Analysis of prognostic genes in the tumor microenvironment of lung adenocarcinoma

Database

The level 3 gene expression profile data of lung adenocarcinoma patients (using Illumina HiSeq_RNASeqV2 lung adenocarcinoma RNA expression profiles) were downloaded from the TCGA website (https://portal.gdc.cancer.gov/). The immune and stromal scores were calculated from each sample using the ESTIMATE algorithm. In addition, patient clinical data were downloaded from the TCGA official website (https://portal.gdc.cancer.gov/repository). The LUAD gene chip expression profile data and the clinical data (GSE37745, GSE11969, and GSE50081) of the samples were downloaded from the GEO official website (https://www.ncbi.nlm.nih.gov/geo/) for subsequent validation analysis. To analyze the mRNA/protein expression of CCR6 in primary LUAD and normal tissues, UALCAN (http://ualcan.path.uab.edu) was used, which was based on the TCGA database and Clinical Proteomic Tumor Analysis Consortium. To validate the expression of CCR6, the Human Protein Atlas (http://www.proteinatlas.org/) was used. Tumor Immune Estimation Resource (TIMER) (https://cistrome.shinyapps.io/timer/) was used to assess the correlation between differentially expressed gene (DEG) expression and the immune cell infiltration level. Immune cell infiltration was evaluated using the TIMER and CIBERSORT method, and stromal cell infiltration was determined using xCell (https://xcell.ucsf.edu/).

Identification of differentially expressed genes (DEGs)

The limma package in R was used for the analysis of differential expression genes (Ritchie et al., 2015), and the cutoff value was set to —log fold change(FC)— > 1, and a p value < 0.05. We plotted a heat map using the cluster analysis results of DEGs in R.

GO and KEGG enrichment analysis

To investigate the molecular functions, biological processes, cell components, and signaling pathways involved in DEGs, we used the DAVID database (v6.8, https://david.ncifcrf.gov/) to perform GO analysis and KEGG pathway enrichment. p value < 0.05 and FDR < 0.05 were marked as significant terms.

Statistical analysis

The data were assessed for normal distribution, and the central tendency was expressed as the mean ± standard deviation for the data with normal distribution. Means were compared using a Student’s t-test. Continuous variables that did not have normal distribution were analyzed using the Wilcoxon rank sum test or Kruskal–Wallis rank sum test. Student’s t-tests were used to compare the distribution of scores in the high and low groups, gender, and M stages. The Wilcoxon rank sum test was used to compare the distribution of scores in age, T stages, and N stages. The Kruskal–Wallis rank sum test was used to compare the differences in the distribution of scores of the four different pathological stages. Spearman’s rank correlation coefficient was used to measure the correlation between lung cancer stage and overall survival. Determination of survival-related genes was performed using univariate Cox regression analysis (with a p value of < 0.05) and the corresponding survival curves were plotted. Multivariate Cox analysis was used to evaluate the contribution of genes as independent prognosis factors for patient survival. All tests were two-tailed, and p values less than 0.05 were considered statistically significant. Al analyses were performed using SPSS for Windows, software version 25.0 (SPSS Inc., Chicago, IL).

Correlation between immune and stromal scores and clinical characteristics of LUAD

We downloaded the gene expression profiles and clinical data of 492 patients with LUAD from the TCGA database, including 225 males (45.7%), 267 females (54.3%), 157 patients under 60 years of age (31.9%), and 335 patients over 60 years of age (68.1%). The ESTIMATE algorithm revealed stromal scores of −1959.31 to 2989.77, and the immune scores of −1355.85 to 2905.3. The results of the relationship between the two scores and clinical data (TNM and pathological stage) are shown in Table 1. There were differences in the distribution of stromal scores between the sexes and across M stages. Immune scores also differed with respect to T and pathological stages.

We found that the immune and stromal scores of female patients were significantly higher than those of from male patients (p = 0.0069 and 0.0049, respectively) (Figs. 1A and 1B). The immune score of T1 was significantly higher than that of T2/T3/T4 stages (p = 0.0002, Fig. 1C); however, but the distribution of stromal scores was not different between the two groups (p = 0.0999) (Fig. 1D). The immune and stromal scores of the M0 stage was higher than those of the M1 stage (p = 0.0098 and 0.0366, respectively) (Figs. 1E and 1F). For the different pathological stages of LUAD, the immune scores were different at each stage (p = 0.0276, Fig. 1G) and gradually decreased from stage I to stage IV (Spearman correlation coefficients: ρ = −0.131, p = 0.004); however, but the difference in stromal scores among the four pathological stages was not statistically significant (p = 0.1525, Fig. 1H).

Next, we evaluated the correlations between immune (or stromal) scores and overall survival. LUADs (N = 492) were divided into high- and low-score groups, with 246 cases in each group. Stromal scores ranged from −959.31 to 36.85 for the low group and from 39.8 to 2098.77 for the high group. Immune scores ranged from −1355.85 to 952.61 for the low group and 964.93 to 2905.3 for the high group. The Kaplan–Meier survival curve analysis for immune scores indicated that the median survival of the low-score group was lower than that of the high-score group (1194 d vs. 1499 d, p = 0.011; log-rank test; Fig. 1I). Similarly, for stromal scores, the median survival of the low-score group was lower than that of the high-score group (1235 d vs. 1492 d, p = 0.179; log-rank test; Fig. 1J), although it was not statistically significant.

