Cross talk of chromosome instability, CpG island methylator phenotype and mismatch repair in colorectal cancer.

Colorectal cancer is a severe cancer associated with a high prevalence and fatality rate. There are three major mechanisms for colorectal cancer: (1) Chromosome instability (CIN), (2) CpG island methylator phenotype (CIMP) and (3) mismatch repair (MMR), of which CIN is the most common type. However, these subtypes are not exclusive and overlap. To investigate their biological mechanisms and cross talk, the gene expression profiles of 585 colorectal cancer patients with CIN, CIMP and MMR status records were collected. By comparing the CIN+ and CIN-samples, CIMP+ and CIMP-samples, MMR+ and MMR-samples with minimal redundancy maximal relevance (mRMR) and incremental feature selection (IFS) methods, the CIN, CIMP and MMR associated genes were selected. Unfortunately, there was little direct overlap among them. To investigate their indirect interactions, downstream genes of CIN, CIMP and MMR were identified using the random walk with restart (RWR) method and a greater overlap of downstream genes was indicated. The common downstream genes were involved in biosynthetic and metabolic pathways. These findings were consistent with the clinical observation of wide range metabolite aberrations in colorectal cancer. To conclude, the present study gave a gene level explanation of CIN, CIMP and MMR, but also showed the network level cross talk of CIN, CIMP and MMR. The common genes of CIN, CIMP and MMR may be useful for cross-subtype general colorectal cancer drug development.

Introduction

Colorectal cancer is one of the most common cancer with leading cause of death (1). Its classical molecular events have been well-studied. The oncogenes in colorectal cancer are ras, scr and c-myc while the tumor suppressor genes are APC and p53. The Wnt pathway is considered to be important in the tumorgenesis of colorectal cancer. In 1990, Fearon and Vogelstein (2) proposed a famous model of colorectal cancer which believes a serials of gene and signaling pathway alterations contribute to the histology changes from normal tissue to adenoma and then to carcinoma. Li et al found that at each stage of colorectal cancer, their gene expression profiles were different (3). Jiang et al found that the early stage colorectal cancer biomarkers and late stage biomarkers were different and they can be connected by signal propagation on the network (4). Many genes were found to be associated with colorectal cancer by gene expression and network analysis (5,6). And many signaling pathways, such as Wnt/β-catenin signaling, epidermal growth factor receptor/Ras signaling, p53 signaling, Notch signaling, Hedgehog signaling, and Hippo signaling, were found to play roles in colorectal cancer (7).

To summary the current understandings of colorectal cancer, there are major mechanisms for colorectal cancer: (1) chromosome instability (CIN), (2) CpG island methylator phenotype (CIMP) and (3) mismatch repair (MMR). In approximately 85% of colorectal cancer patients, the chromosomal instability (CIN) is observed (8). They exhibited genomic instability on the chromosomal level. The CIN patients usually have the poorest prognosis (9). In approximately 15–20% colorectal cancer patients, there are widespread CIMP (10). In approximately 15% colorectal cancer patients, Microsatellite instability (MSI) is detected (11). It is caused by the loss of DNA MMR activity. The MSI patients tend to have a good prognosis (12). These mechanisms are not mutually exclusive. For example, the MMR patients usually also show varying degrees of CIN (8). Different pathways that were used for characterizing each mechanism actually can interact with each other and cross talk (7). Multiple signaling pathways share transcription factors, microRNAs and ligases, such as miR-21, miR-145, FBXW7 and β-TrCP (7).

To systematically investigate the relationship between CIN, CIMP and MMR, we analyzed the gene expression profiles of 585 colorectal cancer patients. These patients were annotated with CIN, CIMP and MMR status. For each status, we applied advanced minimal redundancy maximal relevance (mRMR) and incremental feature selection (IFS) method to select its biomarkers genes. Then we overlapped the CIN, CIMP and MMR biomarker genes. Since they may not directly interact with each other, we used random walk with restart (RWR) method to find the region that the CIN, CIMP and MMR biomarker genes affect and investigated the commonly regulated genes by CIN, CIMP and MMR. The biological functions of these commonly regulated genes were analyzed. Our work found the molecular cross talk among CIN, CIMP and MMR, revealed the internal logic of colorectal tumorgenesis, and provided the emerging therapeutic targets that may be suitable for most colorectal cancer patients rather than a small proportion of patients.

