Characterization and risk association of polymorphisms in Aurora kinases A, B and C with genetic susceptibility to gastric cancer development

Our examined population consisted of 116 GC patients with diagnosed gastric adenocarcinoma from the Clinical Pathology and Cytology at the University Clinical Center Sarajevo, Bosnia and Herzegovina. General status of gastric cancer patients is given in Table 1. Gastric cancer patients in the case group were not subjected to any type of treatment (radiotherapy or chemotherapy).The formalin fixed paraffin embedded (FFPE) cancer tissue sections were collected during surgical procedures. Simultaneously, 203 healthy blood donors (controls) of Bosnian origin (matched to cases for ethnicity) were randomly selected and signed up for the present study. Individuals in the control group had no history of any neoplastic formation, were not related to each other and to the patients group. Three ml of blood was sampled from each control individual and stored at − 80 °C. The study was approved by the Ethical Committee at the University Clinical Centre Sarajevo (No. 0302–36,765). Personal information was encrypted to provide maximum anonymity in compliance with the Helsinki Declaration.

Genomic DNA from FFPE GC tissues was isolated using the Chemagic FFPE DNA Kit special (PerkinElmer Inc., Waltham, MA, USA), according to manufacturer’s recommendations. Automated DNA washing and elution was conducted on Chemagic Magnetic Separation Module I robot (PerkinElmer Inc., Waltham, MA, USA), following manufacturer’s standard programme. All sample transfers were performed with 4-eye principle to avoid sample mix-ups. DNA from lymphocytes (control DNA) was extracted using the Promega™ Wizard™ Genomic DNA Purification Kit Protocol (Promega Corp., Fitchburg, WI, USA), in concordance with the manufacturer’s recommendations. The qualitative and quantitative analysis of extracted DNA was conducted by use of the DropSense96 photometer (Trinean, Gentbrugge, Belgium) and Synergy™ 2 Multi Mode Reader (BioTek, Inc., Winooski, VT, USA).

We selected nine polymorphisms in mitotic genes, namely rs2273535, rs1047972, rs911160 and rs8173 (AURKA), rs2241909 and rs2289590 (AURKB), rs758099 and rs11084490 (AURKC) and rs42873 (PLK1). The positions of selected genetic variants in mitotic genes are presented in Fig. 1. For this purpose, gene structures were extracted from the Research Collaboratory for Structural Bioinformatics (RCSB) Protein Data Bank (PDB) [10]. Selection of the polymorphisms for this study was conducted in accordance with the parameters described below: (a) previously demonstrated association with respect to certain cancer types; (b) minor allele frequency (MAF) of less than or equal to 10% in the population of Utah residents with Northern and Western European ancestry (CEU), as stated by the Phase 31,000 Genomes; and (c) tagging polymorphisms (tagSNPs) status, which was anticipated in silico by use of LD Tag Selection of SNP (tagSNP) (https://​snpinfo.​niehs.​nih.​gov) [11], with the following parameters: 1 kb of the sequences upstream/downstream from gene was selected, linkage disequilibrium (LD) lower limit of 0.8, and MAF range 0.05–0.5 for CEU subpopulation (Table 2 and Fig. 2).

Genotyping was conducted using TaqMan SNP genotyping assays (Applied Biosystems, Foster City, CA). The assay ID numbers are presented in Table 2. The reaction mixtures, GC samples (5 μl) and controls (10 μl), were composed of 20X TaqMan® assay with 2X Master Mix (Applied Biosystems, Foster City, CA), and 20 nanograms of DNA. The polymerase chain reaction (PCR) profile was carried out in concordance with the manufacturer’s recommendations (Initial denaturation at 95 °C for 10 min, 45 cycles at 92 °C for 15 s and 60 °C for 90 s, using the ViiA 7 Real Time PCR System (Applied Biosystems, Foster City, CA). In each plate, at least two negative controls were included. PCR results were analyzed using TaqMan® Genotyper Software (Applied Biosystems, Foster City, CA, USA).

