Introduction

Lung cancer is a major cause of cancer-related mortality worldwide. Non-small cell lung cancer (NSCLC) accounts for approximately 80–85% of all lung cancer cases, with a dismal 5-year survival rate of only 16% when diagnosed at an advanced stage1. Among NSCLC subtypes, lung adenocarcinoma (LUAD) is the most prevalent histological variant, exhibiting a rising incidence in young women and non-smokers. EGFR and KRAS mutations are two common oncogenic drivers in LUAD, with EGFR mutations being more commonly observed than KRAS mutations in East Asian female patients2. The presentation of EGFR mutations is often predictive of the effectiveness of treatment with specific tyrosine kinase inhibitors (TKIs), which are used as first-line therapy in patients with EGFR-mutant (EGFR-MU) LUAD3. Regrettably, a majority of patients eventually acquire resistance to EGFR-TKIs, and despite the exploration of different alternative therapies either as monotherapy or in combination, the results have been suboptimal4.

The utilization of immune checkpoint inhibitors (ICIs) in clinical practice has led to a paradigm shift in the treatment of certain types of cancer, including NSCLC5. Inhibitors of the PD-1/PD-L1 axis have been shown to enhance the cytotoxic activity of CD8+ T cells and stimulate immune responses against cancer6. However, clinical trials have demonstrated that PD-1/PD-L1 inhibitors are largely ineffective in EGFR-MU LUAD7. For instance, in the KEYNOTE-001 trial, EGFR-MU and EGFR-wild-type (EGFR-WT) LUAD patients treated with pembrolizumab exhibited an objective response rate of 4% and 50%, respectively8. Similarly, the results of the CheckMate 153 trial, which investigated the efficacy of nivolumab in patients with lung adenocarcinoma (LUAD), indicated a significantly reduced partial response rate in patients with EGFR mutations than in those with wild-type EGFR9. The reason for the inferior treatment outcome in the former is not clear, and might be partially attributed to a reduced tumor mutational burden in EGFR-MU tumors, i.e., EGFR-MU tumors might be less antigenic than EGFR-WT tumors10.

Recent studies have pointed at a close relationship between EGFR mutations and an immunosuppressive tumor microenvironment (TME)11,12. Nevertheless, the precise mechanism through which EGFR mutations facilitate cancer cell survival while simultaneously inhibiting the immune response remains inadequately elucidated. Here, we performed cancer cell-specific high-dimensional weighted gene coexpression network analysis (hdWGCNA) with scRNA-seq data13. Our findings unveiled key genes and pathways that could potentially play a role in fostering a suppressive tumor immune microenvironment, as well as promoting tumor growth and metastasis in EGFR-mutated tumors. Furthermore, we comprehensively analyzed the immune landscape of EGFR-MU tumors, focusing on T cells, and identified key molecular events underlying T-cell dysfunction and immune suppression. Our findings provide new insights into the mechanism underlying oncogene-related immunosuppression and provide a novel perspective for the development of successful immunotherapeutic approaches.

Results

Identification of malignant cells from scRNA-seq data

Following rigorous quality control measures and data filtering, a total of 39,639 single cells were obtained from the scRNA-seq data. Utilizing canonical cell markers, various cell types were distinguished, encompassing immune cells (myeloids, NK cells, mast cells, B cells, and T cells), fibroblasts, endothelial cells, and epithelial cells (Fig. 1A,B). The UMAP plot illustrates that the distribution of most cell types is similar between EGFR-MU and EGFR-WT tumors (Fig. 1C). For cancer cells, 14 subclusters were further identified (Fig. 1D). EGFR-MU and EGFR-WT cancer cells were clustered according to their transcriptional patterns (Fig. 1E). The cancer cell transcriptional profiles seemed more distinguishable between individual EGFR-WT tumors than between individual EGFR-MU tumors (Fig. 1F), suggesting that EGFR-MU cancer cells are more transcriptionally homogenous. Then, to select malignant cells with high confidence for cell type-specific functional analysis, we obtained somatic large-scale chromosomal copy number variations (CNVs) and calculated CNV scores for each cancer cell cluster. Ciliated cells, endothelial cells, and fibroblasts were used as references (Fig. 1G,H). As expected, most cancer cells exhibited higher numbers of CNVs relative to the reference cells. To eliminate potential normal epithelial cells, we excluded three clusters—specifically clusters 6, 9, and 11—whose median CNV scores were below the average (Fig. 1H). The remaining 11 clusters were subsequently utilized for cancer cell-specific analysis. Alternatively, we considered all 14 clusters and excluded cells with CNV scores in the lowest 5%, as these were likely normal cells. Subsequent analyses using both methods to identify cancer cells yielded highly similar results. For conciseness, the results for the 11 clusters of cancer cells are presented in the main text, whereas the results derived from the alternative method are displayed in Supplemental Figs. 3–5.

Figure 1
figure 1

Identification and clustering of single cells in LUAD samples. (A,C): UMAP plot of the major cell types combined or grouped by EGFR-mutation status. (B): Heatmap of the mean expression of canonical marker genes for ten major cell types. (D–F): UMAP plot of cancer cells grouped by cluster, EGFR-mutation status, and sample, with each group assigned a unique color. (G): Heatmap of copy number variation (CNV) for each cell. (H): CNV scores for each cancer cell cluster and referenced clusters.