Materials and methods

The gene expression profiles of 585 colorectal cancer patients

We downloaded the gene expression profiles of 585 colorectal cancer patients from GEO (Gene Expression Omnibus) with accession number of GSE39582 (13). The expression levels were measured with Affymetrix Human Genome U133 Plus 2.0 Array which had 54,675 probes corresponding to 20,502 genes. The probes corresponding to the same gene were averaged. The gene expression data was preprocessed with quantile normalization. Within the 585 colon patients, there were 369 CIN+ and 112 CIN-, 93 CIMP+ and 420 CIMP-, 77 dMMR and 459 pMMR. For each analysis, the patients with missing status were excluded. For example, for CIN+ and CIN-comparison, the 369 CIN+ and 112 CIN-patients were considered while 104 without CIN information were excluded.

The CIN-associated gene selection

mRMR gene ranking

We used the mRMR method (14) to rank the genes based on their relevance with CIN status and their redundancy between genes. The mRMR method is based on information theory and has been widely used in bioinformatics filed (15–19). To apply mRMR method, we used the C/C++ version mRMR software downloaded from http://home.penglab.com/proj/mRMR/. With mRMR method, we obtained a ranked gene list. The top 500 mRMR genes were analyzed.

IFS

To determine how many genes should be selected from the mRMR gene list, we adopted the IFS method (4,20–24) and constructed 500 support vector machine (SVM) classifiers. In this study, we used the svm function with default parameters from R package e10171 (https://cran.r-project.org/web/packages/e1071/) to build the SVM classifier. Each time, the top k genes in the mRMR list was used to build the SVM classifier. And the performance of the top k-gene classifier was evaluated with leave-one-out cross validation (LOOCV). To objectively evaluate the classifier's performance, Sensitivity (Sn), Specificity (Sp), Accuracy (ACC) and Mathew's correlation coefficient (MCC) were calculated:

where TP, TN, FP and FN stand for true positive (CIN+), true negative (CIN-), false positive (CIN+) and false negative (CIN-), respectively. Since the sizes of positive (CIN+) and negative (CIN-) samples were imbalance in this study, MCC which considered both Sn and Sp, was choose as the major measurement (25). At last, based on the IFS curve in which the number of top genes that were used as x-axis and the LOOCV MCCs of classifiers as y-axis, we can decide how many genes should be used to build a classifier with great performance and small complexity. The peak or the change point of the IFS curve were usually chosen.

The CIMP-associated gene selection

Similarly, we can identify the CIMP-associated genes using mRMR and IFS methods. Since the sample size of CIMP+ and CIMP-patients were also imbalance, the MCC was considered as the key measurement for prediction performance evaluation and was used to plot the IFS curve.

The MMR-associated gene selection

Similarly, we can identify the MMR-associated genes by analyzing the gene expression profiles pMMR and dMMR patients using mRMR and IFS methods. The dMMR and pMMR were considered as positive and negative samples, respectively. The MCC was used to plot the IFS curve since there were much more pMMR than dMMR.

The overlapped genes and common downstream genes of CIN, CIMP and MMR

We would like to known whether there is a general mechanism for CIN, CIMP and MMR. The direct way is to overlap the mRMR and IFS identified CIN associated genes, CIMP associated genes and MMR associated genes.

Since the identified CIN associated genes, CIMP associated genes and MMR associated genes may be incomplete or locate at the upstream of the colorectal cancer signaling pathway, we tried to pin down the area affected by the CIN associated genes, CIMP associated genes and MMR associated genes on the protein-protein interaction network of using RWR method (26–29). The STRING network (version 10.0) (30) is a comprehensive protein-protein functional association network that has been widely used (26,28,31–39). It included 19,247 proteins and 4,274,001 interactions. We constructed the network using the protein-protein interactions with confidence score >0.900 which is the highest confidence interaction in STRING database. Then the n*n adjacent matrix (A) of the network which included n proteins was column-wise normalized to make the column sum to be 1 by assign 1/m to the m interaction proteins of protein j in column j and 0 to other proteins without interactions.

