Uncovering the immune mechanisms underlying the emergence of immunotherapy-induced pneumonitis in lung cancer patients Authors: Xinqing Lin, Chunjie Li, Jiaxi Deng, Ying Wu, Bingbing Xie, Jingyi Bao, Wenwei Mo, Changhao Zhong, Haiyi Deng, Qingjun Pan, Yuan Qiu, Yuwen He, Shiyue Li, Jia Li, Shengbao Suo, Chengzhi Zhou Published: 2025 ABSTRACT AbstractThe clinical application of checkpoint inhibitors in lung cancer has been impeded by the emergence of immune-related adverse effects, such as pneumonitis. To date, the precise immune pathogenesis of immune checkpoint inhibitor-related pneumonitis (CIP) remains elusive. Here, we perform comprehensive single-cell analysis, specifically the combination of scRNA-seq and scTCR/BCR-seq, to profile molecular and cellular changes in CIP tissues and matched noncancerous adjacent tissues from lung cancer patients. CIP patients exhibit disrupted immune homeostasis, marked by expansion of the CD8⁺ tissue-resident memory T cell population, elevated IFNG expression and increased TCR clone sharing with other CD8⁺ T cells. We also identify increased IL-17A levels, robust IgG isotype class switching in B cells and GSDME-mediated macrophage pyroptosis as potential mechanisms involved in CIP. These findings provide valuable insights into the mechanisms underlying CIP and inform potential strategies for further intervention.Subject terms: Translational research, Data mining, Non-small-cell lung cancer, Monocytes and macrophages, CD8-positive T cells Introduction The groundbreaking advent of immune checkpoint inhibitor (ICI) therapy has revolutionized the treatment landscape for various cancer types, with a subset of patients experiencing durable remission and a marked increase in overall survival. The monoclonal antibodies used as ICI therapies act by disrupting inhibitory signalling pathways within the immune synapse through the targeting of key elements such as programmed cell death protein 1 (PD-1), its ligand (PD-L1) and cytotoxic T-lymphocyte antigen-4 (CTLA-4). The outcome is increased immune system activation, thereby enhancing the elimination of tumours1. Despite the promising progress in the use of ICIs in cancer treatment, these therapies can trigger immune-related adverse events (irAE) that may affect different organs, posing substantial obstacles to treatment success. Notably, patients with certain cancer types exhibit distinct ICI-induced irAEs. For example, a pronounced increase in checkpoint inhibitor-related pneumonitis (CIP) has been noted among individuals receiving ICI treatment for non-small cell lung cancer (NSCLC)2. CIP is among the most serious complications of ICI treatment and poses a significant threat to patients’ lives3,4. Real-world studies have consistently reported that the incidence of CIP ranges from 13% to 19%3,4. Compared with other cancer patients, lung cancer patients, particularly those with NSCLC, exhibit the highest incidence of CIP5. In line with the management of other severe irAEs, the current approach for treating CIP involves administering high-dose systemic corticosteroids and discontinuing ICI treatment. However, 10–20% of CIP patients exhibit poor responsiveness to corticosteroids, which is a noteworthy challenge6,7. This lack of response stems from an insufficient understanding of the pathophysiology of refractory CIP, and the absence of effective treatment options contributes to a poor prognosis for this subgroup of patients8. Furthermore, the use of systemic immunosuppressive therapies, such as high-dose corticosteroids, not only complicates CIP management but also raises concerns about the potential negative impact on tumour treatment efficacy7,9. Given the increasing popularity of ICI therapy in the field of advanced lung cancer, an increase in the morbidity and mortality associated with CIP has been projected10. Consequently, there is an urgent need for in-depth research into the mechanisms underlying CIP to improve the understanding of this irAE and develop more effective therapeutic strategies. Deciphering the immunological mechanisms of CIP has been challenging, given the difficulty of obtaining relevant clinical samples and the lack of an appropriate mouse model. To date, research on CIP has primarily relied on blood and bronchoalveolar lavage fluid (BALF) samples and has focused on changes in immune cell proportions, such as the activation of Th1 and Th17 cells and the suppression of regulatory T cells (Treg)11. Studies have also reported increased lymphocyte levels in BALF, particularly CD4+ T cells, along with reduced CTLA-4 and PD-1 expression in CIP Tregs. An elevated IL-1β level in BALF myeloid cells indicates the activation of proinflammatory subsets and the loss of suppressive phenotypes in Tregs12. Among those with acute myeloid leukaemia and myelodysplastic syndrome, compared with BALF samples from infection-related pneumonia patients, BALF