Full size image

EGFR mutation-related genes and modules as well as their functions

We performed hdWGCNA analysis to explore gene modules that are functionally related to EGFR-mutation status. After optimization, the software threshold power was set at 26 to construct a coexpression network (Supplemental Fig. 1). Four modules were identified and each was assigned a unique color as an identifier. Then, the harmonized module eigengenes (hMEs) were calculated to determine the unique distribution area of each module (Fig. 2A). To compare the module expression patterns between EGFR-MU and EGFR-WT cancer cells, we performed differential module eigengene (DME) analysis and found that the yellow and brown modules were enriched in the EGFR-MU group, while the turquoise and blue modules were enriched in the EGFR-WT group (Fig. 2B). Next, the top 25 hub genes for each module were used to calculate signature scores, as depicted in the dimensionality reduction plot (Fig. 2C and Supplemental Fig. 2). No significant positive correlation was observed between the different modules, and gene network analysis suggested that these modules were linked to distinct biological functions (Fig. 2C,D). However, given the small sample size (n = 5 vs. 4; EGFR-MU and EGFR-WT) of heterogeneous lung cancer patients, it is possible that there are other confounding variables in this analysis other than EGFR mutation status.

Figure 2
figure 2

Identification of modules associated with EGFR mutations by hdWGCNA. (A): Visualization of the different coexpression modules by network analysis. (B): UMAP plot of each coexpression module colored in each module’s uniquely assigned color. (C): Differential module eigengene (DME) analysis for different EGFR-mutation groups. (D): Visualization of the correlation between each module.

Full size image

Subsequently, enrichment analysis (KEGG and HALLMARK) was performed to study the potential biological functions of each module (Fig. 3A). The expression levels of the top 20 hub genes for each module in EGFR-MU and EGFR-WT cancer cells are indicated in Fig. 3B and Supplement Fig. 2B. As mentioned before, the yellow and brown modules were more engaged in EGFR-MU cancer cells. The results reveal that the yellow module regulates MAPK or KRAS signaling and the epithelial–mesenchymal transition (EMT) process, both of which play critical roles in cancer cell proliferation, migration, and metastasis. Meanwhile, the brown module was found to be associated with the regulation of various immune-related pathways such as inflammatory response, cytokine-cytokine signaling, and chemokine signaling (Fig. 3A).

Figure 3
figure 3

Hub genes and enrichment analysis of each module. (A): HALLMARK and KEGG enrichment analysis of modules. (B): Comparative analyses of the normalized expression levels of the top 20 hub genes from each module between EGFR-MU and EGFR-WT cancer cells. (C): Gene module scores in EGFR-MU or EGFR-WT tumors in the TCGA-LUAD datasets. A comparative analysis of the normalized expression levels of the top 20 hub genes.

Full size image

The turquoise module was found to be engaged in immunoregulatory processes, such as antigen-processing and presentation, as well as response to IFN-γ. Notably, HLA-B, a hub gene in the turquoise module, seemed to be downregulated in EGFR-MU tumors (Fig. 3B). Although the turquoise module was least differential with respect to EGFR mutation status (Supplemental Fig. 3), we further analyzed the expression of other MHC-I molecules, including B2M and HLA-A. Our findings indicate that their expression is generally reduced in EGFR-MU tumors relative to EGFR-WT tumors (Fig. 5A). Therefore, we hypothesize that EGFR-MU cancer cells may evade CD8+ cytotoxic T-cell-mediated attack by inhibiting MHC-I-dependent antigen presentation. The turquoise module is also involved in the regulation of various metabolic pathways, such as glutathione metabolism, phenylalanine metabolism, and hypoxia (Fig. 3A).

The TNF-α/NF-κB pathway was enriched in the turquoise and blue modules (Fig. 3A). Given that TNF-α is mainly produced by intratumoral immune cells, these data suggest that EGFR-WT tumors may possess a more inflamed or immunologically “hot” tumor milieu. However, immune cells were not excluded from EGFR-MU tumors, as shown in Fig. 1C. Furthermore, distinct inflammation-related pathways were enriched in the brown module (Fig. 3A), implying that immune landscapes differ significantly between EGFR-MU and EGFR-WT tumors.

Finally, although the hdWGCNA analysis revealed gene modules potentially responsible for the suppressive TME, the sample sizes were small. Therefore, we used the TCGA-LUAD bulk RNA-seq dataset to calculate module scores for individual samples based on the expression of hub genes specific to each module. The differential enrichment of these modules based on EGFR-mutation status aligns with previous findings and strengthens the validity of the scRNA-seq analysis results, as depicted in Fig. 3C.

Suppressed T-cell responses and altered MHC-I/II expression in EGFR-MU tumors

To study how EGFR mutations affect the T-cell immune response, we initially characterized major T-cell subpopulations based on canonical marker genes (Fig. 4A,B). While not achieving statistical significance, EGFR-MU tumors numerically exhibited a reduced proportion of CD8+ T cells—including that of effector CD8+ T cells (clusters 1 and 9) and exhausted CD8+ T cells (cluster 5)—relative to EGFR-WT tumors (Fig. 4C). Conversely, two Treg subtypes, Treg1 and Treg2 (clusters 6 and 8), also were numerically more prevalent in EGFR-MU tumors (Fig. 4C) while the difference was statistically not significant. These results are consistent with prior research11,14,15. We then analyzed cell–cell communication using the CellChat software. In brief, cancer cells and T-cell subpopulations were grouped first by EGFR-mutation status. Then, pairwise cell communication was analyzed, and ligand/receptor interactions as well as their strength were compared between EGFR-MU and EGFR-WT groups. The results indicated that although EGFR-WT tumors, in general, had more cell–cell communication, the intensity of this communication was lower than that in EGFR-MU tumors (Supplemental Fig. 6A). Moreover, there is a notable disparity in the strength of particular cell–cell communication pathways, such as MHC-I/II and CXCL, between EGFR-MU and EGFR-WT groups, which will be elaborated upon in the following discussion. Then, specific molecular interactions between cancer cells and T cells were examined. Interactions that were enhanced or suppressed in EGFR-MU tumors relative to EGFR-WT tumors are shown in Fig. 4D–E and Supplemental Fig. 6B. Analysis of the interaction between cancer cells and CD8+ effector T cells (CD8eff_1 and CD8eff_2) indicated that the MHC-I-CD8 interaction was decreased in the EGFR-MU group (Fig. 4D), consistent with the reduced expression of MHC-I genes in EGFR-MU cancer cells (Fig. 5A). These results further support that decreased MHC-I gene expression is essential for the ability of EGFR-MU cancer cells to evade immune attacks from CD8+ cytotoxic T cells.