The random walk procedure repeat in every time tick (t→t+1) from the initial seed genes which were represented as a n length vector with P0 value of 1/k for the k seed genes and value of 0 for other n-k non-seed genes. The state probabilities Pt+1 at time t+1 is calculated as follow: Pt+1=(1-r)t+r0 (5), where Pt is state probabilities at time t, r is the restart probability which is set to 0.7 as suggested by previous studies (26–29,40). It has been reported that if r is in a sizable range (0.5–0.8), the results will have little difference (40). These random walk process will stop when the difference between two steps is smaller than 1e-6. At last, all genes on the network will be assigned with a RWR score which corresponds to the probability of being expanded from the seed genes.

To statically evaluate the significance of RWR score, we randomly chosen the same number of seed genes and calculated their RWR scores for 1,000 times. The significance of actual RWR score can be defined as a permutation P-value of how times the random RWR scores was greater than the actual RWR score over the permutation times which was 1,000 in this study. The genes with permutation P-value smaller than 0.05 were considered as significant RWR expanded genes.

The RWR expanded genes can represent the downstream genes of CIN, CIMP and MMR and be used for common downstream gene analysis. The functions of the common CIN, CIMP and MMR downstream genes were enriched onto KEGG pathways and Gene Ontology (GO) terms using hypergeometric test.

Results and Discussion

The CIN associated genes identified with mRMR and IFS

The top 500 most discriminative genes between CIN+ and CIN-samples were ranked using the mRMR method which considered both their relevance with CIN status, and their redundancy with selected genes. After the genes were ranked by mRMR, we chosen the number of top genes by applying the IFS procedure. Different number of top genes were tried and their prediction performance were evaluated. The IFS curve with the number of genes as x-axis and leave one out cross validation MCC as y-axis was shown in Fig. 1A. It can be seen that when 34 genes were used, the leave one out cross validation MCC was the highest. The leave one out cross validation Sn, Sp, ACC and MCC of these 34 genes were 0.932, 0.696, 0.877 and 0.648, respectively. Therefore these 34 genes were chosen and shown in Table I. As shown in Fig. 2A, the 34 CIN associated genes can cluster the CIN+ and CIN-patients into the right groups. IVD, NDUFAF1, OIP5-AS1, EXOSC9, HSPA4L, RPL22L1, EMC6, NCBP3, CYB5D1, PRPSAP2, RALBP1, ATP9B, ADGRG6, TRIM7, NLRX1, RNF145, CTC1, TMEM102 were highly expressed in CIN-patients while TGFBR2, HERPUD2, KBTBD2, ROCK2, TUFT1, TMEM176A, RHEB, SERINC3, STX16, COMMD7, DYNLRB1, RTFDC1, EIF6, TM9SF4, HEATR4, RRNAD1 were highly expressed in CIN+ patients.

Figure 1.

The IFS curves of CIN, CIMP and MMR. (A) The IFS curve of CIN. The top 34 mRMR genes were chosen and their LOOCV sensitivity, specificity, accuracy and MCC were 0.932, 0.696, 0.877 and 0.648, respectively. (B) The IFS curve of CIMP. The top 19 mRMR genes were chosen and their leave one out cross validation sensitivity, specificity, accuracy and MCC were 0.710, 0.976, 0.928 and 0.744, respectively. (C) The IFS curve of MMR. The top 18 mRMR genes were chosen and their leave one out cross validation sensitivity, specificity, accuracy and MCC were 0.922, 0.985, 0.976 and 0.902, respectively. IFS, incremental feature selection; CIN, chromosome instability; CIMP, CpG island methylator phenotype; MMR, mismatch repair; mRMR, minimal redundancy maximal relevance; LOOCV, leave-one-out cross validation.

Table I.

The 34 chromosome instability-associated genes.

