RNA sequencing data and clinical characteristics of lung squamous cell carcinoma patients in TCGA-LUSC were obtained from The Cancer Genome Atlas (TCGA) database.¹

Univariate Cox analyses were performed to find out the prognostic-related genes in LUSC patients, and the top 20 prognostic genes were presented with a P-value of <0.05. The Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis was performed to identify signaling pathways of prognostic genes in LUSC with the “ClusterProfiler” R package. We further explored the interactions of prognostic genes in LUSC to construct a protein-protein interaction (PPI) network using the STRING website² using R cluster profilter, and cytoscape was performed to present the PPI network.

To explore the relationship between the expression of focal adhesion-related genes and LUSC subtypes, consistent cluster analyses were conducted on the TCGA-LUSC cohort using the R package ConsensusClusterPlus (version 1.54.0). The clustering variable (k) was set from 2 to 6. The heatmap was generated by the pheatmap R package (version 1.0.12). The overall survival (OS) time was compared between the different subgroups through Kaplan-Meier analysis.

The differentially expressed genes (DEGs) between different LUSC clusters were identified using “DEseq2” R package. A P-value of <0.05 and |log 2-fold change| > 1 were regarded as the cutoff criterion. Furthermore, volcano plots were generated by the “ggplot2” R package to visualize the DEGs. The heatmap of DEGs was generated by the pheatmap R package (version 1.0.12). Gene Ontology (GO) and KEGG analyses were conducted by applying the “ClusterProfiler” package.

Based on focal adhesion-related genes, immune activity was evaluated by CIBERSORT in “immunoeconomics”. Eight immune checkpoint-related genes were used to compare the immune activity of two focal adhesion-related clusters: CD274, PDCD1, PDCD1LG2, CTLA4, LAG3, HAVCR2, TIGIT, and SIGLEC15. The heatmap and boxplot were generated by the pheatmap package and the R package ggplot2, respectively. The Wilcox test was used to compare the immune cell infiltration and immune pathway activation between the two groups. A P-value of <0.05 was considered statistically significant.

Drug sensitivity analysis was used to explore the difference between two focal adhesion-related clusters in LUSC using the pRRophetic algorithm. The chemotherapy response of each patient was evaluated by the available pharmacogenomics database.³ The prediction process was performed by the “pRRophetic” R software package. The ridge regression was used to estimate the half-maximal inhibitory concentration (IC50). A P-value of <0.05 was considered statistically significant.

To evaluate the prognostic value of the focal adhesion-related genes, Cox regression analysis was further employed to construct the prognostic model in the TCGA-LUSC cohort with the “glmnet”R package. The variables were non-zero coefficients, and the lambda condition was decided by the minimum criteria. The risk score was as follows: Risk score = sum (expression level of each gene × corresponding coefficient). The TCGA-LUSC patients were divided into low- and high-risk subgroups based on the median risk score. The OS time was compared between the different subgroups through Kaplan-Meier analysis using the “survival” R package. The hazard ratios (HRs) with 95% confidence intervals (CIs) were calculated using the Cox proportional hazards analysis.

To explore the roles of the prognostic genes in LUSC, univariate Cox analyses were performed to find out the prognostic-related genes in TCGA-LUSC patients. There were 444 prognostic genes determined by the univariate Cox analysis in LUSC. The top 20 prognostic genes are shown in Figure 1A and Supplementary Table 1. Moreover, we further explored the roles of the 444 prognostic genes using the KEGG enrichment analysis. The KEGG enrichment results suggested that these pathways, including the TNF signaling pathway, focal adhesion, proteoglycans in cancer, and leukocyte transendothelial migration, might play important roles in LUSC (Figure 1B and Supplementary Table 1).

The PPI networks of these prognostic-related genes were also confirmed. These pathways, which included the TNF signaling pathway, focal adhesion, proteoglycans in cancer, and complement and coagulation cascades, were mainly involved in LUSC (Figure 1C and Supplementary Table 1). The genes related to the KEGG pathway were also identified in Figure 1D, such as focal adhesion (including ITGA3, VAV2, FLNC, FLT4, HGF, ITGB6, THBS1, MYL2, ITGB1, CCND2, PPP1CB, PXN, and PDGFRA) and complement and coagulation cascades (including SERPINE1, SERPIND1, SERPINA5, F10, CD59, CPB2, and A2M).

Considering focal adhesions are closely related to the development of tumors (6, 7), the role of the focal adhesion signaling pathway in LUSC remains to be further studied. The correlations between gene expression and immune score were analyzed using Spearman.

First, we analyzed the connections between focal adhesion-related genes (ITGA3, VAV2, FLNC, FLT4, HGF, ITGB6, THBS1, MYL2, ITGB1, CCND2, PPP1CB, PXN, and PDGFRA) in the focal adhesion signaling pathway, and these genes widely interacted (Figure 2A and Supplementary Table 2), indicating they might play important roles in LUSC. The 13 focal adhesion-related genes were next applied to stratify TCGA-LUSC patients into different subtypes using the consensus clustering analysis.

When the clustering variable (k) was 2, the TCGA-LUSC patients could be well stratified into two clusters with a consistency cluster analysis and principal component analysis (PCA) (Figures 2B–C). The gene expression profiles of 13 focal adhesion-related genes were well separated in a heatmap between the two groups in TCGA-LUSC patients (Figure 2D). Interestingly, there was a statistically significant tendency for better overall survival time in cluster 1 than those in group 2 (HR: 0.665, 95% CI: 0.504–0.877, P = 0.00389; Figure 2E and Supplementary Table 2).

