2.1 Overview of immune cell landscapes in peripheral blood and lung

To characterise the nature of circulating and pulmonary immune cell populations in anti-MDA5+ DM patients with RP-ILD, we performed scRNA-seq (10X Genomics) to analyse paired PBMCs and BALF from five patients. Patients’ clinical data were recorded and presented in Table S1. A high-quality scRNA-seq dataset, comprising 82 819 cells (44 689 cells from PBMCs and 38 130 cells from BALF), enabled us to delineate the immune landscape in both peripheral and lung domains (Figure 1A). We contrasted these profiles with datasets from healthy controls (HCs), particularly with GSE158055 for PBMCs (HC003, HC005, HC008 and HC010) and GSE145926 for BALF (HC1, HC2 and HC3). The demographic information of HCs is provided in Table S2.

In PBMCs, nine major cell types, namely CD4 T cells, CD8 T cells, CD8 T_cycling cells, natural killer (NK) cells, monocytes, B cells, platelets, plasma cells (PCs) and dendritic cells (DCs), were discerned and annotated through established marker genes (Figures 1B,C,D and S1A). Likewise, in BALF, six predominant cell types comprising macrophages, T cells, NK cells, DCs, epithelial cells and B cells were identified and characterised using the same methodology (Figures 1E,F,G and S1C).

Significant differences were evident between HCs and anti-MDA5+ DM patients based on t-distributed stochastic neighbour embedding (t-SNE) projections (Figures 1C,F and S1B,D). In PBMC, the most prominent change was a reduction of NK cell populations, although the difference in NK cell ratio did not reach statistical significance (Figures 1H and S1E). Further analysis of clinical data from 57 additional anti-MDA5+ DM inpatients corroborated these findings, revealing a general decrease in NK and T cell counts, including CD4 T and CD8 T cells (Figure 1I), aligning with previous studies.^{9, 19} On the contrary, in BALF, we found enrichment of NK cells in patients, especially in DM1, DM2 and DM3, diverging from the HCs’ profiles (Figures 1H and S1F). Notably, despite macrophages being the most populous in BALF for both groups, their transcriptional profiles varied significantly (Figure 1F). Together, these data reveal great changes in periphery blood and lung immune cell compartments compared with HCs.

2.2 Regional pulmonary hyperinflammation in contrast to peripheral blood mononuclear cells

Anti-MDA5+ DM-associated RP-ILD could progress to respiratory failure, which is featured by cytokine storm syndrome.²⁰ Histopathological analysis of lung biopsy samples from these patients demonstrated diffuse alveolar damage, fibroproliferative responses and the presence of fibrotic foci (Figure 2A). High-resolution computed tomography (HRCT) further corroborated these findings, showing characteristic patterns of diffuse ground-glass opacities and consolidations (Figure S2A).

We compared gene expressions of cytokine and chemokine in samples from HCs and patients (Figure 2B). The scRNA-seq data from BALF indicated elevated levels of interleukins (IL6, IL10) and TNF, alongside a range of chemokines (CCL2, CCL3, CCL4, CCL7, CCL8, CXCL10 and CXCL11) in patients compared with HCs. This pattern was distinct from that in PBMCs, suggesting a hyperinflammatory state localised to the lung. Luminex assay measurements of BALF and serum in patients aligned with the scRNA-seq data, showing higher concentrations of inflammatory cytokines and chemokines (IL6, IL18, CCL2, CCL4, CXCL1, CXCL8 and CXCL10) in BALF than in paired serum (Figures 2C and S2B). Compared with HCs, the concentrations of inflammatory cytokines and chemokines in patients’ serum were elevated, particularly for IFN-α, IL6, IL18, CCL2, CXCL1, CXCL8 and CXCL10 (Figures 2D and S2C). However, an analysis of the transcriptome data from PBMCs of DM patients did not reveal a significant inflammatory state compared with HCs (Figure 2B). This suggests that the proinflammatory cytokines detected in the serum are likely derived from lung tissue cells rather than from peripheral blood cells.

