CSF2 drives radiation-induced lung injury by eliciting PADI4-high pathogenic neutrophils Authors: Yun Wang, Maidina Abulaihaiti, Yan Pan, Jianghong Zhang, Chunlin Shao Published: 2026 ABSTRACT AbstractBackgroundRadiation-induced lung injury (RILI) represents a severe, therapy-refractory complication of thoracic radiotherapy. Although neutrophils have been implicated in RILI pathogenesis, the specific pathogenic subsets and the molecular mechanisms governing their recruitment and functional specialization remain elusive. This study aims to decode the crosstalk between damaged epithelial cells and neutrophils in the RILI microenvironment.MethodsWe established a murine RILI model using 17 Gy thoracic irradiation and performed integrated single-cell RNA sequencing (scRNA-seq) and bulk transcriptome profiling to map the immune landscape. Neutrophil heterogeneity and developmental trajectories were analyzed using pseudotime algorithms. Mechanistic investigations used an indirect co-culture system of irradiated lung epithelial cells and HL-60 neutrophil-like cells, recombinant CSF2, CSF2-neutralizing antibodies, and in vivo anti-GM-CSF blockade. The therapeutic potential of PADI4 inhibition was evaluated using GSK484 in RILI mice, followed by histopathological, molecular, immunofluorescence, ELISA, and bulk RNA-seq analyses.ResultsIntegrated multi-omics analyses identified PADI4 as a central neutrophil-associated molecule in RILI, with preferential enrichment in immature, NETosis-prone neutrophil subsets mobilized during emergency granulopoiesis. Among PAD family members, PADI4 showed the most prominent neutrophil-restricted and injury-responsive expression pattern. Mechanistically, radiation-damaged epithelial cells secrete CSF2, which induces a PADI4-high activated myeloid phenotype in vitro; conversely, CSF2 neutralization attenuated this response. In vivo GM-CSF blockade reduced PADI4-positive neutrophil infiltration and lung injury severity. Pharmacological PADI4 inhibition with GSK484 suppressed citrullinated histone H3 generation, NETosis-associated programs, neutrophil chemotaxis modules, and tissue damage in the RILI model.ConclusionsThis study delineates an epithelial-myeloid damage-response circuit in RILI and identifies CSF2-driven PADI4 activation in pathogenic neutrophils as a therapeutically actionable mechanism.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12967-026-08413-6.Keywords: RILI, PADI4, Neutrophils, NETosis, CSF2 Introduction Radiation-induced lung injury (RILI) constitutes a major dose-limiting toxicity of thoracic tumor radiotherapy, which significantly impairs patient prognosis and quality of life [1]. Its core pathological manifestations encompass acute inflammation and chronic fibrosis, yet the underlying cellular and molecular mechanisms driving the inflammatory response and consequent tissue damage remain incompletely defined. Recent advances in single-cell technologies have increasingly underscored the pivotal role of the immune system in RILI pathogenesis. Among immune cell populations, neutrophils are recognized as key regulators of tissue damage and repair processes due to their rapid recruitment to injured sites, potent pro-inflammatory properties, and ability to form neutrophil extracellular traps (NETs) [2]. However, the functional diversity of neutrophils in RILI and their dynamic reprogramming within the injured microenvironment have not been systematically elucidated. Recent clinical and translational studies have further suggested that persistent immune dysregulation and inflammatory microenvironment remodeling are key determinants driving the progression of RILI from acute pneumonitis to chronic fibrosis, highlighting the importance of dissecting immune-cell-specific pathogenic mechanisms in irradiated lungs [3]. Peptidyl arginine deiminase 4 (PADI4) is an enzyme highly enriched in neutrophils [4]. By catalyzing histone citrullination, PADI4 acts as a key driver of NET formation and has been shown to extensively modulate chromatin structure and gene transcription programs [5, 6], thereby regulating neutrophil functional states. Although aberrant PADI4 activation is implicated in various inflammatory diseases [7–9], its role in RILI remains unexplored. The signals that drive PADI4-high pathogenic neutrophil subsets in RILI are equally unknown. We hypothesized that within the complex RILI microenvironment, damaged lung epithelial cells, the primary site of injury, may release specific factors that orchestrate this pathogenic neutrophil response. Colony-stimulating factor 2 (CSF2/GM-CSF) [10–13], a potent myelopoietic cytokine produced by various stromal and epithelial cells upon injury, is a critical regulator of granulocyte differentiation, survival, and functional activation. Its potential role in inducing PADI4 expression represents a plausible local regulatory mechanism within the injured lung. Furthermore, we reasoned that such local injury-driven responses might recapitulate systemic pathways of neutrophil mobilization. Granulocyte colony-stimulating factor (G-CSF) is the principal systemic driver of emergency granulopoiesis and neutrophil release from the bone marrow [14–16]. We therefore investigated whether the local damage signal CSF2 and the systemic mobilizing signal G-CSF, though operationally distinct, could converge on a common outcome: the generation and recruitment of immature, PADI4-high neutrophils. However, this proposed model, connecting epithelial injury to pathogenic neutrophil deployment via integrated local and systemic cues, has not been systematically validated in the context of RILI. In this study, we systematically analyzed the dynamic evolution of neutrophil functional states during RILI using a murine model, bulk RNA sequencing, single-cell RNA sequencing, and public datasets. We identified PADI4 as a key neutrophil-associated regulator linked to NETosis and inflammatory tissue injury. Mechanistically, we demonstrated that lung epithelial-derived CSF2 promotes PADI4 upregulation and neutrophil activation programs in neutrophil-like cells, whereas CSF2 blockade suppresses the accumulation of PADI4-positive neutrophils and alleviates lung injury in vivo. Complementary analysis of G-CSF-mobilized human neutrophils indicated that systemic emergency granulopoiesis can also favor the appearance of immature PADI4-high neutrophils. Together, these findings reveal a CSF2-PADI4 neutrophil axis in RILI and provide a rationale for therapeutic targeting of pathogenic neutrophil programs. Materials and methods RILI model of mice Six- to eight-week-old male C57BL/6J mice were housed under specific pathogen-free conditions with free access to food and water. After being anesthetized via intraperitoneal injection of tribromoethanol, the mouse thorax was irradiated with a single dose of 17 Gy X-rays (X-RAD 320, PXI Inc., North Branford, CT, USA; 12 mA, 2 mm aluminum filtration) using a small animal irradiator. The irradiation field was precisely collimated to cover the lungs, while the rest of the body was shielded with lead blocks. Control mice underwent sham irradiation. Lung tissues and peripheral blood were collected 7 days post-irradiation for subsequent analysis. All animal experiments were conducted in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and were approved by the Animal Ethics Committee of Fudan University. Cell culture and irradiation Human promyelocytic leukemia HL-60 cells were cultured in IMDM medium (Gibco, Thermo Fisher Scientific, Waltham, MA, USA), and lung epithelial BEAS-2B cells were cultured in DMEM medium (Gibco, Thermo Fisher