3.1 Cell composition of tumor tissues and brain metastases

After data processing, 57,222 cells from tumor tissues (n = 15) and 29,060 cells from brain metastases (n = 10) were obtained for subsequent analysis (Fig. 1C, F). All cells were divided into 17 subpopulations (Fig. 1A). We annotated each cell subpopulation according to the highly expressed marker genes (Fig. 1D). Eight cell types were identified in 17 subpopulations, including four non-immune cells and four immune cells (Fig. 1B). Non-immune cells included endothelial cells (CLDN5, PECAM1, FLT1, and RAMP2), fibroblasts (DCN, COL1A1, and COL1A2), epithelial cells (KRT18, KRT19, and EPCAM) and oligodendrocytes (OLIG2 and OLIG1) (Fig. 1E). Immune cells included macrophages (MARCO, CD68, and FCGR3A), mast cells (MS4A2, KIT), T/NK cells (GNLY, NKG7, CD3D, CD3E, and CD3D), and B cells (IGHM, CD79A) (Fig. 1E). Then, we calculated the proportion of each cell type in tumor tissues and brain metastases (Fig. 1G). We found that the proportion of T/NK cells and mast cells was significantly lower in brain metastases (Fig. 1H).

Fig. 1

Overview of TME in lung adenocarcinoma and brain metastases. A UMAP visualization of all cells displayed with different colours for clusters. B Eight major cell types in lung adenocarcinoma and brain metastases. C Cell distribution in lung adenocarcinoma and brain metastases. D For violin plots, x-axes stand for the number of clusters, y-axes stand for the relative expression level of marker genes. E Marker genes of eight major cell types identified in this study. F The UMAP plot showing sample origin. G The proportion of each cell type in lung adenocarcinoma and brain metastases. H Percentages of the eight cell types among the two groups. Y-axis: average percentage of samples across the two groups. Groups are shown in different colours

3.2 Heterogeneity of NK/T cells in different lesions

We performed a subclustering analysis of 27,601 NK/T cells and obtained 11 cell subclusters (Fig. 2A). Four clusters (C1, C3, C5, and C7) exhibited high expression of CD8A and CD8B, defined as CD8 + T cells. Three clusters (C0, C4 and C6) defined as Th1/Th17 cells (CD8-, IL7R+). C2 with high expression of FCGR3A and KLRF12 was defined as NK cells. C8 with high expression of IL2RA and FOXP3 was defined as Tregs, and C9 with high expression of SELL and CCR7 was defined as CD4 + naive T cells. since no marker gene was found in the C10 subpopulation, we defined it as Unknown (Fig. 2B, C). To assess the potential functional and immune status of CD8 + T cells and NK cells, we used the AddModuleScore function to calculate the scores of functional modules of CD8 + T and NK cells from different sources (Fig. 2D, E). Notably, CD8 + T cells in brain metastases show lower naive scores and higher exhaustion scores compared to tumor tissues, while NK cells show lower naive scores and cytotoxicity scores (Fig. 2F–K).

Fig. 2

Heterogeneity of NK/T cells in lung adenocarcinoma and brain metastases. A UMAP plot shows eleven subclusters of the NK/T cells. B UMAP plot of NK/T cells revealing nine subtypes. C Canonical cell markers used to identify NK/T cell subtypes. D The UMAP plot showing CD8 + T cells origin. E The UMAP plot showing NK cells origin. F–H Violin plot indicating the naive (left), cytotoxic (middle), and exhausted (right) scores of CD8 + T cells from lung adenocarcinoma and brain metastases. I–K Violin plot indicating the naive (left), cytotoxic (middle), and exhausted (right) scores of NK cells from lung adenocarcinoma and brain metastases

3.3 Heterogeneity of myeloid cells in different lesions

Next, we performed a subclustering analysis of all myeloid cells. A total of 10 cell subpopulations were obtained (Fig. 3A), including 8 clusters of macrophages (APOE, LGMN) (C0, C1, C2, C5, C6, C7, C8, C9), monocytes (C3) (FCN1) and dendritic cells (C4) (CD1C, CLEC10A) (Fig. 3B, C). We then compared the numbers of monocytes, dendritic cells, and individual macrophage subpopulations in different lesions (Fig. 3D). We found that the number of dendritic cells was significantly higher in tumor tissues than in brain metastases, and we also found that macrophage C5 was also significantly higher than in brain metastases (Fig. 3E). Notably, the proportion of macrophage C4 was higher in brain metastases. Moreover, we found that the C4 subpopulation highly expressed cystatin B (CSTB), matrix metalloproteinase 9 (MMP9), chemokine C-C ligand 5 (CCL5), and macrophage migration inhibitory factor (MIF) (Fig. 3F), which have the characteristics of promoting tumor cell migration and invasion. Finally, GSVA results showed that macrophage C4 was mainly enriched in DNA repair, fatty acid metabolism, and oxidative phosphorylation pathways (Fig. 3G). Based on the above results, we hypothesize that macrophage C4 plays an important role in the process of lung cancer brain metastasis.

