Polymorphisms in the AKT1 and AKT2 genes and oesophageal squamous cell carcinoma risk in an Eastern Chinese population

Abstract Ethnic Han Chinese are at high risk of developing oesophageal squamous cell carcinoma (ESCC). Aberrant activation of the AKT signalling pathway is involved in many cancers, including ESCC. Some single nucleotide polymorphisms (SNPs) in genes involved in this pathway may contribute to ESCC susceptibility. We selected five potentially functional SNPs in AKT1 (rs2494750, rs2494752 and rs10138277) and AKT2 (rs7254617 and rs2304186) genes and investigated their associations with ESCC risk in 1117 ESCC cases and 1096 controls in an Eastern Chinese population. None of individual SNPs exhibited an association with ESCC risk. However, the combined analysis of three AKT1 SNPs suggested that individuals carrying one of AKT1 variant genotypes had a decreased ESCC risk [adjusted odds ratio (OR) = 0.60, 95% CI = 0.42–0.87]. Further stratified analysis found that AKT1 rs2294750 SNP was associated with significantly decreased ESCC risk among women (adjusted OR = 0.63, 95% CI = 0.43–0.94) and non‐drinkers (OR = 0.79, 95% CI = 0.64–0.99). Similar protective effects on women (adjusted OR = 0.56, 95% CI = 0.37–0.83) and non‐drinker (adjusted OR = 0.75, 95% CI = 0.60–0.94) were also observed for the combined genotypes of AKT1 SNPs. Consistently, logistic regression analysis indicated significant gene–gene interactions among three AKT1 SNPs (P < 0.015). A three‐AKT1 SNP haplotype (C‐A‐C) showed a significant association with a decreased ESCC risk (adjusted OR = 0.70, 95% CI = 0.52–0.94). Multifactor dimensionality reduction analysis confirmed a high‐order gene–environment interaction in ESCC risk. Overall, we found that three AKT1 SNPs might confer protection against ESCC risk; nevertheless, these effects may be dependent on other risk factors. Our results provided evidence of important gene–environment interplay in ESCC carcinogenesis.


Introduction
Oesophageal cancer, consisting of squamous cell carcinoma (ESCC) and adenocarcinoma, is the 8th most frequently diagnosed cancer worldwide [1][2][3]. Oesophageal squamous cell carcinoma constitutes the majority of the cases (90%) in China with a 5-year survival of less than 20% [2,3]. Therefore, it is urgent to develop more effective prevention strategies for this malignant disease by a better understanding of the aetiology.
The proven-environmental (e.g. lifestyle) risk facts for ESCC are poor nutritional status, low intake of fruits and vegetables, tobacco smoking, alcohol use and drinking hot beverages. Moreover, genetic factors are also implicated in ESCC carcinogenesis. Molecular epidemiological studies have demonstrated that some single nucleotide polymorphisms (SNPs) account, in part, for the variation in cancer susceptibility in the general population [4][5][6], including SNPs in inflammatory response, one carbon metabolism, metabolism of chemical carcinogens and DNA repair pathways as well as some other oncogenes and tumour-suppressor genes [7,8].
Many studies [9,[19][20][21][22] have investigated the effects of SNPs in AKT genes on the risk of cancers in Chinese and shown promising results. However, the contribution of AKT polymorphisms to ESCC risk has not been reported. Therefore, we conducted this case-control study to explore the role of SNPs in AKT genes in the aetiology of ESCC in an Eastern Chinese population.

Study population
This case-control study included 1117 cases and 1096 healthy non-cancer controls. All enrolled cases were newly diagnosed ESCC patients between March 2009 and September 2011, with histopathological confirmation at Fudan University Shanghai Cancer Center. They were all genetically unrelated Han Chinese, residing in Eastern China. Exclusion criteria were as follows: (i) the primary tumour was not oesophageal in origin, (ii) patients with other cancers and (iii) cancers without a definite primary site. Cancer-free controls (without other diseases) were from a large prospective cohort recruited for the Taizhou longitudinal study at the same time period in the Eastern China [23], and frequency matched to cases on age (AE5 years) and sex. While interviewed, all participants were obligated to complete a structured questionnaire including demographic data and environmental exposure history, such as age, sex, ethnicity, body mass index (BMI, calculated by weight in kilograms/height 2 in metres), tobacco use and alcohol intake before treatment. A BMI value of 25 was used as a cut-off point to split participants into two groups with BMI <25 and ≥25, as the World Health Organization sug-gested BMI ≥25 as a cut-off for classification of overweight [24]. Only study participants who signed a written consent form (about 90%) were included in the final analysis. The research protocol of the study was approved by the institutional review board of the Fudan University Shanghai Cancer Center.