Scientific, Waltham, MA, USA). All media were supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin (Gibco, Thermo Fisher Scientific, Waltham, MA, USA). Cells were maintained in a humidified incubator at 37 °C with 5% CO₂. Cells in log-phase were exposed to X-ray at a dose rate of 1 Gy/min at room temperature. Preparation of lung epithelial cell conditioned medium When BEAS-2B cells reached 70–80% confluency, they were subjected to 8 Gy X-ray irradiation. After irradiation, the medium was replaced with fresh complete medium, and cells were cultured for an additional 48 h. The supernatant was then collected, centrifuged to remove cellular debris, and used immediately as conditioned medium (CM). Histopathology, immunohistochemistry (IHC), and quantification Mouse lungs were perfusion-fixed with 4% paraformaldehyde, embedded in paraffin, and sectioned at a thickness of 4–5 μm. For histopathological assessment, sections were stained with hematoxylin and eosin (H&E). Lung injury was scored according to established criteria by two independent pathologists who were blinded to the experimental groups. For immunohistochemistry (IHC), after antigen retrieval, tissue sections were incubated with specific primary antibodies against PADI4 (DF6685, Affinity Biosciences) and S100A8 (GB11421-50, Servicebio, China) overnight at 4 °C. Subsequently, HRP-conjugated secondary antibodies and a DAB chromogen kit were used for signal development. Nuclei were counterstained with hematoxylin. Sections were then mounted, visualized, and imaged under a microscope. IHC staining was quantified using ImageJ software (version 2.16). Multiplex immunofluorescence (mIF), and quantification Paraffin-embedded lung tissue Sects. (4–5 μm) were dewaxed, rehydrated, and subjected to antigen retrieval in citrate buffer (pH 6.0) using microwave heating. After blocking with 5% BSA and 5% normal goat serum in PBS containing 0.3% Triton X-100 for 30 min, sections were incubated overnight at 4 °C with the following primary antibodies: anti-PADI4 (DF6685, Affinity Biosciences), anti-Ly6G (clone 1G5E5, GB12229, Servicebio, China), anti-S100A8 (GB11421-50, Servicebio, China), and anti-citrullinated histone H3 (CitH3; citrulline R2 + R8+R17; clone RM1001, ab281584, Abcam). Alexa Fluor 488-conjugated anti-rabbit, Alexa Fluor 555-conjugated anti-rat, and Alexa Fluor 647-conjugated anti-rabbit secondary antibodies were subsequently applied for multi-color detection. Nuclei were counterstained with DAPI. Sections were then mounted with antifade medium and visualized under a fluorescence microscope. Negative control sections were processed with omission of primary antibodies. mIF staining was quantified using ImageJ software (version 2.16). Transcriptome sequencing and bioinformatic analysis Total RNA was extracted from mouse lung tissues and subjected to sequencing on the DNBSEQ platform or the Illumina NovaSeq X Plus platform (LC Bio Technology Co. Ltd., Hangzhou, China). Following quality control, raw sequencing reads were aligned to the reference genome (mm10), yielding an average alignment rate of 98.72%. Gene count matrices were generated, with 19,437 genes detected. Differential expression analysis was performed using the DESeq2 R package. Gene Set Enrichment Analysis (GSEA) and Gene Set Variation Analysis (GSVA) were conducted using their corresponding R packages. Immune cell composition in mouse lung tissue bulk transcriptomes was quantified using the mMCPcounter R package (v1.1.0). Raw count data were converted to FPKM, log2-transformed with a pseudocount of 1 [log2(FPKM + 1)], and duplicate gene symbols were collapsed by averaging expression values across corresponding transcripts. The mMCPcounter.estimate() function was then applied with default parameters to generate enrichment scores for immune and stromal cell populations. For heatmap visualization, scores were subjected to per-cell-type min-max normalization across samples. Single-cell RNA sequencing data analysis Single-cell RNA sequencing (scRNA-seq) datasets were acquired as follows: lung tissues from a murine RILI model (GSE298566), an HL-60 cell in vitro differentiation model (GSE79044), and an irradiated lung epithelial cell model (GSE166340) were downloaded from the GEO database. scRNA-seq data of neutrophils from healthy human bone marrow and peripheral blood before and after G-CSF mobilization were obtained from the ArrayExpress database (E-MTAB-11188). Raw sequencing data were processed using the Cell Ranger pipeline. Subsequent bioinformatic analyses were conducted in R. The Seurat package was used for quality control, normalization, dimensionality reduction, clustering, and cell type annotation. Cell-cell communication analysis was performed using CellChat. Pseudotime trajectory analysis was carried out using Slingshot. Neutrophil subclustering and state definition. Neutrophils were extracted from the whole-lung scRNA-seq dataset (GSE298566) based on canonical lineage markers (S100a8, S100a9, Lcn2, Retnlg). The subset was re-normalized and subjected to graph-based clustering (Seurat v4.3.0), yielding five transcriptionally distinct clusters that were designated N0–N4 in ascending numerical order. These labels are data-driven cluster identifiers, not a priori functional classifications (e.g., unlike the M1/M2 paradigm for macrophages). Functional identities were assigned post hoc through differential gene analysis, GO/KEGG enrichment, and NETosis module scoring. Quantitative real-time PCR (qPCR) Total RNA was extracted from cells or tissues using Trizol reagent (Invitrogen, San Diego, CA, USA) and reverse-transcribed into cDNA (with gDNase) (Tiangen Biotechnology, Beijing, China). RT-qPCR was performed on the Bio-Rad CFX Opus 96 platform using SuperReal PreMix Plus (SYBR Green) (Tiangen Biotechnology, Beijing, China). β-actin was used as an endogenous control gene. The relative mRNA expression levels of target genes were calculated using the 2^(-ΔΔCt) method, with Actb (β-actin) as the internal control. Details of the primer sequences are available in the Supplementary Materials. Western blotting Total protein was extracted from lung tissues using RIPA lysis buffer (Beyotime Biotechnology, Shanghai, China). Equal amounts of protein were separated by SDS-PAGE and transferred to PVDF membranes (Millipore, Bedford, MA, USA). After blocking with 5% skim milk, membranes were incubated with primary antibodies (anti-PADI4, ab214810; anti-S100A8/A9, ab288715; anti-MPO, ab208670; anti-Vinculin, AF8313) overnight at 4 °C. Following incubation with HRP-conjugated secondary antibodies, signals were detected using an ECL chemiluminescent substrate (Millipore, St. Louis, MO, USA) and captured using an imaging system. In vitro functional validation experiments To investigate the functional role of CSF2, HL-60 neutrophil-like cells were subjected to the following treatments: (1) control medium; (2) conditioned medium (CM) derived from irradiated lung epithelial cells; (3) recombinant human CSF2 protein (TargetMol, TMPJ-01465); (4) CM or recombinant CSF2 combined with a CSF2-neutralizing antibody (Mavrilimumab, CAM 3001; TargetMol, T76690); and (5) CM combined with human IgG4 (S228P) kappa isotype control (10 µg/mL; TargetMol, T78339). After 48–72 h of incubation, cells were harvested, and mRNA expression levels of PADI4, CD11B (ITGAM), and MMP9 were detected by qPCR. In vivo pharmacological intervention To evaluate the therapeutic potential of targeting PADI4, mice were randomly divided into vehicle and GSK484 treatment groups following RILI modeling. GSK484 (4 mg/kg in 0.1 ml; TQ0067, TargetMol) was