Copy number variation of ubiquitin- specific proteases genes in blood leukocytes and colorectal cancer

ABSTRACT

Ubiquitin-specific proteases (USPs) play important roles in the regulation of many cancer-related biological processes. USPs copy number variation (CNVs) may affect the risk and prognosis of colorectal cancer (CRC). We detected CNVs of USPs genes in 468 matched CRC patients and controls, estimated the associations between the USPs genes CNVs and CRC risk and prognosis and their interactions with environmental factors on CRC risk. Finally, we generated five CRC risk predictive models with different CNVs patterns combining with environmental factors (EF). We identified significant association between CYLD deletion and CRC risk (OR_adj = 4.18, 95% CI: 2.03–8.62), significant association between USP9X amplification and CRC risk (OR_adj = 2.30, 95% CI: 1.48–3.57), and significant association between USP11 deletion and CRC risk (OR_adj = 3.49, 95% CI: 1.49–8.64). There were significant gene-environment and gene–gene interactions on CRC risk. The area under the receiver operating characteristic curve (AUC) of EF + SIG (deletion of CYLD and USP11, amplification of USP9X) model was significantly larger than any other models (AUC = 0.75, 95% CI: 0.74–0.77). We did not identify significant associations between CNVs of the three genes and CRC prognosis. CNVs of CYLD, USP9X, and USP11 are significantly associated with the risk of CRC. Gene-gene and gene–environment interactions might also play an important role in the development of CRC.

KEYWORDS:

• Colorectal cancer
• copy number variation
• CYLD
• USP9X
• USP11

Introduction

Colorectal cancer (CRC) is the third most commonly diagnosed cancer, accounting for estimated 1.8 million new cases, over 881 000 cancer deaths worldwide and almost 9.2% of all cancer deaths in 2018.¹ Although the overall survival rate of CRC has been improved, the 5-year survival rate remains at approximately 30–60% in Asia.²

Genetic susceptibility independently or interacted with environmental factors play an important role in the etiology of CRC.^3,4 Copy number variation (CNV), as a molecular biomarker, is implicated in CRC risk and prognosis.⁵ DNA CNVs can be submicroscopic structural variation, large duplication or deletion covers >1kb.⁶

CNVs in the deubiquitinating enzymes (DUBs) of the ubiquitin-proteasome system (UPS) were reported to be associated with tumor risk and prognosis.^7–9 DUBs have a profound impact on the regulation of multiple biological processes including cell cycle control, DNA repair, and cancer-related signaling pathways through its deubiquitinating activity.^10,11 DUBs can be divided into three classes based on the mechanism of catalysis. The largest and most diverse class is the ubiquitin-specific protease (USP).¹² As members of USPs, CYLD, USP9X, USP11 have been proposed to play a vital role in a number of human cancers.^13–16

CYLD, as a tumor suppressor gene, is capable of regulating several signaling pathways such as NF-κB JNK and Akt pathways. Deregulation of CYLD expression has been associated with the risk of CRC.¹⁷ Recent evidence suggests that USP9X is involved in tumor progression through its deubiquitinating activity. Upregulation of USP9X was also associated with CRC risk and poor prognosis.¹⁴ As a member of UBPs, USP11 could deubiquitinate and stabilize some crucial tumor-related proteins and play a critical role in tumor development.^18,19 USP11 was identified to be down-regulated in tumor tissues, suggesting its potential tumor suppressor role in cancer development.¹⁵

Most studies of the three genes focus on gene expression in the cell lines or clinical pathological tissues. The relationship between germline CNVs of the three genes and CRC risk and prognosis have not been elucidated. We therefore carried out this study to explore the associations between germline CNVs of the three genes and CRC risk. We also conducted a prospective follow-up study to estimate the associations between CNVs of the three genes and CRC prognosis in China.

Materials and Methods

Subjects

This study was approved by the ethical committee of Harbin Medical University and a written informed consent was obtained from every patient. We identified CRC patients who were diagnosed with pathology and underwent surgery at the Cancer Hospital of Harbin Medical University. Patients with neuroendocrine carcinoma, malignant melanoma, non-Hodgkin’s lymphoma, gastrointestinal stromal tumors, and Lynch syndrome were excluded. Totally, we recruited 468 CRC patients from 1 November 2004 to 1 May 2010. During the same period, we collected cancer-free participants as controls from the 2nd Affiliated Hospital of Harbin Medical University. We excluded controls with history of gastrointestinal disease according to self-report. Finally, 468 controls matched for age, gender, and residence were recruited.

Data collection

As described in our previous study, each participant was face-to-face interviewed by well-trained interviewers using a structured questionnaire with information on demographic characteristics, family history of CRC, and dietary status during the past 12 months preceding the interview.²⁰ The clinical information of tumor size, TNM stage, chemotherapy, histological and pathological types, and the level of serum carcinoembryonic antigen (CEA) and carbohydrate antigen 19-9 (CA19-9) before surgery were collected from medical records.²⁰ Totally, 310 patients were followed up from November 2004 to March 2014 by hospital records or telephone interview. During follow-up, we obtained information on chemotherapy and radiotherapy protocol, disease progression, recurrence, and the date and cause of death (if deceased). Overall survival (OS) was defined at the primary endpoint in our study. Survival time was calculated from the first diagnosis of CRC to any cause of death or the time of follow-up. We validated the date and cause of CRC death through the medical certification of death and the Harbin death registration system.