samples from CIP patients are enriched in IFNγ+ IL-17- CD8+ T cells and CXCR3+ CCR6+ Th17/Th1 cells13. These findings highlight the critical role of BALF samples in providing insights into immune cell populations and their functions through noninvasive means, with important clinical implications. However, BALF sampling has inherent limitations; although it captures important aspects of the immune response, it may not fully represent the complex spectrum of cellular interactions and processes involved in CIP. In particular, immune cells obtained through BALF are involved in various lung processes beyond inflammation, complicating the identification of CIP-specific characteristics. Furthermore, BALF primarily consists of easily exfoliated immune cells. This composition limits the ability to analyse non-immune cells, such as epithelial cells, which are essential for a comprehensive understanding of CIP mechanisms. To overcome these limitations, obtaining CIP tissue samples could provide more comprehensive insights into the underlying mechanisms of the disease. In our previous study, we performed bulk transcriptomic analysis using lung tissues from NSCLC patients who developed CIP after ICI treatment14. Using deconvolution algorithms, we identified changes in cell composition associated with inflammatory mediators. However, the absence of single-cell resolution data has hindered deeper mechanistic insight into CIP. In this study, we utilize a comprehensive approach integrating single-cell RNA and T-cell receptor/B-cell receptor (TCR/BCR) sequencing (scRNA/TCR/BCR-seq) to investigate the occurrence of CIP in lung tissues. Our study presents a detailed transcriptomic and immune repertoire landscape, shedding light on cell dynamics throughout the pathogenesis of CIP, particularly the roles of CD8+ tissue-resident memory T cell (Trm) expansion and GSDME-mediated macrophage pyroptosis in the development of CIP. By elucidating cellular and molecular pathways, this study addresses crucial gaps in understanding the pathogenesis and underlying mechanisms of CIP, laying a theoretical foundation for clinical diagnosis and targeted interventions. Results Study design and overview of cell populations in CIP We obtained 16 fresh lung tissue samples and performed comprehensive cellular and molecular profiling of both immune and non-immune cell populations from CIP patients (Supplementary Table 1). Nine pneumonitis samples were obtained via bronchoscopic biopsy from patients with advanced lung cancer who developed CIP following PD-1/PD-L1 immunotherapy (Supplementary Fig. 1a). As it is ethically challenging to obtain corresponding adjacent noncancerous tissues from CIP patients as control (CTL) samples, the CTL group consisted of seven adjacent noncancer tissues collected from regions more than 3 centimetres from the tumour margin in patients with operable NSCLC who did not develop any irAEs after PD-1/PD-L1 therapy. This approach avoids the immune microenvironment changes induced by the therapy itself and ensures anatomical comparability by sampling from lung lobes corresponding to those in the CIP group. Fourteen samples were subjected to combined 5’-barcoded scRNA-seq and scTCR/BCR-seq, while two additional CIP samples were subjected to scRNA-seq alone. To extend and cross-validate the findings derived from our single-cell dataset, we further integrated publicly available scRNA-seq data for blood and BALF15,16, as well as bulk RNA-seq data for the lung tissues of CIP patients (Fig. 1a)14. Although bulk RNA-seq lacks single-cell resolution, it still provides valuable complementary insights when interpreted cautiously. Finally, to validate our findings experimentally, we employed immunohistochemistry (IHC) and immunofluorescence (IF) staining and further confirmed the results using both BALF and blood samples from an independent patient cohort along with enzyme-linked immunosorbent assays (ELISA), quantitative reverse transcription polymerase chain reaction (qRT‒PCR), cytometric bead arrays (CBA) and flow cytometry (Fig. 1a, b and Supplementary Tables 2–4). Cellular overview of CIP after immunotherapy. a Workflow showing biopsy collection sites, processing of single-cell suspensions for scRNA-seq and scTCR/BCR sequencing, and validation using blood and BALF samples. Created in BioRender. Lin, X. (2025) https://BioRender.com/viqeiga. b Summary of the patient samples. The figure shows only a subset of the patients validated by IHC/IF, while information on the other validated patients is provided in Supplementary Table 2. c Uniform manifold approximation and projection (UMAP) visualization of all cells captured, colored by major cell type. The left panel displays the overall clustering, while the right panel illustrates the cellular composition of samples from different sources, including the tissue, BALF and blood. d A bar plot illustrating the main