Comparison of gene expression profiles with LUAD immune scores and stromal scores

We performed differential expression analysis of the high- and low- immune scoring patients and found that 884 genes were downregulated and 55 genes were upregulated (Fig. 2A, Table S1). Similarly, differential analysis of the high- and low-stromal scoring groups revealed that 999 genes were downregulated and 28 genes were up-regulated (—logFC—>1, p < 0.05) (Fig. 2B, Table S1). In the Venn diagram, 18 genes were found to be upregulated in both groups (Fig. 2C) and 512 genes were downregulated in both groups (Fig. 2D).

Correlation of individual DEGs expression with overall survival and Functional enrichment analysis of prognostic associated genes

To explore the DEGs associated with overall survival, we performed a univariate Cox regression analysis. The results (Table S2) showed that 286 of the 530 DEGs were survival-related genes (p < 0.05). To further determine the relationship between these prognostic DEGs, we used the STRING online tool to obtain a PPI network. This network has 277 nodes and 2143 edges (Fig. 3). Functional enrichment clustering of 286 prognostic DEGs was also closely related to the immune pathways. GO analysis results showed that 46 terms were statistically significant, and KEGG enrichment results showed that 22 pathways were significantly enriched (p < 0.05, FDR <0.05). The top GO terms included adaptive immune response, T cell co-stimulation, MHC class II receptor activity, and transmembrane signaling receptor activity (Fig. 4A). In addition, most pathways generated from the KEGG pathway enrichment analysis were associated with immune responses, such as cell adhesion molecules, primary immunodeficiency, antigen processing and presentation, and chemokine signaling pathways (Fig. 4B).

Further identification of prognostic genes in the GEO database

To confirm whether the genes identified in the above steps have prognostic functions in other LUAD cohorts, we used the expression profiles and clinical data of three datasets in the GEO database for verification: GSE37745, GSE11969, and GSE50081. A total of 80 genes in GSE37745, 2703 genes in GSE11969, and 34 genes in GSE50081 were associated with survival. CSF2RB, inducible T cell kinase (ITK), and FLT3, and CD79A were found to be common among the GSE37745, GSE11969, and TCGA datasets (Fig. 5A). In addition, CCR4, CCR6, DOK2, AMPD1, and IGJ were found to be common among GSE37745, GSE50081, and TCGA (Fig. 5B). Kaplan–Meier survival curves were generated for the nine prognostic genes (Fig. 5C). The results showed that for all but except CCR4, the overall survival time of the low expression group was significantly lower than that of the high expression group (log-rank p < 0.05). This suggests that reduced expression of the eight genes (CSF2RB, ITK, FLT3, CD79A, CCR6, DOK2, AMPD1, and IGJ) and increased expression of one gene (CCR4) can predict poor survival of patients with for LUAD.

Immune cell infiltration analysis revealed the correlation between the expression of the DEGs and immunocytes

To confirm these findings, we used multivariate Cox regression and found that the five genes, including CCR6, ITK, CCR4, DOK2, and AMPD1, were an independent prognostic indicator for LUAD in their specific data sets (Figs. 6A–6D). We investigated immune infiltration using the TIMER deconvolution approach. Interestingly, the expression of the five genes identified positively correlated to the infiltration level of different immune cells, including B cells, CD4⁺ T cells, CD8⁺ T cells, macrophage cells, neutrophil cells, and dendritic cells (p <0.05). The expression levels of the five genes were negatively correlated with the tumor purity (p < 0.05; Figs. 7A–7II).

Evaluation of the immune and stromal status between the groups with low and high expression of CCR6

Multivariate Cox regression analysis identified that low expression of CCR6 is an independent prognostic factor of poor survival in the TCGA cohort of 484 patients with LUAD (HR 0.29, 95 % CI [0.12–0.72], p = 0.007). According to the results of UALCAN, mRNA and protein expression of CCR6 were all significantly upregulated in primary LUAD tissues as compared with normal samples (all p < 0.05; Figs. 8A and 8B). Immunohistochemistry staining obtained from The Human Protein Atlas database also demonstrated that CCR6 was upregulated (Figs. 8C and 8D). According to the median value of CCR6 expression levels, we divided the 484 LUAD patients into a low expression group and a high expression group. CIBERSORT and xCell were used to depict the immune cell and stromal cell landscape of CCR6 with high and low expression levels. Figs. 8E and 8F show the proportions of immune cells and stromal cells in the 484 LUAD tissues. The differential proportion of infiltration of immune and stromal cells in the group with low CCR6 expression and the group with high CCR6 expression is shown in Figs. 8G and 8H. CIBERSORT analysis suggests that high CCR6 expression is associated with recruitment of memory B cells, resting memory CD4 T cells, monocytes, M1 macrophages, resting dendritic cells, and resting mast cells (Fig. S1). xCell analysis suggests that high CCR6 expression is associated with recruitment of adipocytes, chondrocytes, endothelial cells, and fibroblasts (Fig. S1). A comparison of the tumor infiltration levels among tumors with different somatic copy number alterations (SCNAs) for a given gene was determined using TIMER. The results showed that enrichment of the six immune cell types (B cell, CD4⁺ T cell, CD8⁺ T cell, macrophage cell, neutrophil cell, and dendritic cell) were significantly downregulated in LUAD with SCNA of CCR6 (Fig. 8I).