lung biopsies, explants, blood), which may introduce heterogeneity in gene expression profiles. Although batch correction and normalization were applied, residual biases cannot be fully excluded. Future studies based on tissue-matched and prospective cohorts will help confirm these findings. Third, the sample size used for training the prognostic model, although among the largest BAL-based IPF datasets publicly available, remains limited. Larger, multi-center cohorts are needed to further validate and refine the predictive robustness of the CDS score. Finally, the survival data were obtained from publicly available GEO cohorts that did not provide specific cause-of-death information. As a result, we were unable to conduct a competing risk analysis, and the potential influence of non-IPF-related mortality on survival outcomes cannot be ruled out. Future validation using datasets with well-annotated clinical endpoints will be important to assess the accuracy of the model under competing risk frameworks.

Conclusions

This is the first systematic analysis of information from online open databases, integrating various PCD-related genes to assess their prognostic value in patients with IPF. This study identifies and constructs a prognostic model composed of seven genes (TLR2, PDCD1LG2, AKT3, SLC1A4, ANTXR2, HTRA1 and TIMP1) along with a corresponding nomogram, which can predict survival and prognosis in IPF patients. It is also regarded as a potential biological marker related to anti-IPF drug candidates. This research provides new insights for the precision and personalized treatment of IPF.

Acknowledgments

We thank the editor and the anonymous reviewers for their comments and suggestions.

Footnote

Reporting Checklist: The authors have completed the ARRIVE and TRIPOD reporting checklists. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2024-2173/rc

Data Sharing Statement: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2024-2173/dss