To understand the different mechanisms of the two groups, under the cutoff of P < 0.05 and |log 2-fold change| > 1, 423 DEGs, including 44 upregulated genes and 379 downregulated genes, were identified between the two groups in the volcano plot (G1 vs. G2; Figure 3A and Supplementary Table 3). The gene expression of the top 50 DEGs showed a different tendency between the two clusters in the heatmap (Figure 3B and Supplementary Table 3).

To better explore the biological roles of 423 DEGs, GO and KEGG enrichment analyses were performed. The results of KEGG analysis revealed that 44 upregulated genes were mainly related to metabolism and biosynthesis, for example, tyrosine metabolism, steroid hormone biosynthesis, retinol metabolism, platinum drug resistance, pentose and glucuronate interconversions, metabolism of xenobiotics by cytochrome P450, and drug metabolism-cytochrome P450 (Figure 3C and Supplementary Table 3), while 379 downregulated genes were mainly related to hallmarks of cancer, for example, ECM-receptor interaction, focal adhesion, proteoglycans in cancer, small cell lung cancer, cytokine-cytokine receptor interaction, and TGF-beta signaling pathway (Figure 3E and Supplementary Table 3). Similarly, the GO for BP results showed that the upregulated genes were primarily associated with unsaturated fatty acid metabolic processes, skin development, epidermis development, tertiary alcohol metabolic processes, glycoside metabolic processes (Figure 3D and Supplementary Table 3), while the downregulated genes were primarily associated with extracellular matrix organization, collagen fibril organization, collagen metabolic process, cell-substrate adhesion, positive regulation of cell adhesion, and regulation of cellular response to growth factor stimulus (Figure 3F and Supplementary Table 3). Many studies have proven that focal adhesion, proteoglycans in cancer, and cytokine-cytokine receptor interaction are tumor markers (10, 11).

The above results indicated that the tumor cells of the G2 LUSC subtypes might have stronger tumor migration and proliferation ability.

Many studies have proven that focal adhesion is closely related to immune activity in many cancers (6–8). First, the correlations between the expression of focal adhesion genes and immune score were analyzed using Spearman. The focal adhesion-related genes (ITGA3, VAV2, FLNC, FLT4, HGF, ITGB6, THBS1, MYL2, ITGB1, CCND2, PPP1CB, PXN, and PDGFRA) were closely correlated with many immune cells (Figure 4A and Supplementary Table 4). We further compared the immune activity between two focal adhesion-related clusters in LUSC patients. The boxplots revealed that the immune cells, including the T cells CD8⁺, macrophages, NK cells, and dendritic cells, of the C1 LUSC samples were clearly different from those of the G2 (Figure 4B and Supplementary Table 4).

Furthermore, the boxplots also revealed that 6 of the 8 ICI-related genes (i.e., CTLA4, HAVCR2, PDCD1, PDCD1LG2, TIGIT, and SIGLEC15) were lower in the G1 LUSC sample than the G2 patients (Figure 4C and Supplementary Table 4). These results suggested a close relationship between focal adhesion and immune activity.

We further evaluated the drug sensitivity of the two clusters in LUSC using the pRRophytic algorithm. Six chemotherapy drugs, namely gemcitabine, vinorelbine, paclitaxel, docetaxel, cisplatin, and methotrexate, were selected to perform the drug sensitivity analysis between the two clusters in LUSC. We found that the status of the two clusters was closely associated with the IC50 scores of gemcitabine, vinorelbine, paclitaxel, cisplatin, and methotrexate in LUSC (Figures 5A–F and Supplementary Table 5). The IC50 values of gemcitabine and methotrexate were lower in the high-risk G2 LUSC patients than in the G1 patients (Figures 5A–F and Supplementary Table 5). The IC50 values of vinorelbine, paclitaxel, and cisplatin were higher in the high-risk G2 LUSC patients than in the G1 patients (Figures 5A–F and Supplementary Table 5). These results suggested focal adhesion-related classification might represent a good predictor of response to chemotherapy.

LASSO and Cox regression analyses were further performed to select the 13 focal adhesion-related genes in LUSC. A 10-gene signature was developed based on the optimum λ value. The formula used to calculate the risk score was as follows: risk score = ITGA3 expression × 0.0669 + VAV2 expression × 0.0542 + FLNC expression × 0.0284 + FLT4 expression × 0.0034 + HGF expression × 0.0919 + MYL2 expression × 0.246 + ITGB1 expression × 0.0017 + PDGFRA expression × 0.0297 + CCND2 expression × (−0.0758) + PPP1CB expression × (−0.0847) (Figures 6A, B and Supplementary Table 6). Based on this gene signature, TCGA-LUSC patients were divided into low-risk and high-risk groups (Figure 6C). The overall survival analysis revealed that patients with a low-risk score had better survival than patients with a high-risk score (HR: 2.169, 95% CI: 1.641–2.867, P = 5.35e–08; Figure 6D and Supplementary Table 6). A receiver operating characteristic (ROC) curve was used to test the accuracy of the prognostic model. The area under the curve (AUC) was 0.6126 for 1 year, 0.685 for 3 years, and 0.684 for 5 years (Figure 6E and Supplementary Table 6), suggesting that this model had a fine prognostic effect.

A high TIDE score is associated with a poor response to immune checkpoint blockade (ICB) treatment and a short survival period after receiving ICB treatment. The TIDE score of G2 TCGA-LUSC patients was higher than that of G1 patients, indicating a poor response to ICB therapy and poor overall survival (Figure 7A).

Cancer stem cells (CSCs) are major factors contributing to tumor development, relapse, and metastasis and are responsible for chemotherapy resistance and cancer recurrence. The CSC scores of G2 TCGA-LUSC patients were significantly different from those of G1 patients (Figure 7B). These results suggested that focal adhesion-related classification might represent a good predictor of response to ICB therapies.