New insights into the association between AXIN2 148 C/T, 1365 C/T, and rs4791171 A/G variants and cancer risk

PubMed, Web of Science, Google Scholar, and China Wanfang Databases were systematically searched to identify all eligible published articles on AXIN2 variants and cancer susceptibility. The following terms were utilized for searching abstracts and titles: “Axin OR AXIN2”, “polymorphism OR SNP OR variant”, and “cancer OR adenocarcinoma OR carcinoma OR tumor”. The latest search was conducted on Jan 31, 2019 with no language restrictions. Furthermore, we also carefully screened and manually searched the review or original publications for more eligible studies.

All related information was independently screened by two investigators (L Shi and B Xu) from each enrolled study, including the name of first author, year of publication, country of origin, ethnicity, source of control, genotyping method, cancer type, total number of participants, P value for HWE, age range, genotyping data of AXIN2 148 C/T, 1365 C/T, and rs4791171 A/G variants in cases and controls. Disagreement should be resolved by discussion with a third author (W Zhang). If the controversial content still existed, it should be addressed by all investigators to reach a consensus.

The strength of the relationship between AXIN2 148 C/T, 1365 C/T, and rs4791171 A/G polymorphisms and cancer susceptibility was measured by calculating OR with 95% CI. A total of four genetic models were adopted in the current analysis, including allelic comparison model (M-allele vs. W-allele), homozygote contrast model (MM vs. WW), heterozygote model (MW vs. WW), and dominant model (MM + MW vs. WW). The χ²-test-based Q test was performed to investigate P value for heterogeneity among eligible researches. If P < 0.05, indicating that a significant heterogeneity was found, we employed the random-effects model (DerSimonian–Laird method) [34]. On the other hand, the fixed-effects model (Mantel–Haenszel method) was carried out [35]. We adopted qualitative funnel plot to assess possible publication bias by calculating the standard error of log(OR) for each research plotted against its log(OR). We further conducted quantitative Egger’s test to evaluate funnel plot asymmetry [36]. The web-based program was applied to check for deviations from the Hardy–Weinberg equilibrium (HWE) of distribution frequencies (http://​ihg2.​helmholtz-muenchen.​de/​cgibin/​hw/​hwa1.​pl) [37]. The P value more than 0.05 suggested an HWE balance. Moreover, we applied leave-one-out sensitivity analyses to calculate the stability of pooled results [38]. All of the above analyses were conducted by STATA software v11.0 (Stata Corporation, TX).

An online gene expression database was adopted to investigate the AXIN2 expression in lung and prostate adenocarcinoma tissues and the paracancerous tissues. (http://​gemini.​cancer-pku.​cn/​) [39]. RNA expression profiles of 446 pathologically diagnosed lung adenocarcinoma (including 387 Caucasians, 51 African-Americans, and 8 Asians) and 153 prostate adenocarcinoma tissues (containing 147 Caucasians and 6 African-Americans) were evaluated by this database. The Cancer Genome Atlas (TCGA) samples were also utilized to investigate the high and low expression of AXIN2 on cancer susceptibility and overall survival time. Moreover, the String online server was applied to assess the gene–gene correlation of AXIN2 (http://​string-db.​org/​).

As was shown in Table 1, 15 articles were finally retrieved in the present analysis, which contains 22 case–control studies for AXIN2 148 C/T, 1365 C/T, and rs4791171 A/G variants. There were 2909 cancer subjects and 2907 control volunteers for 148 C/T polymorphism, 587 cancer subjects and 605 controls for 1365 C/T variant, 785 cases and 443 controls for rs4791171 A/G variant. Furthermore, we checked the minor allele frequencies (MAF) of three AXIN2 variants by Trans-Omics for Precision Medicine (TOPMed) online (https://​www.​ncbi.​nlm.​nih.​gov/​snp/​) (Fig. 1). MAF of AXIN2 148 C/T were: in Africans, 0.119; Asians, 0.426; Europeans, 0.526; Americans, 0.561; others (including Pacific Islanders), 0.470; Global, 0.474. MAF of AXIN2 1365 C/T were: in Africans, 0.069; East Asians, 0.192; Europeans, 0.114; Americans, 0.100; others, 0.090; Global, 0.104. Finally, MAF of AXIN2 rs4791171 A/G were: in Africans, 0.267; East Asians, 0.370; Europeans, 0.681; Americans, 0.620; others, 0.670; Global, 0.547. In stratified analysis by ethnicity, seven studies were performed in Caucasian populations, twelve studies were in Asian descendants, and two were done in Arabians and one was in Latin descendants. Eight studies were conducted using population based controls and the rest 14 studies were utilizing hospital based controls. The classical genotyping method, PCR-restriction fragment length polymorphism (RFLP) was adopted in nine of these studies.