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2024-2173/prf

Funding: This study was supported by Shaanxi Provincial Health and Planning Biological Committee, General Project - Social Development Area (No. 2023-YBSF-257, to Q.Z.); First Affiliated Hospital of Xi’an Jiaotong University, Research and Development Funding (No. 2024-MS-15, to Y.Z.); Shaanxi Province Health Research and Innovation Capacity Enhancement Program (No. 2024PT-09, to G.Z.); Shaanxi Province Key Industry Innovation Chain (Cluster)-Social Development Field (No. 2024SF-ZDCYL-02-09, to G.Z.); and National Natural Science Foundation of China (No. 82103467, to Y.Z.).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2024-2173/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. All animal experiments were performed under a project license (No. XJTU1AF-CR-2023-006) granted by Ethics Committee of Xi’an Jiaotong University, in compliance with the national guidelines for the care and use of animals.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Mei Q, Liu Z, Zuo H, et al. Idiopathic Pulmonary Fibrosis: An Update on Pathogenesis. Front Pharmacol 2021;12:797292. [Crossref] [PubMed]
2. Ruwanpura SM, Thomas BJ, Bardin PG. Pirfenidone: Molecular Mechanisms and Potential Clinical Applications in Lung Disease. Am J Respir Cell Mol Biol 2020;62:413-22. [Crossref] [PubMed]
3. Wind S, Schmid U, Freiwald M, et al. Clinical Pharmacokinetics and Pharmacodynamics of Nintedanib. Clin Pharmacokinet 2019;58:1131-47. [Crossref] [PubMed]
4. Kim JS, Murray S, Yow E, et al. Comparison of Pirfenidone and Nintedanib: Post Hoc Analysis of the CleanUP-IPF Study. Chest 2024;165:1163-73. [Crossref] [PubMed]
5. Richeldi L, Collard HR, Jones MG. Idiopathic pulmonary fibrosis. Lancet 2017;389:1941-52. [Crossref] [PubMed]
6. Ley B, Ryerson CJ, Vittinghoff E, et al. A multidimensional index and staging system for idiopathic pulmonary fibrosis. Ann Intern Med 2012;156:684-91. [Crossref] [PubMed]
7. Kim CH, Chung MJ, Cha YK, et al. The impact of deep learning reconstruction in low dose computed tomography on the evaluation of interstitial lung disease. PLoS One 2023;18:e0291745. [Crossref] [PubMed]
8. Tomos I, Roussis I, Matthaiou AM, et al. Molecular and Genetic Biomarkers in Idiopathic Pulmonary Fibrosis: Where Are We Now? Biomedicines 2023;11:2796. [Crossref] [PubMed]
9. Sauler M, Bazan IS, Lee PJ. Cell Death in the Lung: The Apoptosis-Necroptosis Axis. Annu Rev Physiol 2019;81:375-402. [Crossref] [PubMed]
10. Liang Q, Cai W, Zhao Y, et al. Lycorine ameliorates bleomycin-induced pulmonary fibrosis via inhibiting NLRP3 inflammasome activation and pyroptosis. Pharmacol Res 2020;158:104884. [Crossref] [PubMed]
11. Pei Z, Qin Y, Fu X, et al. Inhibition of ferroptosis and iron accumulation alleviates pulmonary fibrosis in a bleomycin model. Redox Biol 2022;57:102509. [Crossref] [PubMed]
12. Fingerlin TE, Murphy E, Zhang W, et al. Genome-wide association study identifies multiple susceptibility loci for pulmonary fibrosis. Nat Genet 2013;45:613-20. [Crossref] [PubMed]
13. Clough E, Barrett T. The Gene Expression Omnibus Database. Methods Mol Biol 2016;1418:93-110. [Crossref] [PubMed]
14. Kanehisa M, Furumichi M, Sato Y, et al. KEGG for taxonomy-based analysis of pathways and genomes. Nucleic Acids Res 2023;51:D587-92. [Crossref] [PubMed]
15. Fishilevich S, Zimmerman S, Kohn A, et al. Genic insights from integrated human proteomics in GeneCards. Database (Oxford) 2016;2016:baw030. [Crossref] [PubMed]
16. Liberzon A, Birger C, Thorvaldsdóttir H, et al. The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell Syst 2015;1:417-25. [Crossref] [PubMed]
17. Fabregat A, Jupe S, Matthews L, et al. The Reactome Pathway Knowledgebase. Nucleic Acids Res 2018;46:D649-55. [Crossref] [PubMed]
18. Subramanian A, Narayan R, Corsello SM, et al. A Next Generation Connectivity Map: L1000 Platform and the First 1,000,000 Profiles. Cell 2017;171:1437-1452.e17. [Crossref] [PubMed]
