This article is mentioned in:

Introduction

Colorectal cancer (CRC) is one of the most common types of cancer in men and women, with ~1.5 million new cases and ~0.5 million mortalities having been reported in 2013 in the United States (1). In recent decades, the mortality caused by CRC has decreased dramatically owing to the great improvement in early diagnosis and treatment (2). However, CRC remains a prominent global health problem that may be attributed to the lack of comprehensive and systemic understanding of the underlying molecular mechanisms of carcinogenesis.

The accumulation of specific genetic and epigenetic changes is considered to be the main molecular mechanism of tumorigenesis, as it can provide a selective growth advantage of tumor cells over neighboring normal cells (3). Among the epigenetic changes, the abnormal methylation of promoter CpG islands leading to the transcriptional inactivation of tumor suppressors is considered to be a common mechanism in several human malignancies including CRC (4). Epigenetic masking may participate in the cancerous transformation of colorectal epithelium by affecting the expression of tumor suppressor genes (4). Recent progress in CRC epigenetics studies indicated DNA methylation occurs in the early phase of cancer formation and in the premalignant phase of the adenoma-carcinoma sequence (5). Thus, identifying the epigenetic alterations would be of great value in the early detection of cancers and cancer relapse, as well as in monitoring the response of cancers to therapies (6).

Epigenetically silenced genes by hypermethylation can be reactivated by 5-aza-2′-deoxycytidine (5-aza-dC), which is able to inhibit DNA methylation (7). In addition, the re-expression of silenced genes caused by 5-aza-dC has been demonstrated in various types of tumors in a dose- and duration-dependent manner (7). The application of 5-aza-dC in expression microarray analysis is considered to be a useful approach for identifying cancer-associated methylated genes (8).

In order to elucidate silenced genes with abnormal methylation in CRC, Khamas et al (9) performed a genome-wide expression screening in 5 CRC cell lines prior and subsequent to 5-aza-dC treatment, and subsequently combined the data with CRC-specific gene expression profiling array. The gene expression data set established by Khamas et al (9) was submitted to the Gene Expression Omnibus (GEO) with the accession number GSE32323. In the present study, the microarray was downloaded and analyzed to identify potential targets for 5-aza-dC by oligonucleotide microarray analysis. A co-expression network of CRC-specific gene expression profile was constructed using the context likelihood of relatedness (CLR) algorithm to identify the signaling pathways in which these targets were involved, thus revealing the function of the selected identified genes.

Materials and methods

Affymetrix microarray data

Transcriptional profile of GSE32323 (9) was extracted from the GEO database (http://www.ncbi.nlm.nih.gov/geo/), which was based on the platform of Affymetrix Human Genome U133 Plus 2.0 Array. A total of 44 chips were available for further analysis, including 17 pairs of cancer and non-cancerous tissues from CRC patients, and expression profiles of 5 CRC cell lines.

Data preprocessing

The raw probe-level data in CEL files were initially converted into expression measures. Robust multiarray average background correction, quantile normalization and probe summarization were subsequently performed in the R (version: 3.0.3, March, 2014) affy package (http://www.bioconductor.org/packages/release/bioc/html/affy.html) (10), and the processed expression matrixes were acquired. For each sample, the expression values of all probes for a given gene were expressed as a single value by taking an average of the values.

Differentially expressed genes (DEGs) analysis

The limma (11) package (http://www.bioconductor.org/packages/2.9/bioc/html/limma.html) in R was used to identify DEGs in the present study. The following thresholds were set for filtering DEGs: |log2 fold-change (FC)|>1.0 and P-value<0.05. The original P-values were adjusted using Benjamini-Hochberg procedure to correct for multiple comparisons. For CRC cell lines, gene differential expression was calculated from each sample prior and subsequent to 5-aza-dC treatment. Only DEGs with co-upregulated or co-downregulated expression in ≥3 cell lines were selected and grouped as ‘DEG1’. For CRC tissues, DEGs in CRC tissue samples compared to non-cancerous tissue were identified and grouped as ‘DEG2’. A comparison was subsequently performed between ‘DEG1’ and ‘DEG2’. The DEGs that simultaneously upregulated in ‘DEG1’ and downregulated in ‘DEG2’, or simultaneously downregulated in ‘DEG1’ and upregulated in ‘DEG2’ were defined as reverse-overlapped DEGs, and were screened for further analysis.

Co-expression network inference and analysis

To identify interactions between genes, the CLR algorithm was used to construct the co-expression network (DEG2.CEN) in the CRC tissue samples. The CLR threshold was set as 2.5. The sub-network (roDEG.CEN) that associated with reverse-overlapped DEGs was selected from DEG2.CEN by employing the package MINET (http://www.bioconductor.org/packages/3.4/bioc/html/minet.html) (12) implemented in R/Bioconductor (version: 3.4; http://www.bioconductor.org/) and subsequently visualized using Cytoscape (version 3.4.0; http://www.cytoscape.org/) (13).

