Hub Genes in Non-Small Cell Lung Cancer Regulatory Networks

Abstract

There are currently no accurate biomarkers for optimal treatment selection in early-stage non-small cell lung cancer (NSCLC). Novel therapeutic targets are needed to improve NSCLC survival outcomes. This study systematically evaluated the association between genome-scale regulatory network centralities and NSCLC tumorigenesis, proliferation, and survival in early-stage NSCLC patients. Boolean implication networks were used to construct multimodal networks using patient DNA copy number variation, mRNA, and protein expression profiles. T statistics of differential gene/protein expression in tumors versus non-cancerous adjacent tissues, dependency scores in in vitro CRISPR-Cas9/RNA interference (RNAi) screening of human NSCLC cell lines, and hazard ratios in univariate Cox modeling of the Cancer Genome Atlas (TCGA) NSCLC patients were correlated with graph theory centrality metrics. Hub genes in multi-omics networks involving gene/protein expression were associated with oncogenic, proliferative potentials and poor patient survival outcomes (p < 0.05, Pearson's correlation). Immunotherapy targets PD1, PDL1, CTLA4, and CD27 were ranked as top hub genes within the 10th percentile in most constructed multi-omics networks. BUB3, DNM1L, EIF2S1, KPNB1, NMT1, PGAM1, and STRAP were discovered as important hub genes in NSCLC proliferation with oncogenic potential. These results support the importance of hub genes in NSCLC tumorigenesis, proliferation, and prognosis, with implications in prioritizing therapeutic targets to improve patient survival outcomes.

Keywords: CRISPR-Cas9; RNAi; biomarkers; hub genes; multi-omics networks; non-small cell lung cancer; patient survival; proliferation; therapeutic targets.

Figure 1

Concordance of correlation coefficients between seven centrality metrics of the selected networks with NSCLC tumorigenesis, proliferation, and patient survival. Tumorigenesis was described with the t statistics (two-sample t tests) of tumor vs. NAT differential expression in mRNA (n_tumor = 51, n_NAT = 49) and protein (n_tumor = n_NAT = 103) datasets in Xu’s LUAD patients [39]. Proliferation was assessed in human NSCLC cell lines with dependency scores in in vitro CRISPR-Cas9 (n = 94) and RNAi (n = 92) genome-wide screening. Patient survival was represented by hazard ratios in univariate Cox modeling of TCGA RNA sequencing data of NSCLC patient tumors (n = 1016). Each cell in the figure showed the number of metrics with concordant significant Pearson’s correlation coefficients in a pair of compared networks: I. CNV–CNV networks (GSE28582 and GSE31800); II. CNV-mediated GE networks (GSE28582 and GSE31800); III. gene co-expression networks (GSE28582 and GSE31800); IV. gene co-expression networks (Xu’s LUAD tumors and NATs [39]); V. mRNA-mediated protein expression networks (Xu’s LUAD tumors and NATs [39]); VI. protein co-expression networks (Xu’s LUAD tumors and NATs [39]).

Figure 2

Distributions of centrality metrics in multi-omics networks with CD27, CTLA4, PD1, or PDL1 ranked within the top 10th percentile. Each subplot represented a centrality metric. (A) Degree centrality. (B) In-degree centrality. (C) Out-degree centrality. (D) Eigenvector centrality. (E) Betweenness centrality. (F) Closeness centrality. (G) VoteRank centrality. Each violin plot showed the distribution of the centrality metric in one specific network: I. CNV-mediated gene expression network in GSE28582; II. mRNA co-expression network in GSE28582; III. CNV-mediated gene expression network in GSE31800; IV. mRNA co-expression network in GSE31800; V. mRNA-mediated protein expression network in tumors of Xu’s LUAD patient cohort [39]; VI. mRNA co-expression network in tumors of Xu’s LUAD patient cohort [39].

Figure 3

The comparison of centrality metrics of two published multi-omics networks vs. randomly selected networks with the same number of genes. Network A contains 30 genes in the CD27, PD1, and PDL1 multi-omics network in NSCLC tumors [28]. Network B contains 66 genes in the multi-omics network of the 7-gene prognostic signature in NSCLC tumors [27]. The p values showed the percentage of randomly selected genes having a larger averaged centrality (except VoteRank) than networks A and B. The p value of VoteRank centrality showed the percentage of randomly selected genes having a lower averaged rank (one-tailed Wilcoxon rank sum test, p < 0.05) than networks A and B. Each column showed a centrality metric: I. degree centrality; II. in-degree centrality; III. out-degree centrality; IV. eigenvector centrality; V. closeness centrality; VI. betweenness centrality; VII. VoteRank centrality.

Figure 4

Gene and protein co-expression network of selected seven genes (BUB3, DNM1L, EIF2S1, KPNB1, NMT1, PGAM1, and STRAP) in NSCLC tumors.