Order	Symbol	Name	Entrez gene	mRMR score
1	STX16	Syntaxin 16	8675	0.161
2	NCBP3	Nuclear cap binding subunit 3	55421	0.062
3	IVD	Isovaleryl-CoA dehydrogenase	3712	0.061
4	DYNLRB1	Dynein light chain roadblock-type 1	83658	0.067
5	EXOSC9	Exosome component 9	5393	0.044
6	ATP9B	ATPase phospholipid transporting 9B (putative)	374868	0.043
7	KBTBD2	Kelch repeat and BTB domain containing 2	25948	0.042
8	EMC6	ER membrane protein complex subunit 6	83460	0.043
9	ADGRG6	Adhesion G protein-coupled receptor G6	57211	0.046
10	OIP5-AS1	OIP5 antisense RNA 1	729082	0.044
11	RNF145	Ring finger protein 145	153830	0.043
12	COMMD7	COMM domain containing 7	149951	0.046
13	TUFT1	Tuftelin 1	7286	0.038
14	NLRX1	NLR family member X1	79671	0.036
15	CYB5D1	Cytochrome b5 domain containing 1	124637	0.038
16	RTFDC1	Replication termination factor 2 domain containing 1	51507	0.037
17	RPL22L1	Ribosomal protein L22 like 1	200916	0.034
18	TMEM102	Transmembrane protein 102	284114	0.032
19	TM9SF4	Transmembrane 9 superfamily member 4	9777	0.035
20	HERPUD2	HERPUD family member 2	64224	0.033
21	RHEB	Ras homolog enriched in brain	6009	0.033
22	NDUFAF1	NADH:ubiquinone oxidoreductase complex assembly factor 1	51103	0.033
23	TGFBR2	Transforming growth factor β receptor 2	7048	0.034
24	TRIM7	Tripartite motif containing 7	81786	0.032
25	PRPSAP2	Phosphoribosyl pyrophosphate synthetase associated protein 2	5636	0.032
26	HEATR4	HEAT repeat containing 4	399671	0.032
27	SERINC3	Serine incorporator 3	10955	0.034
28	HSPA4L	Heat shock protein family A (Hsp70) member 4 like	22824	0.03
29	RALBP1	RalA binding protein 1	10928	0.029
30	RRNAD1	Ribosomal RNA adenine dimethylase domain containing 1	51093	0.029
31	CTC1	CST telomere replication complex component 1	80169	0.03
32	EIF6	Eukaryotic translation initiation factor 6	3692	0.031
33	TMEM176A	Transmembrane protein 176A	55365	0.031
34	ROCK2	Rho associated coiled-coil containing protein kinase 2	9475	0.03

Figure 2.

The heatmaps of CIN, CIMP and MMR. (A) The heatmap of CIN. The 34 CIN associated genes were used to cluster the CIN+ and CIN-patients. (B) The heatmap of CIMP. The 19 CIMP associated genes were used to cluster the CIMP+ and CIMP- patients. (C) The heatmap of MMR. The 18 MMR associated genes were used to cluster the MMR+ and MMR- patients. CIN, chromosome instability; CIMP, CpG island methylator phenotype; MMR, mismatch repair.

The CIMP associated genes identified with mRMR and IFS

Similarly, the CIMP associated genes can be identified using mRMR and IFS methods. As a result, 19 genes were selected based on the IFS curve shown in Fig. 1B and listed in Table II. The 19 genes' leave one out cross validation Sn, Sp, ACC and MCC were 0.710, 0.976, 0.928 and 0.744, respectively. As shown in Fig. 2B, the 19 CIMP associated genes can cluster the CIMP+ and CIMP-patients into the right groups. VANGL2, ZNF665, JUN, FAM84A, ZBTB38, GRM8, DUSP18, PRDX5, HUNK, QPRT, ZNF141, MLH1, MTERF1 were highly expressed in CIMP-patients while PIWIL1, ADGRG6, FOXD1, HOXC6, AFAP1-AS1, HS3ST1 were highly expressed in CIMP+ patients.

Table II.

The 19 CpG island methylator phenotype-associated genes.