DNA extraction and CNV detection

Genomic DNA was successfully extracted from peripheral blood leukocytes of 468 CRC and 468 controls using QIAGEN DNeasy Blood & Tissue Kit. MSI status was determined using PCR-SSCP.²¹

According to the Database of genomic variants (DGV), we selected the CNVs region of the three genes. Custom designed TaqMan Copy Number Assays were used to detect the copy number of genes (Supplementary Table S1). The quantitative assays were performed on the 7500 Fast Real-Time polymerase chain reaction machine in 96-well plates with a 10 µl reaction volume containing 20 ng DNA, 5ul TaqMan Universal PCR Master Mix, 0.5 µl of the CNV assay, and 0.5 µl of the reference RNaseP assay (Applied Biosystems, Carlsbad, Calif).

The PCR conditions were shown as follows: 95°C for 15 seconds and 60°C for 1 minute for 40 cycles. To ensure the accuracy of the results, a sample with two copies of every gene was used as quality control (standard sample) and every sample was repeatedly detected three times. 7500 software v2.0.6 (Applied Biosystems) was used to quantify the amplification cycle. Copy Caller version 2.0 (Applied Biosystems) was used to estimate every gene copy number of every sample. The copy number was calculated using the comparative cycle threshold (Ct) method. The formula was shown as below:

Copy number (CN) = 2 ^{– ΔΔCt}×2

$\mathrm{\Delta }\mathrm{\Delta }Ct=(\mathrm{\Delta }C{t}_{sample}\text{mmdash}\mathrm{\Delta }C{t}_{reference})\text{mmdash}(\mathrm{\Delta }C{t}_{standard}\text{mmdash}\mathrm{\Delta }C{t}_{reference})$

CN = 2 was defined as wild type, CN>2 was defined as amplification type and CN<2 was defined as deletion type.

Candidate genes selection

We conducted a pre-experiment to screen genes for the next phase of the experiment. We detected a total of nine USP genes CNVs including CYLD, USP2, USP4, USP7, USP9X, USP11, USP15 USP28, and UCHL5 in small paired samples (100 matched cases and controls) and calculated the statistical associations between CNVs of the nine genes and CRC risk (Supplementary Table S2). Finally, we chose CYLD, USP9X, and USP11 genes, which were significantly different between CRC cases and controls, as candidate genes to further detect CNVs in the next phase of the experiment.

Model evaluation

We used multivariable logistic regression model to identify significant factors for CRC risk with backward conditional selection method (P values of 0.05 and 0.10 were specified as the thresholds for entry and removal of variables, respectively) (Supplementary Table S3). To identify the predictive model for CRC risk, we constructed five models as follows: Model 1 was an environmental factor (EF) model with significant factors for CRC risk. The following demography and environmental factors were used for model evaluation: occupation, family history of CRC, consumption of roughage, fish, fat meat, egg, leftovers, and sausage. Models 2–4 were the combined models incorporating EF and each of the three genes CNVs (amp vs del+wt), deletion (del vs amp+wt), and variation (amp+del vs wt), respectively; Model 5 (EF+SIG) was the combination model of EF and three genes CNVs (deletion of CYLD and USP11, amplification of USP9X).

Model discrimination capacity was determined by receiver operating characteristic curve (ROC) and AUC. Delong method was used to assess the significant difference in the AUC among the five models. The 95% CI for the AUC was estimated using the MedCalc® statistical software.

Statistical analyses

We calculated and tested the Hardy–Weinberg Equilibrium in controls with Fisher’s exact test. Student’s t-test and Chi-squared test were used for evaluating homogeneity between cases and controls. Multivariate conditional logistic regression analysis was used to estimate the associations between CYLD, USP9X and USP11 gene CNVs and CRC risk with corresponding Odds ratios (OR) and 95% confidence intervals (95% CI). Crossover analyses were used to evaluate gene–environment interaction effects on CRC risk. We defined two copies as wild type (wt), more than two copies as amplification (amp) and less than two copies as deletion (del).²⁰ Three additive models were applied in the conditional logistic regression analysis: del v.s. amp + wt, amp v.s. del + wt and del + amp v.s. wt. Moreover, multifactor dimensionality reduction (MDR) analysis was used to explore the gene–gene interaction on the risk of CRC. We used balanced accuracy in the context of 10-fold cross-validation to assess the model quality and selected an overall best model with the maximum accuracy in the testing data (i.e. testing accuracy). We also recorded the cross-validation consistency (CVC). The Kaplan–Meier survival analysis was used to calculate the 3-year survival rate and 5-year survival rate. The associations between three genes CNVs and the prognosis of CRC were assessed by Cox proportional hazard regression analysis. All the statistical tests were two-sided, P values < .05 were considered statistically significant in the overall analysis, and P values < .025 were considered significant in the stratified analyses by Bonferroni correction. All the statistical analyses were performed by SPSS Statistics version 22.0 (IBM, Inc., USA).

Results

Characteristics of study population

The basic characteristics of the 468 paired cases and controls are summarized in Table 1. The distribution of occupation (P < .001) and family history of cancer (P < .001) were significantly different in cases and controls. Therefore, we adjusted for occupation and family history of CRC in the following analyses.