The genotype frequencies of the investigated variants were tested for Hardy-Weinberg equilibrium (HWE) in the case/control groups separately, using Michael H. Court’s online HWE calculator (http://​www.​tufts.​edu) [12]. The differences in genotype frequencies amongst GC cases and controls were calculated by use of the Chi-square test or Fisher’s exact test. Association between examined polymorphisms and the GC risk was estimated by multinomial logistic regression. Odds ratio (OR) with 95% confidence interval (CI) were computed in order to evaluate the relative risk. For the assessment of each genotype, risk estimates were computed for dominant, overdominant and recessive models using the most frequent homozygote as the reference. Akaike information criterion (AIC) was calculated to define which of the models best fits the data. A combined analysis was performed to evaluate synergistic effect of the studied polymorphisms. All statistical calculations were conducted using SPSS 20.0 software package (SPSS, Chicago, IL, USA). P ≤ 0.05 was chosen as threshold value in significance testing. MAF plot was created by use of the PAST software package, version 3.18 (http://​folk.​uio.​no/​ohammer/​past/​) [13].

Determination of the haplotype block structure and haplotype analysis, which encompassed subsequent corrections for multiple comparisons by 10,000 permutations, were evaluated using the Haploview software, version 4.2 [14]., and SNP tools V1.80 (MS Windows, Microsoft Excel). To construct the haplotype block, the solid spine of the linkage disequilibrium algorithm with a minimum Lewontin’s D′ value of 0.8 was selected.

Impact of the polymorphic DNA sequences (SNPs in introns and untranslated regions (UTRs)) on transcription factors binding sites (TFBSs) was estimated in silico. Bioinformatic functional evaluation was carried out using PROMO software (ALGGEN web-server), which is utilizing data from TRANSFAC database V8.3 [15, 16]. FASTA sequences for the investigated genetic variants were downloaded from Ensembl 90 (www.​ensembl.​org/​index.​html) [17]. Identification of transcription factor binding sites was performed with the following criteria: human species, all sites and factors.

For all of the 9 studied variants, rs2273535 (AURKA), rs1047972 (AURKA), rs911160 (AURKA), rs8173 (AURKA), rs2241909 (AURKB), rs2289590 (AURKB), rs758099 (AURKC), rs11084490 (AURKC), rs42873 (PLK1) was determined to be in HWE in both, case and control populations (P > 0.05). When chi-square test and Fisher exact test were conducted for the frequency distributions at the genotypic level, a significant differences for rs911160 in AURKA (P = 0.044), rs8173 in AURKA (P = 0.018), rs2289590 in AURKB (P < 0.001) and rs11084490 in AURKC (P = 0.009) between the cases and controls for all types of GC were observed (summarized in Table 3).