SNP selection and genotyping
We first retrieved available SNPs in target genes from the National Center for Biotechnology Information dbSNP database (http:// www.ncbi.nlm.nih.gov/projects/SNP) and then selected common, potentially functional SNPs in accordance with these criteria: (i) positioned in exons, the 5 0 near gene, 5 0 untranslated regions (UTR), 3 0 UTR, 3 0 near gene or splice sites; (ii) the minor allele frequency (MAF) should be equal or larger than 5% in Chinese Han population; (iii) SNPinfo software (http://snpinfo.niehs.nih.gov/snpfunc.htm)-identified potentially functional SNPs; and (iv) not studied in the published ESCC genomewide association studies. Moreover, some SNP reported by others was also selected [25]. Haploview software was used to check the linkage disequilibrium (LD) to ensure that selected SNPs were in low LD (c 2 < 0.8) with one another. Ultimately, five SNPs (AKT1: rs2494750, rs2494752 and rs10138277; AKT2: rs7254617 and rs2304186) were included in the study. No SNPs in the AKT3 gene met the defined criteria and thus were not included.
Qiagen Blood DNA Mini Kit (Qiagen Inc., Valencia, CA, USA) was used to acquire genomic DNA from blood specimens, and TaqMan assay was performed to genotype DNA samples as indicated previously [26]. Concisely, allele-specific probes for SNP genotyping were purchased from Applied Biosystems (Foster City, CA, USA). For each of selected SNPs, the probes for the variant and wild-type alleles were labelled with either of the fluorescent dyes VIC and FAM, respectively. The ABI 7900 HT Sequence Detection System (Applied Biosystems) allowed the use of a post-amplification allelic discrimination run on the machine to identify genotype according to the relative fluorescence intensity of VIC and FAM. PCR reactions in 384-well plates was run on the machine, with a total reaction volume of 5 ll for each sample. Individuals involved in genotyping were blind to participants' status.

AKT1 expression analysis based on AKT1 variant genotypes
We further interrogated the impact of the significant polymorphisms on the gene expression by using online databases for 270 individuals from four worldwide populations [CEU: 90 Utah residents with ancestry from northern and western Europe; CHB: 45 unrelated Han Chinese in Beijing; JPT: 45 unrelated Japanese in Tokyo; YRI: 90 Yoruba in Ibadan, Nigeria] [27]. We first obtained genotype information from the international HapMap phase (II+III) release #28 data set, containing genotype data of 3.96 million polymorphisms for 270 individuals (http://www.hapmap.org). mRNA expression information was acquired from the same 270 individuals (http://app3.titan.uio.no/biotools/help.php?app=snpexp) [28], which were derived from GENe Expression VARiation (http://www.san ger.ac.uk/resources/software/genevar/) [29]. Finally, we matched AKT1 polymorphism genotypes and AKT1 mRNA expression levels for each individual to evaluate the correlation between Hapmap genotypes and the gene expression levels.

