CASP7 variants modify susceptibility to cervical cancer in Chinese women

Polymorphisms in Caspase-7 (CASP7) may modulate the programmed cell death and thus contribute to cervical cancer risk. In this case-control study of 1,486 cervical cancer cases and 1,301 controls, we investigated associations between four potentially functional polymorphisms in CASP7 and cervical cancer risk and evaluated their locus-locus interaction effects on the risk. The genotype-phenotype correlation was performed by a generalized linear regression model. We found that the rs4353229 polymorphism was associated with cervical cancer risk (under a recessive model: crude OR = 1.20, 95% CI = 1.02–1.40). Compared with the TT genotype, the rs10787498GT genotype was associated with an increased cervical cancer risk (adjusted OR = 1.19, 95% CI = 1.00–1.41). Combination analysis showed that subjects with four putative risk genotypes had a 1.54-fold increased cancer risk, compared with those who carried three or less putative risk genotypes. We also observed significant locus-locus joint effects on the risk, which may be mediated by the polymorphisms regulating CASP7 mRNA expression. Subsequent multifactor dimensionality reduction and classification and regression tree analyses indicated that the CASP7 genotypes might have a locus-locus interaction effect that modulated cervical cancer risk. Out data suggest that CASP7 polymorphisms may interact to modify cervical cancer risk by a possible mechanism of regulating CASP7 mRNA expression.

C ervical cancer is one of the leading cancers in women worldwide, with 529,800 new-diagnosed cancer cases and 275,100 cancer deaths in 2008 1 . More than 85% of these cases and deaths occur in developing countries, including China 1 . Accumulated molecular epidemiologic data support the hypothesis that persistent infection with oncogenic high-risk types of human papillomavirus (HPV) is the primary, even necessary cause of cervical cancer 1,2 . However, only a small fraction of women with HPV infection eventually develop cervical cancer, suggesting a wide range of inter-individual genetic variability in cervical cancer susceptibility 3 . Recently, two genome-wide association studies showed that some single nucleotide polymorphisms (SNPs) in the major histocompatibility complex region were associated with cervical cancer risk in both Caucasian and Chinese Han populations 4,5 . Despite these successes in identifying genetic variants for cervical cancer risk, the causal variants and/or mechanisms underlying the etiology have been determined for only a small fraction of these associations 6 . Recently, investigations of potentially functional SNPs have now been increasingly advocated across diseases. For example, SNPs at microRNA (miRNA)-binding sites in the 39-untranslated region (UTR) can remarkably alter the biogenesis and/or function of the corresponding miRNAs and thus contribute to cervical carcinogenesis 7 .
Caspases, at the heart of the apoptotic machinery, encode an evolutionary conserved family of cysteine-aspartic acid proteases and coordinate in cellular regulation and execution of apoptosis 8 . Together with caspase-3 and -6, caspase-7 belongs to the subgroup of executioner caspases 9 , and it executes a coordinated program of proteolysis that leads to the final programmed cell death 10 . Besides its activation during apoptosis, proteolytic maturation of caspase-7 has also been observed in inflammatory conditions 11 , which indicates a potential mechanism of caspase-7 involving the process of HPV infection and host immune response in cervical cancer. Previous genetic association studies had revealed that polymorphisms in the Caspase-7 (CASP7) gene may modulate the default programmed cell death, thus leading to genomic instability and contributing to inter-individual variation in cancer susceptibility 12,13 .
To date, no published studies have investigated associations between functional CASP7 SNPs and cervical cancer risk, besides genome-wide association studies. Herein, we performed a relatively large case-control study to test the hypothesis that potentially functional SNPs in the CASP7 39-UTR are independently and/or jointly associated with cervical cancer risk.

Results
Population characteristics. The selected characteristics of the study subjects are listed in Supplementary Table S1. There was no significant difference in distributions of age between the 1,486 cases and 1,301 controls (P 5 0.126) as result of matching. However, the differences in age at primiparity, menopausal status and body mass index (BMI) were significant between cases and controls. Therefore, we subsequently adjusted these variables for any residual confounding effect in multivariate logistic regression analyses.