Patients with rs1047972 (AURKA) CT genotype had a higher risk of GC development in comparison with the reference CC genotype (OR = 1.96, 95% CI = 1.05–3.65, P = 0.033) (Table 4). Genotypes (TT + CT) vs. reference CC genotype in dominant model (OR = 1.92, 95% CI = 1.06–3.48, P = 0.030) and CT vs. reference (CC + TT) genotypes in overdominant model (OR = 1.94, 95% CI = 1.04–3.60, P = 0.036) were associated with higher disease risk (Table 4). Based on Akaike information criterion (AIC), the overdominant model was selected as the model that best fits the data. The rs911160 (AURKA) CC genotype was positively associated with an increased gastric cancer risk in comparison with the reference GG genotype (OR = 5.56, 95% CI = 1.24–24.81, P = 0.025). Also, CC genotype was associated with disease risk in the recessive genetic model (GG + CG) vs. CC genotypes, (OR = 5.26, 95% CI = 1.19–23.22, P = 0.028). However, the confidence intervals in those two cases were wide; therefore, other factors might play a significant role in GC risk in interaction with this polymorphism. Comparison of genotype distributions for rs8173 (AURKA) showed that patients with GG genotype (OR = 0.38, 95% CI = 0.18–0.79, P = 0.010), and CG genotype (OR = 0.60, 95% CI = 0.36–0.99, P = 0.049) had decreased risk of gastric cancer. Analysis of genetic models showed that GG + CG genotypes in comparison with the reference CC genotype in dominant model (OR = 0.54, 95% CI = 0.33–0.87, P = 0.012) and GG vs. reference (CC + CG) (OR = 0.49, 95% CI = 0.25–0.98, P = 0.043) genotypes (recessive genetic model) were associated with decreased GC risk. According to the calculated AIC values, (CC + CG):GG recessive model had more statistical power than dominant model CC:(GG + CG). Analysis of rs2241909 (AURKB) demonstrated that G allele (dominant model: (GG + AG) vs. common AA genotype) was associated with higher risk of GC development (OR = 1.61, 95%CI = 1.01–2.56, P = 0.041). Comparison of the reference AA genotype with AC (OR = 2.41, 95% CI = 1.47–3.98, P = 0.001) and CC (OR = 6.77, 95% CI = 2.24–20.47, P = 0.001) genotypes of rs2289590 (AURKB) also revealed a significant effect of these two genotypes on increased GC risk. CC and AC genotypes in dominant model (OR = 2.78, 95% CI = 1.71–4.51, P < 0.001) as well as CC genotype in recessive model (OR = 4.23, 95% CI = 1.44–12.40, P = 0.009) and AC genotype in overdominant genetic model (OR = 1.77, 95% CI = 1.10–2.85, P = 0.017) were associated with an elevated disease risk. Since recessive genetic model had the lowest AIC value, when compared to the dominant and overdominant models, it was considered to be preferred model. However, in this model the confidence interval was wide, therefore, other factors could influence its effect. For rs11084490 (AURKC) polymorphism, (GG + CG) vs. CC genotypes in dominant model demonstrated statistically significant effect on higher GC risk (OR = 1.71, 95% CI = 1.04–2.81, P = 0.033). Additionally, the five polymorphisms rs1047972, rs911160, rs2241909, rs2289590 and rs11084490, associated with an increased GC risk individually in this study, were subjected to the combined analysis in order to determine polymorphism profiles related to the higher risk of this disease. The results of the synergistic effects of these SNPs are summarized in Table 5. By analyzing various combinations of risk genotypes (two to five combined SNPs), we demonstrated that the additive effect of all combinations significantly affected the risk of GC development, with an odds ratio ranging from (OR = 1.51, 95% CI = 1.03–2.22, P = 0.034) for the combined rs1047972(CT)/rs11084490(CG + GG) risk genotypes to (OR = 2.29, 95% CI = 1.32–3.97, P = 0.003) for the rs1047972(CT)/rs911160(CC) combination. Another interesting combined effect was demonstrated for five-polymorphisms combination rs1047972(CT)/rs911160 (CC)/rs2241909 (AG + GG)/rs2289590(AC + CC)/rs11084490 (CG + GG). In this case, this combination was significantly associated with an increased GC risk (OR = 1.83 95% CI = 1.46–2.29, P < 0.001). No significant effects on gastric cancer susceptibility were revealed for rs2273535 (AURKA), rs758099 (AURKC) and rs42873 (PLK1) polymorphisms (P > 0.05), when patients with both types of GC, intestinal and diffuse, were taken into account.