Figure 4
figure 4

Identification of T-cell subtypes and cell–cell communication analysis. (A): UMAP plot of all T cells colored by subtype and EGFR-mutation status. (B): Heatmap of the mean expression of canonical marker genes for different T-cell subtypes. (C): Proportions of each subcluster in the EGFR-MU and EGFR-WT groups. (D,E): Significantly altered ligand-receptor interactions between cancer cells and effector CD8+ T cells. (D) or Tregs (E) in EGFR-MU or EGFR-WT tumors.

Full size image
Figure 5
figure 5

Examining HLA-DR positive cancer cells by mIHC. (A,B): Violin plots of HLA-I and HLA-II expression in EGFR-MU and EGFR-WT tumors. (C): Representative mIHC images of EGFR-MU (EGFR 19del and EGFR L858R) and EGFR-WT tumors. (D): Clinical features of the LUAD cohort for mIHC analysis. (E): Percentages of HLA-DR+ cancer cells (CD45-PanCK+ HLA-DR+) among the total cancer cells (CD45-PanCK+) in EGFR-WT and EGFR-MU tumors.

Full size image

Next, we analyzed the interaction between cancer cells and Tregs, and found that the interaction between MHC class II molecules (MHC-II) derived from cancer cells and CD4 derived from Tregs was significantly enhanced in EGFR-MU tumors (Fig. 4E). Consistent with this finding, the expression of MHC-II genes, including HLA-DR, HLA-DP, and HLA-DQ, was significantly upregulated in EGFR-MU cancer cells (Fig. 5B and Supplemental Fig. 7A). To verify this finding, we conducted multiplex immunohistochemistry (mIHC) analysis on 48 surgically resected tumor samples from patients with LUAD, including 22 EGFR-WT and 26 EGFR-MU samples (Supplemental Table 1). The results demonstrated that HLA-DR expression was significantly elevated in EGFR-MU cancer cells compared with that in EGFR-WT cancer cells in the LUAD cohort (Fig. 5C–E). Notably, the expression of costimulatory molecules, including CD80, CD86, and ICAM1, was significantly lower in cancer cells than in myeloid cells in EGFR-MU tumors (Supplemental Fig. 7B). Prior research indicates that insufficient CD28-mediated costimulatory signaling can lead to the differentiation of naïve T cells into Tregs16,17,18. Therefore, we speculate that EGFR-MU cancer cells might function as nonclassical MHC-II antigen-presenting cells and thus skew the differentiation of tumor-infiltrating CD4+ T cells towards Tregs.

Previous studies have reported that airway type II (AT-II) epithelial cells express MHC-II and regulate local immune responses in lung tissues19,20,21. To assess whether EGFR-MU cancer cells express MHC-II because of their AT-II cell origin, we analyzed the expression of several surfactant protein genes that indicate the AT-II cell lineage (https://panglaodb.se), including SP-A (SFTPA1, SFTPA2), SP-B (SFTPB), and SP-C (SFTPC)22. In comparison with EGFR-WT cancer cells, EGFR-MU cancer cells expressed high levels of these genes, which positively correlated with HLA-DRB1 expression (Supplemental Fig. 7C). The positive correlation between the expression of surfactant protein and HLA-DRB1 was validated in the TCGA-LUAD dataset (Supplemental Fig. 7D).

Roles of chemokines/chemokine receptors in recruiting Tregs into tumors

Cell–cell communication analysis indicated that several chemokine/chemokine receptor interactions, such as CXCL1-CXCR2, CXCL3-CXCR2, and CXCL16-CXCR6 (Fig. 4E), between cancer cells and Tregs were elevated in EGFR-MU tumors, consistent with the increased expression of these chemokines in EGFR-MU cancer cells (Supplemental Fig. 8A). To further explore the roles of chemokine/chemokine receptors in Treg recruitment as well as their cellular sources, we analyzed the single-cell expression profile of several chemokine receptors that have been reported to be expressed by Tregs and to mediate Treg recruitment, including CCR423,24, CCR725,26, CCR827,28, CCR1029,30, and CXCR431,32 as well as CXCR2 and CXCR6 identified in our analysis. As shown in Fig. 6A and Supplemental Fig. 8B, CCR4 was expressed at higher levels by Treg2 cells in EGFR-MU tumors, although CCR4 expression was not restricted to Tregs. In contrast, CCR8 expression was more restricted to Tregs, and its expression by Treg2 cells in EGFR-WT tumors was more abundant. CCR10 and CXCR2 were generally expressed at low levels by Tregs, while CXCR4 was highly and universally expressed by Tregs, with higher expression in Treg1 cells than in Treg2 cells. CXCR6 was highly expressed by Treg2 cells, and its levels were comparable between EGFR-MU and EGFR-WT tumors. These results reveal a complex pattern of chemokine receptor expression by Tregs in vivo, and suggest that multiple chemokine receptors might be involved in the increased infiltration of Tregs in EGFR-MU tumors, particularly of Treg2 cells that express CCR4.