We next aimed to identify the cellular sources of cytokine production in the patients’ lungs. Cytokine and inflammatory scores were calculated for each cell cluster based on cytokine expression and inflammatory response genes, respectively, to assess their contribution to the inflammatory cytokine storm²¹ (Figures 2E,F). Our analysis revealed significantly elevated expression of these cytokine and inflammatory genes among various immune cells and epithelial cells in patients’ BALF, signifying an intense inflammatory response in the lung (Figure 2F). Notably, macrophages in patients’ BALF samples exhibited substantially higher cytokine and inflammatory scores, indicating their pivotal role in driving inflammatory storm. In contrast, peripheral monocytes demonstrated markedly lower cytokine and inflammatory scores than those in HCs, reflecting their functional impairments and a state of cytokine deficit (Figures 2E and S2D). This disparity underscores the localised nature of the hyperinflammatory response in the lung, which is distinct from the peripheral immune landscape observed in PBMCs.

Given the significant increase in multiple chemokine concentrations in BALF, we assessed gene expression levels of chemokine receptors of PBMC (Figure 2G). The receptor CCR7 of T cells and CX3CR1 of NK cells were substantially upregulated in anti-MDA5+ DM patients, suggesting that the observed reduction in circulating T and NK cells might be attributed to their migration to the lung. Furthermore, monocytes demonstrated elevated expression of CCR1 and CCR2, supporting the hypothesis of monocyte recruitment to the pulmonary microenvironment. Besides, our examination of chemokine expressions across different cell clusters in BALF revealed that lung macrophages in patients are likely key players in local inflammation (Figure 2H). They exhibit high transcriptional levels of chemokines such as CXCL10, CXCL11, CCL2, CCL3, CCL7 and CCL8, which may facilitate the recruitment of additional monocytes and T cells to the lungs, thereby amplifying the inflammatory response.

To identify potential prognostic markers, we analysed differences in cytokine and chemokine expression levels in BALF between surviving and deceased patients. Transcriptomic analysis indicated significantly elevated expression of CXCL10 and CCL2 in BALF macrophages from deceased patients compared with survivors (Figure 2I). Based on the hypothesis mentioned above, the cytokine expression in BALF could be indirectly reflected by serum levels, and since peripheral blood samples are easier to obtain, we used serum cytokine levels to validate the findings. Patients were then further stratified into progression and remission groups based on follow-up pulmonary function data. Although the sample size was limited, CCL2 levels were notably higher in the progression group compared with the remission group (Figure 2J). Receiver operating characteristic (ROC) curve analysis demonstrated that CCL2 effectively distinguished patients with disease progression from those in remission, achieving an area under the curve (AUC) of 0.873 (Figure 2K). Furthermore, univariate logistic regression analysis identified CCL2 as a significant predictor of ILD progression (odds ratio [OR] = 1.012, 95% confidence interval [CI]: 1.003–1.025, p = .0247). In contrast, CXCL10 did not exhibit significant predictive value, potentially due to the small sample size (OR = 1.001, 95% CI: 0.996–1.006, p = .5544).

These findings highlight a discrepancy between peripheral immune dysregulation and severe lung inflammation in anti-MDA5+ DM patients, emphasising the localised nature of the pulmonary response. Peripheral blood serves more as an indirect indicator than a primary disease driver. These results emphasise the importance of focusing on BALF analyses to better understand the mechanisms of anti-MDA5+ ILD.

2.3 Pulmonary monocyte–macrophage lineage mediates inflammation and fibrosis

Given the pivotal role of macrophages in the cytokine storm, we conducted a more detailed analysis of monocyte–macrophages (Mφ) lineages in BALF, identifying nine distinct populations with unique gene expression profiles (Figures 3A,B and S3A). The Monocyte_S100A8 (Mono_S100A8) population showed high levels of CD14, alarmins (S100A8, S100A9, S100A12), IFN-stimulated genes (ISGs) (ISG20, ISG15, IFITM1) and proinflammatory cytokines (CXCL10, CXCL11, CCL2, CCL7), indicating a robust inflammatory response. The neighbouring Mono_Mφ population exhibited a less distinct phenotype, indicating a transitional state of differentiation. Furthermore, we identified three subsets of monocyte-derived alveolar macrophages (Mo-AM): Mφ_LGMN, Mφ_LGALS3 and Mφ_C1QC, primarily expressing CD68 and ISGs (IFI6, IFI20). Additionally, four types of alveolar macrophages (AMφ), AMφ_CCL4, AMφ_NEAT1, AMφ_FBP1 and AMφ_CSTA, were characterised, notably expressing FABP4 and MARCO. In patients, Mono_S100A8, Mono-Mφ and Mo-AMs were predominant, whereas, in HCs, the alveolar compartment was mainly occupied by AMφ (Figures 3C and S3B), highlighting a shift in macrophage populations in the disease state. The discrepancy of the DM5 sample in cell proportion compared with the other four DM BALF samples may be attributed to this sample capturing only a limited number of high-quality cells.