Fig. 3

Heterogeneity of myeloid cells in lung adenocarcinoma and brain metastases. A UMAP plot shows ten subclusters of the myeloid cells. B Canonical cell markers used to identify myeloid cell subtypes. C UMAP plot of myeloid cells revealing ten subtypes. D The proportion of myeloid cell subtypes in lung adenocarcinoma and brain metastases. E Percentages of each myeloid cell subtype among lung adenocarcinoma and brain metastases. F Violin plots of the expression of several marker genes, including CSTB, MMP9, CCL5, and MIF, in macrophage subtypes. G Differences in pathway activities by GSVA among different macrophage subtypes

3.4 Heterogeneity of fibroblasts in different lesions

According to the classical definition of cancer-associated fibroblast (CAF) subtypes, the major subtypes of CAF are myCAF and inflammatory CAF (iCAF). MyCAF is mainly involved in fibrosis, and PTN and RGS5 are marker genes of myCAF. iCAF has greater responsiveness to inflammatory responses and produces large amounts of inflammatory cytokines [20]. As shown in Fig. 4A–D, based on the expression of CAF marker genes, we defined subgroups 0, 1, 2, and 4 as iCAFs and subgroups 3 and 5 as myCAFs. We observed a significant increase of iCAFs in tumor tissues compared with brain metastases, while myCAFs were significantly increased in brain metastases compared with tumor tissues (Fig. 4E, F). By GSVA analysis, we found that iCAFs and myCAFs from brain metastases were highly enriched in pathways that support tumor progression, including glycolysis, oxidative phosphorylation, and EMT, compared with tumor tissues (Fig. 4G, H).

Fig. 4

Heterogeneity of fibroblasts in lung adenocarcinoma and brain metastases. A UMAP plot shows six subclusters of the fibroblasts. B UMAP plot of fibroblasts revealing two subtypes. C Canonical cell markers used to identify fibroblasts subtypes. D The UMAP plot showing fibroblasts origin. E The proportion of fibroblasts subtypes in lung adenocarcinoma and brain metastases. F Percentages of each fibroblasts subtype among lung adenocarcinoma and brain metastases. G Differences in pathway activities scored per cell by GSVA between lung adenocarcinoma and brain metastases iCAFs. H Differences in pathway activities scored per cell by GSVA between lung adenocarcinoma and brain metastases myCAFs.

3.5 Upregulation of angiogenic signaling pathways in endothelial cells in brain metastases

To describe the landscape of endothelial cells in tumor tissues and brain metastases, we re-clustered 884 endothelial cells (Fig. 5A, D). We obtained five cell subpopulations, including four clusters (C0, C1, C2, C4) of tumor endothelial cells (PLVAP, VWA1, HSPG2, and INSR) and lymphatic endothelial cells (C3) (PDPN, CCL21) (Fig. 5B, C). We found that tumor endothelial cells from brain metastases were highly expressed in angiogenesis-related genes such as COL4A1, COL4A2, HSPG2, COL15A1, and SPARC (Fig. 5E), and GSVA results also showed that tumor endothelial cells from brain metastases were enriched in angiogenesis (Fig. 5F), these results suggest that endothelial cells in brain metastases have a higher angiogenic capacity than primary tumor tissues.