The CLR algorithm (14) is an extension of the relevance network approach, which increases the contrast between physical interactions and indirect associations and takes into account the context of each interaction and association. Links are assigned based on the mutual information (MI) that can accommodate non-linear associations between pair-wise gene expression patterns. The most probable interactions are those whose MI scores stand significantly above the background distribution of MI scores. The MI for two discrete random variables X and Y is defined as:

where xi, yj represent ith and jth expression level of X and Y, respectively. P(xi) and P(yj) are the marginal probability distributions. P(xi, yj) is the joint distribution that the expression levels of X and Y are xi and yj, respectively (14).

Pathway enrichment analysis

The Database for Annotation Visualization and Integrated Discovery (15) provides a comprehensive set of functional annotation tools to elucidate biological meaning behind large lists of genes or proteins. Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis was performed in the present study for functional pathway enrichment of reverse-overlapped DEGs with P<0.05 selected as a cut-off criterion. In addition, the enriched functional pathways of reverse-overlapped DEGs and roDEG.CEN were integrated. Thus, a reverse-overlappedDEG would be correlated with a particular enriched functional pathway if the neighboring genes of this particular reverse-overlapped DEG were involved.

Results

Identification of DEGs

For database GSE32323 (|log2 FC|>1.0; P<0.05), a total of 59 DEG1 s in five CRC cell lines prior and subsequent to 5-aza-dC treatment, including 48 upregulated and 11 downregulated genes, were identified. A total of 1,341 DEG2 s with 675 upregulated and 666 downregulated genes were selected when CRC and normal tissue samples were compared. Following comparing between the ‘DEG1’ and ‘DEG2’ groups, 10 reverse-overlapped DEGs were selected (Table I). Among the identified reverse-overlapped DEGs, 6 genes [amine oxidase, copper containing 3, fibulin-2 (FBLN2), uridine phosphorylase 1, cysteine-rich protein 1, protein phosphatase 1, regulatory inhibitor subunit 14A (PPP1R14A; CPI-17) and heat shock 70 kDa protein 2] were downregulated in CRC tissue sample, but upregulated in CRC cell lines following treatment with 5-aza-dC.

Table I. The characteristics of identified reverse-overlapped differentially expressed genes.

Table I.

The characteristics of identified reverse-overlapped differentially expressed genes.

Gene	EntrezID	Name	aDE_State	Degree
AOC3	8,639	Amine oxidase, copper containing 3	down	76
FBLN2	2,199	Fibulin 2	down	66
UPP1	7,378	Uridine phosphorylase 1	down	64
MIPEP	4,285	Mitochondrial intermediate peptidase	up	63
RRM2	6,241	Ribonucleotide reductase M2	up	59
KIF11	3,832	Kinesin family member 11	up	54
PPP1R14A	94,274	Protein phosphatase 1, regulatory inhibitor subunit 14A	down	54
HSPA2	3,306	Heat shock 70 kDa protein 2	down	47
SLC12A2	6,558	Solute carrier family 12 (sodium/potassium/chloride transporter), member 2	up	45
CRIP1	1,396	Cysteine-rich protein 1 (intestinal)	down	39

a DE_State, differential expression state in colorectal cancer tissue. Up, upregulated; down, downregulated, Degree, the number of edges for a differentially expressed genes in the co-expression network.

Co-expression network of DEGs

The co-expression network of DEGs (DEG2.CEN) was constructed by employing the CLR algorithm. The co-expression network was based on the DEG2 expression profile in the CRC tissue samples and the sub-network (roDEG.CEN) that correlated with the identified reverse-overlapped DEGs (Fig. 1). There were 374 nodes and 567 edges in roDEG.CEN. The number of edges emerging from a node was set as the degree of a DEG, as shown in Table I.

Figure 1. Co-expression network of reverse-overlapped differentially expressed genes in colorectal cancer. Red nodes represent the upregulated genes and green nodes represent the downregulated genes. Triangle nodes represent reverse-overlapped genes; circle nodes represent the differentially expressed genes.

Functional pathway analysis of network

Following integrating the roDEG.CEN with the enriched functional pathway, the downregulated genes were enriched in the drug metabolism pathway, while the upregulated genes were enriched in the cellular tumor antigen p53, cell cycle, oocyte meiosis and nucleotide-binding oligomerization domain-like receptors (NLR) signaling pathways (Fig. 2).

Figure 2. Sub-network of associated Kyoto Encyclopedia of Genes and Genomes signaling pathways in colorectal cancer. Red triangle nodes represent upregulated genes and green triangle nodes represent downregulated genes. Red rectangles represent upregulated gene-enriched pathways and green rectangles represent downregulated gene-enriched pathways. AOC3, amine oxidase, copper containing 3; FBLN2, fibulin 2; KIF11, kinesin family member 11; HSPA2, heat shock 70 kDa protein 2; MIPEP, mitochondrial intermediate peptidase; PPP1R14A, protein phosphatase 1, regulatory inhibitor subunit 14A; RRM2, ribonucleotide reductase M2; SLC12A2, solute carrier family 12 (sodium/potassium/chloride transporter), member 2; UPP1, uridine phosphorylase 1.