Table 1. Basic characteristics of cases and controls.

Characteristic	No.of case (%)	No.of controls (%)	P value*
Age	60.56 ± 11.06	59.70 ± 10.68	0.23
Gender			0.39
Male	254 (54.3)	254 (54.3)
Female	214 (45.7)	214 (45.7)
BMI (kg/m^2)	23.77 ± 3.84	23.40 ± 4.33	0.53
Education^a			0.63
Primary school and below	259 (59.4)	248 (56.5)
Junior middle school	88 (20.2)	91 (20.7)
Senior middle school and above	89 (20.4)	100 (22.8)
Occupation^a			< 0.001
White collar	88 (19.7)	57 (13.0)
Blue collar	228 (51.1)	264 (60.4)
Both	130 (29.1)	116 (26.5)
Family history of CRC^a			< 0.001
Yes	35 (9.9)	152 (34.0)
No	318 (90.1)	295 (66.0)
Location of primary tumor
Colon	227 (61.9)
Rectum	145 (38.1)
TNM stage ^a
I	36 (11.0)
II	138 (42.1)
III	119 (36.3)
IV	35 (10.6)
I+ II	174 (53.1)
III+ IV Microsatellite instability ^b MSS MSI-L MSI-H	154 (46.9) 249 (81.9) 37 (12.2) 18 (5.9)

^aMissing data of education: 32 cases, 29 controls; occupation: 22 cases, 31 controls; family history of CRC: 115 cases, 21 controls; location of primary tumor: 96 cases; TNM stage: 140 cases.

^bMSI-L, low-level microsatellite instability, MSI-H, high-level microsatellite instability; MSS, microsatellite stability.

*P < 0.05 was considered statistically significant.

CNVs of CYLD, USP9X, USP11, and the risk of CRC

The CNVs of CYLD, USP9X, and USP11 genes were in Hardy–Weinberg Equilibrium in all controls (P > .05). The three genes CNVs frequencies and their relationships with CRC risk are shown in Table 2.

Table 2. Associations between CYLD, USP9X and USP11 CNVs and the risk of CRC.

Gene	No.of case (%)	No.of controls (%)	Odds Ratio^a	95%Confidence Interval	P Value*
CYLD^b
Wt	286 (62.7)	312 (68.4)	1.00
Del	54 (11.8)	21 (4.6)	4.18	2.02, 8.62	< 0.001
Amp	116 (25.5)	123 (27.0)	0.94	0.62, 1.43	0.79
Del v.s Wt + Amp			4.26	2.10, 8.67	< 0.001
Del + Wt v.s. Amp			1.30	0.88, 1.94	0.19
Del +Amp v.s. Wt			0.76	0.52, 1.10	0.14
USP9X^b
Wt	151 (33.6)	187 (41.6)	1.00
Del	31 (6.9)	45 (10.0)	0.96	0.48, 1.92	0.92
Amp	268 (59.5)	218 (48.4)	2.30	1.48, 3.57	< 0.001
Wt + Amp v.s Del			1.54	0.81, 2.94	0.19
Amp v.s. Del + Wt			2.32	1.53, 3.53	< 0.001
Del + Amp v.s. Wt			1.93	1.28, 2.90	0.002
USP11^b
Wt	245 (55.8)	258 (58.8)	1.00
Del	68 (15.5)	52 (11.8)	3.49	1.49, 8.64	0.007
Amp	126 (28.7)	129 (29.4)	0.91	0.50, 1.66	0.76
Del v.s Wt + Amp			3.52	1.43, 8.71	0.006
Del + Wt v.s. Amp			1.17	0.64, 2.14	0.59
Del + Amp v.s. Wt			1.42	0.87, 2.31	0.16

^aAdjusted for occupation and family history of CRC. ^bMissing value, CYLD, 24; USP9X, 36; USP11, 58. *P < 0.05 was considered statistically significant.

We observed significant associations between CYLD deletion and increased CRC risk (del v.s. wt: OR_adj = 4.18, 95% CI: 2.03–8.62, P < .05; del v.s.wt + amp: OR_adj = 4.26, 95% CI: 2.10–8.67, P < .05), significant associations between USP9X amplification and CRC risk (amp v.s. wt: OR_adj = 2.30, 95% CI: 1.48–3.57, P < .05; amp v.s. del + wt: OR_adj = 2.32, 95% CI: 1.53–3.53, P < .05; del + amp v.s.wt: OR_adj = 1.93, 95% CI: 1.28–2.90, P < .05), as well as significant associations between USP11 deletion and CRC risk (del v.s. wt: OR_adj = 3.49, 95% CI: 1.49–8.64, P < 0.05; del v.s. amp + wt: OR_adj = 3.52, 95% CI: 1.43–8.71, P < .05). Moreover, we identified significant relationships between CNVs of CYLD and USP9X and the risk of rectal cancer in the subgroup analyses stratified by tumor location (Supplementary Table S4).