Order	Name	Gene name	Entrez gene	mRMR score
1	MLH1	mutL homolog 1	4292	0.193
2	HUNK	Hormonally up-regulated Neu-associated kinase	30811	0.069
3	ZNF141	Zinc finger protein 141	7700	0.063
4	DUSP18	Dual specificity phosphatase 18	150290	0.058
5	ADGRG6	Adhesion G protein-coupled receptor G6	57211	0.053
6	FOXD1	Forkhead box D1	2297	0.052
7	FAM84A	Family with sequence similarity 84 member A	151354	0.049
8	AFAP1-AS1	AFAP1 antisense RNA 1	84740	0.047
9	ZBTB38	Zinc finger and BTB domain containing 38	253461	0.052
10	VANGL2	VANGL planar cell polarity protein 2	57216	0.054
11	PRDX5	Peroxiredoxin 5	25824	0.049
12	MTERF1	Mitochondrial transcription termination factor 1	7978	0.05
13	QPRT	Quinolinate phosphoribosyltransferase	23475	0.05
14	HOXC6	Homeobox C6	3223	0.045
15	HS3ST1	Heparan sulfate-glucosamine 3-sulfotransferase 1	9957	0.044
16	PIWIL1	Piwi like RNA-mediated gene silencing 1	9271	0.046
17	JUN	Jun proto-oncogene, AP-1 transcription factor subunit	3725	0.047
18	GRM8	Glutamate metabotropic receptor 8	2918	0.045
19	ZNF665	Zinc finger protein 665	79788	0.046

The MMR associated genes identified with mRMR and IFS

Similarly, the MMR associated genes can be identified using mRMR and IFS methods. As a result, 18 genes were selected based on the IFS curve shown in Fig. 1C and listed in Table III. The leave one out cross validation Sn, Sp, ACC and MCC of these 18 genes were 0.922, 0.985, 0.976 and 0.902, respectively. As shown in Fig. 2C, the 18 MMR associated genes can cluster the MMR+ and MMR-patients into the right groups. CAB39L, H2AFJ, TGFBR2, MLH1, SEC22B, BRD3, FBXO21, FOXO3, INO80D were highly expressed in MMR-patients while EIF5A, RAPGEF6, LYG1, HNRNPL, MTA2, HPSE, STRN3, MIR3916, RAB12 were highly expressed in MMR+ patients.

Table III.

The 18 mismatch repair-associated genes.

Order	Name	Gene name	Entrez gene	mRMR score
1	HNRNPL	Heterogeneous nuclear ribonucleoprotein L	3191	0.285
2	HPSE	Heparanase	10855	0.097
3	CAB39L	Calcium binding protein 39 like	81617	0.081
4	MTA2	Metastasis associated 1 family member 2	9219	0.093
5	RAPGEF6	Rap guanine nucleotide exchange factor 6	51735	0.086
6	LYG1	Lysozyme g1	129530	0.081
7	SEC22B	SEC22 homolog B, vesicle trafficking protein (gene/pseudogene)	9554	0.081
8	BRD3	Bromodomain containing 3	8019	0.076
9	H2AFJ	H2A histone family member J	55766	0.079
10	RAB12	RAB12, member RAS oncogene family	201475	0.072
11	TGFBR2	Transforming growth factor β receptor 2	7048	0.078
12	STRN3	Striatin 3	29966	0.076
13	INO80D	INO80 complex subunit D	54891	0.076
14	MLH1	MutL homolog 1	4292	0.079
15	EIF5A	Eukaryotic translation initiation factor 5A	1984	0.072
16	MIR3916	microRNA 3916	100500849	0.069
17	FOXO3	Forkhead box O3	2309	0.069
18	FBXO21	F-box protein 21	23014	0.069

The direct overlap between CIN associated genes, CIMP associated genes and MMR associated genes

As three major mechanisms of colorectal cancer, we would like to investigate whether there were overlaps between CIN associated genes, CIMP associated genes and MMR associated genes. The Venn diagram of CIN associated genes, CIMP associated genes and MMR associated genes were shown in Fig. 3. It can be seen that none genes were common in these three gene lists. The overlap between CIN and CIMP was ADGRG6, the common gene between CIN and MMR was TGFBR2 and the overlap between CIMP and MMR was MLH1. The references of ADGRG6 was limited and its functions were largely unknown. Interestingly, TGFBR2 has been reported as a candidate driver gene in MSI colorectal cancer (41) and the MMR patients usually also show varying degrees of CIN (8). TGFBR2 may be key of the association of CIN and MMR. The correlation of MLH1 methylation and MMR status has been reported (42) and it confirmed the association of CIMP and MMR.