In order to obtain the differentiation trajectory of circulating and BALF monocyte–macrophages, we used Slingshot for pseudo-time analysis. This revealed that peripheral monocytes (CD14+_HLA^low monocytes, which will be detailed discussed in the next section) underwent a differentiation trajectory toward Mono_S100A8, Mφ_LGALS3, Mφ_LGMN and Mφ_C1QC in the BALF, aligning with the theory of peripheral monocytes recruitment to inflammatory tissues (Figures 3D and S3C). Differential gene expression analysis along the monocyte–macrophage trajectory revealed four key stages (Figure 3E). The initial stage was marked by heightened expression of calprotectin (S100A4, S100A8, S100A9, S100A12) and oxidative stress marker (MT2A), indicating an early inflammatory response. This was followed by upregulation of proinflammatory markers (CCL2, CCL8, CXCL10, CXCL11, CD300E, SELL), antimicrobial genes (DEFB1, LYZ) and ISGs (IFIT1, IFITM1, ISG15, ISG20), suggesting a hyperinflammatory state. In the third stage, an increase in cathepsins (CTSD, CTSL, CTSB) and stress-response marker (HSPB1) was observed. Finally, in the later stage, macrophages expressed genes related to fibrosis (CCL18, LGMN), lipid metabolism (APOE, APOC1, TREM2, FABP5, PLD3), antigen processing and presentation (CD74, LGMN, TREM2) and complement activation (C1QA, C1QB, C1QC, CD74), signalling a shift towards tissue remodelling and immune modulation. To decipher the specific roles of each monocyte and macrophage cluster in severe lung inflammation, we calculated inflammatory and cytokine scores of each group (Figure 3F). Our analysis revealed that the Mono_S100A8, Mono_Mφ and Mo-AM clusters exhibited higher scores than the AMφ, suggesting that these recruited monocytes may be the primary contributors to the cytokine storm in the lungs.

Beyond the acute inflammatory response, anti-MDA5+ DM patients with ILD also suffered from accompanied pulmonary fibrosis, which is linked to worse prognoses.²² Assessing fibrosis-related gene (TGFB1, TGFBI, LGMN and CCL18) expression of each cell cluster, we found elevated levels in Mo-AMs (Mφ_LGMN, Mφ_LGALS3 and Mφ_C1QC) (Figure 3G), pointing to their potential role in fibrogenesis. To further delineate functional polarisation, we assessed the M1 and M2 phenotypic scores of the five clusters. Mono_S100A8 and Mono-Mφ populations predominantly exhibited an M1 phenotype, while Mo-AM clusters showed higher expression of M2 markers (Figure S3D). Notably, macrophages in fibrotic niches of pulmonary fibrosis exhibit shared transcriptional programs.^23-25 To further investigate, we compared the transcriptional profiles of these Mo-AM clusters with publicly available datasets of idiopathic pulmonary fibrosis (IPF). Using gene set expression-based cell scores, we identified significant differences across populations (Figures 3H and S3E). This analysis indicated that Mφ_LGALS3, Mφ_LGMN and Mφ_C1QC populations closely resembled IPF-specific macrophage phenotypes, whereas Mono_S100A8 and AMφ subsets were more similar to homeostatic monocytes and alveolar macrophages, respectively. This continuum of gene expression, transitioning from proinflammatory to profibrotic gene expression along the monocyte–macrophage trajectory, highlights the multifaceted role of these cells in disease progression and tissue repair.

To further assess the functional status of these cells, we analysed their enriched genes using the Gene Ontology Biological Process (GO-BP) database (Figure S3F). This revealed functional heterogeneity and specialisations: Mono_S100A8 and Mono-Mφ clusters were associated with cytokine production and antiviral responses, while Mφ_LGMN and Mφ_LGLAS3 were enriched in energy metabolism and protein/lipid regulation pathways, respectively. Mφ_C1QC was involved in antigen presentation and lipid metabolism.