Interactions between CNVs of CYLD, USP9X, USP11 genes and environmental factors on the risk of CRC

In the del v.s. amp+wt model, we observed marginal antagonistic interaction between CYLD CNVs and fish intake on CRC risk (del v.s. amp + wt: OR_i = 0.20, 95% CI: 0.04–0.99, P = .05) (Table 3). We also identified significant combination effects between CYLD deletion and consumption of roughage (>50 g/week) (OR_eg = 3.12, 95% CI: 1.21–8.08), fat meat (OR_eg = 3.38, 95% CI: 1.01–11.30), egg (>3/week) (OR_eg = 5.36, 95% CI: 2.17 –13.24), sausage (>1 times/month) (OR_eg = 1.98, 95% CI: 1.15–3.43), and leftovers (>3 times/week) (OR_eg = 5.40, 95% CI: 1.83–15.92) on CRC risk. Although there was no significant interaction between USP11 CNVs and environmental factors on CRC risk. We observed significant combination effects between USP11 deletion and consumption of fat meat, egg (>3/week), sausage (>1times/month) and leftovers (>3times/week) (OR_eg = 7.44, 95% CI: 2.28–24.29; OR_eg = 7.83, 95% CI: 2.50–24.57; OR_eg = 8.82, 95% CI: 1.70–45.74; OR_eg = 6.91, 95% CI: 2.34–20.35, respectively) on CRC risk.

Table 3. Interactions between three genes CNVs and environmental factors on the risk of CRC.

CNV genotypes		Environmental factors			Interaction
CYLD		Roughage (g/week)
		≤50		>50
		OReg (95% CI)		OReg (95% CI)		P value*
	Wt + Amp	1.00		0.55 (0.37, 0.81)
	Del	3.37 (1.11, 10.25)		3.12 (1.21, 8.08)	1.70 (0.39, 7.39)	0.48
		Fat meat
		NO		YES
		OReg (95% CI)		ORe g(95% CI)		P value*
	Wt+Amp	1.00		1.19 (0.72, 1.83)
	Del	5.16 (2.15, 12.35)		3.38 (1.01, 11.30)	0.57 (0.13, 2.54)	0.46
		Fish (times/week)
		≤1		>1
		OReg (95% CI)		OReg (95% CI)		P value*
	Wt+Amp	1.00		0.63 (0.41, 0.97)
	Del	10.92 (5.81, 20.49)		1.39 (0.69, 2.83)	0.20 (0.04, 0.99)	0.05
		Egg (/week)
		≤3		>3
		OReg (95% CI)		OReg (95% CI)		P value*
	Wt	1.00		1.72 (1.08, 2.73)
	Del+Amp	1.29 (0.70, 2.36)		2.31 (1.38, 3.86)	1.05 (0.47, 2.34)	0.91
	Wt+Amp	1.00		1.90 (1.29, 2.76)
	Del	7.54 (2.37, 24.03)		5.36 (2.17, 13.24)	0.38 (0.09, 1.61)	0.19
		Sausage (times/month)
		≤1		>1
		OReg (95% CI)		OReg (95% CI)		P value*
	Wt	1.00		1.87 (0.87, 3.99)
	Del+Amp	1.29 (0.82, 2.02)		2.79 (1.29, 6.07)	1.16 (0.38, 3.52)	0.79
	Wt+Amp	1.00		1.33 (0.99, 1.82)
	Del	1.68 (1.15, 2.46)		1.98 (1.15, 3.43)	0.89 (0.44, 1.81)	0.75
		Leftovers^b (times/week)
		≤3		>3
		OReg (95% CI)		OReg (95% CI)		P value*
	Wt	1.00		1.96 (1.24, 3.09)
	Del+Amp	1.60 (0.97, 2.62)		2.23 (1.27, 3.92)	0.71 (0.33, 1.53)	0.39
	Wt+Amp	1.00		1.78 (1.22, 2.62)
	Del	5.32 (2.04, 13.86)		5.40 (1.83, 15.92)	0.57 (0.13, 2.45)	0.45
USP9X		Fish (times/week)
		≤1		>1
		OReg (95% CI)		OReg (95% CI)		P value*
	Wt	1.00		1.00 (0.54, 1.85)
	Del+Amp	3.17 (1.93, 5.19)		1.04 (0.56, 1.93)	0.33 (0.14, 0.78)	0.01
		Fat meat
		NO		YES
		OReg (95% CI)		OReg (95% CI)		P value*
	Del + Wt	1.00		1.59 (0.85, 2.96)
	Amp	1.90 (1.10, 3.27)		3.39 (1.88, 6.11)	1.12 (0.52, 3.42)	0.77
	Wt	1.00		1.55 (0.87, 2.75)
	Del + Amp	2.22 (1.30, 3.78)		3.95 (2.20, 7.10)	1.15 (0.53, 2.47)	0.72
		Egg (/week)
		≤3		>3
		OReg (95% CI)		OReg (95% CI)		P value*
	Del + Wt	1.00		2.83 (1.45, 5.51)
	Amp	3.01 (1.55, 5.86)		4.81 (2.45, 9.44)	0.56 (0.26, 1.24)	0.16
	Wt	1.00		2.59 (1.40, 4.79)
	Del +Amp	3.37 (1.76,6.49)		5.11 (2.69,9.72)	0.59 (0.27, 1.27)	0.18
		Sausage (times/month)
		≤1		>1
		OReg (95% CI)		OReg (95% CI)		P value*
	Del + Wt	1.00		3.27 (0.85, 12.61)
	Amp	2.92 (1.31, 4.02)		3.49 (1.67, 7.25)	0.47 (0.10, 2.10)	0.29
	Wt	1.00		2.76 (0.86, 8.85)
	Del +Amp	2.55 (1.53, 4.25)		3.88 (1.86, 8.07)	0.55 (0.13, 2.40)	0.41
		Leftovers^b (times/week)
		≤3	>3
		OReg (95% CI)	OReg (95% CI)			P value*
	Del + Wt	1.00	1.85 (1.03, 3.31)
	Amp	1.98 (1.20, 3.29)	3.77 (2.09, 6.82)		1.03 (0.50, 2.10)	0.94
	Wt	1.00	2.00 (1.14, 3.48)
	Del +Amp	2.51 (1.49, 4.24)	4.64 (2.25, 8.52)		0.93 (0.45, 1.89)	0.83
USP11		Fat meat
		NO	YES
		OReg (95% CI)	OReg (95% CI)			P value*
	Wt	1.00	1.69 (1.05, 2.73)
	Del +Amp	1.48 (0.78, 2.78)	2.46 (1.28, 4.71)		0.98 (0.45, 2.16)	0.97
	Wt + Amp	1.00	1.64 (1.10, 2.45)
	Del	3.28 (1.12, 9.56)	7.44 (2.28, 24.29)		1.38 (0.43, 4.48)	0.59
		Egg (/week)
		≤3	>3
		OReg (95% CI)	OReg (95% CI)			P value*
	Wt	1.00	1.89 (1.16, 3.10)
	Del + Amp	1.52 (0.80, 2.91)	2.97 (1.54, 5.75)		1.03 (0.49, 2.22)	0.93
	Wt + Amp	1.00	1.89 (1.28, 2.80)
	Del	3.56 (1.26, 10.02)	7.83 (2.50, 24.57)		1.16 (0.36, 3.75)	0.80
		Sausage (times/month)
		≤1		>1
		OReg (95% CI)		OReg (95% CI)		P value*
	Wt + Amp	1.00		1.99 (1.03, 3.85)
	Del	3.34 (1.28, 8.72)		8.82 (1.70, 45.74)	1.33 (0.25, 7.05)	0.74
		Leftovers^b(times/week)
		≤3		>3
		OReg (95% CI)		OReg (95% CI)		P value*
	Wt	1.00		1.18 (0.73, 1.89)
	Del + Amp	0.89 (0.48, 1.65)		2.59 (1.39, 4.83)	2.48 (1.13, 5.41)	0.02
	Wt + Amp	1.00		1.47 (0.98, 2.21)
	Del	2.07 (0.68, 6.32)		6.91 (2.34, 20.35)	2.27 (0.67, 7.77)	0.19

