Distinct signatures of lung cancer types: aberrant mucin O-glycosylation and compromised immune response

Abstract

Background: Genomic initiatives such as The Cancer Genome Atlas (TCGA) contain data from -omics profiling of thousands of tumor samples, which may be used to decipher cancer signaling, and related alterations. Managing and analyzing data from large-scale projects, such as TCGA, is a demanding task. It is difficult to dissect the high complexity hidden in genomic data and to account for inter-tumor heterogeneity adequately.

Methods: In this study, we used a robust statistical framework along with the integration of diverse bioinformatic tools to analyze next-generation sequencing data from more than 1000 patients from two different lung cancer subtypes, i.e., the lung adenocarcinoma (LUAD) and the squamous cell carcinoma (LUSC).

Results: We used the gene expression data to identify co-expression modules and differentially expressed genes to discriminate between LUAD and LUSC. We identified a group of genes which could act as specific oncogenes or tumor suppressor genes in one of the two lung cancer types, along with two dual role genes. Our results have been validated against other transcriptomics data of lung cancer patients.

Conclusions: Our integrative approach allowed us to identify two key features: a substantial up-regulation of genes involved in O-glycosylation of mucins in LUAD, and a compromised immune response in LUSC. The immune-profile associated with LUSC might be linked to the activation of three oncogenic pathways, which promote the evasion of the antitumor immune response. Collectively, our results provide new future directions for the design of target therapies in lung cancer.

Keywords: Co-expression; Differential expression analysis; Lung adenocarcinoma; Lung squamous cell carcinoma; RNA-Seq; Soft clustering; Survival analysis; TCGA.

Fig. 1

The workflow illustrates the steps used in our study. We used the R/CRAN package DiagrammeR v.1.0.1 to illustrate the workflow

Fig. 2

Comparison of DEA results with different DEA protocols. For sake of clarity, we reported the up-regulated genes (a,c) and down-regulated genes in (b,d) in LUAD (a,b) and LUSC (c,d), respectively (see https://github.com/ELELAB/LUAD_LUSC_TCGA_comparison for the whole set of results). The analyses have performed with the R/CRAN package UpSetR v.1.3.3

Fig. 3

Soft-clustering across lung cancer stages of tumor progression. Each cluster describes an expression pattern in the dataset through the four stages of cancer i.e. stages I, II, III and IV. Blue and purple lines correspond to genes with high cluster membership value (i.e., m > 0.56). a Table with the genes belonging to each cluster and their m value is reported in our Github repository. The LUAD (a) and LUSC (c) clusters are showed along with the corresponding dotplots (b and d, respectively) for the top enriched pathways for each cluster. In the dotplots, the plots are colored according to the p-values from blue (high p-values and low enrichment) to red (low p-values and high enrichment). The cluster plots have been generated with the R/Bioconductor package Mfuzz version 2.36.0

Fig. 4

Co-expression modules and their network in LUSC (a) and LUAD (b). The modules which collect genes and pathways that differentiate LUSC from LUAD are shown in the figure. The analyses have been carried out with the R/Bioconductor package CEMITools version 3.10

Fig. 5

Network of transcription factors and their target genes in the co-expressed module 1 of LUAD (a) or LUSC (b). The data have been plotted using Cytoscape version 3.3.0

Fig. 6

In silico validation of the candidate genes using other transcriptomics data for lung cancer. The figure shows the groups of candidate genes that were able to separate LUAD and LUSC samples selected by the two microarray studies used for validation. Panel a is related to the first study (see also Additional file 4: Figure S3), panel b to the second dataset (see In silico validation of the candidate genes on independent cohorts section for more details). The figure has been generated with R/CRAN package gplots version 3.0.1.1

Fig. 7

Illustration of the three oncogenic pathways to evade tumor immune response, which could be activated in LUSC. The pathways that are colored in the figure are the ones where the driving down- and up-regulated genes have been found deregulated also in our dataset of DE genes in LUSC. The illustration has been generated with Adobe Photoshop CC 2014