cellular composition of samples from different sources. e Violin plots display the expression of canonical marker genes across major clusters, with the y-axis representing normalized gene expression values. f Stacked bar plot showing the fractions of major cell types between CTL and CIP groups. g Volcano plot showing the log2 fold change of gene expression across all cells. P values were calculated using a two-tailed Gaussian approximation of the U statistic and adjusted with the Benjamini–Hochberg procedure. h Reactome signaling analysis based on the DEGs identified in (g). P values were derived from the hypergeometric test and corrected using the Benjamini–Hochberg procedure. CIP checkpoint inhibitor-related pneumonitis, BALF bronchoalveolar lavage fluid. After rigorous quality control of data from all sample sources, 229,281 cells were analysed, including 131,323 cells from lung tissues, 66,475 cells from BALF, and 31,483 cells from blood. For our lung tissue samples, we obtained a median of 7500 cells per CTL and 8758 cells per CIP patient (Supplementary Fig. 1b). The normalized scRNA-seq data were integrated, and the cells from each patient were evenly distributed across different clusters (Supplementary Fig. 1c). The cells were grouped by unsupervised annotation into five main clusters and validated using canonical marker genes (Fig. 1c–e and Supplementary Fig. 1d). Notably, the single-cell data from blood samples were exclusively derived from CD3+ T cells15, whereas the cells in BALF were primarily immune cells (Fig. 1c, d)16. In contrast, lung tissue samples included both immune and nonimmune cells (Fig. 1c, d). The integration of single-cell data from lung tissue, BALF and blood yielded results that clearly reflected the expected cell types specific to each source, underscoring the robustness and accuracy of our integration strategy. In lung tissue, lymphoid cells, characterized by the specific expression of PTPRC, CD3D, CD3E and CD2, represented the most abundant cell type (n = 73,301, 55.82% of the total cells; Supplementary Fig. 1e). Importantly, we observed a clear tendency for lymphoid cells to be enriched in CIP patients (Fig. 1f and Supplementary Fig. 1f). Consistent with these observations, compared with those from the CTL group, the BALF samples from CIP patients also had an increased proportion of lymphoid cells (Supplementary Fig. 1g). To characterize the biological features distinguishing CIP from CTL, we performed differentially expressed gene (DEG) and pathway enrichment analyses. CIP patients displayed elevated expression of proinflammatory cytokines (CCL5), secretory proteins and airway-related genes (SCGB1A1 and SCGB3A1) and genes linked to catalytic activity (CTSB and RGS1) (Fig. 1g and Supplementary Table 5). Pathway enrichment analysis revealed the significant enrichment of chemokine receptors that bind chemokines, TCR signalling, cytokine signalling in the immune system, inflammasomes, IFNγ signalling and PD1 signalling pathways in the CIP samples (Fig. 1h). These findings underscore a strong association between CIP and substantial alterations in pulmonary immune cells, particularly those of the lymphoid lineage. CIP-related changes in CD8+ tissue-resident memory cells We first aimed to explore the detailed characteristics of immune cells from different sources to determine the differences between the CIP and CTL groups. Immune cells were categorized into 10 major cell clusters and annotated using established cell markers17, including B cells, CD4+ T cells, CD8+ T cells, other T cells (CD3+CD4-CD8-), natural killer (NK) cells, monocytes, neutrophils, macrophages, dendritic cells (DC) and mast cells (Supplementary Fig. 2a–c). From the perspective of overall cellular composition, we observed that CD8+ T cells were more abundant in the CIP group. In contrast, innate immune cells, such as natural killer (NK) cells and neutrophils, were significantly enriched in lung tissue from the CTL group (Supplementary Fig. 2d). Similarly, in the BALF dataset, we observed the enrichment of CD8+ T cells in the CIP group (Supplementary Fig. 2e). Analysis of the bulk RNA-seq data further revealed significantly increased CD8B expression in patients with CIP (Supplementary Fig. 2f), supporting the increased number of CD8+ T cells in CIP patients and aligning with our single-cell data. CD8+ T cells represented the predominant infiltrating immune cell population (n = 30,227, accounting for 30.4% of the total immune cells), exceeding the reported ratio of 21.7% in normal lung tissue (Supplementary Fig. 2g)18. These findings suggest that ICB treatment may activate CD8⁺ T cells, driving the expansion of antigen-specific populations, their migration into peripheral tissues such as the lung, and the reactivation and proliferation of Trm cells. In some patients, this process manifests as CIP, which is typically characterized by the