Association of CASP7 SNPs with cervical cancer risk. As shown in Table 1, compared with CC/CT genotypes, the rs4353229TT genotype was associated with a significantly increased risk of cervical cancer [crude odds ratio (OR) 5 1.20, 95% confidence interval (CI) 5 1.02-1.40], but after adjustment for age, age at primiparity, menopausal status and BMI, this association was no longer statistically significant. In addition, the rs10787498GT genotype was associated with an increased risk of cervical cancer, compared with the TT genotype (adjusted OR 5 1.19, 95% CI 5 1.00-1.41). No risk association was observed for the other two SNPs (i.e., rs12247479 and rs1127687), nor for the haplotypes of these four CASP7 SNPs (Supplementary Table S2). However, when combining these four SNPs and assuming a dominant genetic model, we found that those women who carried four putative risk genotypes had a 1.54fold increased risk (95% CI 5 1.07-2.22) of cervical cancer, compared with those who carried three or less putative risk genotypes (Table 1). Further stratified analyses showed that the significantly increased risk of cervical cancer associated with the rs10787498 GT/GG genotype was more prominent in women younger at primiparity (adjusted OR 5 1.40, 95% CI 5 1.08-1.82, P for homogeneity test 5 0.014; Supplementary Table S3).
In further logistic regression analyses, we observed a significant locus-locus multiplicative interaction between rs1127687 and rs12247479 as well as between rs1127687 and rs10787498 (P 5 0.016 and 0.007, respectively; data not shown). We then explored their two-locus joint effects. As shown in Table 2, women who carried rs1127687AG/AA-rs12247479AG/AA genotypes and those who carried rs1127687AG/AA-rs10787498GT/GG genotypes had a significantly increased risk of cervical cancer, compared with carriers of  We then calculated false-positive report probability (FPRP) values for all observed significant associations ( Table 3). The rs4353229TT genotype was associated with an increased risk of cervical cancer with a statistical power of 99.8%, compared with CT/CC genotypes. When the assumption of prior probability was 0.1, the association with rs4353229 was noteworthy in all patients and in the subgroup of postmenopausal women (FPRP 5 0.190 and 0.191, respectively), similar for the association of rs10787498GT/GG genotypes in the subgroup of younger at primiparity (FPRP 5 0.043) as well as for that of the four putative risk genotype combination effect (FPRP 5 0.162). Meanwhile, the two-locus joint effect was still noteworthy for rs10787498-rs4353229 and rs1127687-rs10787498 (FPRP 5 0.020 and 0.024, respectively).
Association of high-order interactions with cervical cancer risk. We further performed the multifactor dimensionality reduction (MDR) analysis and found that rs4353229 was the best one-factor model with the highest cross-validation consistency (CVC) (89%) and the lowest prediction error (48.0%) among all four SNPs. Additionally, the four-locus model had a maximal CVC (100%) and a minimal prediction error (46.8%), suggesting a better prediction than other models ( Figure 1A). Subsequent hierarchical cluster analysis placed rs4353229 and rs10787498, rs12247479 and rs1127687 on the same branch ( Figure 1B), suggesting that this fourlocus model might have an interaction effect by modulating cervical cancer risk, which is also supported by the interaction graph ( Figure 1C). Moreover, consistent with the findings in the single locus analysis, rs4353229 (0.13%) and rs10787498 (0.12%) showed a strong effect on cervical cancer risk ( Figure 1C).
By the classification and regression tree (CART) analysis, we found rs4353229 to be the initial split of root nodes, indicating that rs4353229 was the strongest risk factor for cervical cancer among these four SNPs. Further inspection of the tree structure revealed distinct interaction patterns. Women carrying rs4353229TT, rs12247479GG, rs10787498GT/GG and rs1127687AG/AA genotypes [terminal node (TN) 6] had a 1.67-fold increased risk of cervical cancer, compared with the TN3 group at the lowest risk (P 5 0.041; Figure 1D).
