Identification of potential core genes in metastatic renal cell carcinoma using bioinformatics analysis

Abstract

Most cases of mRCC without an early finding are not candidates for curative therapies, which may be one of the reasons for the poor patient prognosis. Therefore, candidate markers to diagnose the disease and treatment with high efficiency are urgently demanded. Three datasets of mRNA microarray have been assessed to discover DEGs between tissues from metastatic RCC and RCC. 111 DEGs in total were identified according to the expression profile result of genes in the database of GEO. Enrichment analyses for GO and KEGG have been conducted to reveal the interacting activities in the DEGs. A network of PPI has been established to reveal the interconnection among the DEGs, and we selected 10 hub genes. Subsequently, the disease-free survival rate and total survival rate analysis for the hub genes have been carried out with the method of Kaplan-Meier curve. RCC patients with CDH11, COL3A1, COL5A1, COL5A2, COL6A3 and COL11A1 alteration showed worse overall survival. Nonetheless, RCC patients with CDH11, COL3A1, COL5A1, COL5A2 and COL11A1 alteration showed worse disease-free survival. In the Jones Renal dataset, mRNA levels of 10 hub genes were associated with metastasis, and the gene expression level in patients with mRCC was higher than that in patients without metastasis. COL5A1, COL6A3 and COL11A1 expression levels were remarkably related to RCC patient survival rate using UALCAN. COL5A1, COL6A3 and COL11A1 were positively correlated with each other in RCC. These genes have been recognized as genes with clinical relevance, revealing that they might have important roles in carcinogenesis or development of mRCC.

Introduction

Throughout the world, about 2.4% of all the malignancy cases are renal cell cancer (RCC). The new cancer cases annually was approximately 337,000 in total [1]. Although most patients with RCC present with early stage renal tumors, up to 30% patients suffered from advanced disease when the diagnosis was made, and the 5-year survival rate was about 12% [2]. Renal cell carcinoma (RCC) is considered to be a group of various histopathology types, among which clear cell RCC is most commonly seen [3]. Approximately 25% of the patients would first show up with incurable and advanced disease, and 1/3 of the patients would ultimately develop into metastatic renal cell carcinoma (mRCC) after initial treatment [4]. During the past ten years, the therapy of metastatic RCC has developed significantly using several agents of the family of vascular endothelial growth factor (VEGF) which specifically aimed at tyrosine kinase inhibitors (TKI) [5,6]. Nevertheless, full responses to the therapies hardly appeared (< 1%), and most patients having initial response would go through cancer development [5,7]. Hence, it is of vital importance to reveal the accurate mechanisms working in the process of proliferation, recurrence and carcinogenesis of mRCC.

In the past ten years, microarray method was used widely to measure the genes (DEGs) that expressed differentially, and also bioinformatics approaches have been applied to obtain the data about profile of gene expression which could be downloaded in the database of Gene Expression Omnibus (GEO). In our research, 3 datasets of mRNA microarray from GEO were downloaded and analyzed in order to select DEGs between tissues from mRCC and RCC. In a word, 10 hub genes and 111 DEGs in total were selected, and they might be potential biomarkers for mRCC.

Material and methods

Data of microarray

GEO is open to public and known as a useful genomics database which contains chips, microarrays and high throughout gene expression level data (http://www.ncbi.nlm.nih.gov/geo) [8]. We downloaded 3 gene expression profiles [GSE22541, GSE85258 and GSE105261] from GEO [9-11]. The GSE22541, GSE85258 and GSE105261 dataset contained RCC samples and mRCC samples (24 vs 20, 15 vs 16 and 9 vs 26, respectively) (Table 1). All the data above could be obtained freely on the internet, and there were no human or animal experiments conducted by any authors in this present research.

Table 1.

Statistics of the three microarray databases derived from the GEO database

Dataset ID	RCC	mRCC	Total
GSE22541	24	20	44
GSE85258	15	16	31
GSE105261	9	26	35

Identification of DEGs

GEO2R (http://www.ncbi.nlm.nih.gov/geo/geo2r/) is a website allowing interaction and it is a tool which can help users make comparisons between at least two datasets in GEO series in order to select DEGs based on conditions of experiments. The genes that satisfied cutoff standard (P < 0.05 and |log FC (fold change)| ≥ 1) were recognized statistically remarkable as DEGs between RCC and mRCC samples. With the web tool Venn diagram, commonly altered DEGs in the data sets have been integrated (http://bioinformatics.psb.ugent.be/webtools/Venn/).