Figure 6
figure 6

Role of chemokines in recruiting Tregs in EGFR-MU LUAD. (A): Violin plots representing chemokine receptor expression in the Treg subtypes. (B): Volcano plot of differentially expressed genes between Treg1 and Treg2 cells; Delta Percent refers to the percentage of (Treg1 vs. Treg2) expression of a specific gene. (C): Developmental trajectory of Treg1, Treg2, and naïve CD4+ T cells. (D): Violin plots representing chemokine ligand expression in all cell types in EGFR-MU and EGFR-WT tumors. (E, F): Association between CCL17/CCL22 and FOXP3 in EGFR-MU LUAD in the TCGA database. (G): Boxplot showing the differential expression of FOXP3 in the TCGA-LUAD samples grouped by median expression of CCL17 and CCL22. (H–I): CCL17/CCL22 expression in EGFR-MU cells (PC9, NCIH3255, NCIH1975, NCIH1650, HCC827, and HCC4006) compared with that in EGFR-WT cells (NCIH69, NCIH522, NCIH520, NCIH460, NCIH292, NCIH1299, A549, and A427) in the CCLE database.

Full size image

To examine whether the Treg1 and Treg2 subtypes were functionally different in terms of immune suppression, we analyzed differentially expressed genes (DEGs) between Treg1 and Treg2 cells. Some genes known to be involved in Treg-mediated immune suppression, including FOXP3, CTLA4, TNFRSF9, and TIGIT, were more highly expressed by Treg2 cells than by Treg1 cells, suggesting that Treg2 cells might be functionally more immunosuppressive (Fig. 6B and Supplemental Fig. 8C). Trajectory analysis of CD4+ naïve T, Treg1, and Treg2 cells suggested that Treg2 cells were terminally differentiated and derived from naïve CD4+ T cells, whereas Treg1 cells were in an intermediate state (Fig. 6C). These findings imply that Treg1 cells might be differentially unstable or exhibit a transitional differentiation status.

Analysis of TF activity revealed that certain TFs involved in cytokine-mediated signal transduction, including STAT1, STAT2, and STAT5, exhibited increased activity in Treg2 cells (Supplemental Fig. 9A). In addition, NFATC2 and SMAD3, which reportedly enhance FOXP3 expression, were also activated in Treg2 cells33. Pathway enrichment analysis revealed that the cytokine/cytokine receptor signaling pathway was significantly enriched in Treg2 cells relative to Treg1 cells (Supplemental Fig. 9B–C). Interestingly, Treg2 cells in EGFR-MU tumors seemed to express less POXP3 and CTLA4 than their counterparts in EGFR-WT tumors (Supplemental Fig. 8C), suggesting that Treg2 cells in EGFR-MU tumors might be relatively less active; this may be attributable to attenuated feedback-triggered activation of Tregs resulting from weaker CD8+ T-cell reactions.

To investigate the chemokine ligands involved in Treg recruitment and their sources, we comprehensively analyzed the expression of major ligands for the critical chemokine receptors (Fig. 6A). Interestingly, CCL1734, which binds to CCR4, was predominantly expressed by myeloid cells and cancer cells in EGFR-MU tumors. CCL22, another CCR4 ligand34, was expressed in a similar pattern as CCL17 (Fig. 6D). Analysis of the TCGA datasets verified that the CCL17 and CCL22 levels positively correlated with FOXP3 expression in EGFR-MU LUAD tumors (Fig. 6E–F). TCGA-LUAD tumors with high levels of both CCL17 and CCL22 (CCL17–CCL22high) expressed significantly more FOXP3 than CCL17–CCL22low LUAD tumors (Fig. 6G). In addition, analysis of the CCLE database (https://sites.broadinstitute.org/ccle/) verified that CCL17 and CCL22 expression levels were higher in EGFR-MU lung cancer cells than in EGFR-WT cells (Fig. 6H–I). These results suggest that CCL17 and CCL22 produced by EGFR-MU cancer cells, at least in part, play a role in Treg recruitment.

Ligands for CCR7, CCR8, and CCR10 were weakly expressed by various cell types, except for CCL18, which was mainly expressed by myeloid cells with similar expression levels observed in EGFR-MU and EGFR-WT tumors (Fig. 6D). Ligands for CXCR2 were mainly expressed by myeloid cells, neutrophils, and cancer cells (Fig. 6D). Considering the weak expression of CXCR2 by Tregs (Fig. 6A), the roles of CXCR2 ligands in Treg recruitment are unclear. CXCL16, the ligand for CXCR6, was mainly expressed by myeloid cells, neutrophils, and mast cells (Fig. 6D). CXCR6 levels were higher in Treg2 than in Treg1 cells (Fig. 6A); however, CXCL16 was not correlated with FOXP3 expression in EGFR-MU LUAD in the TCGA datasets (Supplemental Fig. 8D). Collectively, our results reveal that multiple chemokine/chemokine receptors might mediate Treg recruitment—myeloid and cancer cell-derived CCR4 ligands, CCL17 and CCL22 might be critical in Treg recruitment in EGFR-MU tumors, whereas CCR8 and its ligand CCL18 expressed by myeloid cells might play a more important role in Treg recruitment in EGFR-WT tumors.

CAF and tumor endothelial cell (TEC) subtypes are associated with EGFR-mutation status

Previous studies have revealed that a subpopulation of cancer-associated fibroblasts (CAFs) with antigen-presenting capability (apCAFs) can induce the expansion of Tregs in pancreatic cancer35. We analyzed major CAF subtypes—including myofibroblasts (mFibs), myoCAFs (mCAFs), inflammatory CAFs (iCAFs), apCAFs, and proliferating CAFs (pfCAFs)—based on the expression of classical marker genes (Fig. 7A–C). The findings indicated that the proportion of apCAFs (MHC-II+) was increased while that of iCAFs was reduced in EGFR-MU tumors, relative to EGFR-WT tumors (Fig. 7D). We also analyzed DEGs among CAFs (Fig. 7E) and performed an enrichment analysis (Fig. 7F). The results demonstrated that different CAF subtypes carried out distinct functions—specifically, apCAFs actively participated in antigen processing and presentation, whereas in iCAFs, cytokine- and chemokine-mediated effects, such as TNF/NF-κB and IL-17 signaling, were activated, reflecting that their functions were closely associated with the establishment of an inflammatory environment.