administered intraperitoneally one day before irradiation and three days after irradiation. The vehicle group received an equal volume of vehicle. Lung tissues were collected for histopathological, immunofluorescence, and transcriptomic analyses, and peripheral blood was collected for molecular and ELISA analyses. To evaluate the necessity of CSF2 signaling in RILI pathogenesis, mice were randomly divided into three groups: sham irradiation with isotype control, RILI (17 Gy) with isotype control (Rat IgG2a kappa, HY-P990679, MedChemExpress), and RILI with anti-mouse GM-CSF neutralizing antibody (clone MP1-22E9, HY-P99134, MedChemExpress). The anti-GM-CSF antibody or isotype control was administered intraperitoneally at 250 µg per mouse in 200 µL sterile PBS on days − 1, 1, 3, and 5 related to thoracic irradiation. Lung tissues were collected for histopathological, immunofluorescence, and transcriptomic analyses, and peripheral blood was collected for molecular and ELISA analyses. Enzyme-linked immunosorbent assay (ELISA) Peripheral blood was collected from mice at day-7 post-irradiation. Serum was isolated by centrifugation at 3,000 × g for 10 min and stored at − 80 °C until use. The concentration of PADI4 in serum was measured using a commercial mouse PADI4 ELISA kit (ml058749, Shanghai Enzyme-linked Biotechnology Co., Ltd.), following the manufacturer’s instructions. Circulating MPO-DNA complexes, a surrogate marker of NETosis, were also quantified using a commercially available MPO–DNA ELISA kit (ARE812-MU, Huding Biological Technology, Shanghai, China) without additional plate coating. Serum GM-CSF levels were measured using a commercial mouse GM-CSF ELISA kit according to the manufacturer’s instructions (CLEM006, Laizee Biotech. Absorbance was recorded at 450 nm using a microplate reader, and all samples were assayed in duplicate. Concentrations were calculated based on standard curves provided with the kits. Statistical analysis All in vitro experiments were independently replicated at least three times. Data are presented as mean ± standard error of the mean (SEM), unless otherwise indicated. Statistical analyses were performed using GraphPad Prism software (version 10.1.2). Normality of distribution was assessed using the Shapiro-Wilk test, and homogeneity of variance was evaluated using Levene’s test for multi-group comparisons or the F-test for two-group comparisons. When both assumptions were satisfied, comparisons between two groups were analyzed using an unpaired two-tailed Student’s t-test, and comparisons among multiple groups were analyzed using one-way or two-way ANOVA followed by the appropriate post-hoc test (Tukey’s or Dunnett’s), as specified in the figure legends. If normality or equal variance assumptions were not met, Mann-Whitney U tests or Kruskal-Wallis tests with Dunn’s post-hoc correction were applied. P < 0.05 was considered statistically significant. Results Transcriptome profiling identified PADI4 as the core hub gene driving RILI To elucidate the immunopathological mechanisms of RILI at a system level, we established a murine thoracic irradiation model with a single dose of 17 Gy [17]. Histopathological analysis of hematoxylin and eosin (H&E)-stained lung tissue sections revealed typical late-stage injury pathologies in irradiated mice, characterized by alveolar structure disruption and extensive inflammatory cell infiltration. Concomitantly, the significant increase in injury scores confirmed the successful establishment of the RILI model (Fig. 1A, B). We subsequently performed transcriptome sequencing to profile global transcriptomic alterations in the RILI model. Differential expression analysis identified substantial dysregulation of a large set of genes in irradiated lung tissues. Notably, the most significantly upregulated genes included those encoding the neutrophil-enriched alarmins S100A8 and S100A9, as well as PADI4, a key executor of NET formation (NETosis) (Fig. 1C). Functional enrichment analysis demonstrated that these differentially expressed genes (DEGs) were significantly enriched in pathways related to neutrophil chemotaxis, neutrophil migration, and NET formation (Fig. 1D, E). Gene Set Enrichment Analysis (GSEA) further validated the activation of PADI4-associated pathways (Fig. S1A, B). Consistent with these findings, Gene Set Variation Analysis (GSVA) and immune cell scoring revealed a marked enhancement of the neutrophil signature in the RILI group (Fig. 1F). Collectively, these results indicate that neutrophils and NETosis play critical roles in the pathogenesis of RILI. Transcriptomic profiling implicates PADI4 and neutrophil-driven inflammation in RILI. A Schematic diagram of the mouse RILI model establishment. B Representative H&E-stained lung tissue images and corresponding tissue injury score quantification (n = 6). Scale bar, 200 μm. Data are presented as mean ± SD; statistical significance was determined by unpaired two-tailed Student’s t-test after confirmation of normality and equal variance; *p < 0.05, **p < 0.01, ***p < 0.001. (unpaired two-tailed Student’s t-test). C Volcano plot of differentially expressed genes (DEGs) in lung tissue transcriptomes. Red dots represent significantly upregulated genes, highlighting Padi4, S100a8, and S100a9. D Dotplot of Gene Ontology (GO) biological process enrichment analysis of DEGs. E Lollipop plot of KEGG pathway enrichment analysis of DEGs. F Immune cell profiling of RILI and control lung tissues. Heatmap of immune cell abundances analyzed by MCP-counter (top). GSVA scores showing significantly elevated neutrophil signature in the RILI group (bottom). G Scatter plots of correlation analysis between PADI4 and neutrophil marker expression in GTEx whole blood (n = 337) queried through the GEPIA2 database. Pearson correlation analysis was performed; R, Pearson correlation coefficient Based on these findings, we focused further investigations on PADI4 as a potential molecular marker linking neutrophil activation and tissue-damaging inflammation in RILI. To validate the neutrophil-specific relevance of PADI4 in a pathophysiologically relevant context, we performed tissue-specific correlation analyses using the GTEx transcriptomic database. In whole blood (n = 337), a neutrophil-rich compartment, PADI4 expression exhibited a robust positive correlation with S100A8 and S100A9 (Fig. 1G). Strikingly, this strong association was preserved in GTEx lung tissue (n = 288), corroborating the link between PADI4 and neutrophil-mediated inflammation in the pulmonary context (Fig. S2). We subsequently confirmed the significant upregulation of PADI4 in RILI lung tissues at both the mRNA and protein levels (Fig. 2A-C). Immunohistochemical staining confirmed elevated PADI4 and S100A8 signals in inflammatory lesions (Fig. 2D). To rigorously establish cell-level co-location, multiplex immunofluorescence was performed on the same tissue section, revealing that PADI4 was predominantly expressed within Ly6G+S100A8+ neutrophils in RILI lesions (Fig. 2E). Quantification across multiple samples confirmed significant increases in Ly6G⁺ neutrophil density, PADI4⁺ cell density, and S100A8⁺ cell density in irradiated lungs compared to sham controls (Fig. 2F). PADI4 expression is upregulated in RILI lung tissue and enriched in inflammatory foci. A mRNA relative expression levels of Padi4, S100a8, Il1b, Tnfa in lung tissue detected by qPCR (n = 6). Statistical significance was determined by unpaired two-tailed Student’s t-test after confirmation of normality and equal variance; *p < 0.05, **p < 0.01, ***p < 0.001. B, C Representative Western blot images (B) and quantitative analysis of Padi4, Mpo, S100a8/a9 protein expression levels in lung tissue (C). Vinculin served as the loading control (n = 6). Statistical significance was determined by unpaired two-tailed Student’s t-test after confirmation of normality and equal variance; *p < 0.05, **p < 0.01, ***p < 0.001. D Immunohistochemical analysis showing PADI4-positive and S100A8-positive cells enriched within the inflammatory lesions. Scale bar, 100 μm. E Representative multiplex immunofluorescence images of lung tissue sections stained for DAPI (blue), Ly6G (red), PADI4 (green), and S100A8 (yellow) in sham and RILI mice. Merged panels demonstrate PADI4⁺Ly6G⁺S100A8⁺ triple-positive neutrophils within inflammatory lesions (white arrows). Scale bar, 100 μm. F Quantification of Ly6G⁺ neutrophil density, PADI4⁺ cell density, and S100A8⁺ cell density in lung tissue sections from sham and RILI mice (n = 6; mean ± SD; ****p < 0.0001; unpaired two-tailed Student’s t-test after confirmation of normality and equal variance) Together, these data demonstrate that neutrophils are robustly activated and undergo functional reprogramming during RILI pathogenesis, while PADI4 represents a prominent neutrophil-associated molecular node linked to inflammatory activation and NETosis-associated programs. Single-cell atlas analysis of neutrophil heterogeneity in radiation-induced lung injury To characterize the remodeling of the immune microenvironment in RILI at single-cell resolution, we analyzed scRNA-seq data (GSE298566) from lung tissues of a murine RILI model obtained from the GEO database [18]. Following strict quality control, 80,431 high-quality cells with reliable transcriptomic profiles were retained for subsequent analysis. Unsupervised clustering and cell-type annotation identified 19 distinct cell populations, thereby constructing a comprehensive landscape of the major cellular components in RILI-affected lung tissues (Fig. 3A). Proportion analysis revealed a significant expansion of the neutrophil population in the RILI model compared to nonirradiated control mice, indicating substantial recruitment of neutrophils to inflammatory foci following radiation (Fig. 3B). To further investigate the cell-cell interactions driving inflammation, we performed cell communication analysis. This analysis demonstrated a marked enhancement of immune cell-dominated intercellular communication networks post-irradiation (Fig. 3C, Fig. S3A). The predominant signaling pathways involved included pro-inflammatory cytokines (e.g., TNF), chemokines (e.g., CCL family signaling), and pathways related to antigen presentation and apoptosis [19, 20] (e.g., MHC-I, FASLG) (Fig. 3D, Fig. S3B). These findings indicate that a complex pro-inflammatory and immune activation network is established during RILI pathogenesis. Single-cell atlas reveals immune microenvironment remodeling in RILI (GSE298566). A UMAP visualization of 80,431 high-quality cells from RILI mouse lung tissues, colored by 19 distinct cell types identified through unsupervised clustering and annotation. B River plot comparing the proportional abundance of neutrophils between Sham and RILI groups. C Bar plot showing the top 50 signaling pathways with the most significant changes in communication strength at day 7 post-irradiation (CONV_D7) compared to control, analyzed by CellChat. Pathways are ranked by the difference in communication probability. D Circle plot displaying the intercellular communication network of the four key signaling pathways (TNF, CCL, MHC-I, and FASLG) that were most significantly enhanced in RILI. Line thickness represents the communication strength between cell types We further precisely mapped PADI4 expression at single-cell resolution. Analysis revealed that PADI4 expression was highly cell type-restricted with almost exclusively and highly expressed in neutrophils (Fig. 4A). Quantitative comparison confirmed that PADI4 expression levels in neutrophils were significantly higher than those in macrophages, T cells, and other major immune cell populations (Fig. 4B). Neutrophil heterogeneity and PADI4 expression dynamics in RILI (GSE298566). A UMAP projection of all cells with Padi4 expression overlaid (color scale). B Bar plot quantifying Padi4 expression levels across major immune cell types. C UMAP visualization of neutrophil subsets (N0-N4) after subclustering. D UMAP showing the distribution of neutrophil subsets across control, day 1, and day 7 post-irradiation. E UMAP projection showing Padi4 expression levels across neutrophil subsets N0-N4. F Violin plots displaying the expression patterns of NETosis core module genes across neutrophil subsets. G UMAP projection showing NETosis functional module scores across neutrophil subsets N0-N4 To decipher neutrophil heterogeneity in RILI, we performed sub-clustering analysis and identified five distinct neutrophil subsets (N0-N4) with divergent functional phenotypes (Fig. 4C). The transcriptional distinctness of the five neutrophil substates was validated by subcluster-specific marker expression patterns (Fig. S4). Notably, the distribution of these subsets exhibited clear temporal dynamics. Neutrophils were predominantly in the N0 state during the early period post-irradiation, whereas the N1-N4 subsets became dominant as the injury progressed (Fig. 4D). Critically, PADI4 expression was not uniformly distributed across all neutrophils but was specifically enriched in the N1-N4 subsets that prevail at the later stages of injury (Fig. 4E). Further analysis of a core NETosis gene module (including Padi4, S100a8, S100a9, and Mmp9) [21–26] revealed coordinated activation of these genes in the N1-N4 subsets (Fig. 4F), and these cells exhibited a significantly higher overall NETosis functional score (Fig. 4G), endowing them with strong pro-inflammatory and tissue-damaging capacities. To address whether the observed neutrophil activation reflects a PADI4-specific mechanism or broader PAD family response, we systematically examined the expression patterns of all five PAD family members (Padi1, Padi2, Padi3, Padi4, Padi6) in the RILI scRNA-seq dataset (GSE298566). Dot plot analysis across all major cell types and time points revealed that Padi4 was uniquely enriched in neutrophils, reaching the highest expression intensity and cell proportion among all PAD members at day 7 post-irradiation, whereas Padi1, Padi2, Padi3, and Padi6 remained barely detectable in any lineage (Fig. S5A). When restricted to the neutrophil compartment, violin plot analysis demonstrated that Padi4 was the only PAD gene exhibiting robust and progressive upregulation from sham to day 1 and day 7 post-irradiation. In stark contrast, Padi1, Padi2, Padi3, and Padi6 maintained near-zero expression levels throughout the injury course (Fig. S5B). Quantification of expressing cell frequencies confirmed that Padi4 was detected in 20.0% of sham neutrophils, increasing to 25.0% at day 1 and 58.0% at day 7, in stark contrast, Padi1, Padi2, Padi3, and Padi6 were each present in fewer than 0.3% of neutrophils across all time points (Fig. S5C). Temporal trajectory analysis further validated that Padi4 mean expression rose progressively from 0.30 (sham) to 0.34 (day 1) and 0.78 (day 7), whereas the other PAD genes remained silent at near-zero levels throughout the injury course (Fig. S5D). Although Padi2 has been implicated in inflammatory responses in other disease contexts, this single-cell atlas revealed that its expression in the lung was largely restricted to lymphoid and monocytic lineages, including NK cells, proliferating T cells, monocytes, and M1 macrophages, with