KEGG pathway and GO enrichment analysis for DEGs

GO is considered to be a main bioinformatics method to note on genes and make analysis on biological process of the genes [12]. As a data repository for exploring biological systems and high-level functions from a wide range of molecular datasets, KEGG is established via high-throughput experiment technology methods [13]. Interacting networks between molecules were visualized on Cytoscape (version 3.6.1), which is a software platform with an open bioinformatic source [14]. The ClueGO (Version 2.5.4) in Cytoscape was a plug-in APP to create and visualize the functionally grouped network of terms/pathways [15]. ClueGO was used to calculate analysis for GO annotation and selections of DEGs in analysis for KEGG pathway enrichment in our research. If P < 0.05, then the result was deemed statistically significant.

Construction of PPI network

We used the online database Search Tool for the Retrieval of Interacting Genes (STRING; http://string-db.org) to predict the protein-protein interaction (PPI) network [16]. We used the STRING database to establish PPI network, and an interacting activity with a total score more than 0.4 was deemed statistically significant. Subsequently, we used Cytoscape software to establish relationship network for protein interaction activities.

Hub genes selection and analysis

CytoHubba is a plugin software of cytoscape, which has been utilized to explore the maximal clique centrality (MCC) in every protein node. The genes ranking top ten were selected as the hub genes in our research. A network of the hub genes and their co-expression genes was analyzed using CytoHubba. The illness-free survival and total survival analyses for hub genes have been conducted with the curve of Kaplan-Meier in cBioPortal (website: http://www.cbioportal.org) platform [17,18]. The relationship of gene expression levels between mRCC patients and RCC patients was analyzed using online database Oncomine (http://www.oncomine.com). Correlations between gene levels and patient survival rates have been featured in UALCAN (website: http://ualcan.path.uab.edu/index.html) [19]. Available cancer genome atlas (TCGA) patient survival rate data have been applied for survival analysis using Kaplan-Meier method and to establish the graphs and tables of overall survival. Using multiple-gene signature, we analyzed its ability to predict survival using TCGA with SurvExpress.

Results

Identification of DEGs in mRCC

After standardization of three gene expression profiles, DEGs (1,912 in GSE22541, 750 in GSE85258 and 2,119 in GSE105261) were identified. We performed Venn analysis in order to obtain the DEG profiles intersection (Figure 1). Ultimately, 111 DEGs in total were remarkably expressed differentially in all 3 groups, among which 72 genes were remarkably upregulated and 39 genes were downregulated between tissues from RCC and metastatic RCC.

Figure 1.

Venn diagram of DEGs (1,912 in GSE22541, 750 in GSE85258 and 2,119 in GSE105261) was used to obtain the DEG profiles intersection. 111 DEGs in total were remarkably expressed differentially in all 3 groups.

KEGG and GO enrichment analyses for DEGs

Results (Table 2) of GO analysis demonstrated that variations in biological process (BP) of DEGs have been dramatically rich in extracellular matrix organization. Variations in cellular component (CC) of DEGs have been mostly rich in banded collagen fibril. Variations in functions of molecular function (MF) mostly aggregated in Extracellular matrix structural constituent conferring tensile strength. Furthermore, the data from analysis of KEGG pathway presented that DEGs mostly aggregated in protein digestion and absorption (Table 2).

Table 2.

Significantly enriched GO terms and KEGG pathways of DEGs

Category	Term	Description	Count in gene set	P-value
BP term	GO:0030198	Extracellular matrix organization	18	2.11E-11
BP term	GO:0030199	Collagen fibril organization	7	2.20E-08
BP term	GO:0043200	Response to amino acid	7	1.01E-05
BP term	GO:0060840	Artery development	6	2.51E-05
BP term	GO:0071230	Cellular response to amino acid stimulus	5	4.92E-05
CC term	GO:0098643	Banded collagen fibril	6	1.03E-11
CC term	GO:0005583	Fibrillar collagen trimer	6	1.03E-11
CC term	GO:0098644	Complex of collagen trimers	6	5.86E-10
CC term	GO:0044420	Extracellular matrix component	7	2.49E-08
CC term	GO:0031258	Lamellipodium membrane	3	2.02E-03
MF term	GO:0030020	Extracellular matrix structural constituent conferring tensile strength	7	2.28E-09
MF term	GO:0005539	Glycosaminoglycan binding	13	2.09E-08
MF term	GO:0048407	Platelet-derived growth factor binding	4	2.68E-07
MF term	GO:0008201	Heparin binding	10	4.31E-07
MF term	GO:0019838	Growth factor binding	7	4.25E-05
KEGG pathway	hsa04974	Protein digestion and absorption	7	2.76E-06
KEGG pathway	hsa04512	ECM-receptor interaction	6	2.14E-05
KEGG pathway	hsa05146	Amoebiasis	4	4.54E-03
KEGG pathway	hsa04933	AGE-RAGE signaling pathway in diabetic complications	4	5.25E-03
KEGG pathway	hsa05150	Staphylococcus aureus infection	3	1.20E-02

