Mutational analysis of driver genes with tumor suppressive and oncogenic roles in gastric cancer

Gene list related to gastric cancer

To systematically study the GC-related genes, we downloaded all 1,815 literature-based GC-related genes were downloaded from GCGene for further analysis (Zhao et al., 2016a). The GCGene was constructed by performing an extensive data integration and literature search followed by manual assembly of the data. To provide a more reliable gene list, we also download 683 genes with two or more PubMed abstracts, which represent reliable gene list related from GCGene. The full GC-related gene list provided a basis for integration. By subtraction of those well-studied 683 genes, we also collected 1,132 genes with a single reference, which will help identify unexplored genes and pathways.

Data source for driver identification in gastric cancer

To identify the driver genes in gastric cancer, we used three bioinformatic databases: (i) TSGene (Zhao et al., 2016b), including 1,207 known human TSGs curated from literatures; (ii) ONGene (Liu, Sun & Zhao, 2016), the ONG list with 803 human genes from 8,849 PubMed abstracts; and (iii) DriverDB 2.0 (Chung et al., 2016), a comprehensive cancer driver genes database constructed by integrating 15 published bioinformatics driver identification algorithms.

For TSGene and ONGene, we downloaded all the genes including protein-coding and non-coding genes from corresponding websites. For DriverDB 2.0, we focused on the TCGA GC dataset and predicted putative driver genes in all the GC samples by using the integrated 15 driver identification tools. To obtain a reliable driver gene list, we required that any driver gene should be supported by at least two tools based on the TCGA mutational data.

Pathway and mutational analysis

To assess the function of all the identified GC-related driver genes, we conducted functional enrichment tests using the online tool ToppFunc (Chen et al., 2009). ToppFunc adopts a hypergeometric model in order to measure whether an input gene list has a different annotation frequency to the one that would occur randomly. We conducted chromosome cytoband-based enrichment analysis to identify the genomic regions where the input genes were significantly enriched using all the genes in these regions as background. Similar processes were used to identify enriched gene ontology terms, KEGG and wiki pathways. In these enrichment analyses, all the human genes in ToppFunc were used as background to calculate statistical significance. In addition, the Benjamini–Hochberg method was implemented in the ToppFunc to further exclude false negative results. Finally, a p-value <0.01 was adopted as the cutoff for enriched pathways in KEGG and gene ontology (biological process) and we only considered those representative pathways with two or more genes. For the enrichment analysis on miRNAs, we adopted an online server miEAA (Watanabe et al., 2007). MiEAA offers both over-representation analysis and set enrichment analysis, which is similar to gene set enrichment analysis implemented in ToppFunc.

Throughout the study, the GC-related mutational analyses were conducted using the cBio portal (Cerami et al., 2012). We selected the Stomach Adenocarcinoma (TCGA) dataset. In total, there are 393 tumor samples with single nucleotide mutations, INDELs, or copy number variation (CNV) data. Abnormal gene expression and protein expression were not included. For the DNA copy-number data, the putative discrete values were calculated for all genes, e.g.,  “deeply deleted” or “amplified”. For single-nucleotide variations and INDELs, we excluded those mutations without any functional effect, such as synonymous mutations.

Identification of dual role protein-coding genes as TSGs and ONGs in gastric cancer

To survey how many TSGs and ONGs are involved in the development of gastric cancers, we used the data from three large-scale literature mining databases. The GCGene is the database developed to curate the gastric cancer-related genes from literature. The TSGene and ONGene were the databases to collect known critical cancer genes with literature. By overlapping the protein-coding genes from GCGene to TSGene and ONGene, we identified 313 TSGs,160 ONGs (Fig. 1A) and 48 dual role genes with both TSGs and ONGs effects in other cancers (Table S1).