CI, confidence interval. ^aAdjusted for occupation and family history of CRC. ^b leftovers: leftovers more than 12 hours. *P < 0.05 was considered statistically significant.

In the amp v.s. del + wt model, there were significant combination effects between the amplification of USP9X and consumption of fat meat, egg (>3/week), sausage (>1 times/month) and leftovers (>3times/week) (OR_eg = 3.39, 95% CI: 1.88–6.11; OR_eg = 4.81, 95% CI: 2.45–9.44; OR_eg = 3.49, 95% CI: 1.67–7.25; OR_eg = 3.77, 95% CI: 2.09–6.82, respectively) on CRC risk. However, we did not observe significant interactions between USP9X CNVs and environmental factors on CRC risk.

In the model of del + amp v.s. wt, there were significant combination effects between the del + amp genotype of CYLD, USP9X and USP11 CNVs and consumption of egg (>3/week), and leftovers (>3 times/week) on CRC risk, respectively. We also observed significant combination effects between del + amp genotype of CYLD, USP9X CNVs and consumption of sausage (>1 times/month) on CRC risk, as well as significant combination effects between USP9X, USP11 del + amp genotype and consumption of fat meat on CRC risk. Moreover, we identified a significant antagonistic interaction between the del + amp genotype of USP9X CNV and fish intake on CRC risk (OR_i = 0.33, 95% CI: 0.14–0.78, P < 0.05), and a significant synergistic interaction between the del + amp genotype of USP11 CNV and consumption of leftovers (>3 times/week) on CRC risk (OR_i = 2.48, 95% CI: 1.13–5.41, P < 0.05).

Performance of CRC risk predictive models

Firstly, we constructed EF-only model, with the AUC of 0.72 (95% CI: 0.71–0.74, P < .0001), and then, we added genes CNVs by different variation patterns to the EF-only model. The AUCs for combined-models of EF + amp, EF + del, EF + var and EF+SIG were 0.735 (95% CI: 0.72–0.75, P < .0001), 0.747 (95% CI: 0.73–0.76, P < .0001) and 0.736 (95% CI: 0.72–0.75, P < .0001), and 0.754 (95% CI: 0.74–0.77, P < .0001), respectively (Table 4). The EF + SIG model represented significantly higher discrimination accuracy than any other combined-models (Supplementary Table S5) and with an AUC increase of 2.96% (95% CI: 0.02–0.04, P < .0001) comparing to the EF-only model.

Table 4. The AUC of CRC risk-predicting models.