Figure 3.

The Venn diagram of CIN associated genes, CIMP associated genes and MMR associated genes. None genes were common in these three gene lists. The overlap between CIN and CIMP was ADGRG6, the common gene between CIN and MMR was TGFBR2 and the overlap between CIMP and MMR was MLH1. CIN, chromosome instability; CIMP, CpG island methylator phenotype; MMR, mismatch repair.

The cross talk between CIN, CIMP and MMR

Since there is little overlap between the CIN associated genes, CIMP associated genes and MMR associated genes identified by mRMR and IFS, we would like to investigate whether they have common downstream genes. To verify this, we used the workflow shown in Fig. 4 to investigate the cross talk between CIN, CIMP and MMR. The key is step (C) which identifies the genes that the CIN, CIMP and MMR affects, i.e. the downstream genes of CIN, CIMP and MMR. To do so, first we mapped the CIN associated genes onto the network and then, expanded them using RWR network on the network. At last, by comparing with random permutations, the significant RWR expanded genes were identified as the downstream of CIN. Similarly, the downstream genes of CIMP and MMR can be identified.

Figure 4.

The workflow to investigate the cross talk among CIN, CIMP and MMR. (A) The CIN associated genes, CIMP associated genes and MMR associated genes were identified using mRMR and IFS methods. (B) The direct overlap between CIN genes, CIMP genes and MMR genes were little. (C) The genes that the CIN genes, CIMP genes and MMR genes affect were identified using RWR method. (D) When both the CIN genes, CIMP genes and MMR genes and their RWR genes were considered, the overlap among CIN, CIMP and MMR was significantly increased. (E) The biological functions of the common genes (the red star) were studied. CIN, chromosome instability; CIMP, CpG island methylator phenotype; MMR, mismatch repair; mRMR, minimal redundancy maximal relevance; IFS, incremental feature selection; RWR, random walk with restart.

The numbers of downstream genes of CIN, CIMP and MMR with permutation P-value <0.05 were 745, 709 and 807, respectively. Fig. 5 showed the overlap among CIN, CIMP and MMR and there were 236 common downstream genes of CIN, CIMP and MMR. These 236 genes were shown in Table IV. To statistically evaluate the significance of overlap, we calculated the odds ratio and P-value using R package Super Exact Test (43). The results were shown in Fig. 6. The odds ratio of overlap was 60.3 and the P-value was smaller than 1e-320.

Figure 5.

The Venn diagram of CIN downstream genes, CIMP downstream genes and MMR downstream genes. There were 236 common downstream genes of CIN, CIMP and MMR. CIN, chromosome instability; CIMP, CpG island methylator phenotype; MMR, mismatch repair; RWR, random walk with restart.

Table IV.

Common downstream genes of chromosome instability, CpG island methylator phenotype and mismatch repair.