massive infiltration of CD8⁺ T cells into the alveolar and interstitial spaces14,19. To clarify the role of CD8+ T cells in the pathogenesis of CIP, we performed subclustering analysis and identified 10 distinct CD8+ T cell populations with marked transcriptional heterogeneity (Fig. 2a, b and Supplementary Fig. 3a). Among them, one cluster expressed naïve T cell (Tnaive) markers, such as TCF7, SELL and CCR7. Two subsets of IL7R+ central memory T cells (Tcm) were observed: Tcm-LTB cells, characterized by high LTB expression, and Tcm-FOSB cells, which were enriched in Fos/Jun signalling genes (FOS, FOSB, FOSL2, and JUNB) and CD69, suggesting an active state. Another distinct cluster demonstrated high expression of effector genes such as GZMK, with low expression of TCF7 and IL7R, indicating an effector memory-like subset (Tem_GZMK). A large proportion of CD8+ T cells displayed a tissue-resident phenotype, marked by ITAGE and ZNF683, forming Trm-like, Trm-ZNF683 and Trm-CXCR6 clusters. Notably, their distribution varied across the tissue, BALF and blood samples (Supplementary Fig. 3b, c). Tnaive, Tnk and Tcm-LTB cells were predominantly found in blood samples, which is consistent with their function as immune cells. Tm-RUNX1 cells were enriched in BALF, suggesting the preferential capture of memory-related T cells by BALF. In contrast, the Tcm-FOSB, Trm-CXCR6 and Trm-ZNF683 clusters were enriched in lung tissue, reflecting the specific tissue-resident characteristics of some T cells. Identification of CD8+ Trm cells. a UMAP embedding of the CD8+ T-lineage cells that passed quality control, colored by inferred cell types based on transcriptional profiles. b Expressions of selected marker genes in CD8+ T-cell subsets. c Fraction of CD8+ Trm-CXCR6 cells in CTL (n = 7) and CIP (n = 9) patients. A two-tailed Mann–Whitney U test was performed. d Box plots showing differences in the expression levels of ITGAE and ZNF683 between CTL (n = 29) and CIP (n = 7) patients based on bulk RNA-seq. P values were determined using a one-tailed Mann–Whitney U test. e Box plot showing the scoring of a Trm activation signature in CD8+ T cells from CTL (n = 7) and CIP (n = 9) patients. A two-tailed Mann–Whitney U test was performed. f Representative multiplex immunofluorescence images of CD8+ CD103+ T cells in CTL and CIP patients. g Box plot quantifying multiplex immunofluorescence in (f) for the CTL group (n = 5) and the CIP group (n = 5). P value was determined by a one-tailed Mann–Whitney U test. h Heatmap showing the top genes of Trm-like, Trm-ZNF683 and Trm-CXCR6 clusters. i Dot plot showing relative IFNG and STAT1 expression between CTL and CIP patients. j Representative multiplex immunofluorescence images of CD8+ IFN-γ+ T cells in CTL and CIP patients. k Box plot quantifying multiplex immunofluorescence in (j) for the CTL group (n = 3) and the CIP group (n = 3). P value was determined by a one-tailed Mann–Whitney U test. In the box plots (c–e, g and k), the centre line represents the median, the bounds of the box correspond to the 25th and 75th percentiles, the whiskers extend to the largest and smallest values within 1.5× the interquartile range from the quartiles, and individual points beyond the whiskers indicate outliers. Trm, tissue-resident memory. Since all patients underwent anti-PD-1/PD-L1 treatment, we first compared PDCD1 expression levels across groups in lung tissue. PDCD1 was expressed in Tem-GZMK and Trm cells. Although the difference was not statistically significant, a clear trend towards an increased expression pattern was observed in the CIP group compared with the CTL group (Supplementary Fig. 3d, e). These observations are consistent with those of a published study indicating that the proinflammatory environment in CIP promotes T cell activation, which in turn upregulates PDCD1 mRNA expression in T cells20. Next, we compared cell composition between the CTL and CIP groups. The Trm-CXCR6 cluster was most strongly enriched in lung tissues from the CIP group (Fig. 2c). In contrast, no such enrichment was observed in BALF or blood samples (Supplementary Fig. 3f). Based on published Trm signatures and Trm scores calculated using the GSVA algorithm21, Trm-CXCR6 cells exhibited the strongest Trm cell characteristics (Supplementary Fig. 3g). Bulk RNA-seq analysis of Trm cell-associated genes further revealed higher ITGAE expression in the CIP group. This effect was confirmed by DEG analysis (Fig. 2d, Supplementary Fig. 3h and Supplementary Table 6). Conversely, another Trm marker gene, ZNF683, was more highly expressed in the CTL group (Fig. 2d), indicating the heterogeneity of Trm cell function in CIP. Using the same calculation method as in Supplementary Fig. 3g, we found that compared with the CTL group, the CIP group displayed significantly stronger Trm cell characteristics (Fig. 2e). However, no differences in activity scores for