Single-cell and spatial transcriptome landscapes reveal the spatial immune defense pattern shared by primary and brain metastatic lung adenocarcinoma

Introduction

Lung cancer remains one of the leading causes of cancer-related deaths worldwide, especially lung adenocarcinoma (LUAD) (1). Patients diagnosed with advanced LUAD have a high incidence of brain metastasis (BM) (2). Despite advancements in early diagnosis and treatment, a significant proportion of patients develop brain metastases, which are associated with poor prognosis and a substantial increase in mortality rates (3). The challenge of managing lung cancer brain metastases is compounded by the limited therapeutic options and the complex biological barriers posed by the brain’s unique immune environment (4-6).

Across tumor evolution, LUAD cells progressively acquire immune-evasion capabilities that contribute to therapeutic failure with cytotoxic modalities (7-9). Notably, immune-checkpoint inhibitors (ICIs) can reinvigorate activated T cells that traffic into brain metastases and achieve intracranial responses in selected patients, indicating that effective antitumor immunity is possible behind the blood-brain barrier (BBB) (6,10-12). However, clinical benefit remains variable, underscoring the need to understand how tumor-immune interactions are organized in both primary and intracranial sites (11,13-15).

Most prior studies have emphasized functional or microenvironmental differences between primary LUAD and brain metastases (16-19). While organ-specific immune-evasion mechanisms likely exist in lung versus brain, a key unresolved question is whether shared escape programs also operate across sites and shape metastatic competence. To address this gap, we integrated public single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics to identify spatially organized immune-defense patterns common to primary and brain-metastatic LUAD. Using non-negative matrix factorization (NMF) and signal-distribution landscape analyses, we delineated a myeloid-associated metaprogram (MP1) and a shared tumor immune defense pattern (TIDP) at tumor borders across sites, and we nominate molecules that may help stratify patients and guide immunotherapeutic strategies. We present this article in accordance with the ARRIVE reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1224/rc).

Methods

Data collection

The scRNA-seq data for 11 primary lung cancer samples and 10 BM samples were retrieved from the Gene Expression Omnibus (GEO) database under the accession number GSE131907 (20). Primary LUAD single-cell datasets were derived from patients without clinically/radiologically documented BM at the time of surgical resection (BM-negative primaries; GSE131907). Brain-metastasis samples were obtained from intracranial lesions (BM-positive; GSE131907). For public datasets lacking uniform longitudinal follow-up, BM status refers to the condition at sampling. A per-sample summary (site, histology, stage, BM status) is provided in Table S1. Additionally, the LUAD of The Cancer Genome Atlas (TCGA) cohort, OMIX575 cohort, and two GEO datasets (GSE93157 and GSE135222) were used as external validation cohorts (16,21,22). These datasets were utilized for deconvolution analysis, meta-program correlation evaluation, and the assessment of key molecules in predicting the efficacy of immunotherapy. Spatial transcriptome slices were also obtained from the GEO database, with the accession number GSE189487 and GSE179572 (10,23). Two samples from GSE189487 were included in the study: T1, representing invasive adenocarcinoma (IAC), and T3, representing microinvasive adenocarcinoma (MIA). Two lung cancer BM samples from GSE179572 were included in the study: patient 19 and patient 27. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Data processing of single-cell sequencing and spatial transcriptome data

scRNA-seq data were analyzed using the R package Seurat (v.4.1.0). After quality control, 65,203 single cells were retained for analysis. Data were standardized using the “LogNormalize” method, and highly variable genes were identified using the “FindVariableFeatures” function. Principal component analysis was performed on these variable genes. Clustering was conducted with the “FindNeighbors” and “FindClusters” functions, followed by t-distributed Stochastic Neighbor Embedding (t-SNE) projection into low-dimensional space. Differentially expressed genes were identified with the “FindAllMarkers” function, using the criteria |log(fold change)| >0.5 and P.adj <0.05. Marker genes for subgroup annotation were obtained from the literature. Cell-type proportion analysis. For each sample, the fraction of each annotated subpopulation was calculated relative to total cells. Primary (BM-negative) and brain-metastatic (BM-positive) groups were compared using two-sided Wilcoxon rank-sum tests.

The R package Seurat (v.4.1.0) was used to integrate data from four spatial transcriptome slices (Primary-1, Primary-3, BM-19, BM-27). Data normalization was performed with the “SCTransform” function, and sample integration was done using the “Merge” function, resulting in 8,333 spots. The “ScaleData” function was applied for scaling. Dimensionality reduction and clustering were performed using “RunPCA”, “FindNeighbors”, “FindClusters”, and “RunUMAP”. Gene expression patterns were visualized with the “SpatialFeaturePlot” function.

Expression programs of intratumoral heterogeneity

We employed NMF to identify distinct expression programs within the scRNA-seq data. The NMF method was applied to decompose the gene expression matrix into a set of interpretable components, representing distinct biological processes or gene expression programs (24,25).

The workflow began by pre-processing the scRNA-seq data, which included quality control, normalization, and filtering of low-quality cells and genes with low expression. We then performed dimensionality reduction on the normalized data to reduce the complexity of the dataset. Next, NMF was applied to the reduced expression matrix using the NMF package in R. The number of components (k) to be extracted was determined based on cross-validation using a sparseness criterion, which ensures that the components are biologically interpretable. Five major meta-programs were ultimately obtained by using 15,384 tumor cells from 21 samples.