List of common genes
A1BG	CD248	DEFB131	HECA	LCE1A	NAIP	SEMA4C	TRMU
A1CF	CDKAL1	DEFB134	HES2	LCE1B	NCDN	SERINC3	TSEN15
ABCC5	CEP120	DEFB135	HES3	LCE1D	NCR3	SERINC5	TSEN2
ABHD12	CFAP58	DNAJC9	HGFAC	LCE1E	NCR3LG1	SETDB2	TSEN34
ABHD6	CGREF1	DYSF	HHLA2	LCE3B	NSUN4	SLC16A7	TSEN54
ABI3	CGRRF1	EMB	HHLA3	LCE3C	NTNG1	SLC30A8	UBAP2
ABI3BP	CLASRP	ENAM	HLA-DOA	LCN1	NTNG2	SLC36A2	UNKL
ACOT13	CLEC2A	ETV7	HLA-DOB	LETM1	OR10H1	SLC3A1	UPK1A
ADAT1	CLK2	FAM149B1	HMGN3	LIAS	ORAOV1	SLC51A	UPK1B
ADAT2	CLK3	FAM3C	HOXC13	LIPT2	PANK1	SLC51B	UPK2
ADAT3	CLK4	FAT4	HPCA	LMBR1L	PANK2	SLC6A18	UPK3A
AGR2	CLN6	FBXO38	IGLL1	LRRC4	PANK3	SLC6A19	UPK3B
AGR3	CLN8	FJX1	IGSF3	LRRC4C	PANK4	SLC6A20	VEZT
AMBN	CNBD1	FLCN	IGSF6	LXN	PCTP	SLC6A9	VN1R1
AMICA1	COASY	FNIP2	IGSF9B	LYPD3	PHF11	SLC7A9	VNN2
ANO5	COMMD10	FOXQ1	IKBIP	MARCO	PIP	SMDT1	VPREB1
APOBEC1	COMMD7	FUZ	INTU	MCU	PLXDC1	SP8	XAGE1B
AZGP1	COMMD8	GABRR1	KBTBD6	MDGA1	PPCDC	SPICE1	XAGE2
BCS1L	CPA1	GABRR2	KBTBD7	METTL9	PPCS	SPINK9	YAE1D1
BFSP1	CPA4	GNPTAB	KCNK10	MFSD10	PRLH	SPINT1	YIPF3
BFSP2	CPN1	GNPTG	KCNK2	MICU1	PRLHR	ST14	YIPF4
BSCL2	CPN2	GP2	KCNK4	MICU2	PRSS8	STYX	YRDC
CARHSP1	CRISP3	GRID2	KIAA0319	MMS22L	PTCD3	SUGP2	ZCCHC17
CCDC109B	CRYBA1	GRID2IP	KLF7	MSRA	RASD2	TM2D1	ZFR
CCDC179	CRYBB1	GRXCR1	KLK5	MSRB2	RBBP9	TM2D2	ZNF461
CCDC68	CTAGE5	GSX2	KLRF2	MSRB3	RPUSD4	TMEM126B	ZNF772
CCIN	CXADR	GTPBP1	KRTAP24-1	MTERF4	RSRP1	TMEM19
CD101	CYLC1	GTPBP3	KRTAP25-1	MTO1	SCGB2A2	TMEM27
CD200	DCDC2	HAS1	KRTAP27-1	MUCL1	SCGB3A2	TONSL
CD200R1	DEFB110	HAS3	L3MBTL1	MYO7A	SDCBP2	TOX2

Figure 6.

The significance of overlap among CIN downstream genes, CIMP downstream genes and MMR downstream genes. To statistically evaluate the significance of overlap, we calculated the odds ratio and p value using R package Super Exact Test. The odds ratio of overlap was 60.3 and the P-value was smaller than 1e-320. CIN, chromosome instability; CIMP, CpG island methylator phenotype; MMR, mismatch repair.

The biological functions of the overlapped genes were investigated by enriching them onto KEGG and GO. The enrichment results were summarized in Table V. It can be seen that the significantly enriched KEGG pathways with FDR (false discovery rate) <0.05 were: hsa00770 Pantothenate and CoA biosynthesis, hsa00785 Lipoic acid metabolism and hsa04514 Cell adhesion molecules (CAMs). Similarly, the most significantly enriched GO terms were: GO:0015937 coenzyme A biosynthetic process, GO:0015936 coenzyme A metabolic process, GO:0033866 nucleoside bisphosphate biosynthetic process, GO:0034030_ribonucleoside bisphosphate biosynthetic process and GO:0034033 purine nucleoside bisphosphate biosynthetic process. These results indicated that the CIN, CIMP and MMR all affect biosynthetic and metabolic process and pathway to accelerate the tumorgenesis. In clinic, the metabolic syndrome was found to be able to increase the risk of colorectal cancer (44). And in colorectal cancer cell, there are aberration of various metabolites, such as nucleotides, amino acids, tricarboxylic acid, carbohydrates, and pentose-phosphate (45).

Table V.

Kyoto Encyclopedia of Genes and Genomes and Gene Ontology enrichments of common downstream genes of chromosome instability, CpG island methylator phenotype and mismatch repair.