cytotoxicity, inhibitory receptors or NK cell receptor gene signatures were detected (Supplementary Fig. 3i). To further validate the enrichment of Trm cells in CIP patients, we collected samples from 10 NSCLC patients with or without CIP who received ICB therapy (Supplementary Table 2) and performed multiplex immunofluorescence staining. These results confirmed the enrichment of Trm cells in CIP patients (Fig. 2f, g). To further characterize Trm-CXCR6 cells, we compared them with other ITGAE-positive clusters (Trm-ZNF683 and Trm-like; Supplementary Fig. 3a). The Trm-CXCR6 cluster highly expressed proinflammatory genes such as IFNG and STAT1, indicating enhanced effector function (Fig. 2h). Our analysis demonstrated that the expression of IFNG and STAT1, both of which are critical for IFNγ-dependent signalling, was upregulated in CIP patients (Fig. 2i). The elevated IFNG expression in CD8+ T cells from CIP patients was further confirmed by immunofluorescence (Fig. 2j, k). Consistent with our findings, IFNγ-producing CD8+ Trm cells have been identified specifically in ICI-induced colitis, and treatment with a JAK inhibitor, which inhibits IFNγ signalling, successfully alleviated refractory ICI-induced colitis22. Additionally, a recent phase II clinical trial demonstrated that combining JAK inhibitors with PD-1 inhibitors effectively treated metastatic NSCLC23. Taken together, these results suggest that IFNγ-producing CD8+ Trm cells are promising therapeutic targets that may not only prevent the occurrence of CIP but also increase the efficacy of immunotherapy. Further in vivo and clinical studies are warranted to validate this hypothesis. We next aimed to explore the trajectory of T cells associated with CIP, with a particular emphasis on the potential involvement of Trm cells. This inquiry holds biological significance regarding the effects of CIP on epithelial tissues, given the significant infiltration of Trm cells at these environmental interfaces24,25. Most CD8+ T cell subsets expressed αβTCRs (79% on average; Supplementary Fig. 3j). According to the TCR repertoire, the CIP group exhibited greater clonal expansion and reduced diversity, with the top T-cell clones occupying higher proportions in CIP patients than in CTL patients (Fig. 3a and Supplementary Fig. 3k). Mapping TCR information to single-cell UMAP space revealed that clonally expanded cells were associated mainly with the Tem-GZMK and Trm clusters (Fig. 3b, c). Consistently, CD8+ T cells with hyperexpanded clones highly expressed Trm-related genes, including ITGAE, ZNF683 and CXCR6 (Fig. 3d). To investigate whether clonally expanded CD8+ T cells could be identified across multiple clusters, we analysed the TCR clonotypes shared by two or more clusters in each patient. TCR sharing revealed no significant overlap across clusters in the CTL group, whereas TCR sharing was more pronounced in the CIP group, particularly for Trm cells, which showed significant overlap with Tm-RUNX1 and Tnk cells (Fig. 3e and Supplementary Fig. 3l). While both groups received immunotherapy, a key distinction is that Trm cells were able to transdifferentiate into other clusters only in CIP patients. These findings suggest that Trm cells from CIP patients are activated and may transform into other functional clusters to exert inflammatory effects. However, confirmation of the difference in Trm transdifferentiating ability between groups requires additional data. Taken together, these findings indicate a greater abundance of Trm cells in CIP patients, with the Trm-CXCR6 cluster displaying elevated IFNG activity and stronger clonal expansion. TCR repertoire analysis of CIP patients. a Box plots showing Shannon index values based on TCR clonotype diversity in CTL (n = 7) and CIP (n = 7) patients. The centre line represents the median, the bounds of the box correspond to the 25th and 75th percentiles, the whiskers extend to the largest and smallest values within 1.5× the interquartile range from the quartiles, and individual points beyond the whiskers indicate outliers. P value was determined by a one-tailed Mann–Whitney U test. b UMAP showing the expansion of clonotypes in CD8+ T cells. TCR clonotypes were classified according to their extent of clonal expansion into hyperexpanded (100 12.43%) were filtered by fitting mitochondrial gene expression values to a normal distribution and applying a false discovery rate (FDR) threshold of <0.01. Second, cells with fewer than 300 detected genes were removed to eliminate low-quality cells potentially affected by technical artefacts. Third, DoubletFinder (v. 2.0.3) was used to identify potential doublets, with a cut-off at the 92.5th percentile for the doublet score72. Cells exceeding this threshold were identified as potential doublets and excluded from downstream analysis. After these quality control procedures were performed, 229,281 high-quality single cells