Figure 7
figure 7

Subtypes, functionality, and developmental trajectory of CAFs. (A, B): UMAP plot of all CAF subtypes and EGFR-mutation status. (C): Heatmap of the mean expression of canonical marker genes for different CAF subtypes. (D): Proportion of each CAF subtype in the EGFR-MU and EGFR-WT groups. (E): Differentially expressed genes in each CAF subtype relative to the other CAF subtypes. (F): KEGG and GO enrichment analysis for each CAF subtype. (G): Semisupervised trajectory of CAFs, colored by subtype and pseudotime. (H): Heatmap showing inferred transcription factor activity in different CAF subtypes. (I,J): Boxplot representing the fractions of apCAFs or iCAFs in EGFR-MU and EGFR-WT tumors in the TCGA-LUAD datasets.

Full size image

Trajectory analysis revealed that apCAFs and iCAFs were derived from mFibs, following distinct differentiation routes (Fig. 7G) with the preferential generation of apCAFs or iCAFs in EGFR-MU or EGFR-WT tumors, respectively. Next, we found that compared with other CAF subtypes, iCAFs exhibited a very unique TF activity profile featuring enhanced activity of TFs engaged in NF-κB and cytokine signaling pathways, such as RELA/RELB, NFKB1, and STAT3, suggesting the critical roles of inflammatory cytokines in iCAF differentiation and function (Fig. 7H). In apCAFs, several TFs related to cellular stemness, such as FLI, SIX2, and SOX10, as well as RFX5, which has been reported to promote MHC-II expression, were highly active (Fig. 7H). To overcome the limitation of the small scRNA-seq sample size, we examined whether EGFR mutations affected CAF differentiation by quantifying apCAFs and iCAFs in the TCGA-LUAD dataset with cell type-specific matrixes derived from the scRNA-seq data and CIBERSORT algorithm. Consistently, the results indicated that apCAFs were enriched in the EGFR-MU TCGA-LUAD samples, whereas iCAFs were more abundant in the EGFR-WT samples (Fig. 7I–J). To validate these findings, we examined CAF subsets in the LUAD cohort with mIHC analysis. The results confirmed enrichment of apCAFs and iCAFs in EGFR-MU and EGFR-WT tumors, respectively (Fig. 8A–C). Furthermore, we utilized normalized deviation to evaluate the composition of apCAFs or iCAFs in all CAFs, and found that the proportion of apCAFs was higher than that of iCAFs in EGFR-MU tumors compared with EGFR-WT tumors (Fig. 8D).

Figure 8
figure 8

Analysis of apCAFs and iCAFs by mIHC. (A): Representative mIHC images of EGFR-MU (EGFR 19del and EGFR L858R) and EGFR-WT tumors. (B): Fractions of apCAFs (Pan-CK-CD45-αSMA+ HLA-DR+ IL-6−) among total CAFs (Pan-CK-CD45-αSMA+) in EGFR-MU and EGFR-WT tumors. (C): Fractions of iCAFs (Pan-CK-CD45-αSMA+ HLA-DR-IL-6+) among total CAFs in EGFR-MU and EGFR-WT tumors. D: Normalized deviation between apCAFs and iCAFs formulated by [(% apCAFs—% iCAFs)/% CAFs] in EGFR-MU and EGFR-WT tumors.

Full size image

Tumor endothelial cells (TECs) are critical tumor stromal cells and can serve as APCs in response to inflammatory stimuli. We identified four vascular TEC clusters (Vascular_end_1–4) and one lymphatic TEC cluster (Fig. 9A–C) based on canonical marker genes. Cluster 3 TECs (vascular_end_3) were more abundant in EGFR-MU tumors than in EGFR-WT tumors (Fig. 9D). Furthermore, both DEG and pathway enrichment analyses demonstrated that their function was related to MHC-II antigen presentation (Fig. 9E–F). Notably, the expression of CD86 was markedly lower in TECs and CAFs than in myeloid cells (Supplemental Fig. 7B), suggesting that, similar to apCAFs, vascular_end_3 TECs lack sufficient costimulatory molecules and might be involved in immune suppression instead of immune activation.

Figure 9
figure 9

Subtypes, functionality, and developmental trajectory of ECs. (A–B): UMAP plot of all EC subtypes and EGFR-mutation status. (C): Heatmap of the mean expression of canonical marker genes for different EC subtypes. (D): Proportion of EC subtypes in the EGFR-MU and EGFR-WT groups. (E): Differentially expressed genes in each EC subtype relative to the other EC subtypes. (F): KEGG and GO enrichment analysis for each EC subtype.

Full size image

We also evaluated the fractions of various immune cell types in each TCGA-LUAD sample and found that CD8+ T cells and NK cells were decreased in EGFR-MU samples relative to EGFR-WT samples, whereas Tregs and myeloid cells showed an inverse tendency (Fig. 10). These findings were again consistent with the scRNA-seq analysis results.

Figure 10
figure 10

Analyzing gene modules and immune infiltration with TCGA bulk RNA-seq data. Heatmap showing infiltration of multiple immune cell types in EGFR-MU and EGFR-WT tumors in the TCGA-LUAD datasets.