In the overall analysis, we identified a significant correlation between AXIN2 148 C/T variant and cancer risk (allele contrast: OR = 0.88, 95% CI 0.77–0.99, P_{heterogeneity} = 0.004, P = 0.041; heterozygote comparison: OR = 0.84, 95% CI 0.75–0.95, P_{heterogeneity} = 0.112, P = 0.004; dominant genetic model: OR = 0.82, 95% CI 0.69–0.96, P_{heterogeneity} = 0.022, P = 0.015) (Table 2). In subgroup analysis by race, we observed positive results in Asians (allele contrast: OR = 0.85, 95% CI 0.73–0.98, P_{heterogeneity} = 0.016, P = 0.027; dominant genetic model: OR = 0.80, 95% CI 0.66–0.96, P_{heterogeneity} = 0.030, P = 0.020) and Caucasians (dominant genetic model: OR = 0.76, 95% CI 0.59–0.98, P_{heterogeneity} = 0.701, P = 0.036), (Fig. 2). Moreover, subgroup analysis by cancer type suggested that 148 C/T variant was associated with a decreased cancer risk in lung adenocarcinoma (allele contrast: OR = 0.74, 95% CI 0.65–0.84, P value for heterogeneity = 0.602, P < 0.001; dominant genetic model: OR = 0.70, 95% CI 0.59–0.84, P_{heterogeneity} = 0.803, P < 0.001, Fig. 3). Similar finding was indicated in prostate adenocarcinoma (heterozygote comparison: OR = 0.54, 95% CI 0.35–0.84, P_{heterogeneity} = 0.088, P = 0.006; dominant genetic model: OR = 0.62, 95% CI 0.41–0.93, P_{heterogeneity} = 0.078, P = 0.022). In subgroup analysis by source of control, similar results were also observed in population-based studies. Furthermore, we identified notable correlation between AXIN2 1365 C/T variant and cancer risk (allele contrast: OR = 0.71, 95% CI 0.61–0.98, P_{heterogeneity} = 0.873, P = 0.038; heterozygote comparison: OR = 0.63, 95% CI 0.44–0.91, P_{heterogeneity} = 0.668, P = 0.014; dominant model: OR = 0.66, 95% CI 0.47–0.94, P_{heterogeneity} = 0.775, P = 0.021). For rs4791171 A/G polymorphism, no significant association was indicated (allele comparison, OR = 0.99, 95% CI 0.85–1.17, P_{heterogeneity} = 0.786, P = 0.864; homozygote contrast, OR = 0.94, 95% CI 0.66–1.33, P_{heterogeneity} = 0.873, P = 0.728; heterozygote contrast, OR = 0.86, 95% CI 0.62–1.17, P_{heterogeneity} = 0.522, P = 0.322; dominant model, OR = 0.89, 95% CI 0.66–1.19, P_{heterogeneity} = 0.575, P = 0.429).

Results from in silico tools suggested that AXIN2 expression in normal group was higher than that in lung adenocarcinoma tissue (Fig. 4a). However, no obvious difference was indicated for prostate adenocarcinoma (Fig. 4b). Moreover, we explored whether the AXIN2 expression had an effect on the overall survival time of lung adenocarcinoma patients. However, Kaplan–Meier estimate showed no vital difference of overall survival time between high and low AXIN2 expression groups (P = 0.40, Fig. 5).

Egger’s test and Begg’s funnel plot were utilized to evaluate publication bias in all of enrolled studies. We demonstrated no publication bias for AXIN2 148 C/T polymorphism (allelic contrast, t = − 0.52, P = 0.614; TT vs. CC, t = − 0.66, P = 0.519; heterozygote comparison, t = − 0.30, P = 0.771; TT + TC vs. CC, t = − 0.34, P = 0.741), AXIN2 1365 C/T variant (allelic comparison, t = 2.20, P = 0.159; TC vs. CC, t = 2.18, P = 0.161) and rs4791171 A/G polymorphism (G-allele versus A-allele, t = − 0.55, P = 0.680; homozygote contrast, t = − 0.62, P = 0.645; GA vs. AA, t = − 0.72, P = 0.602; dominant model, t = − 0.78, P = 0.577). As shown in Fig. 6, results from funnel plots appeared symmetrical in the overall analysis under dominant model, which indicated a lack of publication bias. Sensitivity analyses were also utilized to assess the pooled OR by omission of any one study. The results suggested that the current data from pooled ORs were relatively stable. No single study can substantially change the overall OR (Fig. 7).