were retained for further analysis. To integrate individual samples and identify shared sources of variation, we performed robust principal component analysis (RPCA) using default parameters. The data were normalized with the “NormalizeData” function to scale the number of UMIs in each cell to the median UMI count across the dataset, followed by log transformation. The top 2,000 highly variable genes were identified using the “FindVariableFeatures” function. The gene expression matrix was then scaled and centred using the “ScaleData” function. Principal component analysis (PCA) was performed using the “RunPCA” function based on these highly variable genes. The top 30 principal components were selected for uniform manifold approximation and projection (UMAP) to visualize the cellular distribution in two-dimensional space. Clustering was subsequently conducted using the Leiden community detection algorithm, enabling the identification of distinct cell populations on the basis of transcriptional profiles. To optimize clustering performance, multiple trials were conducted for each major cell type using resolutions ranging from 0.4 to 1.4, followed by correlation analysis to assess cluster separation. The final resolution was set to 0.6 for all the clusters. Unless otherwise specified, all the other parameters were used with their default settings. Annotation of cell clusters and data visualization To assign the major cell types for each cluster, we employed a two-step approach. First, differential expression analysis was performed using the “FindAllMarkers” function with default parameters to identify genes whose expression differed significantly between each cluster and all others. Second, cell type identities were assigned by examining the enrichment of known cell type-specific marker genes among the top DEGs in each cluster. Using established reference markers, we determined the most likely cell type corresponding to each cluster. For the annotation of immune cell subclusters, we applied an unbiased cell-type recognition algorithm, SingleR (v.2.0.0)73, which leverages reference transcriptomic datasets of known cell types for automated annotation. SingleR was first used to evaluate the consistency of predicted cell-type labels across multiple reference datasets. On the basis of these results, the final cell type annotation was assigned to each cluster. In parallel, the top-ranked DEGs within each cluster were manually inspected to corroborate and refine the automated annotations. T cell receptor data analysis Raw TCR sequencing data were processed using Cell Ranger (v7.0.0; 10× Genomics). FASTQ reads were aligned to the human GRCh38 V(D)J reference genome (v7.0.0, 10× Genomics) using the “cellranger vdj” pipeline, resulting in the assembly of V(D)J sequences and clonotype identification74. Contig annotations, which provide high-level descriptions of each confidently assembled cellular contig, were refined by excluding records with ‘raw_consensus_id’ labelled ‘none’. Among all the detected clonotypes, 74% contained one TRA sequence and one TRB sequence, 10.5% contained two TRA sequences and one TRB sequence, 3.9% carried one TRA sequence and two TRB sequences, and only 1.0% harboured two TRA sequences and two TRB sequences. The remaining clonotypes included only a single chain. In total, 84% of all CD3+ T cells were successfully assigned αβ TCR sequences. Within the CD4+ and CD8+ subsets, TCR sequences were detectable in 90% and 79% of cells, respectively. Clonally expanded T cell cluster relationships On average, we identified 1969 unique αβ TCR clonotypes per patient, with counts ranging from 777 to 5763 across all patient samples. Among the total 27,568 unique clonotypes detected, clone sizes varied from 1 to 725 cells, and 19% exhibited expansion, defined as being detectable in at least two cells. To statistically assess the relationships among clusters in terms of clonal expansion for all shared clonotypes, we quantified the number of shared clonotypes between clusters and performed a one-sided Fisher’s exact test, followed by Benjamini‒Hochberg multiple test correction to evaluate the significance of clonotype sharing. Finally, Cytoscape (v.3.10.0) was used to visualize clonotype sharing across clusters75. B cell receptor data analysis scBCR-seq data were initially processed using Cell Ranger (v7.0.0) for V(D)J sequence assembly and BCR reconstruction, with the GRCh38 assembly in Ensembl (refdata-cellranger-vdj-GRCh38-alts-ensembl-7.0.0) used as the reference. The Change-O repertoire clonal assignment toolkit was subsequently employed to define B-cell clones76. To evaluate the repertoire similarity between BCR repertoires across groups, we compared the nucleotide sequences of the CDR3 regions. SHMs were identified by reconstructing BCR germline sequences using CreateGermlines.py, after which the number of mutations in each