Construction of PPI network and hub gene selection

With STRING tools and Cytoscape, we established the PPI network related to DEGs. 180 edges and 73 nodes in total were included in the network of PPI (Figure 2). The hub genes ranking top ten were identified in the PPI network by connectivity MCC using CytoHubba (Table 3). The data presented that the collagen type I alpha 1 chain (COL1A1) has been the most significant gene and its connectivity MCC = 92330. All the hub genes have an increased expression in mRCC.

Figure 2.

Protein-protein interaction network was constructed with the differentially expressed genes. Red nodes represent upregulated genes, and green nodes represent downregulated genes.

Table 3.

Top ten hub genes with higher MCC of connectivity

No.	Gene symbol	Full name	Alias	MCC
1	COL1A1	Collagen type I alpha 1 chain	OI1; OI2; OI3; OI4; EDSC; EDSARTH1	92330
2	POSTN	Periostin	PN; OSF2; OSF-2; PDLPOSTN	92313
3	COL1A2	Collagen type I alpha 2 chain	OI4; EDSCV; EDSARTH2	92304
4	COL3A1	Collagen type III alpha 1 chain	EDS4A; EDSVASC; PMGEDSV	92216
5	COL5A2	Collagen type V alpha 2 chain	EDSC, EDSCL2	87120
6	DCN	Decorin	CSCD; PG40; PGII; PGS2; DSPG2; SLRR1B	85807
7	COL6A3	Collagen type VI alpha 3 chain	BTHLM1, DYT27, UCMD1	85800
8	COL5A1	Collagen type V alpha 1 chain	EDSC, EDSCL1	85706
9	CDH11	Cadherin 11	CAD11, CDHOB, ESWS, OB, OSF-4	46807
10	COL11A1	Collagen type XI alpha 1 chain	CO11A1, COLL6, STL2	45384

Analysis for hub genes

We have listed the symbol of the gene, full name and alias for the hub genes in Table 3. A hub genes network and the co-expression genes have been detected by CytoHubba (Figure 3). Subsequently, the disease-free survival rate and overall survival rate analysis of these hub genes have been conducted with the Kaplan-Meier curve using the online platform in cBioPortal. RCC patients with CDH11, COL3A1, COL5A1, COL5A2 and COL11A1 alteration showed worse disease-free survival (Figure 4A-J). Nonetheless, RCC patients with CDH11, COL3A1, COL5A1, COL5A2, COL6A3 and COL11A1 alteration showed worse overall survival (Figure 4K-T). In the Jones Renal dataset [20], mRNA levels of ten hub genes were associated with metastasis, and the gene expression level in patients with mRCC was higher than that in patients without metastasis (Figure 5). Among CDH11, COL3A1, COL5A1, COL5A2, COL6A3 and COL11A1, analyses for survival rate with Kaplan-Meier method presented that COL5A1, COL6A3 and COL11A1 expression levels were remarkably related to RCC patient survival rate Using UALCAN (Figure 6C, 6E and 6F). So, COL5A1, COL6A3 and COL11A1 have been recognized as genes related to clinical conditions, revealing that the genes might have important roles in the carcinogenesis or development of mRCC.

Figure 3.

A network of the hub genes and their co-expression genes was analyzed using CytoHubba. Blue nodes represent the co-expression genes, and the others represent hub genes. Red node represents the highest MCC gene, and yellow node represents the lowest MCC gene.

Figure 4.

Disease-free survival and overall survival analyses of hub genes were performed using cBioPortal online platform. RCC patients with CDH11, COL3A1, COL5A1, COL5A2 and COL11A1 alteration showed shorter disease-free survival (P < 0.05) (A-J). Nonetheless, RCC patients with CDH11, COL3A1, COL5A1, COL5A2, COL6A3 and COL11A1 alteration showed shorter overall survival (P < 0.05) (K-T).