Gene set enrichments analysis

Corresponding Gene Set Variation Analysis (GSVA) analysis was performed on cell subtypes using relevant functional gene sets to obtain relevant functional scores (26). This part is done through the gsva function in the GSVA R package, with parameters set to default values. All P values are false discovery rate (FDR) corrected.

Deconvolution of immune cell populations in bulk RNA-seq data

To explore the immune microenvironment in LUAD, we utilized deconvolution algorithms, specifically TIMER and CIBERSORT (ABS) (27,28). TIMER estimates the abundance of immune cell populations in bulk RNA-seq data by using gene expression signatures of immune cells, providing insights into the relative proportions of immune infiltrates such as CD8⁺ T cells, macrophages, and dendritic cells. We also applied CIBERSORT in its Absolute mode, which calculates both relative and absolute immune cell proportions from bulk RNA-seq data (368 TCGA LUADs) using a reference gene signature matrix (LM22) that includes 22 immune cell types.

Tumor boundary recognition

We employed the Cottrazm algorithm to evaluate the tumor boundaries of primary and metastatic brain samples from LUAD (29). By integrating spatial transcriptomics and histological images, a detailed spatial profile of the tumor microenvironment was constructed. The core area of the tumor is identified and the boundary area is delineated by analyzing copy number variations (copy number variations) and gene expression characteristics. Default parameters are used during the operation process.

Cell-cell communication

We utilized CellPhoneDB v5 to assess intercellular interactions among single-cell subpopulations in our study (30,31). We processed the scRNA-seq data to obtain normalized gene expression matrices and corresponding cell type annotations. Using the cellphonedb Python package, we executed the interaction analysis by specifying the input gene expression matrix and cell type metadata. The results were visualized using the igraph R package.

Spatial integration analysis

We utilized the stLearn algorithm to conduct unsupervised clustering of spatial transcriptomic data from LUAD and BM samples (32). By combining histological morphology and gene expression profiles, we reduced technical noise and improved clustering accuracy, thereby achieving a more comprehensive understanding of tissue and cellular diversity in the samples. Furthermore, we adopted the Pseudo-Time-Space (PSTS) method to reconstruct the spatiotemporal trajectory of tumor evolution and simulate the spatiotemporal dynamics of tumor cells.

Animal BM model

All animal experiments were performed under a project license (No. IACUC-2304023) granted by Institutional Review Board of Jiangsu Cancer Hospital, in compliance with the national guidelines for the care and use of animals. A protocol was prepared before the study without registration. We transfected PC9-luc cells, which stably express luciferase, with siRNA targeting CAMP (LL37) [si-CAMP (LL37)] or a control empty vector. Transfection was performed using a standard lipid-mediated transfection protocol. The knockdown efficiency of CAMP (LL37) in the transfected cells was confirmed by western blotting, ensuring that CAMP (LL37) expression was effectively reduced in the experimental group. Subsequently, six nude mice aged 6–8 weeks were used to establish a BM model, with three mice in the control group and three mice in the experimental group. This model preserves myeloid compartments (e.g., macrophages/microglia) and allows assessment of CAMP (LL37)-related effects on colonization, but it does not permit evaluation of T-cell-dependent mechanisms. The transfected PC9-luc cells were injected into the left ventricle of the heart via intracardiac injection, allowing the cells to circulate through the bloodstream and metastasize to distant organs, including the brain (33,34). Two weeks post-injection, tumor growth and metastasis were monitored non-invasively using bioluminescence imaging, which detects the luciferase activity expressed by the injected cells. The imaging provided real-time, quantitative assessment of tumor formation and progression. The intracardiac injection model was chosen as it mimics the natural hematogenous spread of lung cancer to the brain, allowing for the assessment of metastatic tumor formation and the impact of LL37 gene modulation on the metastatic process. Bioluminescence imaging provided real-time, quantitative data on tumor growth, while the western blot analysis validated the efficiency of LL37 gene editing, ensuring the success of our experimental approach.

Statistical analysis

Statistical analysis was performed using R (v4.1.0). Chi-squared or Fisher’s exact tests were used to compare categorical variables. All P values were two-sided, with a significance threshold of <0.05. Figures were generated using the R packages ggplot2 and RColorBrewer. Flowcharts and mechanism diagrams were created using BioRender (http://biorender.com/).

Results

High-resolution single cell landscape of primary LUAD and BM

To investigate the potential mechanisms of immune escape and intracranial colonization in mid- to late-stage LUAD, we conducted a thorough analysis of the single-cell transcriptome profiles from 11 primary lung cancer and 10 BM samples (20). The primary LUAD samples analyzed represent BM-negative cases at resection and serve as the comparator group for cross-site analyses versus BM-positive intracranial lesions. After performing quality control on the single-cell data, we constructed cell profiles for both primary lung cancer and BM, comprising a total of 65,203 single cells. Using UMAP clustering and marker genes specific to each cell type, we classified the cells into nine major types: tumor cells (EPCAM, KRT18, KRT19), endothelial cells (CLDN5, RAMP2), oligodendrocytes (ALCAM, CLDN11), mast cells (KIT, MS4A2, GATA2), monocytes (RETN, MCEMP1, FBP1, MEMP1), macrophages (MARCO, FCGR3A, PECAM1), T lymphocytes (CD3D, CD3E, CD3G), B lymphocytes (CD79A, MZB1), and fibroblasts (COL1A1, DCN, COL1A2, THY1) (Table S2). We then performed a subpopulation analysis of the tumor immune microenvironment (TIME) cells, myeloid cells, T- and B-lymphocytes, and fibroblasts, dividing the entire dataset into 34 subpopulations (Figure 1A, Figure S1).