Type	Gene set	FDR
KEGG	hsa00770 Pantothenate and CoA biosynthesis	4.35E-11
	hsa00785 Lipoic acid metabolism	0.0226
	hsa04514 Cell adhesion molecules (CAMs)	0.0476
GO BP	GO:0015937 coenzyme A biosynthetic process	9.66E-08
	GO:0015936 coenzyme A metabolic process	1.32E-06
	GO:0033866 nucleoside bisphosphate biosynthetic process	1.72E-06
	GO:0034030 ribonucleoside bisphosphate biosynthetic process	1.72E-06
	GO:0034033 purine nucleoside bisphosphate biosynthetic process	1.72E-06
	GO:0008033 tRNA processing	6.32E-05
	GO:0009451 RNA modification	0.000240
	GO:0033865 nucleoside bisphosphate metabolic process	0.000267
	GO:0033875 ribonucleoside bisphosphate metabolic process	0.000267
	GO:0034032 purine nucleoside bisphosphate metabolic process	0.000267
	GO:0015804 neutral amino acid transport	0.000561
	GO:0006865 amino acid transport	0.00215
	GO:0015807 L-amino acid transport	0.00215
	GO:0046942 carboxylic acid transport	0.00218
	GO:0000379 tRNA-type intron splice site recognition and cleavage	0.00218
	GO:0006399 tRNA metabolic process	0.00233
	GO:0015849 organic acid transport	0.00254
	GO:0036444 mitochondrial calcium uptake	0.00277
	GO:0015711 organic anion transport	0.00408
	GO:0008544 epidermis development	0.00408
	GO:0031424 keratinization	0.00458
	GO:0030855 epithelial cell differentiation	0.00458
	GO:0006820 anion transport	0.00458
	GO:1905039 carboxylic acid transmembrane transport	0.00473
GO MF	GO:0000213 tRNA-intron endonuclease activity	5.22E-05
	GO:0004594 pantothenate kinase activity	5.22E-05
	GO:0015171 amino acid transmembrane transporter activity	0.000748
	GO:0008514 organic anion transmembrane transporter activity	0.000748
	GO:0046943 carboxylic acid transmembrane transporter activity	0.000760
	GO:0008509 anion transmembrane transporter activity	0.00112
	GO:0005342 organic acid transmembrane transporter activity	0.00112
	GO:0015175 neutral amino acid transmembrane transporter activity	0.00162
	GO:0016892 endoribonuclease activity, producing 3′-phosphomonoesters	0.00230
	GO:0004549 tRNA-specific ribonuclease activity	0.0128
	GO:0015179 L-amino acid transmembrane transporter activity	0.0221
	GO:0005328 neurotransmitter:sodium symporter activity	0.0291
	GO:0005212 structural constituent of eye lens	0.0333
	GO:0016894 endonuclease activity, active with either ribo- or deoxyribonucleic acids and producing 3′-phosphomonoesters	0.0458
	GO:0008251 tRNA-specific adenosine deaminase activity	0.0462
GO CC	GO:1990246 uniplex complex	0.000199
	GO:0000214 tRNA-intron endonuclease complex	0.00661
	GO:0005886 plasma membrane	0.0114
	GO:0071944 cell periphery	0.0114
	GO:0031526 brush border membrane	0.0125
	GO:0098590 plasma membrane region	0.0125
	GO:0098862 cluster of actin-based cell projections	0.0148
	GO:0044459 plasma membrane part	0.0168
	GO:0001533 cornified envelope	0.0242

As a complex disease, the colorectal cancer can be caused by several different mechanisms. The three well-known one were CIN, CIMP and MMR. They were different but not exclusive. We investigated the genes that were associated with CIN, CIMP and MMR, separately using mRMR and IFS methods. Then by direct overlapping the CIN associated genes, CIMP associated genes and MMR associated genes, they share little common genes. Therefore, they were highly possible to interact with each other indirectly. To verify this idea, we identified the downstream genes that the CIN associated genes, CIMP associated genes and MMR associated genes may affect using RWR method. After the RWR analysis, the overlap between CIN, CIMP and MMR become significantly greater and the common downstream genes were involved in biosynthetic and metabolic process and pathway. These results can help explain the non-exclusiveness of CIN, CIMP and MMR and why they may co-occur from a protein-protein interaction network view. What's more, the common genes of CIN, CIMP and MMR can be possible targets of new broad-spectrum anti-cancer drugs that can treat more patients.