19. Zhang X, Liu H, Hock T, et al. Histone deacetylase inhibition downregulates collagen 3A1 in fibrotic lung fibroblasts. Int J Mol Sci 2013;14:19605-17. [Crossref] [PubMed]
20. Diwan A, Wansapura J, Syed FM, et al. Nix-mediated apoptosis links myocardial fibrosis, cardiac remodeling, and hypertrophy decompensation. Circulation 2008;117:396-404. [Crossref] [PubMed]
21. Yang L, Wang Y, Pan Z, et al. Tetraspanin 1 inhibits TNFα-induced apoptosis via NF-κB signaling pathway in alveolar epithelial cells. Inflamm Res 2018;67:951-64. [Crossref] [PubMed]
22. Wang G, Zhao H, Zheng B, et al. TLR2 Promotes Monocyte/Macrophage Recruitment Into the Liver and Microabscess Formation to Limit the Spread of Listeria Monocytogenes. Front Immunol 2019;10:1388. [Crossref] [PubMed]
23. Chen Y, Pu Q, Ma Y, et al. Aging Reprograms the Hematopoietic-Vascular Niche to Impede Regeneration and Promote Fibrosis. Cell Metab 2021;33:395-410.e4. [Crossref] [PubMed]
24. Kreuter M, Lee JS, Tzouvelekis A, et al. Monocyte Count as a Prognostic Biomarker in Patients with Idiopathic Pulmonary Fibrosis. Am J Respir Crit Care Med 2021;204:74-81. [Crossref] [PubMed]
25. Zhao W, Wang L, Yang J, et al. Endothelial cell-derived MMP19 promotes pulmonary fibrosis by inducing E(nd)MT and monocyte infiltration. Cell Commun Signal 2023;21:56. [Crossref] [PubMed]
26. Monnier J, Zabel BA. Anti-asialo GM1 NK cell depleting antibody does not alter the development of bleomycin induced pulmonary fibrosis. PLoS One 2014;9:e99350. [Crossref] [PubMed]
27. Han QJ, Mu YL, Zhao HJ, et al. Fasudil prevents liver fibrosis via activating natural killer cells and suppressing hepatic stellate cells. World J Gastroenterol 2021;27:3581-94. [Crossref] [PubMed]
28. Kishore A, Petrek M. Roles of Macrophage Polarization and Macrophage-Derived miRNAs in Pulmonary Fibrosis. Front Immunol 2021;12:678457. [Crossref] [PubMed]
29. Pokhreal D, Crestani B, Helou DG. Macrophage Implication in IPF: Updates on Immune, Epigenetic, and Metabolic Pathways. Cells 2023;12:2193. [Crossref] [PubMed]
30. Koike A, Hayashi K, Fujimori K. Involvement of necroptotic cell death in macrophages in progression of bleomycin and LPS-induced acute exacerbation of idiopathic pulmonary fibrosis. Eur J Pharmacol 2024;972:176572. [Crossref] [PubMed]
31. Tang PM, Nikolic-Paterson DJ, Lan HY. Macrophages: versatile players in renal inflammation and fibrosis. Nat Rev Nephrol 2019;15:144-58. [Crossref] [PubMed]
32. Sanders YY, Hagood JS, Liu H, et al. Histone deacetylase inhibition promotes fibroblast apoptosis and ameliorates pulmonary fibrosis in mice. Eur Respir J 2014;43:1448-58. [Crossref] [PubMed]
33. Patra S, Praharaj PP, Klionsky DJ, et al. Vorinostat in autophagic cell death: A critical insight into autophagy-mediated, -associated and -dependent cell death for cancer prevention. Drug Discov Today 2022;27:269-79. [Crossref] [PubMed]
34. Korfei M, Mahavadi P, Guenther A. Targeting Histone Deacetylases in Idiopathic Pulmonary Fibrosis: A Future Therapeutic Option. Cells 2022;11:1626. [Crossref] [PubMed]
35. Rajagopalan S, Dobre M, Dazard JE, et al. Mineralocorticoid Receptor Antagonism Prevents Aortic Plaque Progression and Reduces Left Ventricular Mass and Fibrosis in Patients With Type 2 Diabetes and Chronic Kidney Disease: The MAGMA Trial. Circulation 2024;150:663-76. [Crossref] [PubMed]
36. Agarwal R, Kolkhof P, Bakris G, et al. Steroidal and non-steroidal mineralocorticoid receptor antagonists in cardiorenal medicine. Eur Heart J 2021;42:152-61. [Crossref] [PubMed]
37. Zhao J, Yu L, Xue X, et al. Corrigendum to "Diminished α7 nicotinic acetylcholine receptor (α7nAChR) rescues amyloid-β induced atrial remodeling by oxi-CaMKII/MAPK/AP-1 axis-mediated mitochondrial oxidative stress" Redox Biol 2025;82:103604. [Redox Biol 59 (2023) 102594].