Figure 1 The composition map of single-cell subsets in the primary lesion of LUAD and brain metastases. (A) Single-cell landscape of primary tumors and brain metastases LUAD (the asterisk of label indicates that there are differences in the proportions of each sample subgroup between the primary lesion group and the brain metastasis group). (B) Bar chart of tumor cells and various components of the tumor microenvironment in each sample. (C) The UMAP landscape of LUAD, colored by the primary lesion and brain metastases. BM, brain metastasis; CAF, cancer-associated fibroblast; LUAD, lung adenocarcinoma; NKT, natural killer T; Treg, regulatory T cell; UMAP, Uniform Manifold Approximation and Projection.

In the subpopulation composition of both LUAD primary and BM samples, we observed that, aside from tissue-specific components such as oligodendrocytes, differences in subpopulation distribution between primary LUAD and BM were primarily associated with tumor cell occupancy within cancer nests and specific immune microenvironment components (Figure 1A,1B, Figure S2). Eleven of the 24 subpopulations, including mast cells, exhibited differences in occupancy between primary and brain metastatic sites (Figure 1A, Figure S2). Despite these proportional differences, immune-related components were consistently detectable across different tissue sources (Figure 1C). The presence of shared immune subpopulations suggests that tumor cells in both primary lung cancer and brain metastases may engage with a common set of immune cell types, although the magnitude and functional consequences of such interactions are likely influenced by their relative abundance. This prompted us to further investigate whether tumor cells themselves exhibit conserved immune-related programs under immune pressure.

Meta-programs within the LUAD malignant compartment

To investigate the common characteristics of tumor cells in different tissues under immune stress, we first evaluated the cell state diversity of LUAD tumor cells. Consistent with previous studies, 20,168 tumor cells displayed distinct clustering patterns of inter-tumor heterogeneity based on their tissue origin (Figure 2A,2B) (35). Applying NMF to each tumor cell, we identified five distinct metaprograms (MP1–5), which reflect a shared pattern of intra-tumor heterogeneity across different tumors (Figure 2C). Notably, MP3 was predominantly associated with epithelial tumor features (epithelial cytokeratin), MP4 exhibited fibroblastic characteristics, and MP5 highlighted neural stemness differentiation and glucose metabolism signals. In contrast, MP1 and MP2 were linked to myeloid-related features, immune infiltration, and cell proliferation signals in tumor cells (Table S3). Tumor cells were scored based on the marker molecules of MP1-5 and subsequently annotated according to their scores (Figure 2D). Except for MP5, the distribution of tumor cells with different metaprograms did not show significant differences between primary tumors and brain metastases (Figure 2C right, Figure 2E). These findings suggest that immune-related features in MP1–2 and epithelial/fibroblastic phenotypes associated with MP3–4 are shared between primary and metastatic brain lesions.

Figure 2 Transcriptome heterogeneity in tumor cells of the primary lesion of LUAD and brain metastases. (A) The UMAP landscape of tumor cells in LUAD, colored by the primary lesion and brain metastases. (B) The UMAP landscape of unsupervised clustering of tumor cells in LUAD. (C) Hierarchical clustering of pairwise similarities between NMF programs identified across tumor cells from all the analyzed NSCLC patients. The left panel shows tumor MP identified from consensus NMF programs. Similarities between NMF programs were quantified by Jaccard index over the programs’ signature genes. (D) UMAP plots of tumor cells colored by the signature scores of each tumor MP. (E) Bar chart of different MP ratio in tumor cells of each sample. *, P<0.05. BM, brain metastasis; LUAD, lung adenocarcinoma; MP, metaprogram; NMF, non-negative matrix factorization; NSCLC, non-small cell lung cancer; UMAP, Uniform Manifold Approximation and Projection.

MP1 exhibit distinct correlations with TIME

We further analyzed tumor cells annotated by metaprograms (MPs). MP1 and MP2 exhibited higher expression of immune-related signals, while MP3 and MP4 were enriched in epithelial and fibroblastic markers. Differentially expressed genes within each subpopulation reflected the cellular functional characteristics of MP1–5 (Figure 3A). We then evaluated the correlation between the proportion of MP1–5 in single cells and the components of the tumor microenvironment. MP1 showed a significant positive correlation with T cell, macrophage, and monocyte subpopulations, except for MP4 (Figure 3B). This finding was validated in TCGA LUAD samples, where MP1 was strongly correlated with M2-type macrophages, an immunosuppressive component, as assessed by both the TIMER and CIBERSORT (ABS) deconvolution algorithms (Figure 3C) (27,28). Notably, MP1 expressed a pan-cancer immunity signature compared to the normal tumor cell signature of MP3 and exhibited concurrent immunosuppressive signaling in contrast to the immunocidal response of MP2. These results suggest that MP1 may represent a unique, previously unreported myeloid immunity-related malignant epithelial state.

Figure 3 The correlation between MP1–5 and microenvironment components. (A) Bubble plot of characteristic expression molecules in various subsets of tumor cells annotated with the MP scores. (B) The correlation heatmap among the components in LUAD. (C) Correlation analysis between MP1–5 and each immune microenvironment component in the LUAD samples of TCGA [the scores of immune microenvironment components were evaluated by the deconvolution algorithm TIMER and CIBERSORT (ABS)]. *, P<0.05; **, P<0.01. CAF, cancer-associated fibroblast; LUAD, lung adenocarcinoma; MP, metaprogram; NA, not applicable; NKT, natural killer T; TCGA, The Cancer Genome Atlas; Treg, regulatory T cell.