sequence was calculated with the observedMutations function (Shazam v1.1.0). Finally, paired scBCR-seq data were integrated with scRNA-seq data via matched unique cell barcodes using the scRepertoire package77. Bulk RNA-seq data analysis Bulk-level transcriptomes were obtained from an independent study14. The raw sequencing reads underwent several preprocessing steps. First, quality control was performed using FastQC (v.0.11.9) to assess read quality. Subsequently, the adapter sequences were trimmed using Trim Galore (v.0.6.7). The clean reads were subsequently aligned to the human reference genome (GRCh38) using HISAT2 (v.2.2.1). SAMtools (v.1.17) was used to index and merge the BAM files. For transcript quantification, StringTie (v.2.1.7) was employed, utilizing GTF annotation and BAM alignment to estimate transcript expression levels. The prepde.py3 script from StringTie was subsequently used to extract count data and calculate transcripts per million (TPM) values for each gene. TPM normalization accounts for both gene length and the number of reads mapped to each gene, providing a standardized measure of gene expression. Differential expression analysis DEG analysis between CTL and CIP patients was performed using the scRNA-seq data and the presto (v. 1.0.0) package with default parameters. Genes with an adjusted p < 0.05 and log2(fold change) > 1 were considered significantly differentially expressed, unless otherwise specified. Functional enrichment analysis To determine whether specific gene sets or pathways were significantly enriched, the “enrichGO” function of the clusterProfiler package (v. 4.6.2) was used to evaluate the fold enrichment score across various GO categories. We conducted fast gene set enrichment analysis (fGSEA; v.1.11.1)78 to identify common biological pathways using the hallmark gene sets from the molecular signatures database. For all enrichment analyses, an adjusted P value < 0.05 was considered to indicate the significant enrichment of specific gene sets or pathways associated with CIP. Calculation of gene set activity scores To evaluate the enrichment of a specific or related gene sets among highly expressed genes in each cell, we performed gene coexpression network analysis with the AUCell package (v1.22.0) to identify enriched pathways for each cell type73. On the basis of the area under the curve (AUC) values for the selected signatures, gene expression rankings were generated for each cell to estimate the proportion of highly expressed gene sets. Cells with higher expression of genes within a given gene set had higher AUC values. The “AUCell_exploreThresholds” function was used to determine the cut-off for identifying cells with active gene sets. The Trm cell, cytotoxicity, inhibitory receptor and NK cell receptor signatures were obtained from a previous publication36. Angiogenesis and phagocytosis signatures were computed following the methodology described in a previously published study32. Ligand‒receptor analysis To investigate the differences in intercellular communication between CTL and CIP patients, we employed CellChat (v1.6.1) to construct ligand‒receptor communication networks30. Significant ligand‒receptor interactions were identified in an unbiased manner from the merged scRNA-seq dataset by following the official CellChat tutorial (https://github.com/sqjin/CellChat). To analyse B cell recruitment in CIP, we manually incorporated known B cell attractant–receptor pairs, including CXCL13–CXCR5, CCR7–CCL19/CCL21, CXCR4–CXCL12, CXCR3–CXCL9/10/11, and CCR6–CCL2079. Survival analysis We utilized data from the OAK study49, an NSCLC cohort treated with anti-PD-L1 therapy and docetaxel, for survival analysis to explore the prognostic significance of the Fibro-POSTN signature within this cohort. The Fibro-POSTN signature was defined using the FindAllMarkers function in Seurat (v.4.3.0), revealing DEGs between Fibro-POSTN and other fibroblast subsets. The top 10 genes were selected as the Fibro-POSTN cluster signature because this size captures key biological signals while balancing practicality and reproducibility, making it a widely adopted gene signature. Overall survival curves were generated using the Kaplan–Meier method, and statistical significance was assessed by the log-rank test. Statistical analysis Statistical analyses were performed using R software (v.4.2.2). For normally distributed data, Student’s t test was used to assess differences between the CIP and CTL groups. For nonnormally distributed data, the Mann‒Whitney U test was applied. Unless otherwise specified, all the statistical tests were two-tailed. For multiple comparisons, P values were adjusted using the Benjamini–Hochberg (BH) FDR method. DEGs in the scRNA-seq data were identified using the nonparametric Wilcoxon rank-sum test, followed by Bonferroni’s correction on the basis of the total number of genes analysed. