Identification of common tumor signatures based on gene set enrichment analysis

Abstract

The identification of common tumor signatures can discover the shared molecular mechanisms underlying tumorgenesis whereby we can prevent and treat tumors by a system intervention. We identified tumor-associated signatures including pathways, transcription factors, microRNAs and gene ontology categories by analyzing gene sets for differential expression between normal vs. tumor phenotypes classes in various tumor gene expression datasets. We obtained the common tumor signatures based on their identified frequencies for different tumor types. Some shared signatures important for various tumor types were uncovered and discussed. We proposed that the interventions aiming at both the shared tumor signatures and the tissue-specific tumor signatures might be a potential approach to overcoming cancer.

1 Introduction

A large amount of studies have revealed that cancer has been associated with the genetic and genomic changes [1–2]. As the microarray technology have enabled the simultaneous measurement of the expression levels of tens of thousands of genes in a single experiment [3], the use of microarray technology to analyze gene expression changes in tumor tissues is a powerful tool for uncovering the molecular mechanisms underlying cancer [4]. At the same time, the biology of cancer is extremely intricate so that a simple genetic or genomic perspective is insufficient to understand it. Only by attaining more complete cancer-associated molecular profiles such as pathways and transcriptional regulatory circuits, could we comprehend the disease more clearly.

Gene expression profiling has been widely used for identification of cancerous biomarkers whereby we can improve cancerous diagnosis, treatment and prognosis [5–19]. Moreover, since it has been recognized that a gene set could be more biologically significant than individual genes considering gene interactions, the microarray-based gene set enrichment analysis has been investigated on the assumption that it could provide additional insights into the cancer biology [20–22]. Generally speaking, cancer is a systems biology disease [23–24]. To understand the disease at a system level, identification of common tumor signatures among multiple tumor tissues is a critical avenue, although a substantial number of tumor signatures might be tissue-specific.

In the present study, we identified the common tumor signatures closely associated with various tumor types. The signatures include four types: pathways, transcriptional factors (TFs), microRNAs (miRNAs) and gene ontology (GO) categories, which were identified through the gene set enrichment analysis based on gene expression profiling. The signatures suggested some basic molecular mechanisms underlying tumor, and might imply potential routes of interventions for cancerous diagnosis and treatment.

2 Methods and Materials

2.1 Methods

We identified important pathways, TFs, miRNAs and GO categories by analyzing gene sets for differential expression between normal vs. tumor phenotypes classes. The LS or KS permuation test and Efron-Tibshirani’s GSA maxmean test were used to determine the significant gene sets at 0.05 significance level for identification of pathways, TFs and miRNAs, and 0.0001 significance level for GO categories. The pathways (BioCarta) related to the significant gene sets were identified. The TFs were identified by the gene sets, in each of which all genes were experimentally verified to be targets of the same transcription factor (TF). Each miRNA potentially targeting all the genes in one of the gene sets was identified. The identification of important pathways, TFs and miRNAs was performed with the gene set expression class comparison tool in BRB-ArrayTools, which is an integrated software package for the visualization and statistical analysis of DNA microarray gene expression data [25].

2.2 Materials

We analyzed 23 human gene expression datasets involving 15 tumor types (Table 1) [26]. For each dataset, we carried out class comparison algorithm to identify informative pathways, TFs, miRNAs and GO categories relevant to the tumor(s).

Table 1.

Summary of human tumor gene expression datasets

Tumor Type	# Datasets	Reference
Brain Cancer	1	[83]
Cervical Cancer	1	[84]
Embryonal Cancer	1	[85]
Esophageal Cancer	1	[86]
Gastric Cancer	2	[87–88]
Head and Neck Cancer	1	[89]
Lung Cancer	2	[90–91]
Melanoma	1	[92]
Mesothelioma	1	[93]
Pancreatic Cancer	1	[94]
Prostate Cancer	5	[95–99]
Renal Cancer	2	[100–101]
Soft Tissue Sarcoma	1	[102]
Thyroid Cancer	1	[103]
Uterine Leiomyoma	2	[104–105]

3 Results and Analysis

3.1 Identification of tumor-associated pathways

In the total of 26 class comparisons, we identified 25 pathway sets significant at 0.05 threshold level. The 25 sets encompassed 304 different pathways, 17 of which appeared at least in 10 different sets, suggesting that they were associated with at least 10 different types of tumors. Table 2 lists the 17 most frequent identified pathways. The complete 304 pathways identified are presented in the supplementary Table S1. From Table 2, we can see that the most common tumor-associated pathways are often involved in cell cycle regulation, mitogen-activated protein kinase (MAPK) signaling, epidermal growth factor receptor (EGFR), metabolism, oxidative stress, cell motility etc. Many studies have come to the similar conclusions [27–43].