In the primary lesions of LUAD and brain metastases, MP1 exhibits a shared immune defense pattern in the tumor spatial structure

Transcriptional heterogeneity of malignant epithelial cells interacts with TIME formation specificity to determine tumor fate (36). To further explore the spatiotemporal characteristics of MP1 during tumor invasion, we selected four spatial transcriptome sections of lung cancer: two from primary lung cancer (Primary-1, Primary-3) and two from brain metastases (BM-19, BM-27) (10,23). Using the Cottrazm algorithm, we identified tumor regions (Mal), tumor boundaries (Bdy), and paracancerous regions (nMal) in the four samples (Figure 4A,4B, left; Figure S3A,S3B, left) (29). We then mapped the expression of MP1–5 marker molecules in space. Interestingly, MP1, which is associated with immunosuppression, was enriched at the tumor borders in both primary foci and brain metastases, surrounding the cancer nests characterized by MP3 (Figure 4A,4B, right; Figure S3A,S3B, right). Furthermore, MP1 expression in the tumor border regions was significantly higher than in the tumor and paracancerous regions (Figure 4C,4D; Figure S3C,S3D, right; Figure S3E). Consistently, MP1 was also detectable at the border of the early-stage sample Primary-3 (MIA), although the enrichment was attenuated compared with Primary-1 (IAC) (Figure 4C; Figure S3C). Notably, the significance of MP1 signals at the Primary-3 border was slightly less pronounced than at the Primary-1 border, likely due to the different stages of tumor development: Primary-1 was in the IAC stage, whereas Primary-3 was in the MIA stage, with Primary-1 being more aggressive (Figure 4C; Figure S3C). Considering MP1’s association with myeloid-related immunosuppressive functions and its distribution at the tumor margin, we propose that MP1 represents a shared TIDP that evolves during tumorigenesis in response to early immune exposure and killing (Figure 4E). Survival analysis integrating multiple public datasets showed that high MP1 expression is associated with poor prognosis in LUAD (Figure 4F; Figure S3C). Together, these results indicate that the MP1-defined TIDP is stage-spanning, being detectable in early (MIA) lesions and persisting in advanced disease, and is associated with immune escape at invasive margins and with brain colonization.

Figure 4 The spatial signal distribution landscape of MP1 related to immunosuppression. (A) Primary-1 (spatial transcriptomics slide). Left: H&E-stained tissue image with Cottrazm segmentation overlay: malignant (Mal, red), non-malignant/paracancerous (nMal, yellow), and tumor boundary (Bdy, blue). Scale bar: 1 mm. Right: spatial maps of MP1–5 per-spot scores (z-score; warmer colors indicate higher signal). Blue dashed contours highlight MP1-enriched regions; red dashed contours highlight MP3-enriched cancer nests. (B) BM-27 (spatial transcriptomics slide). Left: H&E-stained tissue image with the same Cottrazm overlays (Mal/red, nMal/yellow, Bdy/blue). Scale bar: 1 mm. Right: spatial distribution of MP1–5 scores on the Visium spot grid (z-score; warmer colors denote higher signal). Blue dashed contours mark MP1 enrichment; red dashed contours mark MP3 enrichment. (C,D) The expression differences of MP1 signals in tumor areas, boundary and adjacent areas (Primary-1 and BM-27). (E) Hypothesis diagram: Tumor lesions gradually form an immune defense pattern located around the cancer nest after immune exposure. Created in https://BioRender.com. (F) MP1 signaling is associated with a poor prognosis of LUAD (LUAD samples are derived from TCGA). ***, P<0.001. Bdy, tumor boundaries; CAF, cancer-associated fibroblast; CI, confidence interval; H&E, hematoxylin and eosin; HR, hazard ratio; LUAD, lung adenocarcinoma; Mal, tumor regions; MP, metaprogram; NK, natural killer; nMal, paracancerous regions; TCGA, The Cancer Genome Atlas.

Cell-cell communication molecules related to myeloid cells and T lymphocytes are involved in the tumor molecular biological function of TIDP

Based on the molecular characterization of MP1, TIDP formation may be associated with myeloid cells and T lymphocytes. To explore this, we assessed the frequency of interactions and detailed ligand-receptor signaling between MP1 subpopulations and corresponding immune subpopulations using CellPhoneDB (30,31). The results revealed that MP1 exhibited the highest interaction intensity with the monocyte CD1E⁺CCL17⁺ and macrophage CCL18⁺ subpopulations (Figure 5A). Notably, MP1 had the highest signal output intensity, while the monocyte CD1E⁺CCL17⁺ and macrophage CCL18⁺ subpopulations received the most signals. These two myeloid immune subpopulations played key roles in chemoinflammatory and immunosuppressive activities within the tumor microenvironment, respectively (Figure 5B) (37-39). Additionally, MP1 engaged in a high number of ligand-receptor exchanges with various immune subpopulations, including chemotactic inflammatory interactions (CCR3-CCL26, CCR2-CCL26, IL18R-IL18), inhibitory cell proliferation (EGFR-MIF), and immunomodulatory interactions (FPR2-CAMP) (Figure 5C,5D) (40-47). These ligand-receptor pairs may contribute to the functional properties and early formation of TIDP.