Overall, these data highlight a notable accumulation of monocytes and Mo-AMs in the lungs of anti-MDA5+ DM patients with ILD. These cells appear central to innate immune activation, driving both inflammation and tissue fibrosis progression.

2.4 Peripheral expansion of myeloid-derived suppressor cell-like monocytes

We then delved into the molecular features of peripheral monocytes in anti-MDA5+ DM patients. Our transcriptome analysis identified a distinct expression profile with 747 genes upregulated and 950 genes downregulated compared with controls. GO-BP pathway analysis of the upregulated differentially expressed genes (DEGs) highlighted their involvement in the antiviral mechanism, supporting the hypothesis that viral infections may be a trigger for anti-MDA5+ DM pathogenesis.^{26, 27} In addition, these monocytes showed enrichment in type I IFN and IFN-γ signalling pathways, as well as energy metabolism pathways (Figure 4A). In contrast, the downregulated DEGs were associated with response to lipopolysaccharide and lymphocyte activation, implying possible functional impairments in these cells (Figure 4A). Further analysis through re-clustering led to the identification of five distinct monocyte subtypes: CD14+_HLA^low monocyte, CD14+_CLDN11+ monocyte, CD14+_JUN+ monocyte, CD14+_HLA+ monocyte and CD16+ non-classical monocyte (ncMon) (Figures 4B,D and S4A). The t-SNE analysis showed perturbed transcriptome features in monocytes from anti-MDA5+ DM patients compared with controls (Figure 4C). Specifically, monocytes from patients expressed higher levels of PLAC8 and MPO, markers associated with immature states, alongside anti-inflammation markers such as CD163, SELL and PLBD1²⁸ (Figure S4C).

The composition of monocyte clusters in anti-MDA5+ DM patients differed significantly from that of HCs (Figures 4C and S4B). A distinctive lack of ncMon (Figure S4B) was noted in anti-MDA5+ DM patients, which was previously linked to inflammation resolution.²⁹ Flow cytometry results further confirmed a reduced presence of CD16+ ncMon in these patients (Figure S4D), consistent with the earlier study.³⁰ Intriguingly, the percentage of CD16+ ncMon in monocytes was inversely correlated with IL-6 levels, reflecting the disease activity and inflammation severity (Figure S4D).

The CD14+_HLA^low cluster, which formed the largest proportion of monocytes in anti-MDA5+ DM patients, exhibited a unique profile with overexpression of canonical ISGs such as IFI27, IFI6, IFITM1 and IFITM3, while showing reduced expression of cytokine/chemokine genes like IL1β and CXCL8 (Figure 4D). Additionally, it was characterised by lower expression of MHC II molecules (HLA-DPA1, HLA-DPB1 and HLA-DR), a hallmark of myeloid-derived suppressor cells (MDSCs) (Figure 4D). MDSCs are a heterogeneous group of immature myeloid cells that expand in response to inflammatory conditions, known for their immunosuppressive properties and capability to suppress T cell responses.³¹ The distinctive features of CD14+_HLA^low monocytes, including lower composite scores for MHC II molecules, higher scores for calprotectin (Figures 4E and S4E), and functional annotations revealed by GO enrichment analysis (Figure 4A), closely resembled those of MDSCs. This cluster also had the highest MDSC-like score among all five clusters (Figure 4F).

Notably, the five patients included in our study had received immunosuppressive therapy for a relatively long duration, raising the question of whether the emergence of MDSC-like monocytes is a consequence of the treatment or an intrinsic feature of anti-MDA5+ DM-ILD. To address this, we integrated scRNA-seq data of PBMCs from five treatment-naïve anti-MDA5+ DM-ILD patients with our own PBMC data and publicly available HCs. The detailed clinical data of these patients can be found in Table S3. Using our established clustering approach, we re-analysed the monocyte populations (Figure S4F). The t-SNE projection showed a substantial overlap in monocyte distributions between treated and untreated patients, suggesting shared features (Figure S4G). Moreover, four out of the five untreated patients exhibited nearly 100% CD14+_HLA^low monocytes, a proportion significantly higher than that observed in HCs (Figures 4G and S4H). These findings further confirm a fundamental shift in the functional state and subtype composition of peripheral monocytes in patients with anti-MDA5+ DM-ILD.