Next, we estimated the effects of genotypes on GC subtypes (presented in Table 4). CT genotype of rs1047972 (AURKA) was more frequent in patients with intestinal type (OR = 2.53, 95% CI = 1.02–6.30, P = 0.045) in comparison with the reference CC genotype. Likewise, (TT + CT) genotypes vs. reference CC (OR = 2.39, 95% CI = 1.02–5.63, P = 0.045) and CT vs. common (CC + TT) genotypes (OR = 2.50, 95%CI = 1.01–6.22, P = 0.047) were associated with higher risk for the development of intestinal subtype. According to the AIC values, (CC + TT):CT overdominant genetic model displayed stronger statistical confidence than dominant model CC:(TT + CT). The rs8173 (AURKA), GG genotype, in comparison with the reference CC genotype, was underrepresented in patients with diffuse GC type (OR = 0.32, 95% CI = 0.13–0.77, P = 0.012). Furthermore, both (GG + CG) genotypes as compared to its common CC genotype in dominant model (OR = 0.49, 95% CI = 0.27–0.89, P = 0.021) and GG vs. reference (CC + CG) genotypes in recessive model (OR = 0.44, 95% CI = 0.20–0.98, P = 0.044) were associated with the decreased diffuse type GC risk. In order to discriminate between these two competing models, in accordance with AIC, recessive model represents the preferred model in comparison with the dominant model. In stratified analysis for rs2241909 (AURKB), we found that carriers of AG genotype had elevated risk of developing intestinal type GC as compared to its reference AA genotype (OR = 2.23, 95% CI = 1.16–4.27, P = 0.016). Carriers of (GG + AG) genotypes had more frequently intestinal type of GC when compared to the carriers of the more common AA genotype in dominant model (OR = 2.38, 95% CI = 1.27–4.46, P = 0.007). In overdominant model (OR = 1.93, 95%CI = 1.02–3.67, P = 0.042) individuals with AG genotype had more frequently intestinal type GC in comparison with reference genotypes (AA+GG). According to the calculated AIC values, overdominant model had more statistical power than dominant, therefore it represents the model that better fitted the data. The higher risk for intestinal type GC development was also detected for the patients with CC genotype of rs2289590 (AURKB) (OR = 5.19, 95% CI = 1.14–23.56, P = 0.033). Dominant genetic model revealed that patients with (CC + AC) genotypes when compared to the AA genotype (OR = 2.04, 95% CI = 1.07–3.88, P = 0.028) had significantly more frequently intestinal GC subtype. AC genotype (OR = 3.12, 95% CI = 1.68–5.80, P < 0.001) was more frequently observed in patients with diffuse subtype. Regarding genetic models, (CC + AC) genotypes in dominant model (OR = 3.58, 95% CI = 1.96–6.52, P < 0.001), CC genotype in recessive model (OR = 4.72, 95%CI = 1.09–20.43, P = 0.038) and AC genotype in overdominant model (OR = 2.27, 95% CI = 1.24–4.13, P = 0.007) were associated with the increased risk of diffuse subtype, with a recessive model being the one that best suited the data (according to the AIC value), however, the confidence interval in this model was also the largest. For rs11084490 (AURKC), dominant model (GG + CG) vs. CC (ref.) genotypes reveled a significant effect of GG and CG genotypes on the higher risk of intestinal subtype (OR = 2.03, 95% CI = 1.02–4.04, P = 0.043).

For genotypes of rs2273535 (AURKA), rs911160 (AURKA), rs758099 (AURKC) and rs42873 (PLK1) no significant effect on any of the GC histological subtypes was noted (P > 0.05).

Raw genotyping data for the studied polymorphisms rs2273535, rs1047972, rs911160 and rs8173 in AURKA gene were used to perform haplotype analysis. Using the Haploview software, our results showed that no haplotype block was created with an average Lewontin’s D < 0.8 (Fig. 3) thus, no haplotypes were available for the analysis of their potential association with GC risk.

Our in silico analysis suggested that polymorphic sequences in transcription factors binding sites (TFBSs), within AURKA, AURKB, AURKC and PLK1 genes, bind various transcription factors (TFs). In this regard, the region comprising G allele of rs911160 in AURKA was linked with C/EBPalpha, C/EBPbeta and GR-beta proteins, whereas for C allele, additional binding sites for NF-Y, NFI-CTF and NF-1 were identified (Table 6). For rs2289590 in AURKB, an additional motif for YY1 binding was recognized when C allele was present. The region near C allele of rs758099 was associated with binding sites for NF-1, NF-Y, XBP-1, ENKTF-1, CTF, PEA3 and POU2F1, whereas in the presence of T allele NF-1, NF-Y, GATA-1 and TFII-I sequence-specific DNA-binding factors were recorded. Only in the case of rs11084490 in AURKC there were no changes in transcription factor binding site motif (XBP-1), if different alleles, either C or G, were present. The G allele of rs42873 in PLK1 was linked with an additional recognition motif for c-Jun transcription factor.