Figure 5.

Association between the expression of hub genes and metastasis in the Jones Renal dataset. The hub gene expression level in patients with mRCC was higher than that in patients without metastasis. M0, renal cell carcinoma without metastasis; M1, renal cell carcinoma with metastasis.

Figure 6.

Kaplan-Meier survival plot of hub genes using data of 531 renal cell carcinoma patients from TCGA database. CDH11, COL3A1 and COL5A2 expression levels were not remarkably related to survival rates of RCC patients (A, B, D). COL5A1, COL6A3 and COL11A1 expression levels were remarkably related to survival rates of RCC patients (C, E, F).

Using three-gene signature, we analyzed its ability to predict survival using TCGA with SurvExpress. In our three-gene signature, the prognostic index (PI) of the 415 patients was 0.603-2.479, with the optimal cut-off value of 1.214. PI < 1.214 was divided into low risk group (n = 208), while PI > 1.214 high risk group (n = 207). The analysis demonstrated that a high risk was correlated with high expressions of COL5A1, COL6A3 and COL11A1, shorter survival and worse prognosis, while a low risk was correlated with low expressions of COL5A1, COL6A3 and COL11A1, longed survival and better prognosis in RCC (Figure 7A). Moreover, Kaplan-Meier survival curves showed that patients with a predicted high risk (n = 207) had significantly shorter OS than those with a low risk (n = 208) in RCC (P < 0.05) (Figure 7B). Moreover, we detected the gene expression level of COL5A1, COL6A3 and COL11A1 in high risk and low risk groups in RCC. Our results displayed that the gene expressions of COL5A1, COL6A3 and COL11A1 were higher in high risk group than those in low risk group, and all had significant differences in the three-gene signature (P = 1.02e-91, P = 4.29e-73 and P = 1.85e-34, respectively) (Figure 7C).

Figure 7.

The three-gene signature predicted survival of RCC patients. A. SurvExpress was used to analyze the association of the three-gene signature with the predicted risk, survival and prognosis. B. Kaplan-Meier survival curves showed that patients with a predicted high risk had significantly shorter OS than those with a low risk. C. The gene expression level of COL5A1, COL6A3 and COL11A1 was detected in high risk and low risk groups. The gene expression of COL5A1, COL6A3 and COL11A1 was higher in high risk group than that in low risk group.

We also analyzed the association among COL5A1, COL6A3 and COL11A1 by the LinkedOmics database, and it was found that COL5A1 was both positively correlated with COL6A3 (R = 0.8917, P < 0.05) and COL11A1 (R = 0.684, P < 0.05), and COL6A3 was positively correlated with COL11A1 in RCC (R = 0.6421, P < 0.05) (Figure 8A). Furthermore, it was verified using the GEPIA dataset; surprisingly, COL5A1 was both positively correlated with COL6A3 (R = 0.77, P < 0.05) and COL11A1 (R = 0.67, P < 0.05), and COL6A3 was positively correlated with COL11A1 in RCC (R = 0.47, P < 0.05) (Figure 8B).

Figure 8.

The correction between COL5A1, COL6A3 and COL11A1 in RCC. The correction between COL5A1, COL6A3 and COL11A1 in RCC was analyzed by LinkedOmics (A) and GEPIA (B), and COL5A1, COL6A3 and COL11A1 were positively correlated with each other.

Discussion

The 5-year survival rate of mRCC is no more than 10%, and the aim of the therapy is slowing down the progression of disease and managing symptoms [21]. The treatment of mRCC has changed dramatically using several agents of the TKI which is targeted by the VEGF family. Nevertheless, full responses to the therapies hardly appeared (< 1%), and most patients having initial response would suffer cancer development. Therefore, candidate markers to diagnose the disease and treatment with high efficiency are urgently demanded. Microarray technology gives us a chance to explore the gene variations in mRCC, and it was recognized as a helpful method to identify new biomarkers in other diseases.