Flow cytometry further confirmed an increased percentage of HLA-DR^low cells in CD14+ monocytes from anti-MDA5+ DM patients (Figure 4H,I), correlating positively with serum ferritin and IL-6 levels (Figure 4J). The relations between the percentage of HLA-DR^low monocytes and other inflammatory indexes are shown in Figure S4I.

Given the elevated levels of monocyte-attracting chemokines (e.g., CCL2) in BALF (Figure 2C) and upregulation of CCR1 and CCR2 in circulating monocytes (Figure 2G), we then analysed CCR1 and CCR2 expression across monocyte clusters to identify the cluster most likely to migrate to the lung. Among these, the CD14+_HLA^low cluster exhibited the highest expression of CCR1 and CCR2 (Figure 4K). Moreover, the Mono_S100A8 cluster, which represents recruited monocytes in the lung, closely resembles CD14+_HLA^low monocytes, characterised by low MHC II expression and elevated levels of calprotectin (S100A8, S100A9, S100A12) (Figure 3B). Taken together, these findings support the hypothesis that MDSC-like monocytes are actively recruited to the lung, where they further differentiate into Mo-AMs.

Overall, these findings indicate that the circulating monocytes in anti-MDA5+ DM patients exhibit significant functional alterations, characterised by heightened ISG responses, reduced expression of cytokines and diminished MHC II molecule levels, suggesting a shift toward an immunosuppressive phenotype. Furthermore, these monocytes are actively recruited to the lungs, where they differentiate into Mo-AMs, potentially contributing to local immune dysregulation and the pathogenesis of anti-MDA5+ DM-ILD.

2.5 Dysregulated cell death and active migration to the lung of peripheral NK and T cells

We investigated the dynamics of NK and T lymphocytes, observing reduced absolute cell counts in anti-MDA5+ DM patients (Figure 1I).^{9, 19} Our detailed clustering of those cells identified 13 subsets in PBMCs, revealing significant heterogeneity (Figures 5A,B and S5A). This diversity is evident in the varied ratios of each cell cluster (Figure S5B,C).

Further exploration into the T and NK lymphopenia (Figure 1I), involved comparing the functional phenotypes of these subsets, focusing on enriched pathways in upregulated genes. In addition to IFN signalling and immune response pathways, we also detected a marked increase in cell death pathways, including apoptosis, pyroptosis and necrosis (Figure 5C). Consistently, the percentage of late apoptotic NK cells was higher in anti-MDA5+ DM patients of both active and remission states than in HCs, with the percentage of late apoptotic T cells being a little higher in patients though not reaching statistical significance (Figure 5D). This finding suggests an elevated incidence of cell death among NK and T cells in anti-MDA5+ DM.

Utilising T-cell receptor sequencing (TCR-seq), we assessed the clonal expansion and clonotype sharing between PBMC and BALF samples in patients. Compared with the periphery, pulmonary T cells showed more pronounced clonal expansion and shared clonotypes with their paired peripheral blood counterparts (Figure 5E). This, combined with higher chemokine levels in BALF, suggests a likely recruitment of T cells from the peripheral to the lungs.

On the whole, these findings indicate that the reduction in circulating NK and T cells in anti-MDA5+ DM patients may be attributed to both augmented cell death and active migration to the lungs.

2.6 B cell dynamics and abnormal autoantibody response in BALF

Anti-MDA5+ DM is characterised by the dysregulated autoantibody response, with anti-MDA5 antibody titres linked to poor prognosis.³² To better understand the role of abnormal B cell response in this process, we subclustered them into five subsets according to canonical B cell markers: B naïve cells, B naïve_IFN_stimulated (B naïve_IFN_stim) cells, B_memory cells, PCs and PC_RRM2 (Figures 6A,B and S6A). In BALF samples of DM4 and DM5, we could hardly capture any B cells, so we only present the results for DM1-3. Notably, plasma B cells were found to be more abundant in BALF than in PBMCs (Figures 6C and S6B), suggesting their potential role in autoantibody production. The functional analysis of upregulated genes in BALF B cells, as opposed to those in PBMCs, showed a particular enrichment in protein complex assembly and protein-transport-related pathways, likely reflecting the heightened synthesis of immunoglobulins in these cells (Figure 6D).