Many genes could function as both TSG and ONG depending on the cancer type, stage of development, or interaction partners (Zhao et al., 2016b). For example, as an oncogene, SIRT1 can promote cancer progression by negative control of the TGF-beta signaling pathway (Lamouille & Derynck, 2009). However, SIRT1 can also interact with promyelocytic leukemia protein to express its tumor suppressor property by stabilizing TP53 and inducing cell senescence. To provide a global functional distribution of the 48 dual role genes in GC, we performed a functional enrichment analysis of gene ontology, KEGG pathway, genomic location, and protein family (Figs. 1B–1C, Table S2). We found 35 genes are active regulators in cell proliferation (GO:0042127, corrected P-value = 2.89E–22). More interesting, there are 32 genes involving in “response to endogenous stimulus” (GO:0009719, corrected P-value = 6.70E–18). Consistently, the subcellular localization also mainly group into nuclear chromosome (11 genes, GO:0000228, corrected P-value = 9.13E–06) and the plasma membrane region (12 genes, GO:0098590, corrected P-value = 1.57E–04). Those genes located on chromosome or chromatin are mainly from transcription factor complex. For example, we found five p53-like transcription factors: RUNX1, RUNX3, STAT3, TP63, and TP73 (InterPro domain IPR008967, corrected P-value = 3.65E–05). By mapping to the pathways (Fig. 1C), we were able to locate those genes in the critical cancer pathways. The dual role genes are associated with cancer pathways (corrected P-value = 5.75E–20), microRNA (corrected P-value = 2.72E–10) and transcriptional regulation (corrected P-value = 9.64E-07). Cell cycle (corrected P-value = 1.22E–08) and response to DNA damage (corrected P-value = 1.24E–07) may be controlled by the dual role genes which are also competing in the androgen receptor (corrected P-value = 3.74E–09), TGF-beta (corrected P-value = 1.84E–08) and p53 signaling (corrected P-value = 4.60E–08) pathways. In summary, our integrative analysis revealed that those genes with both TSG and ONG roles may group into two main functional clusters in gastric cancer: transcriptional regulation inside nuclear, and response to endogenous signals in plasma membrane region. The competition of these genes in critical pathways, such as the cell cycle p53 signaling pathway, may be critical for GC progression.

The majority of genes are unique to some pathways (Fig. 1C). However, several genes, including CDKN1A and CDKN1B, are involved in multiple oncogenic pathways (Fig. 1C). Due to their dual functions in these critical oncogenic pathways, we need to be cautious about drawing conclusions concerning their effects on GC cells. By overlapping those genes with TCGA somatic mutation data, we investigated the potential functions of the 48 genes with dual roles in TCGA GC samples. A few genes did not have dual roles according to their mutational pattern (Fig. 2). For example, SALL4, JUP, NOTCH1 and CDK6 are all with frequent amplification in multiple cancer samples, which may indicate their oncogenic roles in GC. In contrast, more copy number deletions were observed in CDH1, RHOA, PLK1, and MST1R, which may imply a TSG role in GC. For the two genes with broad effects, CDKN1A and CDKN1B, both gene copy gain and loss were found, which means they may have dual roles in GC. The mutational pattern also confirmed some of the genes, such as CDKN1A and CDKN1B, may have dual roles in the TCGA GC cohort.

The functional and mutational features of protein-coding driver TSGs and ONGs in TCGA GC cohort

To explore the driver TSGs and ONGs, we utilized 15 driver mutational detection tools to identify the protein-coding drivers in the TCGA GC cohort. For a driver gene, we required two positive results from two or more driver detection tools. In total, we found 874 genes with driver mutations in the TCGA data. By overlapping with the gastric cancer-related genes in GCGene, we identified those driver genes with and without TSG and ONG roles. We found 30, 18, 6, and 84 driver genes as TSGs, ONGs, dual role, and non-TSG-ONG respectively (Fig. 3A). We considered that those 138 driver genes were well-studied in gastric cancer. However, based on the number of literature evidence in GCGene database, we found there are only 56 driver genes with two or more literature evidences (Fig. 3B). All the six genes with dual roles in cancers are supported by at least two references, which confirmed their important roles in GC. By using TCGA GC mutation data, we explored the mutational frequency of the six genes (Fig. 3C). In total, there were 105 instances with at least one somatic mutation (copy number variation included) in 393 sequenced patients (27%). Some of the six genes had only one function based on the mutational pattern. For example, STAT3 is amplified in the majority of mutated samples suggesting that the activation of STAT3 signaling genes supports GC cell survival (Kanda et al., 2004). Similarly, PLK1 and RHOA were deleted in a number of patients and behave more like TSGs. However, the remaining three genes (PTPN11, VIM, and CDH1) are more likely to have dual roles with both mutations for gain-of-function and loss-of-function.

To discover some novel driver genes not extensively studied in GC, we focused on 82 predicted driver genes with one reference in the GCGene database (Figs. 3A, 3B). There were 12 TSG and 7 ONG driver genes associated with a single PubMed abstract. Among the 82 non-TSG-ONG driver genes (Fig. 3A), 63 are not well studied in GC (Fig. 3B and Table S3). To validate the functions of these genes in GC it is important to check over the mutational frequency, we investigated the 82 putative drivers in the TCGA GC mutational data (Fig. 4). All of the seven ONG driver genes are highly mutated with a highest mutational frequency of 14% on GLI3 and the lowest mutational frequency 4% on SMO. Only the ONG NEDD9 had sporadic deletions, which is not consistent with its oncogenic role. For 12 of the TSG driver genes, we found sporadic amplifications on AXIN1, CSMD1, LRP1B, NEDD4L, NF1, PARK2, and RHOBTB2. Furthermore, two of the TSG driver genes (SOCS3, PTPRT) have concordant amplifications which implies a ONG role in the TCGA GC cohort. For the remaining driver genes, which have no TSG and ONG functions, the majority have both amplifications and deep deletions in multiple cancer samples.