In this research, GO enrichment analyses for the DEGs have mainly aggregated in the extracellular matrix organization, collagen fibril organization, banded collagen fibrils, and dramatically aggregated in the KEGG terms protein digestion and absorption. A network of PPI has been established to reveal the interconnection among the DEGs, and we selected 10 hub genes. All of these genes were upregulated in mRCC. RCC patients with CDH11, COL3A1, COL5A1, COL5A2, COL6A3 and COL11A1 alteration showed worse overall survival. Nonetheless, RCC patients with CDH11, COL3A1, COL5A1, COL5A2 and COL11A1 alteration showed worse disease-free survival. According to the method of Kaplan-Meier curve, alteration of all those mentioned genes has been associated with poor prognosis in renal cell carcinoma patients. The mRNA levels of ten hub genes were associated with metastasis, and the gene expression level in patients with mRCC was higher than that in patients without metastasis. Survival analyses using Kaplan-Meier method presented that COL5A1, COL6A3 and COL11A1 expression levels were remarkably related to RCC patient survival rate using UALCAN. Using TCGA with SurvExpress, expression levels of COL5A1, COL6A3 and COL11A1 were higher in high risk group than those in low risk group, and all had significant differences in the three-gene signature. The correction between COL5A1, COL6A3 and COL11A1 in RCC was analyzed by LinkedOmics and GEPIA, and COL5A1, COL6A3 and COL11A1 were positively correlated with each other. COL5A1, COL6A3 and COL11A1 have been recognized as genes with clinical relevance, revealing that they might have important roles in carcinogenesis or development of mRCC.

Upregulation of COL5A1 was discovered to involve in several tumors, including renal carcinoma, breast carcinoma and tongue squamous cell carcinoma [22]. Some researches proposed that upregulation of COL5A1 might be related to the unfavorable prognosis in the carcinomas. Feng et al. [23] showed that after COL5A1 was knocked down, cell proliferation process was remarkably suppressed, cell apoptosis process increased, cell migration ability and invasion capability enhanced in cell experiments, and tumor development was repressed in animal models. Hence, COL5A1 might become a new biomarker for prognosis and a potential therapy target in the RCC management. Yu et al. [24] presented that LMCD1-AS1 exerting oncogenic function is partly related to expression of COL6A3. Liu et al. [25] reported that COL6A3 was clinically relevant to the progression of colorectal malignancy via knocking out COL6A3. COL6A3 might become a potential target and biomarker in prognostic and therapeutic method in colorectal tumor. Arafat H et al. [26] suggested that identifying COL6A3 subtypes as pancreatic ductal adenocarcinoma-specific shed a light on future researches to discover the oncogenic function and the diagnostic capability of the selective splicing activities. Li et al. [27] suggested that COL11A1 might take part in modulating various genes which are involved in cell development process and/or invasion ability. COL11A1 may be a tumor promoter gene in gastric carcinoma and is a potential therapy target in the treatment of cancer. Shen et al. [28] suggested that COL11A1 might be a biomarker in advanced non-small cell lung cancer, and it can work as a predictor for post-surgery resection recurrence. Vázquez-Villa F et al. [29] reported that COL11A1 has been a significant biomarker in indicating tumor development and human malignancy-related stromal cells. Galván JA et al. [30] reported that high COL11A1 expression in colon adenocarcinoma patients has been related to nodal invasion, the progression of metastases to a distant part, and poor Dukes stages.

In our study, except for COL5A1, COL6A3 and COL11A1, we detected other seven hub genes associated with metastatic renal cell carcinoma, including COL1A1, POSTN, COL1A2, COL3A1, COL5A2, DCN, and CDH11. Overexpression of these genes has also been found in the breast, pancreas, colorectal, gastric, bladder and renal cancers, and may be regarded as a valuable diagnostic biomarker, treatment and prognostic marker of tumors [31-34]. COL5A1, COL6A3 and COL11A1 were overexpressed in mRCC compared to RCC tissues. Therefore, alteration in COL5A1, COL6A3 and COL11A1 was involved in worse overall and disease-free survival, indicating that these genes may function as potential prognostic markers and therapy targets in the management of mRCC. Nonetheless, further research is required to discover the function of these genes in mRCC.

Conclusion

This research aimed to select DEGs which might play a part in the carcinogenesis or progression of mRCC. DEGs and hub genes selected in our research give us a chance to have better understanding of the underlying mechanisms in the carcinogenesis and progression of mRCC. COL5A1, COL6A3 and COL11A1 may be key genes for metastatic renal cell carcinoma and might become potential targets in the diagnostic method and treatment of mRCC. Further research is required to discover the function of COL5A1, COL6A3 and COL11A1 in the treatment of mRCC.

Disclosure of conflict of interest

None.