Furthermore, we assessed the B cell response by analysing BCR repertoires (Figure 6E). This analysis revealed a heterogeneous distribution of BCR clonal sizes among patients. For instance, one patient showed no clonal expansion in either peripheral or BALF B cells, while another exhibited significant clonal expansion in both compartments. Additionally, a third patient demonstrated increased clonal expansions in BALF B cells. This variability indicates that the humoral immune response of B cells is not spatially confined and can occur both in the lung and periphery. However, these findings, derived from a limited number of patients, necessitate further investigation to fully elucidate the dynamics of BCR repertoires in anti-MDA5+ DM.

4.1 Study design

This study aimed to gain insights into the disrupted immune landscape spanning both the peripheral and pulmonary domains in individuals afflicted with severe anti-MDA5+ DM and identify potential therapeutic targets in anti-MDA5+ DM. We employed single-cell RNA sequencing to examine cellular constituents within patients’ BALF and paired PBMCs. Luminex assay and flow cytometry were used to further validate the findings. Peripheral blood and BALF samples were collected from patients meeting the 2017 EULAR/ACR Classification Criteria for Idiopathic Inflammatory Myopathies.⁶⁴ All subjects provided written informed consent.

4.2 Sample preparation

PBMCs were isolated from blood using density gradient centrifugation and subsequently resuspended in phosphate-buffered saline (PBS).

The collection of BALF is performed by a qualified and experienced physician with specialised training in an intensive care unit (ICU). The operation follows the clinical practice guidelines of the American Thoracic Society from 2012.⁶⁵ The most affected lung segment, as identified on HRCT, is selected for the lavage, which is performed under local anaesthesia. Normal saline at room temperature (100–300 mL, divided into 3–5 aliquots) was instilled through the bronchoscope. A minimum of 5% of the instilled volume was retrieved. Fresh BALF was filtered through a 100-µm nylon cell strainer to remove aggregates and debris, then centrifuged. The resulting cell pellets were resuspended in chilled RPMI 1640 complete medium. The supernatant was collected for cytokine detection. Both BALF and PBMC samples were analysed fresh.

4.3 MDSC and ncMono Flow cytometry

PBMC samples for flow cytometry were collected during routine follow-up visits (every 3–6 months), with most patients in remission and a small subset experiencing relapse. For cell-surface labelling, 1 × 10⁶ cells were blocked with Fc-block reagent (BD Biosciences) and True-Stain Monocyte Blocker (BioLegend). The following antibodies were then added and incubated for 30 min: CD14-PE (BioLegend; 63D3), HLA-DR-APC-Cy7 (BioLegend; L243), CD3-APC (BioLegend; HIT3a), CD8-APC (BioLegend; RPA-T8), CD19-APC (BioLegend; HIB19), CD56-APC (BioLegend; HCD56) and CD16-FITC (BioLegend; 3G8). To distinguish live from dead cells, the cells were incubated with 7-AAD (Biogems; 61410-00). After incubation, the samples were washed and resuspended in PBS for flow cytometric analysis using a BD FACSAria II. CD14 and HLA-DR were used to gate MDSCs, while CD3, CD8, CD19, CD56, CD14 and CD16 were used to gate ncMonos.⁶⁶

4.4 Cell death detection by flow cytometry

Cell death was assessed using the Annexin V APC Apoptosis Detection Kit (Biogems; 62700) according to the manufacturer's instructions. Briefly, cells were incubated with Annexin V-APC and 7-AAD in 1× Annexin V binding buffer for 15 min at room temperature. After staining, 400 µL of 1× binding buffer was added, and the samples were analysed on a BD FACSAria II.

4.5 Cytokine detection

Cytokines in the serum or BALF supernatant were detected using the multiplexed Luminex xMAP assay (eBioscience) following the manufacturer's instruction. To study inflammation in BALF and serum, we selected cytokines based on their established roles as key drivers of inflammation, ensuring a focused analysis of immune and inflammatory dynamics.