Figure 5 The communication feature between MP1 and immune components. (A) Heatmap of cellular interaction between tumor MP1 components and immune cell subsets. (B) The distribution of ligand-receptor signals of tumor MP1 components and various immune cell subsets. (C) The shell plot of the cellular interaction between the MP1 component of the tumor and the immune cell subsets. (D) Significant ligand-receptor communication signals between tumor MP1 components and immune cell subsets. MP, metaprogram; NKT, natural killer T; Treg, regulatory T cell.

LL37 related to the formation of TIDP, is associated with the efficacy of immunotherapy and intracranial tumor colonization

We used stLearn, which integrates histomorphological and transcriptomic features of the spatial transcriptome samples, to assess the progression trajectory within the tumor (32). The temporal and tumor evolutionary trajectories revealed that tumor boundary regions were homogenized through unsupervised clustering into a distinct subpopulation, located at the terminal stage of the temporal sequence, all of which originated directly from the center of the cancer nests (Figure 6A,6B). We then examined the expression trends of TIDP-associated ligand-receptor molecules with significant communication in the evolutionary trajectories and found that the expression of IL18RAP, IL18BP, and CAMP progressively increased throughout the trajectory (Figure 6C). In the external cohort, IL18RAP and CAMP expression was significantly higher in brain metastases compared to paired primary samples (Figure 6D) (16). Together, these analyses indicate that TIDP/MP1 constitutes a conserved border program across primary and brain-metastatic LUAD, while brain lesions display a context-specific amplification of select nodes. In particular, IL18RAP and CAMP are significantly higher in brain metastases than in matched primaries within MP1-high cells/spots, consistent with enhanced IL18RAP/IL18 and CAMP/FPR2 signalling in the intracranial microenvironment (Figure 6C,6D). Additionally, in both immunotherapy cohorts, IL18RAP and CAMP were better predictors of immunotherapy efficacy (Figure 6E) (21,22). CAMP encodes the human cathelicidin precursor (hCAP18), whose C-terminal peptide LL37 is the active, secreted effector (45). Given the prominent FPR2-CAMP (LL37) interaction within the MP1/TIDP axis, we prioritized CAMP (LL37) for in vivo experiments. We transfected PC9-luc cells, which stably express luciferase, with siRNA targeting CAMP (LL37) [si-CAMP (LL37)] or a control empty vector, and confirmed CAMP (LL37) knockdown efficiency in transfected cells via western blotting (Figure 6F, left). Subsequently, 6–8-week-old nude mice were used to establish a BM model. Transfected PC9-luc cells were injected into the left ventricle by intracardiac injection, allowing the cells to circulate through the bloodstream and metastasize to distant organs, including the brain. Bioluminescence imaging, performed two weeks after injection, revealed that knockdown of CAMP (LL37) significantly reduced the efficiency of tumor colonization in brain tissue (Figure 6F, right). These analyses and experiments suggest that CAMP (LL37) plays a critical role in TIDP formation and is closely associated with the malignant behavior of tumors in the middle and late stages. Because the intracardiac metastasis model was performed in athymic nude mice, these in vivo data primarily interrogate tumor-intrinsic and myeloid-associated facets of the TIDP/MP1 axis; T-cell-mediated effects cannot be inferred from this experiment and are instead supported by our human single-cell/spatial analyses.

Figure 6 The clinical association between MP1 differentiation-related molecules and the efficacy of immunotherapy. (A,B) Temporal spatial distribution and spatial trajectory of pattern progression in Primary-1 and BM-27. (C) Heatmap of the expression of significant ligand-receptor communication molecules of MP1 in the evolutionary trajectories of Primary-1 and BM-27. (D) The expression differences of IL18RAP and CAMP (LL37) between primary lesions and brain metastases (43 pairs of paired samples of primary lesions and brain metastases from OMIX575-20-01). (E) The differences in the efficacy of immunotherapy among lung cancer patients with different expression levels of IL18RAP and CAMP (LL37). (F) Left: western blot was used to analyze the expression of CAMP (LL37) in PC9-luc cells stably expressing luciferase, PC9-luc-si-Control cells, and PC9-luc-si-CAMP (LL37) cells with CAMP (LL37) knockout. Right: representative images of the bioluminescence activity of tumor luciferase (luc) after injection of PC9-luc-si-CAMP (LL37) and PC9-luc-si-control cells stably expressing luciferase. BM, brain metastasis; PD, progressive disease; PR, partial response; Pri, primary.

Discussion

We provide novel insights into the immune defense mechanisms shared by primary and brain metastatic LUAD, highlighting the role of tumor cell transcriptional heterogeneity and immune evasion in cancer progression. By integrating scRNA-seq and spatial transcriptomics, we identified a TIDP, driven particularly by myeloid-associated immunosuppressive features, represented by metaprogram MP1, which was present in both primary and metastatic sites. The application of NMF and signal distribution landscape analysis enabled us to track the temporal and spatial progression of immune responses during tumor invasion, revealing that MP1 plays a key role in shaping the immune microenvironment in both contexts. Moreover, we uncovered critical ligand-receptor interactions, such as IL18RAP, CAMP (LL37), and other chemotactic inflammatory and immunomodulatory molecules, which contribute to immune evasion and tumor progression, particularly at the tumor boundaries.