Model	AUC	SE	95% CI	*P value
EF	0.724	0.00730	0.711, 0.737	<0.001
EF + AMP	0.735	0.00717	0.722, 0.747	<0.001
EF + DEL	0.747	0.00703	0.734, 0.759	<0.001
EF + VAR	0.736	0.00719	0.723, 0.748	<0.001
EF + SIG	0.754	0.00695	0.741, 0.766	<0.001

AUC, the area under the receiver operating characteristic curve; SE, standard error; CI, confidence interval. *P < 0.05 was considered statistically significant.

Gene–gene interaction using MDR

We used MDR analysis to explore the gene–gene interactions between CYLD, USP9X, and USP11 on the risk of CRC. Associations of gene–gene higher-order interactions on CRC risk were summarized in Supplementary Table S6. The MDR model with best testing accuracy consisted of CNVs of USP9X and USP11 (TA = 0.5707, P = .0107). The interaction had a maximum testing accuracy of 57.07% and a maximum CVC (cross-validation consistency) of 10 out of 10 followed by statistical significance of 1,000-fold permutation test (P < .01).

We also estimated the interaction between CNVs of USP9X and USP11 on CRC risk in the multivariate logistic regression analyses. There were significant synergistic interactions between USP9X and USP11 CNVs on the risk of CRC (Supplementary Table S7).

CNVs of CYLD, USP9X, USP11 and CRC prognosis

A total of 310 patients completed the follow-up. Of the 310 patients, 170 were males (45.2%) and 140 were females (54.8%), with the mean age of 59.31 ± 10.86. The mean overall survival (OS) of CRC patients was 54.59 ± 1.53 months. The clinical and pathological features of the 310 CRC patients are shown in Supplementary Table S8. There were no associations between CNV changes v.s. TNM stage and MSI status (Supplementary Table S9). The pathological type of CRC, preoperative CA19-9 and CEA level, and TNM stage were adjusted for in the univariate Cox proportional hazards regression analysis for the associations between CYLD, USP9X and USP11 CNVs and prognosis of CRC, due to their significant associations with CRC prognosis.

We did not observe statistically significant associations between CNVs of CYLD, USP9X, and USP11 and CRC prognosis (Table 5), and colon or rectal cancer prognosis in the subgroup analyses stratified by tumor location (Table 6). Moreover, in the stratified analyses by MSI status, we did not observe significant associations between CNVs of the three genes and the prognosis of CRC (Supplementary Table S10).

Table 5. The relationships between gene CNVs and prognosis of CRC.

CNV genotypes	Patients (%)	5-year survival (%)	3-year survival (%)	OS (Mean (SD)) (month)	HR (95% CI)^a	P value*
CYLD^b
Wt	192 (62.3)	63	76	54.09 (1.98)	1.00
Del	41 (13.3)	80	80	53.79 (3.14)	0.85 (0.27, 2.67)	0.78
Amp	75 (24.4)	52	79	53.83 (2.60)	0.69 (0.29, 1.63)	0.39
Del v.s Wt +Amp					1.10 (0.35, 3.47)	0.87
Amp v.s. Del + Wt					1.44 (0.61, 3.41)	0.41
Wt v.s Del +Amp					0.73 (0.35, 1.53)	0.41
USP9X^b
Wt	102 (33.2)	64	64	50.37 (2.36)	1.00
Del	22 (7.2)	95	95	63.10 (2.84)	0.10 (0.01, 1.13)	0.11
Amp	183 (9.6)	52	69	53.56 (2.05)	0.56 (0.26, 1.23)	0.10
Del v.s Wt +Amp					6.07 (0.61, 60.34)	0.12
Amp v.s. Del + Wt					1.22 (0.58, 2.57)	0.61
Wt v.s Del +Amp					0.53 (0.24, 1.16)	0.11
USP11^b
Wt	160 (55.7)	88	92	53.28 (1.87)	1.00
Del	51 (17.8)	67	75	52.22 (3.54)	1.35 (0.51, 3.57)	0.54
Amp	76 (26.5)	60	67	53.02 (3.31)	1.14 (0.53, 2.44)	0.75
Del v.s Wt +Amp					1.29 (0.51, 3.28)	0.59
Amp v.s. Del + Wt					0.94 (0.45, 1.95)	0.87
Del + Amp v.s. Wt					0.83 (0.42, 1.64)	0.60

CI, confidence interval; HR, hazard ratio; OS, overall survival. ^aAdjusted for CEA and CA19-9 level before surgery, TNM stage, pathological type, and metastasis. ^bMissing value: CYLD, 2; USP9X, 3; USP11, 23. *P < 0.05 in the survival analysis was considered statistically significant.

Table 6. The relationships between gene CNVs and prognosis of colon cancer and rectal cancer.