Correlation between CASP7 genotypes and mRNA expression levels. In 270 HapMap individuals whose mRNA expression data were available, although there was no correlation of CASP7 mRNA expression levels with the risk loci, we did observe a board-line significant correlation of CASP7 mRNA expression levels with the joint effect of rs10787498 and rs12247479 [generalized linear model (GLM), P 5 0.056; Figure 2G]. Moreover, CASP7 mRNA expression levels showed an increased trend for rs10787498TT-rs12247479GG carriers and a decreased trend for rs1127687AG/AA-rs10787498GT/ GG carriers (Student's t test, P 5 0.018 and 0.064, respectively; Figure 2G, 2J).

Discussion
In this case-control study of 1,486 cervical cancer cases and 1,301 female controls, we found that the rs4353229TT genotype was associated with an increased risk of cervical cancer with a statistical power of 99.8%. Moreover, we also observed significant joint effects and locus-locus interactions of the CASP7 SNPs on cervical cancer risk. This is, to the best of our knowledge, the first report that describes the associations between potentially functional SNPs in CASP7 and cervical cancer risk. Our study is also among the few that have examined the locus-locus interaction in the etiology of cervical cancer. CASP7, located at chromosome 10q25, encodes a member of cysteine peptidase and has been identified as one of the three downstream effectors in the apoptosis pathway in mammalian cells 9 , involved in the execution-phase process of cellular apoptosis. Previous data demonstrated that genetic variations in apoptosis genes might modulate the programmed cell death in various biological systems and alter tissue response to irradiation and cytotoxic chemotherapy 14 , thus eventually leading to genomic instability and tumorigenesis in humans 15 . It is of note that the resistance to apoptosis is an important indicator related to cervical carcinogenesis 16 . In cervical cancer cells, the lack of caspase-mediated apoptosis due to unresponsiveness to pro-apoptotic stimuli causes uncontrolled cell proliferation 17 .
Recently, Wang et al. reported that the rs4353229TT genotype was associated with 0.83-fold decreased risk of gastric cancer 13 . Inversely, in the current study, we found a possibly increased risk of cervical cancer for the rs4353229TT genotype. This discrepancy might be partly due to tumor specificity and population stratification. On the other hand, we also observed that this risk association might be modified by environmental variables and that the effect of one single CASP7 locus on cervical cancer risk might be weak. Indeed, for cancer biology, the functional characterization of risk loci as well as the complex interplay among multiple loci in many cancers poses a particular exciting challenge for the era of post genome-wide association study.
In the present study, we did find that potentially functional SNPs at CASP7 39-UTR might be jointly associated with cervical cancer risk. Further genotype-phenotype analyses suggested an association of CASP7 mRNA expression levels with the joint effect between rs10787498 and rs12247479 as well as between rs1127687 and rs10787498. Consistently, the locus-locus joint effect association analyses demonstrated that there was a super-multiplicative joint effect between rs1127687 and rs10787498 as well as possibly between rs10787498 and rs12247479 on cervical cancer risk. These findings indicated that CASP7 SNPs might interact to modify cervical cancer risk by affecting CASP7 mRNA expression. Subsequent high-order interaction analyses also helped to explain this paradigm. The best interaction model revealed that the four CASP7 SNPs interacted with a maximal CVC and a minimal prediction error, which was more evident in the interaction entropy analysis. Additionally, the CART analysis identified subsets of individuals with cervical cancer risk based on various combinations of genotypes, and the OR for individuals in each TN ranged from 1.11 to 1.67, which also suggests a synergistic interaction between these four SNPs.
Despite the strengths and biologic plausibility of the associations observed in the current study, several limitations need to be addressed. Firstly, there may be selection and information bias originated from a retrospective study design, which may have been minimized by frequency-matching for cases and controls as well as the adjustment for potential confounding factors in multivariate analyses. Secondly, the P value of Hardy-Weinberg equilibrium (HWE) was 0.034 for rs10787498, but given that the deviation from HWE among controls was defined as a significance level of a ,10 23 or 10 24 , all the SNPs in our analyses were in agreement with HWE. Finally, because the lack of routine HPV screening for all cases and controls in our hospital, we could not evaluate HPV infection as the potential confounder in risk estimates of cervical cancer.
In summary, in the current case-control study of 1,486 cases and 1,301 controls, we found that CASP7 SNPs might be associated with cervical cancer risk in Eastern Chinese women. There were substantial joint effects and locus-locus interactions among these SNPs, and such effects may contribute to cervical cancer risk by affecting CASP7 mRNA expression. However, well-designed, larger, and prospective studies with detailed information about HPV infection are warranted to validate our findings.