In line with the growing potential of ICIs in the treatment of brain metastases from LUAD, a substantial body of research now focuses on the mechanisms of immune escape in brain metastatic lung cancer. These studies mainly emphasize the role of tumor cell diversity in evading brain immune surveillance and the dynamic interactions between metastatic cancer cells and brain immunity (11,13-15). Our findings are consistent with the above research conclusions but differ in that we specifically examined the shared TIDP of primary and metastatic LUAD, providing a more comprehensive view of tumor-immune interactions across different tissue sites. However, conservation does not imply identical magnitude; rather, the architecture is shared while node weights are reweighted by the brain milieu (e.g., myeloid skewing, astrocyte-derived cues, BBB-related constraints). This approach is particularly important, given the unique challenges posed by brain metastases, where the immune environment is markedly different from that in other tissues.

We acknowledge extensive prior reports of immunosuppression/immune exclusion at invasive fronts across cancers. Our contribution is to provide a single-cell-derived, spatially validated metaprogram (MP1) that is conserved across primary LUAD and brain metastases, topographically anchored at the border, and operationalized by a reproducible ligand-receptor architecture [e.g., IL18RAP/IL18, CAMP (LL37)/FPR2]. Moreover, MP1/TIDP shows a brain-metastasis–associated configuration with selective intensification in intracranial lesions compared with matched primaries. Thus, while we do not claim novelty for the general concept of border immunosuppression, we delineate a cross-site, spatially organized program and its convergent signalling axes that appear particularly relevant to brain colonization. Prognostic associations are presented as supportive context rather than as a novelty claim.

Our work highlights the role of key molecules such as IL18RAP and CAMP (LL37), which are co-expressed in both primary and brain metastatic tumors and significantly upregulated in brain metastases and the TIDP. These molecules play a crucial role in immune response regulation and may serve as potential biomarkers for predicting the efficacy of immunotherapy in patients with brain metastases from lung cancer. Notably, the upregulation of IL18RAP and CAMP (LL37) in brain metastases underscores their potential as therapeutic targets, offering opportunities to improve treatment strategies for patients who currently have limited treatment options. Additionally, MP1 plays a critical role in immune evasion and tumor progression, suggesting that modulating this pathway could enhance immunotherapy efficacy and potentially inhibit the drug resistance mechanisms in advanced lung cancer.

Furthermore, this study emphasizes the importance of considering the spatiotemporal dynamics of tumor-immune interactions, particularly in metastatic settings. We demonstrate that the tumor border regions in both primary and brain metastatic tumors are enriched in immune-related signals, which play a key role in tumor progression and immune escape. The spatial aspect of immune evasion, combined with the temporal progression of immune responses, provides essential insights into how tumors adapt to immune stress over time. Understanding these dynamics could enable the development of more targeted and effective therapeutic strategies tailored to the unique immune environment in brain metastases, potentially improving clinical outcomes for patients with LUAD.

However, our study is limited by the sample size, and the TIDP proposed here requires validation through preclinical and clinical studies. Moreover, the exact mechanisms underlying the maintenance and transformation of the TIDP state warrant further investigation. The exploratory approach combining high-resolution multi-omics with spatial morphological feature distribution may offer new insights into other heterogeneity-related studies. We observed plastic alterations in the tumor cells across both MP1 and MP3 states, supported by corresponding histologic morphology and MP1/3 scores. Because public datasets lacked uniform longitudinal follow-up, ‘BM-negative’ denotes absence of BM at the time of sampling; prospective cohorts will be required to determine whether MP1/TIDP levels in primary tumors predict subsequent brain dissemination. Future research is needed to extend these findings to tissue samples from different developmental stages and to determine whether the observed heterogeneity arises from the induction of myeloid immune components. Another key limitation is that our in vivo validation employed athymic nude mice, precluding direct assessment of T-cell contributions to TIDP/MP1. Accordingly, the functional data mainly support tumor-intrinsic and myeloid-associated components [e.g., CAMP (LL37)/FPR2 signaling]. T-cell involvement is inferred from human single-cell/spatial datasets and deconvolution analyses, and should be tested in immunocompetent syngeneic or humanized models in future studies.

Conclusions

In this study, we identified a shared TIDP in primary and brain metastatic LUAD, characterized by myeloid-driven immune evasion at tumor borders. Key molecules IL18RAP and CAMP (LL37) were significantly upregulated in brain metastases and may serve as biomarkers for immunotherapy response and potential therapeutic targets.

Acknowledgments

We appreciate the release of the relevant research data. We thank all the colleagues involved in the current research projects in our laboratory for facilitating this study.

Footnote

Reporting Checklist: The authors have completed the ARRIVE reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1224/rc

Data Sharing Statement: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1224/dss

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1224/prf

Funding: The work was supported by the National Science Foundation of China (Nos. 82073235 and 81872378).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-1224/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. All animal experiments were performed under a project license (No. IACUC-2304023) granted by Institutional Review Board of Jiangsu Cancer Hospital, in compliance with the national guidelines for the care and use of animals.