Table 2.

Seventeen pathways frequently identified in tumors

Pathway Symbola	Pathway Namea	Frequencyb
Vitcb	Vitamin C in the Brain	14
Mhc	Antigen Processing and Presentation	13
No1	Actions of Nitric Oxide in the Heart	13
Keratinocyte	Keratinocyte Differentiation	12
Cbl	Mediated ligand-induced downregulation of EGF receptors	12
Biopeptides	Bioactive Peptide Induced Signaling Pathway	12
Cell Cycle	Cyclins and Cell Cycle Regulation	11
Cftr	Cystic fibrosis transmembrane conductance regulator (CFTR) and beta 2 adrenergic receptor (b2AR) pathway	11
Erad	ER-associated degradation Pathway	11
Arenrf2	Oxidative Stress Induced Gene Expression Via Nrf2	11
Vobesity	Visceral Fat Deposits and the Metabolic Syndrome	11
PlateletApp	Platelet Amyloid Precursor Protein Pathway	11
Mcalpain	mCalpain and friends in Cell motility	10
Ranms	Role of Ran in mitotic spindle regulation	10
P38 Mapk	p38 MAPK Signaling Pathway	10
Tcra	Lck and Fyn tyrosine kinases in initiation of TCR Activation	10
Dsp	Regulation of MAP Kinase Pathways Through Dual Specificity Phosphatases	10

^aBioCarta pathway symbol and name.

^bThe occurrence times of the pathway in the 25 pathway sets identified.

3.2 Identification of tumor-associated TFs

We identified 26 sets of TF targets significant at 0.05 threshold level. There were 99 different TFs identified relevant to the 26 sets, 22 of which were associated with more than 1/3 of the 26 target sets (Table 3). The most frequently identified TF was c-Myc with 62% occurrence rate, and the next ones were E2F-4, MYB and TP53 all with 58% occurrence rate. All the 99 TFs and their occurrence rates were provided in the supplementary Table S2.

Table 3.

Twenty-two TFs frequently identified in tumors

TF	Frequency
c-Myc	16
E2F-4	15
MYB	15
P53	15
SMAD1	14
TAL1	13
TFAP2A	11
JUN	11
PPARA	11
MYBL2	10
SP2	10
EPAS1	10
FLI1	10
STAT3	10
E2F-1	9
PPARD	9
STAT5B	9
NFIC	9
POU2F2	9
PGR	9
ETS2	9
HIF1A	9

Evidently, c-Myc is one of the most important TFs relevant to cancer [44]. Since c-Myc target genes are often involved in the critical mechanisms underlying cancer like cell cycle regulation, apoptosis, metabolism etc., the dysregulation of c-Myc greatly contributes to cancer [45–47]. Table 3 shows that two members of the MYB family of TFs: MYB and MYBL2, have important relevance to cancer. Indeed, many studies have strongly suggested that they played a role in tumorgenesis [48–58]. The two members of the E2F TF family: E2F-1 and E2F-4, have been revealed to be associated with cancer [59–63]. An extremely important tumor-associated TF p53 is also presented in Table 3. The role played by p53 in tumorgenesis has been well-recognized [64–66].

3.3 Identification of tumor-associated miRNAs

We identified 24 sets of miRNA targets significant at 0.05 threshold level. The 24 sets were involved in 587 different miRNAs, 34 of which were associated with at least one half of the 24 sets. The 34 miRNAs are listed in Table 4 and the 587 miRNAs are provided in the supplementary Table S3. The most frequently identified three miRNAs were miR-29b, miR-29c and miR-29a, members of miR-29 miRNA gene family. The miR-29 family has been proven to be strongly involved in cancer [67–71]. Table 4 shows that another miRNA gene family miR-30 seems to be closely associated with various tumors. There has been some evidence to support this conclusion [72–74]. In addition, miR-19 and miR-526 miRNA gene families appear to be involved in various tumors (see Table 4). Some literatures have suggested their roles in tumorgenesis [75–77]. The other miRNAs with high frequencies like miR-181c, miR-590, miR-212, miR-338 and miR-202 have also been reported to be associated with tumorgenesis [78–82]. It should be noted that most of the cited support literatures were from recent publications, while all the gene expression datasets studied were from earlier publications, indicating that our inference and prediction were reliable to a certain degree.