38. Roth M. Airway and lung remodelling in chronic pulmonary obstructive disease: a role for muscarinic receptor antagonists? Drugs 2015;75:1-8. [Crossref] [PubMed]
39. Li S, Geng J, Xu X, et al. miR-130b-3p Modulates Epithelial-Mesenchymal Crosstalk in Lung Fibrosis by Targeting IGF-1. PLoS One 2016;11:e0150418. [Crossref] [PubMed]
40. Hu X, Gan L, Tang Z, et al. A Natural Small Molecule Mitigates Kidney Fibrosis by Targeting Cdc42-mediated GSK-3β/β-catenin Signaling. Adv Sci (Weinh) 2024;11:e2307850. [Crossref] [PubMed]
41. Espindola MS, Habiel DM, Narayanan R, et al. Targeting of TAM Receptors Ameliorates Fibrotic Mechanisms in Idiopathic Pulmonary Fibrosis. Am J Respir Crit Care Med 2018;197:1443-56. [Crossref] [PubMed]
42. Cai Z, Guo H, Qian J, et al. Effects of bone morphogenetic protein 4 on TGF-β1-induced cell proliferation, apoptosis, activation and differentiation in mouse lung fibroblasts via ERK/p38 MAPK signaling pathway. PeerJ 2022;10:e13775. [Crossref] [PubMed]
43. Hu W, Xu Y. Transcriptomics in idiopathic pulmonary fibrosis unveiled: a new perspective from differentially expressed genes to therapeutic targets. Front Immunol 2024;15:1375171. [Crossref] [PubMed]
44. Sha JF, Xie QM, Chen N, et al. TLR2-hif1α-mediated glycolysis contributes to pyroptosis and oxidative stress in allergic airway inflammation. Free Radic Biol Med 2023;200:102-16. [Crossref] [PubMed]
45. Wang Y, Yang H, Su X, et al. SREBP2 promotes the viability, proliferation, and migration and inhibits apoptosis in TGF-β1-induced airway smooth muscle cells by regulating TLR2/NF-κB/NFATc1/ABCA1 regulatory network. Bioengineered 2022;13:3137-47. [Crossref] [PubMed]
46. Li P, Hao X, Liu J, et al. miR-29a-3p Regulates Autophagy by Targeting Akt3-Mediated mTOR in SiO(2)-Induced Lung Fibrosis. Int J Mol Sci 2023;24:11440. [Crossref] [PubMed]
47. Zheng J, Gong J. SLC1A4 Promotes Malignant Transformation of Hepatocellular Carcinoma by Activating the AKT Signaling. Anal Cell Pathol (Amst) 2025;2025:1115184. [Crossref] [PubMed]
48. Huang X, Zhang L, Luo W, et al. Endothelial anthrax toxin receptor 2 plays a protective role in liver fibrosis. Front Cell Dev Biol 2023;11:1278968. [Crossref] [PubMed]
49. Xia Y, Lei C, Yang D, et al. Construction and validation of a bronchoalveolar lavage cell-associated gene signature for prognosis prediction in idiopathic pulmonary fibrosis. Int Immunopharmacol 2021;92:107369. [Crossref] [PubMed]
50. Xia J, Wang F, Wang L, et al. Elevated serine protease HtrA1 inhibits cell proliferation, reduces invasion, and induces apoptosis in esophageal squamous cell carcinoma by blocking the nuclear factor-κB signaling pathway. Tumour Biol 2013;34:317-28. [Crossref] [PubMed]
51. Yu T, Chen CZ, Xing YQ. Inhibition of cell proliferation, migration and apoptosis in blue-light illuminated human retinal pigment epithelium cells by down-regulation of HtrA1. Int J Ophthalmol 2017;10:524-9. [Crossref] [PubMed]
52. Li L, Ye K, Wang D. Upregulation of HTRA1 mediated by the lncRNA NEAT1/miR-141-3p axis contributes to endometriosis development through activating NLRP3 inflammasome-mediated pyroptotic cell death and cellular inflammation. In Vitro Cell Dev Biol Anim 2023;59:166-78. [Crossref] [PubMed]
53. Wang T, Yamashita K, Iwata K, et al. Both tissue inhibitors of metalloproteinases-1 (TIMP-1) and TIMP-2 activate Ras but through different pathways. Biochem Biophys Res Commun 2002;296:201-5. [Crossref] [PubMed]
54. Xu W, Li F, Zhu L, et al. Pacenta polypeptide injection alleviates the fibrosis and inflammation in cigarette smoke extracts-induced BEAS-2B cells by modulating MMP-9/TIMP-1 signaling. J Biochem Mol Toxicol 2023;37:e23453. [Crossref] [PubMed]
55. Perez-Favila A, Garza-Veloz I, Hernandez-Marquez LDS, et al. Antifibrotic Drugs against Idiopathic Pulmonary Fibrosis and Pulmonary Fibrosis Induced by COVID-19: Therapeutic Approaches and Potential Diagnostic Biomarkers. Int J Mol Sci 2024;25:1562. [Crossref] [PubMed]