Methods
Study subjects. The recruitment of the cases and controls was partly described previously 7 . Briefly, all subjects were unrelated ethnic Han Chinese and residents in the Eastern China. The 1,486 newly diagnosed and histopathologically confirmed primary cervical cancer patients were consecutively recruited and collected by the tissue bank of Fudan University Shanghai Cancer Center (FUSCC). The 1,301 frequency-matched healthy controls without history of cancers were recruited from women who had come to FUSCC for breast cancer screening. After a written informed consent was obtained, all subjects were interviewed to collect their demographic and risk factor information. Because most Chinese women did not smoke cigarettes or drink alcohol, all participants included in the analysis were nonsmokers and non-drinkers, and provided a one-time 10 mL of venous blood sample (after diagnosis and before the initiation of treatment for cases). The experimental and research protocols were approved by the Institutional Review Board of FUSCC, and all experiment on humans was performed in accordance with relevant guidelines and regulations.
SNP selection and genotyping. By searching the NCBI dbSNP database (http://www. ncbi.nlm.nih.gov/projects/SNP) and the International HapMap Project database (http://hapmap.ncbi.nlm.nih.gov/), we found that there were 22 SNPs in CASP7 39-UTR, of which four were finally selected for genotyping, based on the following criteria: 1) minor allele frequency of at least 5% in Chinese populations, 2) with low linkage disequilibrium by using an r 2 threshold of ,0.8 for each other, 3) predicted to www.nature.com/scientificreports SCIENTIFIC REPORTS | 5 : 9225 | DOI: 10.1038/srep09225 be potentially functional by the SNP function prediction platform (http://snpinfo. niehs.nih.gov/snpinfo/snpfunc.htm), and 4) not included and published in genomewide association studies. Thus, the selected SNPs were rs4353229 T. C, rs12247479 G. A, rs10787498 T. G and rs1127687 G. A. Genomic DNA extraction and genotyping were conducted as described previously 18 . As a result, the discrepancy rate in all positive controls (i.e., duplicated samples, overlapping samples from previous studies and samples randomly selected to be sequenced) was less than 0.1%.
Genotype-phenotype correlation analysis. To evaluate biological plausibility of our findings, we used the data on CASP7 genotypes and CASP7 mRNA expression levels both available for 270 HapMap subjects by SNPexp online tool (http://app3.titan.uio. no/biotools/help.php?app5snpexp) and conducted genotype-phenotype correlation analysis as described previously 18,19 .
Statistical analysis. HWE was tested by x 2 -test for each SNP. We performed the Pearson's x 2 -test for the differences in selected variables between cases and controls. The association of CASP7 genotypes with cervical cancer risk was estimated by computing ORs and their 95% CIs from both univariate and multivariate logistic regression models. We also evaluated the associations in subgroup and joint effect analyses. The PROC HAPLOTYPE procedure in SAS software was applied to infer haplotype frequencies among the four SNPs. To avoid false positive associations in this study, we calculated the FPRP with the assumption of different prior probabilities (0.0001, 0.001, 0.01, 0.1 and 0.25). FPRP values ,0.2 were considered to be noteworthy 20 . We used GLM for the genotype-phenotype correlation, and used student's t test and analysis of variance test to evaluate the differences in the relative mRNA expression levels among different genotype groups.
The MDR and CART analyses were conducted by the MDR V2.0 beta 8.2 program (http://www.multifactordimensionalityreduction.org/) and SAS software (version 9.1; SAS Institute, Cary, NC), respectively, as described previously 21 . Briefly, we enrolled the four risk loci in the MDR analysis to identify the best n-factor interaction model. Then, we performed the interaction dendrograms and graphs 22 . The color of branches and lines is referred to the type of interaction, green-to-yellow-to-red indicates a weak-to-strong interaction. CART creates a decision tree that depicts how well each genotype predicts disease and ends up with TNs.
All statistical analyses were performed with SAS 9.1 software (SAS Institute, Cary, NC), unless stated otherwise. All P values were two-sided with a significance level of P ,0.05.