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary informations Transparent Peer Review file Reporting Summary Source data Source Data Acknowledgements The authors would like to thank all patients who provided samples. This work is supported in part by National Natural Science Foundation of China (32370972, S.S.), National Key Research and Development Program (2024YFF0509000, S.S.), Guangzhou Science and Technology Major Clinical Project (2023C-DZ06, C.Z.), Guangdong Basic and Applied Basic Research Foundation (2024B1515020052, S.S.; 2023A1515011783, S.S.; 2020A1515011293 Y.Q.), Fundamental and Applied Fundamental Research Project of City-School (Institute) Joint Funding Project, Guangzhou Science and Technology Bureau (202201020466, Y.H.), Beijing Xisike Clinical Oncology Research Foundation (Y-pierrefabre202101-0023, X.L.), China Primary Health Care Foundation (BH001496, C.Z.), autonomous Research Project of the National Key Laboratory of Respiratory Diseases (SKLRD-Z-202206, X.L.), the Union Project from Guangzhou National Laboratory and State Key Laboratory of Respiratory Disease, Guangzhou Medical University (GZNL2024B01004, S.S.) and Major Project of Guangzhou National Laboratory (GZNL2023A02007, S.S.; GZNL2023A03005, S.S.). Author contributions C.Z., S.S. and X.L. conceptualized and designed the study; X.L., C.L. and J.D. performed the methodology; C.L. performed bioinformatics data analysis, generated the majority of the figures, and wrote the manuscript; J.D collected clinical samples, performed immunofluorescence and flow cytometry, and contributed to writing the manuscript; Y.W., B.X., J.B., W.M., C.Z., H.D., Q.P., Y.Q., Y.H., S.L., and J.L. participated in the interpretation and gave practical advice; S.S. and X.L. performed the review and editing, S.S. and C.Z. supervised the study. All authors read and approved the manuscript. Peer review Peer review information Nature Communications thanks Andreas Pircher, Karthik Suresh who co-reviewed with Mohammad Ghanbar and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Data availability The single-cell transcriptomic data generated in this study have been deposited in the National Genomics Data Center of China and are publicly available under the accession code HRA006607. This research reanalyses existing, publicly available genomic datasets, including data from ArrayExpress under accession numbers E-MTAB-6149 and E-MTAB-6653, the GEO database under accession number GSE216329, Mendeley Data under accession number 8c3x28r5hk (https://data.mendeley.com/datasets/8c3x28r5hk/1), and the EGA European Genome-Phenome Archive under accession number EGAS00001005013. Flow cytometry and immunofluorescence imaging data have been deposited in Zenodo (10.5281/zenodo.17318429). All the data are included in the Supplementary Information or available from the authors, as are the unique reagents used in this article. The raw numbers for charts and graphs are available in the Source Data file whenever possible. Source data are provided with this paper. Code availability The codes for these analyses adhered to the vignettes for all the packages mentioned in the methods section. The source code used to analyse the data is available upon request to the corresponding authors. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Xinqing Lin, Chunjie Li, Jiaxi Deng. Contributor Information Shengbao Suo, Email: suo_shengbao@gzlab.ac.cn. Chengzhi Zhou, Email: zhouchengzhi@gird.cn. Supplementary information The online version contains supplementary material available at 10.1038/s41467-025-66509-x. References Associated Data Supplementary Materials Supplementary informations Transparent Peer Review file Reporting Summary Source Data Data Availability Statement The single-cell transcriptomic data generated in this study have been deposited in the National Genomics Data Center of China and are publicly available under the accession code HRA006607. This research reanalyses existing, publicly available genomic datasets, including data from ArrayExpress under accession numbers E-MTAB-6149 and E-MTAB-6653, the GEO database under accession number GSE216329, Mendeley Data under accession number 8c3x28r5hk (https://data.mendeley.com/datasets/8c3x28r5hk/1), and the EGA European Genome-Phenome Archive under accession number EGAS00001005013. Flow cytometry and immunofluorescence imaging data have been deposited in Zenodo (10.5281/zenodo.17318429). All the data are included in the Supplementary Information or available from the authors, as are the unique reagents used in this article. The raw numbers for charts and graphs are available in the Source Data file whenever possible. Source data are provided with this paper. The codes for these analyses adhered to the vignettes for all the packages mentioned in the methods section. The source code used to analyse the data is available upon request to the corresponding authors.