Discussion

Significant progress has been achieved in the diagnosis and treatment of CRC. However, CRC remains the third most common cancer worldwide (16). In the present study, the mRNA expression profile of GSE32323 was downloaded and DEGs were analyzed. The DEGs in 5 CRC cell lines prior and subsequent to 5-aza-dC treatment were combined with the CRC-specific gene expression profiling array. A total of 6 reverse-overlapped DEGs were obtained. These reverse-overlapped DEGs were downregulated in the CRC tissue samples but upregulated in CRC cell lines following 5-aza-dC treatment. The CLR algorithm was employed to construct a co-expression network of CRC-specific gene expression profile and the sub-network that correlated with reverse-overlapped DEGs was selected. Furthermore, functional pathway analysis identified the reverse-overlapped DEGs enriched in a number of critical cellular pathways, including p53, cell cycle and the NLR signaling pathway.

The 6 reverse-overlapped DEGs identified in the present study were involved in a variety of cellular functions. Among them, two genes (FBLN2 and PPP1R14A) have been previously reported to be hypermethylated (17,18). FBLN2, an extracellular matrix (ECM) protein, is recognized as a multifunctional binding protein (19). Due to its ability to mediate interactions between diverse ECM components, FBLN2 plays an important role in the maintenance of extracellular structures such as the basement membranes, as well as contacts between cells and ECM (20,21). FBLN2 has also been reported to have the opposite effects in pathological conditions including cancer. The pro-tumor effects of FBLN2 were demonstrated in pancreatic cancer cells (22); however, an increasing number of studies indicate that FBLN2 may act as an anti-angiogenic factor in various types of cancer, including nasopharyngeal carcinoma (18,23,24), as well as an anti-tumor factor in breast cancer cells (25). In addition, FBLN2 has been previously demonstrated to be epigenetically silenced in B cell acute lymphoblastic leukemia (26) and methylated in breast and other epithelial cancer types (27). In the present study, FBLN2 was downregulated in the CRC tissue sample but upregulated in CRC cell lines following treatment with 5-aza-dC.

PPP1R14A was also identified to be methylated in CRC. Following treatment with 5-aza-dC, the expression of PPP1R14A increased significantly. PPP1R14A is a phosphorylation-dependent inhibitory protein of smooth muscle myosin phosphatase activity (28), which has been reported to be an epigenetic biomarker in CRC (29). The PPP1R14A gene has also been reported to be associated with growth arrest and DNA damage (30). Following treatment with anti-cancer drugs, including Fluorouracil, PPP1R14A is upregulated (31). In addition, a previous study has demonstrated that PPP1R14A is aberrantly methylated in human esophageal squamous cell carcinoma (18) and various types of B-cell non-Hodgkin lymphoma (32).

A co-expression network based on the data of the CRC-specific gene expression profile was constructed using the CLR algorithm and the sub-network corresponding to reverse-overlapped DEGs was selected. Bias from uneven conditions of sampling, upstream regulation and inter-laboratory variations in microarray can make it difficult to infer network between genes (14). CLR algorithm increases the contrast between the physical interactions and the indirect associations by taking the context of each interaction and association into consideration, thus minimizing the bias from these factors (14). Therefore, the CLR algorithm is an attractive method to use for the identification of indirect links and for uncovering associations between genes within co-regulated communities. The CLR algorithm estimates a likelihood of the MI score for a particular pair of genes by comparing the MI values for that particular pair of genes to a background distribution of MI values. The most probable interactions are those whose MI scores are significantly above the background distribution of MI scores (11). Following KEGG pathway enrichment, the reverse-overlapped DEGs identified in the present study were demonstrated to be enriched in several pathways, including p53, cell cycle and NLR signaling pathways. As previously reported, p53, cell cycle and NLR receptor signaling pathways are closely associated with tumorigenesis and metastasis (33–35). The results of co-expression network analysis indicated the identified reverse-overlapped DEGs may be important tumor suppressors and are inactivated by methylation in CRC.

Compared to the previous study published by Khamas et al (9), the criteria used for selecting DEGs in the present study were different. Khamas et al (9) selected probe sets from cell lines using a combination of two criteria: Upregulation of gene expression in ≥4 CRC cell lines and FC>1.6 in at least one cell line. In the present study, DEGs were selected if they were co-upregulated or co-downregulated in ≥3 cell lines and at the same time FC>2 in at least one cell line. Due to the difference in threshold selection for DEGs, the identified genes in the present study were different from the previously published report (9). Furthermore, a co-expression network was constructed using the CLR algorithm and the sub-network correlated with the identified genes was selected. Pathway enrichment analysis was performed to reveal the function of identified genes.

There were a number of limitations in the present study. The expression of the identified targets (FBLN2 and PPP1R14A), as well as the association of the methylation status of these genes with the development of CRC, remains to be confirmed by future investigations.

In the present study, two silenced genes FBLN2 and PPP1R14A with abnormal methylation in CRC were identified. Furthermore, the co-expression network of identified DEGs in the CRC tissue samples was constructed by employing the CLR algorithm and a sub-network of reverse-overlapped DEGs was selected. Functional pathway analysis indicated that the identified reverse-overlapped DEGs were enriched in a number of pathways, including p53, cell cycle and NLR signaling pathway. The results of the present study may provide novel targets for the treatment of CRC.

References