Statistical methods
The chi-squared test was used to evaluate whether there was any difference in the frequency distributions of certain demographic variables, risk factors and genotypes of the studied SNPs between the cases and controls. A goodness-of-fit chi-squared test was used to detect possible deviation from Hardy-Weinberg equilibrium (HWE) in controls. The crude and adjusted odds ratios (ORs) and 95% confidence intervals (CIs) for the association of ESCC risk with SNPs of interest were determined by univariate and multivariate logistic regression analyses controlling for co-variates (e.g. age, sex, smoking, drinking and BMI). The stratification analyses were also performed to identify the associations by age, sex, BMI, and smoking and drinking status. Moreover, a combination of rs2494750, rs2494752 and rs10138277 genotypes in the AKT1 gene was considered as a haplotype. Unphased genotype data were used to determined haplotype frequencies and individual haplotypes. Logistic regression analysis was performed to calculate ORs and 95% CIs for the association of haplotypes with ESCC risk. All tests were twosided with a significance level of P < 0.05. All statistical analyses were performed with SAS software (version 9.1; SAS Institute, Cary, NC, USA). Furthermore, the high-order gene-gene or gene-environment interactions were established in the association with cancer risk using the multifactor dimensionality reduction (MDR) software (V2.0 beta 8.2), as described elsewhere [30]. A model with the minimum average prediction error and the maximum cross-validation consistency (CVC) was considered the best candidate interaction model.
Finally, we performed mini meta-analyses to evaluate the association of AKT1 rs2494750 and AKT2 rs7254617 SNPs with ESCC risk. Briefly, relevant studies were searched with defined search terms from the common public database (MEDLINE and EMBASE) and screened with inclusion and exclusion criteria in accordance with previous procedure [31][32][33]. Chi-square-based Q-test was performed to test heterogeneity assumption. The fixed-effects model (the Mantel-Haenszel method) was used to calculate the pooled OR estimates. If the study had high heterogeneity, the random-effects model (the DerSimonian and Laird method) would be chosen as an alternative. The funnel plot and Egger's linear regression test were used while detecting potential publication bias. Sensitivity analysis was performed to assess the effect of single studies on pooled risk estimates. We were not able to perform meta-analysis for the remaining SNPs, because of very few publications having investigated the association of these SNPs and cancer risk. All the statistical tests were performed with STATA (version 11.0; Stata Corporation, College Station, TX, USA). Two-sided P-values were applied, and a P < 0.05 was used as the significance level.

Characteristics of ESCC patients and controls
In this study, cases and controls were well matched by age (P = 0.338) and sex (P = 0.072; Table 1). Distributions of smokers and drinkers were found to be significantly different between cases and controls. As expected, the percentages of smokers and drinkers in cases were higher than in controls (smokers: 61.2% versus 54.2%, P < 0.0009; drinkers: 44.3% versus 32.9%, P < 0.0001). Moreover, mean BMI was significantly smaller in cases than in controls (mean BMI AE SD: 23.46 AE 7.36 versus 26.88 AE 7.52, P < 0.0001). Along with univariate analyses, multivariate logistic regression analyses adjusted for these variables were subsequently performed to control for potential confounding effect.

Association between AKT1/AKT2 SNPs and ESCC susceptibility
First, the genotype distributions of the five SNPs in controls were consistent with those expected from the HWE. Second, the MAFs of the genotyped SNPs in controls were comparable to those identified in the CHB data from HapMap or reported in Asians [25]: 0.315 versus 0.267 (rs2494750), 0.266 versus 0.220 (rs2494752), 0.104 versus 0.083 (rs10138277), 0.135 versus 0.149 (7254617) and 0.447 versus 0.54 (rs2304186). We calculated ORs using logistic regression analyses after adjustment for age, sex, drinking status, smoking status and BMI ( Table 2). In the single-locus analysis, comparison of genotype frequency distributions revealed no significant difference between ESCC cases and controls, indicating that none of these SNPs was independently associated with ESCC risk in this study population.
Next, we explored whether combined analysis of multiple genetic variants facilitated the identification of high-risk individuals. We combined variant genotypes of the five SNPs (variant heterozygotes and homozygotes) under investigation to scrutinize whether these SNPs would collaboratively contribute to ESCC risk. Once again, participants carrying one to five variant genotypes have ESCC risk as high as those carrying wild-type genotypes. Furthermore, all participants were split into two groups based on the presence or absence of variant genotypes, with one group having only the wild-type genotype as reference and the other having at least one variant genotype. Likewise, we found carriers of one or more variant genotypes did not show altered risk (OR = 0.94, 95% CI = 0.68-1.28, P = 0.683) for ESCC, when compared with non-carriers. However, the combined analysis with only three AKT1 SNPs found that having one AKT1 variant genotype was associated with a protective effect (adjusted OR = 0.60, 95% CI = 0.42-0.87, P = 0.007, statistical power = 0.353) for developing ESCC, which is likely because of a chance.