virtually no detectable expression in neutrophils (Fig. S6A-B). Co-expression analysis further demonstrated minimal overlap between Padi4 and Padi2 within individual neutrophils (Fig. S6C). Taken together, these data demonstrate that PADI4 is the predominant, neutrophil-specific, and injury-responsive PAD enzyme in RILI, fully justifying our focused mechanistic investigation. To systematically define the functional propensities of these neutrophil subsets, we performed GO and KEGG enrichment analyses. GO enrichment analysis revealed distinct functional signatures for each subset (Fig. S7A): the N0 state was associated with classical acute inflammatory responses such as TNF and IL-1 signaling; the N1 state exhibited immunoregulatory features linked to T cell proliferation and activation [27, 28]; the N2 state was enriched for innate immune system activation; the N3 state was uniquely associated with antigen presentation processes and MHC-I molecules [29]; and the N4 state was linked to IL-4 and IFN-γ signaling pathways [30]. KEGG pathway analysis further validated this functional heterogeneity (Fig. S7B) and uncovered potential associations between distinct neutrophil subsets and human disease-related pathways. Significant enrichment of the TNF signaling pathway in the N0 state corroborated its early pro-inflammatory role. Notably, pathways enriched in the N1, N2, and N4 subsets are well-documented to play pivotal roles in various chronic inflammatory disorders. Furthermore, the N3 state was significantly associated with graft-versus-host disease (GVHD) [31]. Overall, these results suggest that neutrophil subset differentiation, driven by emergency granulopoiesis in RILI, not only determines immediate effector functions but also links these cells to a broader spectrum of acute and chronic pathological processes. In summary, single-cell transcriptomic analysis not only confirmed PADI4 upregulation in RILI but also precisely identified its cellular source. PADI4 is not uniformly expressed across neutrophils but is highly concentrated within specific pro-inflammatory subsets activated by the damaged lung microenvironment. This finding reveals that the PADI4-driven immunopathological mechanism in RILI exhibits a high degree of cell-state specificity. Analysis of the function and mechanism of PADI4high neutrophil subsets To investigate the potential evolutionary relationships between neutrophil functional states and the role of PADI4, we performed pseudotemporal trajectory analysis. This inference positioned neutrophil populations isolated on day 7 post-irradiation (N1-N4) at the initiation of the trajectory, whereas the day 1 neutrophil subset (N0) localized to the terminal end (Fig. 5A). Analysis of pseudotime-dependent gene expression profiles revealed distinct dynamic trends: positively correlated genes exhibited an initial decrease followed by a subsequent increase in expression, while negatively correlated genes displayed the reciprocal pattern (Fig. 5B, C). Strikingly, PADI4 expression exhibited a bell-shaped trajectory, with a transient increase in the early phase followed by a gradual decline (Fig. 5D). PADI4-high neutrophils were enriched along an early/intermediate pseudotemporal state associated with immature and stress-mobilized neutrophil features, consistent with an emergency granulopoiesis-associated phenotype that may facilitate rapid inflammatory responses to radiation-induced tissue damage. Pseudotime analysis reveals conserved PADI4 expression dynamics during neutrophil development (A-D: GSE298566; E-H: E-MTAB-11188; I: GSE79044). A Pseudotime trajectory of neutrophil subsets from RILI model. Cells are colored by collection time (control, day 1, and day 7 post-irradiation), showing N1-N4 subsets at the start and N0 subset at the end of the trajectory. B Expression trends of the most significantly changed genes along pseudotime. C Expression trends of three representative genes from pseudotime differential analysis: positively correlated genes displaying a U-shaped pattern (left), and negatively correlated genes displaying an inverted U-shaped pattern (right). D Padi4 expression pattern along the pseudotime trajectory, displaying a characteristic increasing-then-decreasing trend. E UMAP visualization of human bone marrow (BM) and peripheral blood (PB) neutrophils from single-cell RNA sequencing data. F, Violin plots comparing PADI4 and other key genes expression levels between human bone marrow and peripheral blood neutrophils. G, Pseudotime trajectory of human neutrophil development, with bone marrow neutrophils at the start and peripheral blood neutrophils at the end. H PADI4 expression dynamics along human neutrophil developmental trajectory, showing a bell-shaped curve. I PADI4 expression kinetics during ATRA-induced neutrophil differentiation of HL-60 cells in vitro. Statistical significance between the two groups was determined by unpaired two-tailed Student’s t-test after confirmation of normality and equal variance To assess the conservation of this PADI4 expression pattern in humans, we analyzed scRNA-seq data from healthy human bone marrow and peripheral blood neutrophils [32] (Fig. 5E). This analysis confirmed that PADI4 expression was significantly higher in bone marrow neutrophils compared to their mature peripheral blood counterparts (Fig. 5F). Further pseudotime analysis of neutrophil development positioned bone marrow-derived neutrophils at the initiation of the differentiation trajectory, with mature peripheral blood neutrophils localized to the terminal end (Fig. 5G). Along this developmental continuum, PADI4 expression exhibited a characteristic dynamic rise-and-fall pattern, increasing during early differentiation and subsequently declining as neutrophils matured (Fig. 5H). This precise kinetic profile establishes PADI4 as a key dynamically regulated molecule during early human neutrophil development and maturation. Consistent with this in vivo observation, PADI4 expression displayed a comparable pattern of initial upregulation followed by downregulation during the in vitro differentiation of human HL-60 cells from promyelocytes to mature neutrophils [33, 34] (Fig. 5I). Altogether, these results demonstrate that the PADI4-high neutrophils infiltrating RILI lesions are fundamentally immature populations, mobilized under stress conditions and prematurely recruited to the injured lung tissue. PADI4 targeting attenuates neutrophil-driven inflammatory programs in RILI To directly evaluate the functional involvement of PADI4 in RILI pathogenesis using the appropriate radiation model, we performed bulk RNA-seq on lung tissues from mice subjected to 17 Gy thoracic irradiation and treated with the PADI4 catalytic inhibitor GSK484 or vehicle. Differential expression analysis revealed substantial transcriptomic reprogramming after PADI4 inhibition, with prominent suppression of neutrophil activation and NETosis-associated genes in GSK484-treated lungs compared with vehicle-treated RILI controls (Fig. 6A, B). Heatmap visualization of core pathogenic genes, including Padi4, S100a8, S100a9, Mpo, and Mmp9, confirmed coordinated downregulation of the neutrophil/NETosis module (Fig. 6C). GSVA analysis demonstrated reduced enrichment of the neutrophil inflammatory module after GSK484 treatment (Fig. 6D). KEGG- and GO-based GSEA further demonstrated significant attenuation of NETosis- and neutrophil-associated inflammatory pathways after GSK484 treatment (Fig. 6E, F). These transcriptomic findings were corroborated by qPCR validation of representative inflammatory genes, including