Full size image

Discussion

There are many open questions regarding how oncogenic alterations, such as EGFR mutations, promote cancer cell growth and concomitantly cause immune suppression. Here, we comprehensively analyzed LUAD samples by the hdWGCNA method, which is tailored for scRNA-seq data, and identified vital gene modules in cancer cells that are differentially enriched in EGFR-MU or EGFR-WT tumors. Functional enrichment analysis revealed that two modules that are associated with EGFR mutations (yellow and brown) simultaneously regulate MAPK (RAS/RAF/MEK/ERK) signaling (Fig. 3A), which is the major protumorigenic pathway downstream of many receptor tyrosine kinases (RTKs)4,36. These findings suggest that multiple regulatory mechanisms converge on this critical pathway. We noticed that the expression of DUSP4, a hub gene in the blue module, was significantly lower in EGFR-MU cancer cells (Fig. 3B). DUSP4 functions to attenuate the activation of the MAPK pathway through the dephosphorylation of threonine and tyrosine residues on MAP kinases ERK1 and ERK237,38. The low expression of DUSP4 might be critical for the sustained activation of the MAPK pathway required for EGFR-MU tumor growth37,38. Intriguingly, we did not observe enrichment of the PI3K/AKT pathway—another major downstream pathway of EGFR activation39—in either EGFR-MU or EGFR-WT cancer cells. This finding implies that it is the MAPK pathway rather than the PI3K/AKT pathway downstream of EGFR signaling that plays a dominant role in cancer cell survival and proliferation.

Prior research has demonstrated that EGFR-MU tumors exhibit an immunosuppressive or “cold” TME characterized by reduced CD8+ T-cell infiltration, leading to a poor response to ICI treatment15,40. Our previous investigation indicated that B2M gene expression determined the levels of MHC-I molecules expressed by cancer cells, and that enhancing B2M expression in cancer cells heightened the tumor’s responsiveness to PD-1 blockade immunotherapy41. In this study, we found that MHC-I-related genes, including B2M, were markedly downregulated in EGFR-MU cancer cells, suggesting that B2M inhibition in EGFR-MU cancer cells may be responsible for the diminished CD8+ T-cell response in these cells. While the exact mechanism by which EGFR mutations lead to B2M downregulation remains unclear, previous studies have indicated that EGFR mutations inhibit MHC-I expression even in the presence of IFN-γ42. Furthermore, it has been demonstrated that activation of the MAPK pathway, rather than the PI3K pathway, is necessary for MHC-I downregulation in EGFR-MU cancer cells42. We have shown that the targeting of mutant KRAS in lung cancer cells can lead to an increase in B2M expression, suggesting a connection between the activation of the KRAS/MAPK pathway and the downregulation of B2M41. Collectively, these results suggest that the aberrant activation of the EGFR/MAPK axis in EGFR-MU tumors is critical in suppressing MHC-I antigen presentation and the response to ICI treatments.

Interestingly, we found that EGFR-MU cancer cells express high levels of MHC-II molecules. Cancer cell-specific MHC-II (cMHC-II) expression was reported at various frequencies in some cancers43, including lung cancer44. It has been suggested that MHC-II+ cancer cells might function as nonprofessional APCs and promote tumor-specific CD4+ T-cell activation43. Notably, in tumors where cMHC-II expression was associated with better prognosis or better response to immunotherapy, CD4+ and CD8+ T-cell infiltration and IFN-γ pathway activation were found to be prominent45,46. However, this is not the case for EGFR-MU LUAD, where CD4+ Th1 and CD8+ cytotoxic T cells are scarce. Extensive research on the involvement of cMHC-II in tumor progression indicates that the upregulation of MHC-II expression by cancer cells typically promotes anti-cancer immune responses47,48,49. However, discrepancies still exist: for example, the introduction of CIITA, a master regulator of MHC-II genes, into a lung cancer cell model leads to cMHC-II expression but accelerates tumor growth because of the absence of costimulatory signaling50. We found that cancer cells from EGFR-MU tumors express MHC-II but lack CD80 and CD86 expression, implying that their ability to stimulate CD4+ T cells might be impaired.

Recent studies have reported that some CAFs and tumor lymphatic endothelial cells express MHC-II but lack costimulatory molecules, and, as a result, induce the expansion of intratumoral Tregs and enhance their immunosuppressive activities51,52. Other studies have reported that MHC-II expression by pulmonary stromal cells is critical for Treg generation, and loss of MHC-II in stromal cells enhances the local inflammatory response and inhibits metastatic lung tumors19,53. Whether MHC-II+ cancer cells can function similarly and induce the expansion of intratumoral Tregs in EGFR-MU tumors warrants further investigation. Recently, increased cMHC-II expression was reported to enhance the efficacy of anti-CTLA-4 immunotherapy in a breast cancer mouse model54. However, it was shown to induce resistance to anti-PD-1 immunotherapy via LAG-3/FCRL6 engagement in melanoma and lung cancer models55. More studies are required to examine whether the increased MHC-II expression in EGFR-MU cancer cells would cause a poor response to PD-1-based immunotherapy9,10.

Although the mechanism of regulation of MHC-II expression in EGFR-MU cancer cells remains unclear, our analysis reveals that it might reflect the AT-II cell lineage of the cancer cells. Our analysis is consistent with a recent study demonstrating that environmental particulate matter measuring ≤ 2.5 μm (PM2.5) promotes malignant transformation of AT-II cells with EGFR mutations, and this process is closely associated with the inflammatory stimulation of IL-1β (encoded by the IL1B gene)56. Interestingly, we found that IL1B gene expression is markedly upregulated in EGFR-MU cancer cells (Fig. 3B). Furthermore, both IL1B and HLA-DR are hub genes in the brown module identified by the WGCNA, suggesting that they may be functionally connected (Fig. 3B). A previous study suggested that IL-1β can enhance MHC-II expression by innate lymphoid cells57. However, whether IL-1β derived from cancer cells promotes cMHC-II expression in an autocrine manner requires more investigation.