Table 4.

Thirty-four miRNAs frequently identified in tumors

miRNA	Frequency
miR-29b	18
miR-29c	17
miR-29a	16
miR-30e-3p	14
miR-547	14
miR-181c	13
miR-30e-5p	13
miR-590	13
miR-212	13
miR-603	13
miR-669b	13
miR-338	13
miR-202	13
miR-1	12
miR-103	12
miR-181a	12
miR-19a	12
miR-19b	12
miR-200b	12
miR-21	12
miR-30a-5p	12
miR-30c	12
miR-30d	12
miR-330	12
miR-518a	12
miR-526c	12
miR-562	12
miR-31	12
miR-128a	12
miR-30a-3p	12
miR-182	12
miR-669c	12
miR-526b	12
miR-95	12

3.4 Identification of tumor-associated GO categories

We identified 25 sets of GO categories significant at 0.0001 threshold level. The 25 sets were involved in 2273 different GO terms, 39 of which were concerned with at least five different tumor types. The 39 GO terms are listed in Table 5 and the complete 2273 GO terms are provided in the supplementary Table S4.

Table 5.

Thirty-nine GO terms frequently identified in tumors

GO Terms	Frequency
antigen processing and presentation of peptide antigen	7
MHC protein complex	7
antigen processing and presentation of peptide antigen via MHC class I	6
MHC class I receptor activity	6
cytosolic ribosome	6
extracellular matrix part	6
collagen	6
fibrillar collagen	6
cell proliferation	6
regulation of cell proliferation	6
regulation of biological quality	6
basement membrane	5
collagen metabolic process	5
ribosomal subunit	5
extracellular matrix structural constituent	5
peptide cross-linking	5
glycosaminoglycan binding	5
small ribosomal subunit	5
blood vessel development	5
vasculature development	5
response to external stimulus	5
response to wounding	5
anatomical structure morphogenesis	5
tissue development	5
response to organic substance	5
cell differentiation	5
response to chemical stimulus	5
cell development	5
organ development	5
cellular developmental process	5
proteinaceous extracellular matrix	5
extracellular space	5
extracellular matrix	5
extracellular region part	5
structural molecule activity	5
cytoskeletal protein binding	5
cytoskeleton	5
identical protein binding	5
response to stress	5

Table 5 shows that the genes involved in immune, metabolism, development, cell proliferation and differentiation, damage response etc., are most relevant to tumorgenesis.

4 Discussion and Conclusions

The microarray analysis of gene expression profiling of tumor tissues can not only discover the marker genes relevant to tumor malignancies, but also identify the informative gene sets to reveal the molecular mechanisms underlying tumor. The gene set enrichment analysis is a strong supplement to the individual gene analysis as it can potentially make use of the gene interaction information, which is often missed by the individual gene analysis. In this study, we used the gene set enrichment analysis to identify the shared tumor signatures whereby we could reveal the common mechanisms underlying different types of tumors, and therefore might provide a basic reference to tumor prevention and treatment.

Our tumor signatures included tumor-associated pathways, TFs, miRNAs and GO catagories. Each of the tumor signatures was related to multiple genes and identified based on gene set comparison. Therefore, these kinds of signatures may imply the mechanisms underlying tumor at a close system level. Since it has been recognized that cancer was a systems biology disease, the systems interventions aiming at cancer prevention and treatment could contribute to conquering cancer. Of course, some tumors might involve tissue-specific signatures, and therefore the tissue-specific interventions are necessary for treating the tumors in addition to the systems interventions.

The reliability of the results obtained by the present study is mainly affected by two factors: the quality of microarrays and the statistical power. Microarrays, especially poor-qualified microarrays, often contain a large amount of noises, which are prone to result to identification of false signatures. In addition, the discovery of common tumor signatures by their occurrence frequencies in different tumor types is not based on sufficiently strong statistical power so that some signatures might be identified by chance. This is a work needed to be improved in the future.

Supplementary Material

Table S1. The list of 304 pathways identified

Table S1. The list of 304 pathways identified

Table S2. The list of 99 TFs identified

Table S2. The list of 99 TFs identified

Table S3. The list of 587 miRNAs identified

Table S3. The list of 587 miRNAs identified

Table S4. The list of 2273 GO terms identified

Table S4. The list of 2273 GO terms identified