Padi4, S100a8, S100a9, Il1b, and Tnfα (Fig. 6G). Pharmacological PADI4 inhibition rewires neutrophil-associated inflammatory programs in RILI at the transcriptomic level. Mice were subjected to 17 Gy thoracic irradiation and treated with GSK484 (4 mg/kg, i.p.) or vehicle control. Lung tissues were collected at day 7 post-irradiation for bulk RNA sequencing and downstream analyses. A Principal component analysis (PCA) of transcriptomic profiles from sham, RILI + vehicle, and RILI + GSK484 groups. B Volcano plot of differentially expressed genes (DEGs) between GSK484-treated and vehicle-treated RILI lungs. C Heatmap of representative NETosis- and neutrophil-associated inflammatory genes. D GSVA scores of neutrophil-associated inflammatory signatures. E, F KEGG- and GO-based GSEA of neutrophil-associated inflammatory pathways. NES, normalized enrichment score. G qPCR validation of representative inflammatory genes in lung tissues (n = 12 per group). Data are presented as mean ± SD. Statistical significance was determined by one-way ANOVA followed by Tukey’s post-hoc test. **p < 0.01, ***p < 0.001, ****p < 0.0001 Subsequently, we assessed whether these molecular changes translated into tissue-level protection. GSK484 treatment markedly improved alveolar architecture and reduced histopathological lung injury scores in irradiated mice (Fig. 7A, B). Immunohistochemical staining confirmed a significant reduction in PADI4- and S100A8-positive inflammatory cells within lung lesions (Fig. 7C, D). To directly evaluate PADI4 catalytic activity in situ, immunofluorescence staining for citrullinated histone H3 (CitH3) was performed to detect NETosis-associated chromatin citrullination. Vehicle-treated RILI lungs exhibited abundant CitH3 immunoreactivity, whereas GSK484 treatment markedly reduced CitH3 immunoreactivity (Fig. 7E, F). Circulating PADI4 and MPO-DNA complexes were also decreased after GSK484 treatment (Fig. 7G, H), and Western blot analysis showed significantly reduced levels of PADI4, MPO, and S100A8/A9 in lung homogenates (Fig. 7I, J). Because GSK484 is a catalytic inhibitor, these data support suppression of PADI4 enzymatic activity as the primary protective mechanism, whereas reductions in PADI4 transcript or protein abundance are interpreted as secondary to decreased neutrophil infiltration and inflammatory amplification. Therapeutic efficacy and mechanistic validation of PADI4 inhibition in RILI. A Representative H&E-stained lung tissue sections. Scale bar, 200 μm. B Histopathological lung injury scores (n = 12). C Representative immunohistochemistry (IHC) images of lung sections stained for PADI4 and S100A8. Scale bar, 200 μm. D Quantification of PADI4- and S100A8-positive cells (n = 10). E, F Representative multiple immunofluorescence images and quantification of CitH3 staining in lung tissues from sham, RILI + vehicle, and RILI + GSK484 groups. GSK484 treatment markedly reduced CitH3 immunoreactivity in irradiated lungs. Nuclei were counterstained with DAPI. Scale bar, 100 μm. G, Serum PADI4 levels measured by ELISA (n = 12). H Circulating MPO-DNA complex levels measured by ELISA (n = 12). I Representative Western blot images of PADI4, MPO, and S100A8/A9 in lung tissues. J Quantification of protein expression levels normalized to Vinculin (n = 6). Data are presented as mean ± SD. Statistical significance was determined by one-way ANOVA followed by Tukey’s post-hoc test. *p < 0.05, **p < 0.01, ***p < 0.001 Lung epithelial-derived CSF2 promotes PADI4-high neutrophil activation and contributes to RILI in vivo To identify the upstream signals driving PADI4 expression in neutrophils, we established an indirect co-culture system using lung epithelial cells and neutrophil-like cells (Fig. 8A). Conditioned medium from irradiated BEAS-2B bronchial epithelial cells significantly upregulated the mRNA expression of neutrophil activation and emergency granulopoiesis–associated markers, including PADI4, CD11B, and MMP9, in HL-60 neutrophil-like cells (Fig. 8B). Although HL-60 is a promyelocytic leukemia cell line and cannot fully recapitulate primary neutrophils, it is widely used as a tractable model for neutrophil-like differentiation and activation studies [35–37]. These findings suggest that soluble factors released by radiation-damaged epithelial cells can drive myeloid cells toward an activated, PADI4-high phenotype. Lung epithelial cells induce neutrophil PADI4 expression through CSF2 (GSE166340). A Schematic diagram of the indirect co-culture system between lung epithelial cells and neutrophils. (Created by using BioRender, https://BioRender.com/u8jwp6c). B mRNA expression levels of PADI4, CD11b, and MMP9 in HL-60 cells treated with conditioned medium from irradiated BEAS-2B cells, measured by qPCR. Statistical significance among groups was determined by one-way ANOVA with Tukey’s post-hoc test; *p < 0.05, **p < 0.01, ***p < 0.001. C Volcano plot of differentially expressed genes in human bronchial epithelial cells after irradiation, highlighting candidate molecules including CSF2, IL1A, IL1B, PTX3, and SERPINE1. D qPCR validation of candidate gene expression in BEAS-2B cells after irradiation. Statistical significance was determined by unpaired two-tailed Student’s t-test after confirmation of normality and equal variance; *p < 0.05, **p < 0.01, ***p < 0.001. E, F, G Functional enrichment analyses of irradiation-induced epithelial transcriptomic alterations: (E) GO biological processes enrichment analysis, (F) KEGG pathways enrichment analysis, and (G) Gene Set Enrichment Analysis (GSEA) of CSF2-related pathways. H Functional validation of CSF2 signaling in neutrophil-like HL-60 cells. Relative mRNA expression levels of PADI4, CD11B (ITGAM), and MMP9 were measured by qPCR following treatment with recombinant human CSF2 protein in the presence or absence of CSF2-neutralizing antibody. Data are presented as mean ± SD. Statistical significance was determined by one-way ANOVA followed by Tukey’s post-hoc test. *p < 0.05, **p < 0.01, ***p < 0.001. I Functional validation of epithelial-derived CSF2 signaling in neutrophil-like HL-60 cells. Relative mRNA expression levels of PADI4, CD11B (ITGAM), and MMP9 were measured by qPCR following 48-hour treatment with control medium, conditioned medium (CM) from 8 Gy-irradiated BEAS-2B cells, CM co-treated with CSF2-neutralizing antibody, or CM co-treated with isotype control antibody. Data are presented as mean ± SD. Statistical significance was determined by one-way ANOVA followed by Tukey’s post-hoc test. *p < 0.05, **p < 0.01, ***p < 0.001 To systematically identify candidate epithelial-derived mediators, we analyzed transcriptome data from irradiated human bronchial epithelial cells (GSE166340) [38]. Differential expression analysis identified several significantly upregulated secreted or inflammatory factors, including CSF2, IL1A, IL1B, PTX3, and SERPINE1 (Fig. 8C). Their upregulation was further confirmed by qPCR in the BEAS-2B irradiation model (Fig. 8D). Given that CSF2 (GM-CSF) is a well-established regulator of myeloid cell survival, activation, and granulocytic differentiation [39–42], we further performed enrichment analyses, which demonstrated significant association of CSF2-related programs with immune-inflammatory processes, including cytokine signaling and cytokine activity (Fig. 8E-G). Based on these findings, we selected CSF2 for functional validation. Recombinant CSF2 recapitulated the effect of conditioned medium by inducing PADI4, CD11B, and MMP9 expression in HL-60 neutrophil-like cells (Fig. 8H). Conversely, CSF2-neutralizing antibody treatment effectively attenuated