	Colon cancer						Rectal cancer
	Patients (%)	5-year survival (%)	3-year survival (%)	OS (Mean (SD)) (month)	HR (95% CI)	P value*	Patients (%)	5-year survival (%)	3-year survival (%)	OS (Mean (SD)) (month)	HR (95% CI)	P value*
CYLD^a
Wt	61 (59.2)	72	80	53.67 (2.97)	1.00		125 (64.1)	60	75	52.29 (2.42)	1.00
Del	12 (11.7)	100	100		0.02 (0.00, 1116.04)	0.49	28 (14.4)	71	71	49.78 (4.26)	1.70 (0.46, 6.34)	0.42
Amp	30 (29.1)	69	78	51.53 (4.61)	1.16 (0.22, 6.11)	0.86	42 (21.5)	58	84	54.79 (3.22)	0.49 (0.13, 1.86)	0.29
Del v.s Wt+Amp					641.18 (0.00, 93629.36)	0.47					0.54 (0.14, 1.99)	0.26
Amp v.s. Del +Wt					0.68 (0.13, 3.48)	0.64					2.14 (0.57, 8.02)	0.69
Wt v.s Del+Amp					0.77 (0.17, 3.52)	0.77					0.83 (0.32, 2.13)	0.54
USP9X^a
Wt	34 (33.3)	73	73	51.35 (4.47)	1.00		66 (33.8)	60	60	47.63 (2.63)	1.00
Del	10 (8.8)	100	100		0.01(0.00, 4719.17)	0.51	13 (6.7)	92	92	60.92 (4.87)	0.39 (0.13, 8.18)	0.97
Amp	59 (57.8)	40	76	52.91 (3.12)	0.32(0.05, 2.27)	0.25	116 (59.5)	58	67	53.77 (2.46)	0.60 (0.25, 1.42)	0.25
Del v.s Wt+Amp					159.20 (0.00, 1.26E+15)	0.74					0.71 (0.09, 5.48)	0.74
Amp v.s. Del+Wt					1.80 (0.34, 9.44)	0.49					1.67 (0.72, 3.88)	0.23
Wt v.s Del+Amp					0.31 (0.04, 2.26)	0.25					0.62 (0.26, 1.45)	0.27
USP11^a
Wt	62 (62.6)	43	71	53.67 (2.71)	1.00		95 (52.8)	65	65	52.71 (2.42)	1.00
Del	16 (16.2)	92	92	61.33 (4.47)	0.00 (0.00, 3873.97)	0.30	33 (18.3)	55	69	47.14 (4.07)	3.37 (1.15, 9.87)	0.03
Amp	21 (21.2)	40	40	41.85 (7.30)	3.50 (0.26, 47.58)	0.35	52 (28.9)	64	73	55.51 (3.52)	1.63 (0.63, 4.18)	0.31
Del v.s Wt+Amp					41740.8 (0.00, 1.15E+19)	0.53					0.36 (0.14, 0.96)	0.04
Amp v.s. Del+Wt					0.61 (0.11, 3.43)	0.56					2.06 (0.90, 4.71)	0.69
Wt v.s Del+Amp					0.28 (0.04, 1.82)	0.18					0.84 (0.36, 1.98)	0.09

Note: In the analysis stratified by tumor location, there was a total of 300 cases included due to the 10 cases who missed detail information about tumor location. ^aMissing value, CYLD, 2 in colon cancer; USP9X, 2 in colon cancer; USP11, 6 in colon cancer and 15 in rectal cancer. *P < 0.025 in the stratified survival analysis by location was considered statistically significant.

Discussion

In our study, we investigated the relationships between CNVs of CYLD, USP9X and USP11 genes and CRC risk and prognosis. To the best of our knowledge, this is the first study on the associations between germline CNVs of the three genes and CRC risk and prognosis. There were significant associations between the CNVs of CYLD, USP9X, and USP11 and increased CRC risk, significant gene–environment interactions on the risk of CRC: CNVs of USP9X interact with fish consumption and USP11 CNVs interact with leftover consumption. In the CRC risk prediction model evaluation, EF+SIG model has the best performance in CRC risk predicting. Moreover, there were gene–gene interactions between CNVs of USP9X and USP11 on CRC risk.

As a tumor suppressor gene, CYLD is widely reported to be linked to the development of various human malignancies including CRC because of its deubiquitinating activity.^17,22 Significant associations were reported between copy number reduction of CYLD and risk of uterine cervix carcinoma, kidney cancer, and hepatocellular carcinoma.^7,23,24 Suppressed mRNA and protein expression of CYLD also contributed to tumorigenesis including colon, hepatocellular carcinoma, and melanoma.^17,25 Reduced or undetectable CYLD mRNA expression was also observed in 10/10 colon carcinoma tissues. Moreover, compared with normal colonic epithelial cells, loss of CYLD protein expression was observed in two colon carcinoma cell lines.¹⁷ In our study, we also observed that CYLD deletion increased CRC risk by 4.2-fold compared to wt genotype and 4.3-fold compared to wt+amp genotype. The following mechanisms may explain our results: CYLD is capable of controlling cell proliferation, differentiation, and migration by regulating cancer-related signaling pathways (JNK, Akt, and Wnt pathways). CYLD could also negatively regulate NF-κB signaling by deubiquitinating NEMO and several IKK upstream regulators to play its tumor suppressor function.^26–28 Low expression of CYLD was reported to be associated with poor prognosis in many cancers.^16,29,30 However, we did not observe significant associations between CNVs of CYLD and prognosis of CRC. The lack of associations between CNVs of CYLD and CRC prognosis in our study may reflect the peripheral blood sample source instead of tumor tissues, cell lines, and animal models.