MicroRNA TSGs and OGCs in gastric cancer

Recent studies have reported that some micro-RNAs (miR), single-stranded, small noncoding RNA genes, can function as TSGs and ONGs. Evidence from our GCGene shows that there are 111 miRNAs related to GC. By intersecting with those curated TSG and ONG miRNAs, we found 55 TSGs, 14 ONGs and 19 dual role miRNAs (Fig. 5A). Among these genes, there are 29 TSGs, 9 ONGs and 12 dual role miRNAs with two or more references in GC (Fig. 5B). Some of the dual role miRNAs are confirmed as having dual roles in gastric cancer. For example, miR-223 is overexpressed in metastatic GC cells and stimulates non-metastatic GC cells migration and invasion by directly targeting its 3′-untranslated regions of EPB41L3 (Li et al., 2011). In addition, miR-223 functions as an oncogene in human GC by targeting FBXW7/hCdc4 (Li et al., 2012) and targets oncogene STMN1 (Kang et al., 2012). However, the majority of the miRNAs with dual roles in other cancers are not associated with GC. For instance, miR-335 was only reported as TSGs to target Bcl-w and specificity protein 1 (Xu et al., 2012).

By running an enrichment analysis on the 19 identified dual role miRNAs using miEAA (Watanabe et al., 2007), we confirmed that these miRNAs are enriched in various cancers (Table 1). Among the 19, four (miR-20a, miR-18a, miR-16-1, miR-17) are located on chromosome 13 (Corrected P-value = 0.00376467) and three of these belong to the miR-17 family. Although the mir-17/92 cluster are known as oncogenes, recent studies suggest that their dual roles depend on the targeting genes (Xiang & Wu, 2010).

By subtracting those genes shown in Fig. 5B from those in Fig. 5A, we found there are seven dual role miRNAs with single literature evidence in GCGene, which may warrant further investigation. For example, the miR-150 functioned as ONG to target a TSG EGR2 (Wu et al., 2010) but there is no information concerning its TSG role in gastric cancer. By overlapping to TCGA CNV data, we found there are four miRNAs with copy numberchanges: miR-31, miR-182, miR-135a-1, and miR-150. We found that miR-31 has both amplification and deletion functions in different tumor samples and this suggests dual roles in gastric cancer. The miR-182 and miR-150 only have amplifications but miR-135a-1 was all deleted in 1.5% TCGA GC cohort. In summary, we found several potential dual role miRNAs in GC with different CNV pattern.

As shown in Figs. 5A and 5B, there are 55 gastric cancer-related miRNAs with TSG functions. A few miRNAs have been suggested as potential TSGs via suppressing different oncogenes. For instance, the miR-148a (Zheng et al., 2011), miR-195 and miR-375 (Ding et al., 2010) may function as TSG in GC, and theirs anti-oncogenic activity may involve the direct targeting and inhibition of ONG ROCK1, CDK6, and JAK2. Some miRNAs may not function in cancer development, but cancer metastasis original from GC tissues. The interaction between Robo1 on the Slit2Slit-Robo1 pathway triggers tumor metastasis of GC, which can be suppressed by miR-218 (Tie et al., 2010). By subtracting those genes shown in Fig. 5B from Fig. 5A, we found there are 26 miRNA TSGs with single literature evidence in GCGene. Examples include miR-486 (targeting OLFM4) (Oh et al., 2011), miR-378 (suppressing VEGF) (Deng et al., 2013), and miR-101 (targeting EZH2, Cox-2, Mcl-1 and Fos) (Wang et al., 2010b). Among the 26, there are 19 miRNAs with CNVs in TCGA GC cohort. For example, the miR-148b could have a TSG role by targeting CCKBR (Song et al., 2011), which was mutated in 1.5% of 393 TCGA patients. Although most of these CNVs for miR-148b are deletions, the amplifications were still observed in a few samples. A few of the miRNA TSGs were amplified in all the samples with mutations, such as miR-196b, miR-206, miR-let-7i, miR-100, and miR-29c, which either implies the potential oncogenic role or those amplifications may not have dosage effects on these miRNAs.

In general, miRNA ONGs are located in the amplified regions in human cancers and tend to cleave target mRNAs more frequently (Wang et al., 2010a). For example, the miR-21 (Zhang et al., 2012), negatively regulates the tumor suppressors PTEN which, in-turn, promote gastric tumor proliferation and invasion. All five ONGs with a single study recorded in GCGene have more amplifications than deletions in the TCGA cohort (Fig. 5C); this confirms their critical oncogenic roles in GC.