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

References

1. Chen W, Zheng R, Baade PD, et al. Cancer statistics in China, 2015. CA Cancer J Clin 2016;66:115-32. [Crossref] [PubMed]
2. Arrieta O, Salas AA, Cardona AF, et al. Risk of development of brain metastases according to the IASLC/ATS/ERS lung adenocarcinoma classification in locally advanced and metastatic disease. Lung Cancer 2021;155:183-90. [Crossref] [PubMed]
3. Peters S, Bexelius C, Munk V, et al. The impact of brain metastasis on quality of life, resource utilization and survival in patients with non-small-cell lung cancer. Cancer Treat Rev 2016;45:139-62. [Crossref] [PubMed]
4. Kudo Y, Haymaker C, Zhang J, et al. Suppressed immune microenvironment and repertoire in brain metastases from patients with resected non-small-cell lung cancer. Ann Oncol 2019;30:1521-30. [Crossref] [PubMed]
5. Wang Y, Chen R, Wa Y, et al. Tumor Immune Microenvironment and Immunotherapy in Brain Metastasis From Non-Small Cell Lung Cancer. Front Immunol 2022;13:829451. [Crossref] [PubMed]
6. Vogelbaum MA, Brown PD, Messersmith H, et al. Treatment for Brain Metastases: ASCO-SNO-ASTRO Guideline. J Clin Oncol 2022;40:492-516. [Crossref] [PubMed]
7. Pernot S, Evrard S, Khatib AM. The Give-and-Take Interaction Between the Tumor Microenvironment and Immune Cells Regulating Tumor Progression and Repression. Front Immunol 2022;13:850856. [Crossref] [PubMed]
8. Mitra A, Kumar A, Amdare NP, et al. Current Landscape of Cancer Immunotherapy: Harnessing the Immune Arsenal to Overcome Immune Evasion. Biology (Basel) 2024;13:307. [Crossref] [PubMed]
9. Spella M, Stathopoulos GT. Immune Resistance in Lung Adenocarcinoma. Cancers (Basel) 2021;13:384. [Crossref] [PubMed]
10. Sudmeier LJ, Hoang KB, Nduom EK, et al. Distinct phenotypic states and spatial distribution of CD8(+) T cell clonotypes in human brain metastases. Cell Rep Med 2022;3:100620. [Crossref] [PubMed]
11. Ranjan T, Podder V, Margolin K, et al. Immune Checkpoint Inhibitors in the Management of Brain Metastases from Non-Small Cell Lung Cancer: A Comprehensive Review of Current Trials, Guidelines and Future Directions. Cancers (Basel) 2024;16:3388. [Crossref] [PubMed]
12. Teixeira Loiola de Alencar V, Guedes Camandaroba MP, Pirolli R, et al. Immunotherapy as Single Treatment for Patients With NSCLC With Brain Metastases: A Systematic Review and Meta-Analysis-the META-L-BRAIN Study. J Thorac Oncol 2021;16:1379-91. [Crossref] [PubMed]
13. Dong W, Sheng J, Cui JZM, et al. Systems immunology insights into brain metastasis. Trends Immunol 2024;45:903-16. [Crossref] [PubMed]
14. Wang X, Bai H, Zhang J, et al. Genetic Intratumor Heterogeneity Remodels the Immune Microenvironment and Induces Immune Evasion in Brain Metastasis of Lung Cancer. J Thorac Oncol 2024;19:252-72. [Crossref] [PubMed]
15. Liu W, Powell CA, Wang Q. Tumor microenvironment in lung cancer-derived brain metastasis. Chin Med J (Engl) 2022;135:1781-91. [Crossref] [PubMed]
16. Li M, Hou X, Sai K, et al. Immune suppressive microenvironment in brain metastatic non-small cell lung cancer: comprehensive immune microenvironment profiling of brain metastases versus paired primary lung tumors (GASTO 1060). Oncoimmunology 2022;11:2059874. [Crossref] [PubMed]
17. Song SG, Kim S, Koh J, et al. Comparative analysis of the tumor immune-microenvironment of primary and brain metastases of non-small-cell lung cancer reveals organ-specific and EGFR mutation-dependent unique immune landscape. Cancer Immunol Immunother 2021;70:2035-48. [Crossref] [PubMed]
18. Xiao Y, Hu F, Chi Q. Single-cell RNA sequencing and spatial transcriptome reveal potential molecular mechanisms of lung cancer brain metastasis. Int Immunopharmacol 2024;140:112804. [Crossref] [PubMed]
19. Wang Z, Wang Y, Chang M, et al. Single-cell transcriptomic analyses provide insights into the cellular origins and drivers of brain metastasis from lung adenocarcinoma. Neuro Oncol 2023;25:1262-74. [Crossref] [PubMed]
20. Kim N, Kim HK, Lee K, et al. Single-cell RNA sequencing demonstrates the molecular and cellular reprogramming of metastatic lung adenocarcinoma. Nat Commun 2020;11:2285. [Crossref] [PubMed]
21. Prat A, Navarro A, Paré L, et al. Immune-Related Gene Expression Profiling After PD-1 Blockade in Non-Small Cell Lung Carcinoma, Head and Neck Squamous Cell Carcinoma, and Melanoma. Cancer Res 2017;77:3540-50. [Crossref] [PubMed]
22. Jung H, Kim HS, Kim JY, et al. DNA methylation loss promotes immune evasion of tumours with high mutation and copy number load. Nat Commun 2019;10:4278. [Crossref] [PubMed]