We found that EGFR-MU tumors are not necessarily “cold” and devoid of immune cell infiltration; rather, the inflammatory responses in EGFR-MU tumors are mechanistically and functionally different from those in EGFR-WT tumors, as indicated by the distinct immune-related pathways enriched in EGFR-MU vs. EGFR-WT tumors (Fig. 3A). Our analysis suggests that the immune composition differs significantly between EGFR-MU and EGFR-WT tumors. Consistent with previous studies4,11, more Tregs but fewer T effector cells (CD4/CD8+ Teff) were present in EGFR-MU tumors. This suggests that the increased Treg infiltration in EGFR-MU tumors might not be because of feedback on Treg recruitment by inflamed tumors—instead, EGFR-MU cancer cells might directly recruit peripheral Tregs by secreting more chemokines or inducing local Treg differentiation and expansion. This notion is supported by studies showing that after EGFR-TKI treatment, the population of Tregs was reduced while that of Teffs was unchanged in EGFR-MU tumors15.

The CCR4/CCL17/CCL22 axis may be critical in Treg recruitment in EGFR-MU tumors, and cancer cells might be the primary source of CCL17/CCL22. Consistent with a previous lung cancer scRNA-seq study58, two major subtypes of Tregs were identified in both EGFR-MU and EGFR-WT tumors. Treg2 cells express higher levels of FOXP3 and some inhibitory molecules, reminiscent of the phenotypes of inflammation-induced functional Tregs59 or highly suppressive effector Tregs identified in tumor models60. Interestingly, CCR4 is preferentially expressed by Treg2 cells in EGFR-MU tumors, whereas CCR8 expression is more significant in Treg2 cells in EGFR-WT tumors, although the underlying mechanism is unknown. Previous studies have shown that targeting CCR461,62 or CCR863,64 in various cancer models could decrease intratumoral Treg accumulation, suggesting critical roles for these chemokine receptors in Treg recruitment. However, Treg recruitment can only be suppressed by CCR8 antibodies capable of inducing antibody-dependent cell-mediated cytotoxicity65,66, which largely depends on the cytotoxic effects of NK cells. Further, CCR8 deficiency does not affect Treg migration and tumor progression27, implying that although CCR8 is dispensable for Treg migration and function, it can serve as a relatively specific Treg marker and drug target. Collectively, our results suggest that CCR4+ Tregs play critical roles in establishing an immunosuppressive TME in EGFR-MU tumors and that targeting the CCR4/CCL17/CCL22 axis is a potential strategy to stimulate immune responses in EGFR-MU tumors.

CAFs are heterogeneous populations, and the identification of various CAF subsets in the TME has been enabled by advances in single-cell technology67,68. By comparing CAFs between EGFR-MU and EGFR-WT tumors, we found that apCAFs, a subpopulation of CAFs expressing MHC-II and CD74, are more abundant in EGFR-MU tumors. Although the roles of apCAFs in pancreatic and breast cancers seem to be immune-suppressive69, a recent study using a lung cancer model derived from the LLC cell line (EGFR-wild-type) suggests that the MHC class II antigen immune response—which is critical for an effective anticancer response— depends on intratumoral apCAFs. Therefore, apCAFs can exert protumorigenic or antitumorigenic activities depending on the specific context. We still know very little about the origin, differentiation, and function of apCAFs. Our trajectory analysis revealed that myofibroblasts tend to differentiate toward apCAFs in EGFR-MU tumors or iCAFs in EGFR-WT tumors. We further identified some important signaling pathways and TFs engaged in the differentiation and function of CAFs, which provide potential targets for manipulating CAF differentiation in future studies. Previous studies have shown that the introduction of essential lineage-related transcription factors into cancer cells can suppress oncogene expression and enhance lineage-related gene expression, thereby prompting cancer cell reprogramming towards a more normal phenotype70. We propose that this approach can be applied to induce the transformation of pro-tumorigenic CAFs into anti-tumorigenic phenotypes, thereby offering great promise for potential therapeutic applications in lung cancer.

We also found a group of TECs expressing substantial amounts of MHC-II and CD74. Previously, MHC-II expression has been found either in tumor lymphatic ECs or vascular ECs with a capillary phenotype71; however, the roles of intratumoral MHC-II+ TECs are unclear. Given they are positioned at the front line of the tumor-host interface and are constantly influenced by tumor-derived factors, we speculate that MHC-II+ TECs could significantly impact the activity of CD4+ T cells as they migrate towards tumor sites. Notably, TECs and apCAFs lack costimulatory molecules, and we suspect that, in addition to MHC-II+ cancer cells, apCAFs or MHC-II+ TECs may also contribute to Treg enrichment in EGFR-MU tumors.

When interpreting our findings, it is important to acknowledge the potential limitations stemming from differences in sample size between scRNA-seq and bulk RNA sequencing datasets. scRNA-seq provides a detailed characterization of cellular heterogeneity and diversity within samples, which may not be fully captured by bulk RNA sequencing that averages gene expression across numerous cells. Nevertheless, the small sample sizes typically seen in scRNA-seq studies can lead to increased statistical noise and variability. Furthermore, it is challenging to definitively exclude the potential confounding effects of concomitant mutations frequently encountered in lung cancers, such as TP53 and STK11 mutations. These factors can hinder precise identification of authentic biological signals and compromise the reproducibility of the findings.

In conclusion, our findings indicate that EGFR-MU LUAD is characterized by suppressed CD8+ T-cell immune responses, increased Treg infiltration, and abnormal differentiation of CAFs and TECs. These factors may orchestrate a highly immunosuppressive TME in EGFR-MU LUAD. These immunosuppressive components are intertwined and gradually intensify as tumors progress. Therefore, although EGFR-TKI treatment seems to reverse immune suppression to a certain extent, it may not be sufficient to induce an overarching immune response to effectively eliminate cancer cells. Our study provides new insight into the mechanism underlying tumor immune escape in EGFR-MU LUAD and may contribute to the design of novel immunotherapies to improve treatment outcomes in the future.