4.6 Single-cell RNA sequencing

Single-cell RNA-seq libraries were prepared using the Chromium Single Cell 3ʹ Reagent Kits v3 (10× Genomics), following the manufacturer's protocol. Gene expression libraries were sequenced on a NovaSeq platform (Illumina) to generate 150 bp paired-end reads.

4.7 Processing of scRNA-seq data

The datasets of HCs are from GSE158055 (HC003, HC005, HC008 and HC010 were chosen as controls for PBMCs) and GSE14592 (HC1, HC2 and HC3, were chosen as controls for BALF). All single-cell transcriptome sequencing data were aligned and quantified using Cell Ranger (v6.0.2) with the GRCh38 human reference genome obtained from the 10× Genomics official website. The filtered feature matrixes were then imported into Seurat (V 4.1.2) for quality control. All functions were executed with default parameters unless specified otherwise. We retained cells with between 300 and 5000 expressed genes and a mitochondrial percentage of less than 10%. Principal component analysis (PCA) was performed using the 2000 most variable features identified by the FindVariableFeatures. The PCA matrix was then imported into Harmony (V0.1.1, R) to integrate the single-cell gene expression data and correct for batch effects.

4.8 Cell clustering and annotation

UMAP was performed on the top 20 principal components to visualise the cells. The first round of unsupervised clustering (resolution = 0.5) was used to identify the main cell types, including the NK cells, CD4 T cells, CD8 T cells, B cells, PCs, monocytes, DCs, platelets, macrophages and epithelial cells. To further define subclusters within each major cell type, a second round of clustering was performed separately on T cells, NK cells, B cells, monocytes and macrophages (resolution = 1).

4.9 Differential gene expression and pathways enrichment analyses

FindMarkers function (Seurat) was used to identify DEGs (|Log2FoldChange | > 1, adjusted p < .05). To investigate the functions of these DEGs, we performed GO pathway enrichment analysis using ClusterProfiler (v4.6.0) in R. Enrichment analysis of the B cell part was conducted using the Metascape web tool (www.metascape.org).

4.10 Inflammatory score, cytokine score, MHC class II score, calprotectin protein score, MDSC-like score, M1 score and M2 score

Scores were calculated using the AddModuleScore function in the Seurat package. Inflammatory score, cytokine score,²¹ MHC class II score, calprotectin protein score and MDSC-like score⁶¹ were calculated according to published literature. The reference genes used for calculating these scores are provided in Table S4.

4.11 Pseudo-time trajectory analysis

Trajectory analysis for monocyte–macrophages was performed with slingshot.⁶⁷ To integrate monocyte–macrophage data from PBMCs and BALF, cell trajectories were inferred with peripheral monocytes as the starting point.

4.12 Calculation of gene expression program scores

To assess the transcriptional similarity between macrophages in our dataset and those from previously published pulmonary fibrosis studies, we used gene set module scoring. Specifically, gene expression program scores were calculated using the AddModuleScore function in Seurat (seed = 1993). Reference gene sets were derived from three published pulmonary fibrosis datasets.^{23, 25, 68} For each reference dataset, the top 50 marker genes were used to define the module. In cases where fewer than 50 genes were available, all genes from the respective reference were included. This approach allowed for the quantification of transcriptional profiles indicative of profibrotic macrophage phenotypes, as described in Ref. ⁶⁹.

4.13 BCR and TCR repertoire analysis

On the basis of TCR and BCR CDR3 amino acid sequences, we compared the clone size distribution and clonotype in the periphery and BALF of the same patient.

4.14 Histological analysis

The samples of lung tissue were fixed in 4% formaldehyde and embedded in paraffin (FFPE). FFPE blocks were sectioned into 4 µm slices, deparaffinised, rehydrated and stained with haematoxylin–eosin and Masson's trichrome following standard protocols.

4.15 Statistics

Quantitative data were expressed as the mean ± standard error of the mean. Statistical analysis was performed by Prism 9.0 (GraphPad). For comparisons between two groups, independent-sample t-tests or paired t-tests were used for normally distributed variables and Mann–Whitney tests or Wilcoxon rank-sum tests were applied for non-normally distributed variables. For comparisons of more than two groups, the Kruskal–Wallis test was used for data not following a normal distribution. Spearman's correlation was applied for correlation analyses. All statistical tests were two-tailed, with a p value < .05 considered significant.