conditioned medium-induced gene upregulation (Fig. 8I). Together, these findings support CSF2 as a major epithelial-derived upstream regulator of the PADI4-high activated myeloid phenotype. Importantly, because additional inflammatory cytokines such as IL1A and IL1B were also induced after irradiation, our findings do not exclude potential cooperative contributions from other epithelial-derived mediators. We then investigated whether CSF2 signaling contributes to pathogenic neutrophil accumulation in vivo. Thoracic irradiation induced a significant systemic increase in serum GM-CSF levels compared with sham controls (Fig. 9A). In vivo GM-CSF neutralization significantly attenuated radiation-induced lung injury, as demonstrated by improved alveolar architecture and reduced histopathological injury scores compared with isotype-treated RILI mice (Fig. 9B, C). Immunofluorescence staining further revealed markedly reduced infiltration of PADI4-positive inflammatory cells within irradiated lung lesions following GM-CSF blockade (Fig. 9D, E). Consistently, qPCR and Western blot analyses demonstrated reduced Padi4 expression and decreased PADI4, MPO, and S100A8/A9 protein levels in lung tissues after anti-GM-CSF treatment (Fig. 9F-H). In vivo GM-CSF blockade attenuates PADI4-high neutrophil accumulation and lung injury in RILI. Mice subjected to 17 Gy thoracic irradiation received anti-GM-CSF neutralizing antibody or isotype control. Lung tissues and peripheral blood were collected at day 7 post-irradiation. A Serum GM-CSF/CSF2 concentrations measured by ELISA in control and irradiated mice (n = 4). Data are presented as mean ± SD; statistical significance was determined by unpaired two-tailed Student’s t-test; **p < 0.001. B, C Representative H&E-stained lung tissue sections (B) and semiquantitative histopathological injury scores (C; n = 6). Scale bar, 200 μm. Data are presented as mean ± SD; statistical significance was determined by one-way ANOVA with Tukey‘s post-hoc test; *p < 0.05, **p < 0.01, ***p < 0.001. D, E Representative immunofluorescence images (D) and quantification (E) showing PADI4 (green), and DAPI (blue) in lung tissue (n = 6). Scale bar, 100 μm. Data are presented as mean ± SD; statistical significance was determined by one-way ANOVA with Tukey’s post-hoc test; *p < 0.05, **p < 0.01, ***p < 0.001. F Relative mRNA expression levels of Padi4 in lung tissues (n = 8). Data are presented as mean ± SD; statistical significance was determined by one-way ANOVA with Tukey’s post-hoc test; p < 0.05, *p < 0.01, **p < 0.001. G, H Representative Western blot images (G) and semiquantitative protein analysis (H; n = 6) of PADI4, MPO, and S100A8/A9 expression in lung tissues. Vinculin served as the loading control. Data are presented as mean ± SD; statistical significance was determined by one-way ANOVA with Tukey’s post-hoc test; *p < 0.05, **p < 0.01, ***p < 0.001 Having identified CSF2 as a key local inducer of PADI4, we next investigated whether systemic hematopoietic mobilization signals exert a similar regulatory effect. We analyzed single-cell sequencing data (E-MTAB-11188) from bone marrow and peripheral blood neutrophils obtained from healthy donors before and after G-CSF mobilization [32] (Fig. 10A, B). Under homeostatic conditions, bone marrow neutrophils exhibited high PADI4 expression, which was significantly downregulated upon maturation and release into the peripheral blood (Fig. 10C, D). However, G-CSF mobilization was associated with the appearance of immature, PADI4-high neutrophils in peripheral blood [43]. Consequently, the average PADI4 expression in peripheral blood neutrophils was significantly elevated in the G-CSF-treated group compared to the untreated group (Fig. 10C, D). Further pseudotime analysis revealed that the developmental trajectory of neutrophils from G-CSF-treated donors was markedly altered, with a substantial proportion of cells entering the systemic circulation while retaining a high-PADI4 immature phenotype (Fig. 10E-G). These results indicate that G-CSF interrupts the normal neutrophil maturation and promotes systemic accumulation of immature neutrophil populations [44]. G-CSF systemically mobilizes PADI4-high immature neutrophils (E-MTAB-11188). A UMAP visualization of neutrophil subpopulations in bone marrow and peripheral blood. B UMAP projection comparing neutrophil distributions before and after G-CSF treatment. C PADI4 expression levels in bone marrow versus peripheral blood neutrophils under normal conditions and after G-CSF treatment. D UMAP visualization of PADI4-high neutrophil distribution in peripheral blood before and after G-CSF treatment. E Pseudotime trajectory analysis of neutrophil development (colored by subpopulation) under normal conditions and after G-CSF treatment. F Pseudotime trajectory analysis of neutrophil development (colored by pseudotime value) under normal conditions and after G-CSF treatment. G, PADI4 expression dynamics along pseudotime trajectories, showing interrupted maturation program in G-CSF-treated conditions Together, these data suggest that both local tissue injury signals (CSF2) and systemic hematopoietic mobilization signals (G-CSF) can converge on the generation or accumulation of immature PADI4-high neutrophils during emergency granulopoiesis. Discussion Consistent with recent studies emphasizing the central role of innate immune activation in clinically relevant RILI progression, our findings further support the concept that persistent neutrophil-driven inflammatory signaling constitutes a critical pathological driver of radiation-induced pulmonary damage [45]. The profound heterogeneity and functional plasticity of neutrophils in physiological and pathological contexts have gained growing recognition [2, 46]. Within this framework, our study systematically delineates the dynamic evolution and regulatory mechanisms of neutrophils in RILI, a pathology characterized by maladaptive inflammatory amplification after thoracic irradiation. By integrating multi-omics sequencing, cross-species validation, and functional experiments, we identify a previously under-characterized pathogenic axis in RILI: radiation-damaged lung epithelial cells release CSF2, which promotes the emergence and pulmonary accumulation of immature, PADI4-high neutrophils. These pathogenic neutrophils are associated with NETosis, inflammatory tissue damage, and impaired lung architecture. This work identifies PADI4 as a critical molecular node linking tissue damage signals to dysregulated myeloid immunity, thus highlighting its substantial potential as a therapeutic target. Our investigation commenced with an unbiased systems biology approach. Transcriptomic profiling analysis of RILI identified PADI4 as a top upregulated gene, whose expression was strongly correlated with neutrophil infiltration and NETosis-associated programs [22]. This suggested that a fundamental shift in neutrophil functional states, rather than a mere numerical increase, is central to disease pathogenesis [47]. Our subsequent analysis of single-cell transcriptomic atlas directly validated this hypothesis. We unveiled extensive heterogeneity of neutrophils in RILI, identifying five distinct subsets with divergent functional phenotypes. The N1-N4 subsets, which predominate the later stages of injury, are characterized by the specific high expression of PADI4 and core NETosis modules [24], effectively constituting a terminal effector population that mediates tissue damage. These findings advance the understanding of neutrophils in RILI from a homogeneous population to a collection of functionally