As a member of the peptidase C19 family, USP9X participated in a large number of biological processes which played a pivotal role in human cancers, both as oncogene or tumor suppressor gene.^14,31-34 USP9X was identified as a tumor-suppressor gene in pancreatic duct adenocarcinoma (PDA), inactivated in over 50% of the tumors. Moreover, disruption of USP9X was observed to accelerate tumorigenesis in mouse model, suggesting USP9X might suppress tumor formation in some epithelial tissues in PDA.³¹ In contrast, a considerable amount of studies have demonstrated that elevated expression and increased activity of USP9X were associated with tumor progression in a number of cancers including multiple myeloma, esophageal squamous cell carcinoma, non-small cell lung cancer, and colon cancer.^14,32-34 We also identified that USP9X amplification could increase 2.3-fold CRC risk and USP9X CNVs (del + amp genotype) could increase 1.93-fold CRC risk as compared with wild genotype of USP9X. The role of USP9X in cancer development is controversial, which may be explained by the tissue specificity of USP9X in different tumors.³³ We speculated that USP9X may function as an oncogene in the development of CRC. Previous studies have reported increased expression in colon tissues, which were consistent with our results.¹⁴ Meanwhile, high USP9X expression may regulate certain genes, such as MCL1 and β-catenin, which may also illuminate the oncogene role of USP9X in CRC.^14,32,35 Both elevated and low expression of USP9X in tumor cells have been reported to serve as independent poor prognostic factor for many cancers.^31–33 Accumulating evidence demonstrate that associations between genetic alterations including CNVs and CRC risk differ for right-sided colon cancer and left-sided colon cancer (including rectal carcinomas), suggesting that carcinogenetic mechanism and progression of CRC may differ with tumor location.^36–38 In our study, we observed that the risk of colon and rectal cancers is different for CYLD and USP9X, which may provide the similar conclusion that different genetic pathways of carcinogenesis exit in right and left-sided colon cancer. In our study, we did not observe a significant association between CNVs of USP9X and prognosis of CRC. Larger number of CRC patients may be needed to shed light on the relationship between CNVs of USP9X and CRC prognosis. Moreover, we observed antagonistic interaction between CNVs of USP9X and consumption of fish on CRC risk, which may be explained by the protective effects of fish consumption in cancers including CRC due to the enrichment of omega-3 polyunsaturated fatty acids (PUFAs) in fish.^39,40

We observed that USP11 deletion could significantly increase 3.49-fold CRC risk. Previous studies have reported tumor suppressive role of USP11 in brain tumor tissues,¹⁵ skin cancer tissues, and lung adenocarcinoma cells.^41,42 While there is no report about USP11 CNVs in CRC cells or tumor tissue, especially in population samples. To our knowledge, it is the first study to reveal the relationship between CNVs of USP11 and risk of CRC. USP11 was involved in the regulation of inflammation, immunity, cell proliferation, transformation, and apoptosis.^19,41-44 USP11 could interact with and stabilize p53,¹⁹ negatively regulate TNFα-mediated NF-кB activation through targeting IκBα,⁴³ and control VGLL4 protein stability by promoting its deubiquitination and act as a tumor suppressor gene by modulating VGll4/YAP-TEADs regulatory loop.⁴⁴ USP11 could also regulate BRCA2 expression level through deubiquitinating and participate in the cellular response to DNA damage.¹⁸ Recent studies have revealed the association between abnormal expression of USP11 and prognosis of breast cancer.⁴⁵ However, we did not observe any significant association between USP11 CNVs and CRC prognosis. Small sample size of CRC patients presented with USP11 CNVs may limit the statistical power. Significant synergistic interaction existed between CNVs of USP11 and leftovers consumptions on CRC risk, which may be explained by the risk effect of leftover consumption on cancers due to the nitroso compounds or other carcinogens generating in leftovers.^46–48

In our study, we developed five CRC risk-predicting models. The AUC of EF-only model was 0.724, which was comparable to other CRC risk models in previous studies (AUCs ranging from 0.61 to 0.76).^49–51 The EF+SIG model significantly improved the discriminatory performance by 0.0296 unit with the AUC of 0.754 comparing to EF-only model, suggesting its potential to identify individuals at high risk of CRC. However, the discriminatory performance of EF+SIG predictive model was only moderate, which was similar to the combination model of EF and DNA methylation in our previous study.⁵² In the gene–gene interaction exploring by MDR analysis, we identified significant synergistic interactions between USP9X and USP11 variants on the risk of CRC. As the members of USPs, USP9X and USP11 may affect biological processes through some common pathways leading to the superposition of susceptibility effects on CRC.

We must consider the limitations in deriving conclusions as case-control and prospective survival study. Recall bias and selection bias might exist since this is a hospital-based case-control study. Moreover, the small sample size in the prospective survival study may limit the statistical power to reveal the associations between the three genes CNVs and CRC prognosis.

In summary, our study suggested that CNVs of CYLD, USP9X, and USP11 are significantly associated with the risk of CRC. Gene–gene and gene–environment interactions might also play an important role in the development of CRC.

Disclosure of potential conflicts of interest

The authors have no conflicts of interest.

Supplementary material

Supplemental data for this article can be accessed on the publisher’s website.

Additional information

Funding

This work was supported by Natural Science Foundation of China [Grant No. 81302483]; Fifty-second batch of the Postdoctoral Science Foundation of P. R. China [Grant No. 2012M520773]; Postdoctoral Science Foundation of the government of Heilongjiang Province [Grant No. LBH-Z11070].