23. Zhu J, Fan Y, Xiong Y, et al. Delineating the dynamic evolution from preneoplasia to invasive lung adenocarcinoma by integrating single-cell RNA sequencing and spatial transcriptomics. Exp Mol Med 2022;54:2060-76. [Crossref] [PubMed]
24. Fan J, Lu F, Qin T, et al. Multiomic analysis of cervical squamous cell carcinoma identifies cellular ecosystems with biological and clinical relevance. Nat Genet 2023;55:2175-88. [Crossref] [PubMed]
25. Chen Y, Wang D, Li Y, et al. Spatiotemporal single-cell analysis decodes cellular dynamics underlying different responses to immunotherapy in colorectal cancer. Cancer Cell 2024;42:1268-1285.e7. [Crossref] [PubMed]
26. Hänzelmann S, Castelo R, Guinney J. GSVA: gene set variation analysis for microarray and RNA-seq data. BMC Bioinformatics 2013;14:7. [Crossref] [PubMed]
27. Li T, Fan J, Wang B, et al. TIMER: A Web Server for Comprehensive Analysis of Tumor-Infiltrating Immune Cells. Cancer Res 2017;77:e108-10. [Crossref] [PubMed]
28. Newman AM, Liu CL, Green MR, et al. Robust enumeration of cell subsets from tissue expression profiles. Nat Methods 2015;12:453-7. [Crossref] [PubMed]
29. Xun Z, Ding X, Zhang Y, et al. Reconstruction of the tumor spatial microenvironment along the malignant-boundary-nonmalignant axis. Nat Commun 2023;14:933. [Crossref] [PubMed]
30. Efremova M, Vento-Tormo M, Teichmann SA, et al. CellPhoneDB: inferring cell-cell communication from combined expression of multi-subunit ligand-receptor complexes. Nat Protoc 2020;15:1484-506. [Crossref] [PubMed]
31. Troulé K, Petryszak R, Cakir B, et al. CellPhoneDB v5: inferring cell-cell communication from single-cell multiomics data. Nat Protoc 2025; Epub ahead of print. [Crossref]
32. Pham D, Tan X, Balderson B, et al. Robust mapping of spatiotemporal trajectories and cell-cell interactions in healthy and diseased tissues. Nat Commun 2023;14:7739. [Crossref] [PubMed]
33. Wang L, Chen X, Dong C, et al. Protocol for generating brain metastatic tumor cells through repeated intracardiac injections in mice. STAR Protoc 2025;6:103531. [Crossref] [PubMed]
34. Mizutani R, Hirata E. Intracardiac Injection Mouse Model to Study Cancer Cell Dormancy in Brain Metastasis. Methods Mol Biol 2024;2811:113-22. [Crossref] [PubMed]
35. Puram SV, Tirosh I, Parikh AS, et al. Single-Cell Transcriptomic Analysis of Primary and Metastatic Tumor Ecosystems in Head and Neck Cancer. Cell 2017;171:1611-1624.e24. [Crossref] [PubMed]
36. Gavish A, Tyler M, Greenwald AC, et al. Hallmarks of transcriptional intratumour heterogeneity across a thousand tumours. Nature 2023;618:598-606. [Crossref] [PubMed]
37. Wu Y, Yang S, Ma J, et al. Spatiotemporal Immune Landscape of Colorectal Cancer Liver Metastasis at Single-Cell Level. Cancer Discov 2022;12:134-53. [Crossref] [PubMed]
38. Zhang S, Fang W, Zhou S, et al. Single cell transcriptomic analyses implicate an immunosuppressive tumor microenvironment in pancreatic cancer liver metastasis. Nat Commun 2023;14:5123. [Crossref] [PubMed]
39. Zou Y, Ye F, Kong Y, et al. The Single-Cell Landscape of Intratumoral Heterogeneity and The Immunosuppressive Microenvironment in Liver and Brain Metastases of Breast Cancer. Adv Sci (Weinh) 2023;10:e2203699. [Crossref] [PubMed]
40. Shou J, Peng J, Zhao Z, et al. CCL26 and CCR3 are associated with the acute inflammatory response in the CNS in experimental autoimmune encephalomyelitis. J Neuroimmunol 2019;333:576967. [Crossref] [PubMed]
41. Ding L, Fu Y, Zhu N, et al. OXTR(High) stroma fibroblasts control the invasion pattern of oral squamous cell carcinoma via ERK5 signaling. Nat Commun 2022;13:5124. [Crossref] [PubMed]
42. Lutz V, Hellmund VM, Picard FSR, et al. IL18 Receptor Signaling Regulates Tumor-Reactive CD8+ T-cell Exhaustion via Activation of the IL2/STAT5/mTOR Pathway in a Pancreatic Cancer Model. Cancer Immunol Res 2023;11:421-34. [Crossref] [PubMed]
43. Cao F, Xiao Z, Chen S, et al. HDAC/MIF dual inhibitor inhibits NSCLC cell survival and proliferation by blocking the AKT pathway. Bioorg Chem 2021;117:105396. [Crossref] [PubMed]
44. Zheng Y, Li X, Qian X, et al. Secreted and O-GlcNAcylated MIF binds to the human EGF receptor and inhibits its activation. Nat Cell Biol 2015;17:1348-55. [Crossref] [PubMed]
45. Amagai R, Takahashi T, Terui H, et al. The Antimicrobial Peptide Cathelicidin Exerts Immunomodulatory Effects via Scavenger Receptors. Int J Mol Sci 2023;24:875. [Crossref] [PubMed]
46. Lou X, Chen H, Chen S, et al. LL37/FPR2 regulates neutrophil mPTP promoting the development of neutrophil extracellular traps in diabetic retinopathy. FASEB J 2024;38:e23697. [Crossref] [PubMed]
47. Yasuda K, Nakanishi K, Tsutsui H. Interleukin-18 in Health and Disease. Int J Mol Sci 2019;20:649. [Crossref] [PubMed]