Stratification analysis
We thereafter explored the gene-environment interaction by determining the potential association of ESCC risk with the SNPs in the stratified analyses by age, sex, smoking status, drinking status and BMI. Among all the tested SNPs, we found that AKT1 rs2294750 might exert a protective effect on ESCC risk; in particular, this effect was significant for women (adjusted OR = 0.63, 95% CI = 0.43-0.94, P = 0.024, statistical power = 0.925) and non-drinkers (OR = 0.79, 95% CI = 0.64-0.99, P = 0.042, statistical power = 0.995) under the dominant model (Table 3A). Moreover, the stratification analyses did not identify any other significant association (Table 3A and B).

AKT1 haplotypes and ESCC risk
We further investigated whether the haplotypes of three AKT1 SNPs were associated with ESCC risk. As shown in Table 4, four AKT1 haplotypes were determined in the study population. We defined the haplotype consisting of wild-type alleles (G-A-C) as the reference group. The protective association was found between haplotypes C-A-C and ESCC susceptibility (adjusted OR = 0.70, 95% CI = 0.52-0.94). However, the results need to be further validated.

High-order interactions in ESCC risk by MDR analysis
The MDR analysis was carried out to further explore the high-order interactions of SNPs and environmental factors in ESCC risk. Five studied SNPs and five risk factors (i.e. age, sex, smoking status, drinking status and BMI) entered the analysis. BMI was shown to be the best one-factor model, as it had the highest cross-validation consistency (CVC, 100%) and the lowest prediction error (39.4%) out of all 10 factors. It indicated that among all factors, BMI conferred the highest ESCC risk in the study population. Moreover, when compared to other models (e.g. five-factor mode and seven-factor model), the 10-factor model, having a maximum CVC (100%) and a minimum prediction error (33.7.0%), could yield a better prediction for ESCC risk (Table 5).
Correlation between AKT1 rs2494750 genotypes and AKT1 mRNA expression levels Finally, 264 of 270 individuals were informative for analysis, of whom there were 63, 90 and 111 carriers of GG, CG and CC genotypes respectively. We found that AKT1 rs2949750 variant C allele was significantly associated with increased AKT1 gene expression levels under the additive model (one-way ANOVA, P = 0.0006) and recessive model (Student's t-test, P = 0.0001; Fig. 1A). Further analysis by population group indicated that significant impact of the variant on gene expression was only observed among YRI ( Fig. 1B; one-way ANOVA, P = 0.0058; Student's t-test, P = 0.0013), rather than CEU, CHB and JPT populations (data not shown).  Meta-analysis for the association of AKT1 rs2494750 and AKT2 rs7254617 with cancer risk Thus far, three publications have reported conflicting results on the associations of AKT1 rs2494750 and AKT2 rs7254617 with cancer risk [9,14,19]. With the inclusion of all these studies and our data, we carried out a mini meta-analysis composed of 2606 cases and 2783 controls. Pooled analysis provided no evidence of the association of these two SNPs and cancer susceptibility (rs2494750 under dominant model: OR = 0.99, 95% CI = 0.93-1.06; rs7254616 under the dominant model: OR = 1.02, 95% CI = 0.94-1.11) (Fig. 2). No publication bias was detected for AKT2 rs7254617, but significant publication bias was detected for rs2494750.