specialized subsets, providing a precise cellular foundation for targeted therapeutic interventions. Crucially, pseudotime trajectory analysis further suggested that PADI4-high neutrophils in RILI are closely related to immature populations generated during stress-induced emergency granulopoiesis [2]. This pattern was conserved in human bone marrow-peripheral blood neutrophil developmental trajectories (E-MTAB-11188) and was also reflected in HL-60 neutrophil-like differentiation. These observations support the concept that PADI4-high neutrophils represent an immature and functionally activated myeloid state rather than a uniform expansion of mature circulating neutrophils. Direct pharmacological intervention with GSK484 in the RILI model further demonstrated that PADI4 activity is functionally involved in sustaining neutrophil chemotaxis, NETosis-associated programs, and tissue-damaging inflammation. Thus, integrated developmental, comparative, and pharmacological evidence collectively establishes PADI4 as a critical molecular node governing the pathogenic neutrophil program in radiation-induced lung injury [48]. Mechanistically, we link the initial radiation-induced epithelial insult to downstream myeloid activation through the epithelial-myeloid CSF2-PADI4 axis. Injured lung epithelial cells function as local inflammatory signal sources and release CSF2, which is sufficient to induce PADI4 and activation-associated genes in neutrophil-like cells. Conversely, CSF2 neutralization blunted conditioned medium-induced PADI4 upregulation in vitro, and in vivo GM-CSF blockade reduced PADI4-positive neutrophil infiltration, inflammatory cytokine production, and histopathological lung injury. Nevertheless, our data also show that other epithelial-derived cytokines, including IL1A and IL1B, are induced after irradiation. Therefore, CSF2 should be regarded as a major upstream regulator rather than the only mediator of epithelial-neutrophil communication in RILI. Future studies should define how CSF2 cooperates with other cytokines to shape pathogenic neutrophil states. Building upon this mechanistic foundation, pharmacological inhibition of PADI4 with GSK484 provides direct preclinical evidence that interrupting this pathway can mitigate RILI. Because GSK484 is a catalytic inhibitor of PADI4, its primary protective mechanism is most directly supported by the observed suppression of CitH3 immunoreactivity in injured lung tissue. The accompanying reductions in Padi4 mRNA and PADI4 protein are interpreted as secondary consequences of reduced pathogenic neutrophil infiltration and inflammatory feedback, rather than as direct transcriptional effects of GSK484. These findings support PADI4 enzymatic activity as a therapeutically actionable component of neutrophil-driven lung injury. Several limitations should be acknowledged. First, although HL-60 cells provide a tractable model for neutrophil-like differentiation and activation, validation in primary immature neutrophils derived from bone marrow would further strengthen the physiological relevance of the CSF2-PADI4 regulatory axis. Second, the present study mainly focuses on early inflammatory events in RILI, whereas the long-term contribution of PADI4-high neutrophils to fibrosis progression remains to be determined. Third, although our experimental data support CSF2 as a major upstream regulator, additional cytokines such as IL1A and IL1B may cooperate with CSF2 in shaping pathogenic neutrophil states. Finally, larger multicenter clinical cohorts will be necessary to evaluate the prognostic value and translational potential of the CSF2-PADI4 axis in human RILI. In summary, this work elucidates an epithelial-myeloid pathogenic pathway in RILI, linking radiation-induced epithelial damage to CSF2-dependent activation of PADI4-high neutrophils and downstream NETosis-associated tissue injury. Pharmacological inhibition of PADI4 and in vivo GM-CSF blockade both attenuate neutrophil-driven inflammation and lung damage, supporting the CSF2-PADI4 neutrophil axis as a promising therapeutic target for RILI intervention (Fig. 11). Recent clinical and translational studies have highlighted the urgent need for mechanism-based therapeutic strategies targeting persistent inflammatory and immune-remodeling processes in RILI, particularly during the transition from acute pneumonitis to chronic fibrotic progression [49]. CSF2-PADI4 axis drives pathogenic neutrophil-mediated radiation-induced lung injury. Schematic model illustrating the pathogenic circuit and therapeutic intervention. Thoracic irradiation causes lung epithelial damage and CSF2 release. CSF2 drives emergency granulopoiesis in the bone marrow, generating and mobilizing immature PADI4-high neutrophils. These cells infiltrate the lungs and execute pathogenic NETosis, resulting in alveolar structure destruction and tissue damage. The specific PADI4 inhibitor GSK484 effectively blocks this pathogenic axis, ameliorating NETosis and tissue injury. (Created by using BioRender, https://BioRender.com/u8jwp6c) Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2 Acknowledgements We appreciate Dr. Feizhen Wu (Institute of Biomedical Science, Fudan University) for his valuable advice in the bioinformatic analyses, and appreciate the Medical Science Data Center of Shanghai Medical Collage for the support and assistance in the bioinformatic analysis. Abbreviations Analysis of Variance Bone Marrow Conditioned Medium Colony-Stimulating Factor 2 (GM-CSF) Differentially Expressed Genes Enzyme-Linked Immunosorbent Assay Fas Ligand Granulocyte Colony-Stimulating Factor Gene Expression Omnibus Gene Expression Profiling Interactive Analysis Granulocyte-Macrophage Colony-Stimulating Factor Gene Ontology Gene Set Enrichment Analysis Gene Set Variation Analysis Graft-Versus-Host Disease Hematoxylin and Eosin Immunohistochemistry Interleukin-1, Interleukin-4, Interleukin-17, etc Interferon Gamma Integrin Subunit Alpha M Kyoto Encyclopedia of Genes and Genomes Lipocalin-2 Microenvironment Cell Populations-counter Major Histocompatibility Complex Class I Matrix Metallopeptidase 9 Myeloperoxidase Neutrophil Extracellular Traps Neutrophil Extracellular Trap Formation Peptidyl Arginine Deiminase 4 Peripheral Blood Pentraxin 3 Transcription Factor PU.1 Quantitative Real-Time Polymerase Chain Reaction Radiation-Induced Lung Injury Single-Cell RNA Sequencing Standard Deviation Standard Error of the Mean Serpin Family E Member 1 S100 Calcium Binding Protein A8/A9 Transforming Growth Factor Beta Tumor Necrosis Factor (Alpha) Uniform Manifold Approximation and Projection Western Blot Author contributions Conceptualization: Yun Wang, Chunlin Shao. Methodology: Yun Wang, Maidina Abulaihaiti. Investigation, Data Analysis, Formal Analysis, and Visualization: Yun Wang. Writing-Original Draft: Yun Wang. Writing‐Review & Editing: Chunlin Shao. Supervision: Chunlin Shao. Funding acquisition: Yan Pan, Jianghong Zhang, Chunlin Shao. Funding This study is supported by National Natural Science Foundation of China (Nos. 12235004, 32171235, 12175044, and 12375338). Data availability The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. Declarations Ethical approval All animal experiments were conducted in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and were approved by the Animal Ethics Committee of Fudan University. Consent for publication Not applicable. Competing interests The authors declare that they have no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Supplementary Material 2 Data Availability Statement The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.