Methods

Processing of RNA sequencing data

Lung cancer single-cell RNA and bulk RNA sequencing data were obtained from the Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/)72. The GEO number of the single-cell RNA sequencing (scRNA-seq) dataset was GSE171145, which included five EGFR-MU and four EGFR-WT LUAD samples11. Cells meeting any of the following criteria were excluded: < 500 or > 6000 unique genes expressed or > 30% of reads mapping to mitochondria-associated genes. The R package “Seurat” (version 4.0) was used to construct the analysis object and normalize the scRNA-seq data73. For scale data, unique molecular identifier counts were variance-stabilized using SCTransform with 3000 variable features. Next, the batch effect from different samples was eliminated by using the R package “harmony”74. The top 30 principal components were used to construct the spiking neural network graph and perform UMAP embedding. For cell identification, we selected a resolution of 0.8 and canonical cell type-specific markers. Uniformly standardized TCGA-LUAD datasets (n = 535) were downloaded from the UCSC database (https://xenabrowser.net/), and samples were grouped based on EGFR-mutation status for further analysis.

hdWGCNA

Weighted gene coexpression network analysis (WGCNA) of the scRNA-seq data was performed with the R package “hdWGCNA” designed for analyzing high-dimensional transcriptome data13. scRNA-seq data normalized with SCTransform were used as the input data, and recommended parameters were applied for analysis. Hub genes were identified based on their significant network connectivity, as indicated by the kME (eigengene connectivity) values. The expression levels of these hub genes were normalized using the SCTransform method and compared across different groups.

Cell–cell communication analysis

Following the official tutorial, the R package CellChat (version 1.13, https://github.com/jinworks/CellChat)75 was used for cell–cell communication analysis. Ligand-receptor pairs with p values < 0.05 were filtered to evaluate the relationship between different cell types.

Trajectory analysis

The R package “Monocle 2” was used to perform single-cell trajectory analysis76. The scRNA-seq profiles of fibroblasts were used to reconstruct the single-cell trajectory analysis. The pseudotemporal order of the cells was constructed using the “reduce-Dimension” and “orderCells” functions, and DDRTree was used to develop tree-like trajectories.

Transcription factor (TF) and pathway enrichment analysis

DoRothEA, which is a gene regulatory network containing signed TFs and their target gene interactions77, was used to perform TF activity analysis. KEGG and HALLMARK enrichment analyses for each gene set module were performed with the R package “enrichR”. Functional annotation of fibroblast and endothelial cell subtypes was conducted with the R package “clusterProfiler”78.

Correlation analysis of scRNA- and bulk RNA-seq data

Each gene set module score for TCGA samples was calculated using the R package “GSVA” designed for performing gene set variation analysis79. CIBERSORTx software (online, https://cibersortx.stanford.edu/)80 was utilized to evaluate the proportions of specific immune cell types in bulk RNA-seq datasets.

Multiplex immunohistochemistry (mIHC)

The study was approved by the Ethical Committee of the Affiliated Hospital of Nantong University. All methods were performed in accordance with the relevant guidelines and regulations. Forty-eight paraffin-embedded sections from patients with LUAD with or without EGFR mutations were obtained from the Affiliated Hospital of Nantong University (Nantong, China), and a tissue microarray was prepared for mIHC staining. Clinicopathological characteristics of the LUAD cohort are provided in Supplemental Table 1. The mIHC staining was performed to assess the expression of pan-cytokeratin (Pan-CK, cancer epithelium marker), α-SMA (pan-CAFs marker), CD45 (hematopoietic cell marker), IL-6, and HLA-DR. Briefly, the slices were treated with xylene, briefly washed in a graded series of ethanol (100%, 95%, 85%, 80%, 75% ethanol), and treated with AR9 solution (Akoya Biosciences, Marlborough, MA) at 100 °C for 20 min for antigen retrieval. Next, the slices were treated with 3% hydrogen peroxide at room temperature for 10 min to eliminate endogenous peroxidase. Subsequently, slices were treated with goat serum for blocking at room temperature for 10 min. Staining was performed with the BOND Polymer Refine Detection Kit (DS9800, Leica Biosystems, Deer Park, IL) in accordance with the manufacturer’s instructions. Antibodies used in the assay included anti-IL6 (GTX17623, Genetex), anti-HLA-DR (ab92511, Abcam), anti-CD45 (BX50068, Biolynx), anti-α-SMA (19,245, CST), anti-Pan-CK (BX50143-C3, Biolynx), and the polymer HRP secondary antibody from the DS9800 kit. Coverslips were mounted with anti-fade mounting medium, and images were captured using the Pannoramic MIDI fluorescent slide scanner (3DHISTECH, Budapest, Hungary). Finally, the VisioPharm software (Visiopharm Corporation, Broomfield, CO; version 2023.01.213695, https://visiopharm.com/) was used to perform fluorescence colocalization analysis81. For data analysis, the percentages of HLA-DR+ cancer cells (CD45-, Pan-CK+, HLA-DR+) among total cancer cells for each sample were calculated and compared between EGFR-WT and EGFR-MU tumors. In addition, the percentage of CAFs (CD45−, Pan-CK−, α-SMA+), apCAFs (CD45−, Pan-CK−, α-SMA+, HLA-DR+, IL6−), and iCAFs (CD45-, Pan-CK−, α-SMA+, IL-6+, HLA-DR−) among total cells were calculated to examine the composition of CAF subsets in EGFR-WT and EGFR-MU tumors.

Statistical analysis

Statistical analysis was performed with R v.4.1.0. The correlation between two continuous variables was evaluated using Pearson’s correlation coefficients. The chi-squared test was used to compare categorical variables, while the Wilcoxon rank-sum test or t-test was used to compare continuous variables (*p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001).

Ethics approval and consent to participate

The study was approved by the Ethical Committee of the Affiliated Hospital of Nantong University, and signed informed consent were obtained from all participants.