Discussion
Oesophageal squamous cell carcinoma, with a 5-year survival rate of less than 20% [1,3], is the fourth leading cause of cancer-related death in China [34]. Excessive activity of the PI3K-AKT pathway is involved in carcinogenesis. AKT acts as a serine/threonine kinase downstream of PI3Ks. It is frequently constitutively activated in a wide spectrum of human cancers, including ESCC [18]. Previous studies have reported that SNPS in PI3K and mTOR genes within the AKT pathway modulate the risk of various cancers [8-11, 19, 35-38]. SNPs that influence the activity of AKT may also modify the risk of ESCC. Therefore, we searched potentially functional SNPs in the AKT genes and studied for their association with ESCC susceptibility. The single-locus analysis did not provide evidence of statistically significant association between ESCC risk and the five studied SNPs. Moreover, our meta-analysis observed no association of AKT1 rs2494750 and AKT2 rs7254617 and ESCC risk. However, the combined analysis of three AKT1SNPs identified that individuals carrying only one of three AKT1 variant genotypes might have decreased risk to develop ESCC cancer in comparison with non-carriers, but this finding could be because of chance. It was noted that significant publication bias was detected in the mini meta-analysis for rs2494750. One reason for the publication bias was that medical findings with statistical significance have greater chance to be polished than those not significant. The limited number of eligible studies might be another reason for publication bias. The resulting bias could cause erroneous conclusions [39]. Thus, our meta-analysis results call for further validation.
The effects of some AKT SNPs on cancer risk have been investigated previously [9,[19][20][21][22]. A study conducted among Caucasians reported that two SNPs in the AKT3 gene had profound effects on bladder cancer susceptibility [20]. AKT3 rs2994329 was shown to significantly increased bladder cancer risk, while AKT3 rs12045585 exhibited reverse association [20]. The same group also reported that AKT2 3730050 was significantly associated with the survival of muscle invasive and metastatic bladder cancer patients [40]. When compared with those with the wild-type genotype, patients carrying one or two AKT2 3730050 variant alleles had an increased death risk up to 2.99-fold [40]. Recently, one study demonstrated that AKT1 rs1130214 and rs3803300 were associated with oral squamous cell carcinoma in Chinese Han Population [21]. Zhang et al. genotyped five AKT1 SNPs (rs3803300, rs1130214, rs3730358, rs1130233 and rs2494732) in 593 nasopharyngeal carcinoma cases and 480 controls [22]. Although none of individual SNP had significant effect on the risk of nasopharyngeal carcinoma, a two-SNP haplotype, consisting variant alleles of AKT1 rs1130233 and rs2494732, was significantly associated with increased nasopharyngeal carcinoma risk [22]. Moreover, both AKT1 rs2294750 and AKT2 rs7254617 polymorphisms have been investigated in cancers among Chinese populations [9,19], but results are conflicting. Cao et al. reported there was no association between renal cell cancer risk and these two SNPs, but a stratification analysis was not performed [19]. Chen et al. reported that AKT2 rs7254617, but not AKT1 rs2294750, significantly increased the risk of prostate cancer [9]. Taken together, the majority of studies [19][20][21][22] support AKTs as cancer susceptibility genes. The inconsistency among results may be because of the discrepancies in the sampling, different ethnicity or the fact that    [41], is one of most remarkable risk factor for ESCC carcinogenesis. The underlying mechanism of how alcohol affects the development of ESCC remains unclear. It may directly irritate the epithelium of oesophagus, enhance vulnerability to another carcino-gen or cause nutrition deficiencies that are also a recognized risk factor for ESCC [42]. Despite lack of the mechanism, epidemiologic evidence has consistently shown that alcohol use is associated with an increased ESCC risk [41][42][43][44][45]. As an example, alcohol consumption exceeding the recommended U.S. dietary guidelines is significantly associated with elevated ESCC risk [41]. The protective effects of AKT1 rs2294750 on non-drinker observed in this study is in accordance with the perception of cancer susceptibility, which represents a genetic attribute that modify the possible cancer risk under the influence of environmental conditions or lifestyles, such as smoking, drinking and diet. Given the aetiological role of drinking in the development of ESCC, the moderate protective effect of AKT1 rs2294750 on drinker is probably overridden by the potent carcinogenic effect of alcohol. Alternatively, among non-drinkers without alcohol's damaging effects, the SNP was able to significantly decrease ESCC risk.