Fig. 5

Heterogeneity of endothelial cells in lung adenocarcinoma and brain metastases. A UMAP plot shows five subclusters of the endothelial cells. B UMAP plot of endothelial cells revealing two subtypes. C Canonical cell markers used to identify endothelial cells subtypes. D The UMAP plot showing endothelial cells origin. E Violin plots of the expression of several marker genes, including COL4A1, COL4A2, HSPG2, COL15A1, and SPARC, in endothelial cells from lung adenocarcinoma and brain metastases. F Differences in pathway activities scored per cell by GSVA between lung adenocarcinoma and brain metastases endothelial cells

3.6 Epithelial cells in brain metastases have a higher degree of malignancy

To analyze the CNV of epithelial cells from different sources, we applied the inferCNV algorithm to analyze the CNV of epithelial cells using the copy number of fibroblasts and endothelial cells in the data as a control. We found extensive CNV in epithelial cells in both tumor tissues and brain metastases (Fig. 6A). Notably, epithelial cells derived from brain metastases had significantly higher CNV levels than epithelial cells from tumor tissues (Fig. 6B). These results suggest to us that multiple copy number variants exist in malignant epithelial cells and that genomic reprogramming occurs in malignant epithelial cells in brain metastases, but the specific mechanisms of alteration still need to be further investigated.

Fig. 6

Epithelial cells in brain metastases are more malignant. A Copy number variation in Epithelial cells, endothelial cells, and fibroblasts were used as a reference, red represents overexpression of genes, and blue represents low expression. B Violin plot indicating the CNV scores of Epithelial cells from lung adenocarcinoma and brain metastases. C Heatmap showing the top 50 differential expressed genes between the two groups. D Differences in pathway activities by GSVA between lung adenocarcinoma and brain metastases epithelial cells

3.7 Analysis of up-regulated genes in epithelial cells from brain metastases by scRNA-seq data

To understand the molecular differences between epithelial cells of different origins, we analyzed up-regulated genes in epithelial cells from brain metastases (Fig. 6C, Supplementary Data 1), which can eliminate the influence of other cells in the tumor microenvironment. Next, we performed the GSVA, which revealed that many metabolic pathways were significantly upregulated in epithelial cells from tumor tissues and brain metastases, including fatty acid metabolism, xenobiotic metabolism, and bile acid metabolism (Fig. 6D). Notably, inflammatory responses as well as interferon alpha response and interferon gamma response pathways were significantly suppressed in epithelial cells in brain metastases compared to tumor tissues. This suggests that the immune function of epithelial cells in brain metastases is more suppressed than in tumor tissue.

3.8 Validation of up-regulated genes in bulk RNA-sequencing data

To investigate the impact of these up-regulated genes on the prognosis of lung adenocarcinoma patients, we selected lung adenocarcinoma data from the TCGA and the GSE68465 for prognostic analysis. We found that high expression of twelve genes (Figure S1) simultaneously in these two databases predicted a worse prognosis of lung adenocarcinoma. These twelve genes included PERP, DDIT4, CCT6A, F12, CNIH1, CDA, PKM, PSMD1, S100A9, PSMB5, TCN1 and SEC61G (Fig. 7A, B).

Fig. 7

Prognostic role of genes identified by scRNA-seq in TCGA lung adenocarcinoma cohort and GSE68465. A Overexpression of genes predicts poor prognosis in TCGA lung adenocarcinoma cohort. B Overexpression of genes predicts poor prognosis in GSE68465.

3.9 DDIT4 promotes the proliferation, invasion, and migration of lung cancer cells

Among the twelve genes mentioned above, DDIT4 has been less studied in lung adenocarcinoma, so we chose it for the next study. It has been reported that DDIT4 plays an essential role in tumor progression [21]. In our study, the upregulation of DDIT4 expression was associated with poorer OS in lung adenocarcinoma. Taken together, we hypothesized that DDIT4 plays a key role in the process of lung cancer metastasis. Finally, we found that the downregulation of DDIT4 significantly inhibited the proliferation of lung cancer cells PC9 and A549 (Fig. 8C, D). Furthermore, in wound healing assay, DDIT4 downregulation significantly inhibited wound healing in PC9 and A549 (Fig. 8E). In transwell migration and invasion assays, migration and invasion of cells transfected with DDIT4 were significantly reduced compared to controls (Fig. 8F). Thus, our results revealed that DDIT4 plays a crucial role in the metastasis of lung cancer.

Fig. 8

DDIT4 promotes the progression of lung cancer. A, B The mRNA expression of DDIT4 in A549 and PC9 cells transfected with siRNAs (siDDIT4) or siRNA control (siCon) was measured by qRT-PCR. Means ± SD are shown. ***p < 0.01 by unpaired Student’s t-test. C, D Downregulation of DDIT4 slowed down A549 and PC9 cell proliferation. E siDDIT4 significantly inhibited wound closure compared to the corresponding controls. F Migratory and invasive cells were dramatically reduced in A549 and PC9 cells transfected with siDDIT4