Moreover, we found that AKT1 rs2294750 had a protective effect on women against ESCC risk. Previous epidemiology studies demonstrated the conspicuous male preponderance of ESCC [1,2,46], which suggests that males appear to be predisposed to environmentally induced ESCC, compared with female. Comparable to the results observed in stratified analysis by drinking status, the protective impact of AKT1 rs2294750 was also more predominant in low-risk subgroup (women) than in high-risk subgroup (males). These data may suggest that the protective effect of this SNP on men might be superseded by unknown sex-related environmental aetiology, which could be resulted from gene-environment interaction [47] that needs to be detected in a large study. In the current studies, significant associations were only observed in women and non-drinkers, indicating the importance in considering other factors when investigating genotypic impact on cancer susceptibility. Alternatively, these results could be because of chance, which call for larger and validation studies. The relative gene expression analysis by HapMap genotypes demonstrated that AKT1 rs2949750 variant C allele was significantly associated with elevated AKT1 gene expression levels among the general population and the YRI population but not other three subpopulations.
Finally, although there was no association between ESCC susceptibility and any of AKT1 variants in the single-locus analysis, our results revealed that three AKT1 SNPs might collectively pro-  tect individuals from developing ESCC. First, among women and non-drinkers, the observed combined protective effect of the three AKT1 SNPs was stronger than each of individual SNPs. Second, significant gene-gene interaction among three AKT1 SNPs was identified by logistic regression analysis. Third, a three-AKT1 SNP haplotype was significantly associated with ESCC risk. The lack of main effect of AKT1 variants might suggest that the effect size of any of the variants under investigation was small and the current sample size was not large enough to detect such small effects. It might also suggest that these SNPs were low penetrance variants that modulate cancer susceptibility through gene-gene and/or environment-gene interactions. Moreover, the combined analysis is able to amplify the moderate effect of each individual SNP and enhance the predictive power. The identification of multiple risk variants may therefore improve risk prediction and could conceivably be applied to assessment of an individual's ESCC risk. As indicated by the online tool SNPinfo software, AKT1 rs10138277 and rs2494750 are SNPs in the transcription factor-binding site of the gene and these SNPs may alter the binding capacity of the related transcription factors. AKT1 rs2494752 was selected because that it was reported to be associated with chemotherapy response in advanced non-small cell lung cancer among a Chinese Population [25], and it is also a SNP in the transcription factor-binding site of the AKT1 gene. The MDR analysis further validated the observed gene-gene and gene-environment interaction by logistic regression analysis, in which 10-factor model consisting of SNPs and environmental factors could more accurately predict ESCC risk than any SNP or environmental factor alone. ESCC is known as a complex, multifactorial disease, in which interplay between genetic and environment factors may play a crucial role, and one single SNP is not adequate to predict the overall risk. However, the combination of susceptible loci in multiple biological pathways and environmental factor may help health profession improve predictions of the overall risk and clinical outcome, identification of high-risk subpopulation and early detection for ESCCs.
There are some limitations in this study. First, although age, sex, smoking, drinking and BMI were considered and adjusted for in the multivariate analysis, many other factors (nutrition, diet, socio-economic status, etc.) that may also modulate predisposition to ESCC were not available for the analysis; Lack of the detailed data limited our ability to explore gene-gene and geneenvironment interactions. Second, ESCC patients were only recruited from Fudan University Shanghai Cancer Center, the case-control study might suffer from selection bias and information bias. Third, this study only had moderate sample size, which might compromise our ability to detect relatively weak main effect or interactions of some potentially functional SNPs. Fourth, the statistical power for the stratification analysis and determination of gene-gene and gene-environmental interaction might be limited. Moreover, our findings from observational association studies may require in vitro and in vivo experiments to further provide biological evidence of the observed protective effects of AKT1 SNPs on ESCC risk, which would unravel the underlying molecular mechanisms. As a result, our results should be carefully interpreted.
In summary, we found that AKT1 rs2294750 alone or together with other two AKI SNPs may modify the susceptibility to ESCC risk; nevertheless, these effects were largely dependent on the presence of other risk factors, i.e. sex and drinking status. Our results draw attention to the importance of gene-gene and geneenvironment interactions in determining the ESCC susceptibility. These genetic variants may cause an individual susceptible to certain effects of environmental factors. Larger population-based studies, with a focus on gene-environment interaction, are needed to substantiate our findings.